SOUTHWESTERN ECONOMIC REVIEW
Anne Macy, Editor College of Business West Texas A&M University
Volume 42, Number 1 Spring 2015 ISSN 1941-7683
i SOUTHWESTERN ECONOMIC REVIEW
EDITOR: Anne Macy, West Texas A&M University
EDITORIAL BOARD: Salem AboZaid, Texas Tech University Abul Alamgir, University of Texas Adnan Daghestani, Barry University M. Kabir Hassan, University of New Orleans Nicholas Hill, Jackson State University Marshall J. Horton, Ouachita Baptist University Shari B. Lawrence, Nicholls State University Joshua Lewer, Bradley University Warren Matthews, LeTourneau University Chu Nguyen, University of Houston - Downtown Mihai Nica, University of Central Oklahoma Bruce Payne, Barry University Elizabeth Rankin, Centenary College Neil Terry, West Texas A&M University Susanne L. Toney, Savannah State University Ata Yesliyaprak, Alabama A&M University Geungu Yu, Jackson State University
ISSN: 1941-7683
The Southwestern Economic Review is published annually by the Southwestern Society of Economists and the College of Business, West Texas A&M University. The journal is indexed in Cabell's, EBSCO, and Econ Lit.
Please visit the website, http://swer.wtamu.edu for information on submission process and manuscript guidelines. Please direct questions to the editor at the email provided below: Anne Macy, Editor Southwestern Economic Review West Texas A&M University Box 60187 Canyon, TX 79016 Tel: 806-651-2523 E-Mail: [email protected]
• Copyright© 2015 by the Southwestern Society of Economists. All rights reserved. No part of this publication may be reproduced in retrieval systems, or transmitted in any form or by any means, electronically, photocopying, recording, or otherwise without prior written permission of the publisher.
• No responsibility for the views expressed by authors in the Southwestern Economic Review is assumed by the editor, the publisher, West Texas A&M University, or the Southwestern Society of Economists.
ii
SOUTHWESTERN ECONOMIC REVIEW
Volume 42, Number 1 Spring 2015 ISSN 1941-7683
TABLE OF CONTENTS
Transfer Pricing: How to Apply the Economics of Differential Pricing to Higher Education Marshall Horton, John Cox and James Files ...... 1 The Determinants of Box Office Revenue for Documentary Movies Neil Terry, Leigh Browning and Lisa Mitchell...... 15 A Canonical Correlation Approach to Investigate the Determinants of Investment Safety Roman Wong, Nichole Castater and Bruce Payne...... 29 How Progressive Is the U.S. Tax System? Syed Shahabuddin ...... 45 Determinants of the Soaring Growth of U.S. Non-Marital Births Frederic L. Pryor ...... 69 Behavior of the Vietnamese Equity Premium Chu V. Nguyen...... 83 The Economic Impact of Oklahoma Tobacco Settlement Spending on Research Fritz Laux, Brian Jackson and Justin Halpern ...... 99 How Non-Profit Inspection Services Can Correct for Credence Good Type Market Failures John McCollough...... 113 The Impact of Changes in the Dow Jones Industrial Average List on Prices and Trading Volumes Geungu Yu, Phillip Fuller and Patricia A. Freeman...... 125 US-Australia Trade Balance and Exchange Rate Dynamics Matiur Rahman and Muhammad Mustafa...... 135
Saving for Sustainability: Why a 10% Personal Saving Rate Is Too Low Laura L. Coogan, John Lajaunie and Shari Lawrence...... 147 An Empirical Re-Examination of the Fisher Hypothesis: Panel Cointegration Tests Swarna (Bashu) Dutt and Dipak Ghosh ...... 161 Digitization and Recorded Music Sales Whither the 'Long Tail'? Ian Strachan...... 175 Louisiana Motion Picture Incentive Program: How Well Is It Working? Anthony J. Greco ...... 197
iii INSTITUTION APPRECIATION
The Editor of the Southwestern Economic Review and the Southwestern Society of Economists would like to thank the following institutions for supporting our members' involvement in the organization and contribution of time and effort in reviewing articles for this issue of the Review.
Alabama A&M University Park University American University in Dubai Perryman Group Arkansas State University Savannah State University Barry University South Carolina State University Baylor University Southeastern Louisiana University Binghampton University Texas A&M University Boston College Texas A&M University - Commerce Bradley University Texas A&M University - San Antonio California State University - Los Angeles Texas Christian University Cameron University Texas Southern University Centenary College of Louisiana Texas Tech University Columbus State University University of Alabama Creighton University University of Arizona DeVry University University of Arkansas Dillard University University of Arkansas - Pine Bluff Emporia State University University of Buffalo Frostburg State College University of Central Oklahoma Georgia Regents University University of Denver Henderson State University University of Georgia Heritage Foundation University of Guadalajara Ibrid National University, Jordan University of Hartford Jackson State University University of Houston – Downtown King Fahd University University of Incarnate Word Lamar University University of Michigan LeTourneau University University of Nebraska Louisiana State University University of New Mexico Lynchburg University University of New Orleans McMaster University University of North Carolina- Pembroke McNeese State University University of Oklahoma Mercer University University of Pittsburgh Metropolitan State University of Denver University of Texas Morehead State University University of Texas at Brownsville National Chengchi University University of West Alabama New Mexico State University University of West Georgia Nicholls State University University of Wisconsin - Whitewater Norfolk State University Utah Valley University North Carolina A&T State University Valdosta State University Old Dominion University Washington State University Ohio State University West Texas A&M University Ouachita Baptist University West Virginia University
iv PREFACE
The publication of this issue of the Southwestern Economic Review was made possible by the support and efforts of many dedicated people. The faculty and staff of the College of Business at West Texas A&M University provided much needed support, which ensured the success of this endeavor. In particular, Emily Gross provided critical review, editing, and formatting of the articles. The reviewers of the manuscripts submitted to the Southwestern Economic Review provided much needed expertise in helping to decide which papers to accept for publication. The editor is grateful for all of this help and hopes this issue of the Southwestern Economic Review will promote and advance our organization, the Southwestern Society of Economists, to new educational and professional horizons.
The Southwestern Society of Economists is affiliated with the Federation of Business Disciplines. The annual membership fee of the Society is $49 and includes a subscription to the Southwestern Economic Review. There are no submission fees or publication fees associated with the Review, but authors are encouraged to join the Southwestern Society of Economists.
The website for the Southwestern Economic Review is http://swer.wtamu.edu. It contains past issues and manuscript guidelines.
Once again, many thanks to those who helped bring this publication to fruition.
Anne Macy, Editor Southwestern Economic Review
Prior host institutions of the Southwestern Economic Review include Baylor University, Ray Perryman, editor; Arkansas State University, John Kaminarides, editor; and Texas Christian University, Ed McNertney, editor.
v vi TRANSFER PRICING: HOW TO APPLY THE ECONOMICS OF DIFFERENTIAL PRICING TO HIGHER EDUCATION Marshall Horton, Ouachita Baptist University John Cox, Ouachita Baptist University James Files, Ouachita Baptist University
ABSTRACT
This paper outlines how transfer pricing is applied in modern American colleges and universities. Economic concepts from marketing, such as efficiency templates, are employed to illustrate the practical nature of budgeting and how it falls short. The implications of differential price elasticities and limited information reporting are explored, with recommendations as to how the process may be optimized. JEL Classification: D4
INTRODUCTION
American universities have traditionally resisted charging different prices for different academic subjects.1 Some have found this practice objectionable, since it leads to “implicit cross-subsidy across major fields that results from the conventional practice of charging similar prices.” (Strange (2013), p. 5) Differential tuition is increasing in American higher education (Ehrenberg (2012), p. 211). Although most differential tuition rates are charged based on whether students are lower-level or upper-level, more programs such as business, engineering, and nursing are charging premiums for their classes at flagship, research-oriented, state universities. Strange (2013) performed an event-study which indicated that raising tuition for engineering courses had a significant, negative, effect on engineering enrollment, but that the effect of raising tuition for business and nursing courses had negligible or even positive effects, respectively. The purpose of the current study is to analyze tuition decisions, i.e., pricing policy, in the context of transactions costs economics. Differentiating institutions according to operational goals and contrasting them should allow us to model pricing policy and make some simple predictions about the direction of the marketplace.
TRANSFER PRICING APPLIES TO REVENUES AS WELL AS TO COSTS
How would a university go about deciding on differential prices for academic programs? The practice of transfer pricing (Baldenius and Reichelstein (2006)) as it has
1 been applied to higher education (Johnson and Turner (2009), Horton and Faught (2008))2 would indicate some areas as cost centers, some as expense centers, some as revenue centers, some as profit centers, and others as investment centers ((Brickley, Smith and Zimmerman (1995)).3 It would also provide a profit-maximizing rationale for differential pricing. To illustrate how transfer pricing would be applied in differential pricing and what kinds of reporting requirements and information would be required is the purpose of this paper. To this end, we propose a simple model of templates incorporating elasticity of supply (cost) and demand (revenue) which are the core concepts of managerial optimality.
A Simple Generalization of How Institutions are Operated and Why They Need Transfer Pricing
Some of the anomalies reported by Johnson and Turner (2009) include • little evidence that the economics rather than politics of decision- making explained “disparities in student-faculty ratios across fields and disciplines.” (p. 169) • a positive relationship between the salaries of a department’s faculty members and the student-faculty ratio (p. 179) • a negative relationship between faculty salary expenditures per student and individual faculty pay (p. 179) • a long-run inelastic relationship between changes in enrollment by academic area and faculty hiring (p. 184) • a dearth of institutions that “use pricing mechanisms to influence the allocation of students across fields.” (p. 186)
To try to accommodate these and other anomalies, we somewhat arbitrarily divide higher- education institutions into two types:4
Type A is entirely centralized in decision-making, hiring, and strategic planning. Type B is decentralized and overlays a more-or-less autonomous group of individual units.
Type A institutions tend to be smaller and “teaching-oriented.” They are run from the top down with professional managers as administrators. The academic programs are strongly mission-based with a centralized sense of identity. Loyalty is prized over marketability. Type A institutions view themselves as monolithic and use commonly-accepted measures to compare themselves with similar institutions, so far as a pure-play may be found. The leadership in such institutions views itself, correctly or not, as a customer in a single labor market and a provider in a single product market. This is consistent with the model used by Strange (2013) in which individual departments are treated as price-takers who are merely quoting the price dictated by a higher authority (board of trustees or regents). We conjecture that virtually all institutions evolved from something like Type A.5 Type B institutions are large enough that coordination problems preclude micro- management of enrollments, hiring decisions, etc. In contrast to Type A schools, they are run like loosely-amalgamated independent agencies as in an M-form corporation. 6 As such, the decentralized divisions within the multi-divisional institution are allowed (encouraged) to self-select into profitable markets vis-à-vis Proctor & Gamble or General Motors. When competing head-to-head, a division of institution B possesses both advantages
2 and disadvantages relative to institution A. Some advantages are deep pockets from a large, parent, institution and name recognition and the ability to offer complementary programs. Some disadvantages are (1) lack of institution-wide commitment to recruiting, and (2) lack of market-niche status (i.e., “liberal arts” institution with individual attention). But the most prominent differences between the two are in (1) hiring specialized labor (faculty resources) and (2) recruiting students. The biggest disadvantage that Type A institution has is its treatment of all faculty resources as perfectly substitutable and all students as homogeneous. It practices neither price discrimination between its various academic programs nor differential salaries between faculty hires in various fields. The lack of departmental price discrimination is reflected in a relatively elastic demand, while the tendency to pay all faculty resources the same because of a strong mission commitment is reflected in a relatively inelastic supply. Administrators of Type A institutions may think that they are optimizing at the institutional level by employing a mark-up over cost. But they are failing in at least two regards. First, they tend to treat all programs alike as to revenue potential, discarding potentially valuable information and “leaving money on the table.” Second, they tend to make little effort to estimate marginal cost, even if the concept is known to them, much less equate it to marginal cost in order to find the optimal markup. Instead, they likely begin with a pool of scholarship funds with which they can give discounts, sparingly at first, liberally at last as their desperation to meet enrollment expectations, fill dorm rooms, etc, mounts.7 Rather than using discounts strategically from the first in order to practice first-degree price discrimination, they make path-independent assignments of financial aid based on enrollment projections that will be needed to meet the budget (“feeding the beast”). The budget is the anchor that dooms Type A institutions to tie all decisions to historical cost and thereby live a permanent hand-to-mouth existence, never getting control of their discount rates. Why does this matter? Consider an exogenously-imposed cost increase on the institution. This is reflected by an upward shift in the supply curve for each program on campus. In the Type A institution on the left, all departments bear the increase in cost equally. For simplicity, both supply and demand are characterized as linear, although a constant elasticity of either is unlikely. Who will bear the burden of cost increases? Similar to the analogy of tax burden, the Type B institution will be able to pass along more of its cost increase to students in the form of higher (less discounted) tuition, while the Type A institution will have to absorb more of its costs by forgoing raises, curtailing supplies and services, etc. The Type A institution must frequently subsidize operations with undesignated gifts and endowment income, while the Type B institution finds itself in the enviable position of actually adding to its endowment in good years from surplus operating funds. Which is more viable in the long-run? The question is purely rhetorical, since the Type A institution’s sustainability is a matter of fundraising and hope while the Type B institution’s sustainability is a matter of sound management. But practicing price discrimination requires (1) market power, (2) knowledge of elasticities, and (3) barriers to entry that effectively preclude low-cost arbitrage profits. Through accreditation, universities have effective barriers to entry, so the third requirement appears to have been met. What about the first two requirements? Type B institutions tend to deal with the problem by simply letting individual departments operate autonomously, mimicking independent contractors as closely as
3 they can. As in the M-form organization, departments compete with one another and are responsible for getting their estimates of elasticities right. Type B administrators are not necessarily smarter than Type A administrators, but they are more decentralized, which not only allows but also requires deans and department heads to price their products optimally. But Type A institutions have no such Coasian-type market discovery process at their disposal. Being monolithic, they cannot afford the time and expertise needed to estimate and implement the proper analysis of market elasticities. Given that the data needs for a type A institution are greater than it can obtain, the institution must attack its problem in another dimension: It must treat itself as the sole reporting entity in two competitive marketplaces: that for students as a supplier, and that for faculty members as a demander.
Transfer Pricing Templates
Efficiency templates incorporating activity-based costing, market power analysis, and revenue projection have found rich application in business.8 We present two generic transfer pricing templates that might face a typical Type A institution: a cost (supply side) template (Table 1) and a revenue (demand side) template (Table 2). The use of such templates transcends the nature of an organization’s business and lends itself to the analysis of the production and marketing functions of educational organizations i.e., colleges, universities, for profit, nonprofit, private, public, etc. It should be understood at the outset that the templates we present are presented as examples only and will likely vary significantly between specific institutions. To explain some of the details of each table, we turn first to Table 1. The entries in the table are organized along two dimensions: each section of an accounting degree plan is grouped row-wise; each academic department’s course offerings are weighted column- wise according to their composition of each row group. For example, for a typical (Type A) institution, a forty-six semester credit hour general education core requirement of an accounting major would cost $6,769.58 for instruction, based on national salaries in the fields of those faculty members who teach in the core. That cost comprises a little over 27 percent of the instructional cost of an accounting major in such an institution. Of that $6,769.58, the History department costs 21.6 percent while the Accounting department costs nothing, since it has no courses in the general education core. However, for the major emphasis (accounting), the History department costs nothing and the Accounting department accounts for 100 percent of the instructional cost. The template notes that non- instructional areas and their expenses can be appended to the table using additional rows. Table 2 analyzes the marketing flows necessary to bring the product to market and provide revenues: promotion, financing, negotiation, ordering, risking, and payment. These are organized by rows while the departments and office involved are organized by columns: the Accounting Department, Development Office, Admissions Office, Financial Services, Administration, and Student Services (see a standard text on marketing channels such as Coughlan, et. al. (2001)). The benefit potential of the channel flow column demonstrates value-added by each channel flow as estimated in the previous column (proportionate channel cost of flow). Further interpretation of this table is provided a few paragraphs below as part of a scenario analysis. Both templates provide a standardized way of looking at the activities of an organization. The cost template is a mechanism for relating organizational costs to the production function for a particular organization and a particular functional area. The
4 revenue template is a mechanism for relating organizational revenues and costs to the marketing function for a particular functional area. Each template can be used for scenario analysis in which an analyst posits the effect on costs and/or revenues of changing production or marketing activities and observes the impact on costs or revenues, respectively. In economic terms, a cost template requires that judgments be made concerning relative price elasticities of supply, while a revenue template requires judgments concerning relative price elasticities of demand. Taken together, judgments can be made concerning a theoretical “profit maximizing” blend for a portfolio of “products” based upon the coefficients that comprise the template matrix. These coefficients are based on the activity-based costing technique and the analyst’s reasonable estimates of relative supply and demand elasticities under a set of constraining resources. For a host of reasons, including regulatory oversight, “groupthink,” and internal political battles, most type A institutions have not structured their accounting practices and management information systems to develop data useful for estimating reasonable parameters for either cost or revenue templates. There is no central reporting authority or group (not even the National Center for Education Statistics (NCES)) that asks member institutions to report data that would be helpful in formal estimation of elasticities. This means that the templates must be interactive and flexible enough that they will allow institutions to engage in scenario analysis where statistical data may be absent. The intuitive conclusion is that pricing all products the same regardless of varying supply costs and marketing all products the same regardless of market demand does not represent an efficient profit maximizing blend of an organization’s product portfolio. In most cases, this conclusion should be confirmed by the transfer pricing analysis.9 To the extent that the revenue parameters used in the demand template do not reflect price elasticities of demand for individual programs, the decisions based on such a template will be no more optimal than the decisions made without such a template. The tendency of Type A institutions, we suspect, would be to simply solve for a particular point on a demand curve consistent with the cost structure that the institution’s existing budget would impose on the supply template. This reinforces the truth that transfer pricing is no magic bullet: garbage in-garbage out. In such a situation of budget worship that belies administrative convenience alluded to in the previous paragraph, the best use of templates would be to perform a scenario analysis in which the institution could find the advantages and/or disadvantages of decentralizing its current resources. As a very simple example, in Table 2, if the given institution were to require the accounting department to take more ownership of its own program, the coefficient under the Accounting column and corresponding to the Promotion row might change from 0.2 to 0.7, which the coefficient under the Development column and corresponding to the Promotion row might change from 0.7 to 0.2. This change would affect the Normative Channel Revenue Share row for the Accounting department from 0.158 to 0.533 and for the Development office from 0.544 to 0.169, resulting in compensation transferred from administration to accounting faculty. Since accounting faculty members are presumably better able to communicate the advantages to potential students of studying accounting than are members of the overall institution’s Development staff, overall revenue should rise.10
5 CONCLUSION
In conclusion, several points are instructive. Some of these points lend themselves to testable hypotheses with data are generally available. Others, with data that are available at this time, do not. As James (1990) and others have pointed out, universities are still very guarded against releasing data (this is addressed again in point six, below). First, the type A (teaching) institutions that do not practice differential tuition policies between departments are not optimizing resources. In a non-competitive marketplace they may be able to do this for many years, but in an industry that is becoming increasingly competitive, they cannot do it for long. Second, there are substantial cost differentials between degrees in the same institution. This hold for both type A (teaching) and type B (research) schools. Third, cost differentials are identifiable through supply elasticities. Institutions have, if they choose to use it, sufficient information to find average cost parameters, although marginal cost parameters are more difficult to obtain because they require understanding of opportunity cost rather than just relying on the inertia of historical average cost information from last year’s budget. Fourth, in the absence of differential tuition (particularly in the tuition-driven, type A, institutions), marginal revenue is treated as constant for each major as a practical matter. The unfortunate result is that the institution loses the ability to compete in different product lines. Fifth, in theory at least, price elasticities of demand may be used to construct optimizing efficiency templates for revenue centers. In practice, rougher estimates for internal allocation of resources will do. Sixth, because of a lack of accountability to a market-driven source of funding, institutions tend not to keep, much less report, necessary information for estimating meaningful demand elasticities. Seventh, without the ability or willingness to use price elasticity of demand information by department and engage in contribution margin maximization, type A institutions cannot determine, much less charge, optimal markups. This is before non-instructional activities are even considered. The lack of requiring reporting data that is market-meaningful to markets helps decentralized (type B) institutions, since they let (require) individual departments on large campuses “duke it out” with their own budgets and funding sources. Such departments are often large enough to possess sufficient slack to take over administering marketing efforts, product development, etc., while Type A institutions are not. The lack of managerial slack in Type A institutions precludes their optimal behavior. The best hope for teaching institutions is to decentralize and try to become research institutions.
6 REFERENCES
Ansary, Asim, S. Siddarth, and Charles B. Weinberg (1996) “Pricing a Bundle of Products or Services: The Case of Nonprofits,”Journal of Marketing Research 33:1 (February), 86-93. Baldenius, Tim, and Stefan Reichelstein (2006) “External and Internal Pricing in Multidivisional Firms,” Journal of Accounting Research 44:1 (March), 1-28. Becker, William E. (1990) “The Demand for Higher Education,” Chapter 7 in The Economics of American Universities: Management, Operations, and Fiscal Environment, Stephen A. Hoenack, Editor, pp. 155-187. Albany: State University of New York Press. Bergmann, Barbara R. (1991) “Bloated Administration, Blighted Campuses,” Academe (November-December), 12- 16. Bhaskar, V., Alan Manning, and Ted To (2002) “Oligopsony and Monopsonistic Competition in Labor Markets,” Journal of Economic Perspectives 16:2 (Spring), 155-174. Brickley, James A., Clifford W. Smith, and Jerold L. Zimmerman (1995) “Transfer Pricing and the Control of Internal Corporate Transactions,” Journal of Applied Corporate Finance (Summer), 60-67. Brinkman, Paul T. (1990) “Higher Education Cost Functions,” Chapter 5 in The Economics of American Universities: Management, Operations, and Fiscal Environment, Stephen A. Hoenack, Editor, pp. 107-128. Albany: State University of New York Press. Burnett, Sharon, and Darlene Pulliam (2014) “Transfer Pricing Seven Years after Glaxo Smith Kline,” Southwestern Economic Review 41:1, 99-108. Coase, Ronald (1937) “The Nature of the Firm,” Economica 4:16, 386-405. Coughlan, Anne T., Erin Anderson, Louis W. Stern, and Adel I. El-Ansary (2001) Marketing Channels: Sixth Edition. Upper Saddle River, NJ: Prentice-Hall. Ehrenberg, Ronald G. (2012) “American Higher Education in Transition,” Journal of Economic Perspectives 26:1 (Winter), 193-216. Giardini, Valerie (1983) Internal Transfer Pricing of Bank Funds. Rolling Meadows, Illinois: Bank Administration Institute. Ginsberg, Benjamin (2011) The Fall of the Faculty: The Rise of the All-Administrative University and Why it Matters. New York: Oxford University Press. Hoenack, Stephen A. (1990) “An Economist’s Perspective on Costs within Higher Education Institutions,” Chapter 6 in The Economics of American Universities: Management, Operations, and Fiscal Environment, Stephen A. Hoenack, Editor, pp. 129-154. Albany: State University of New York Press. Holstrom, Bengt, and Jean Tirole (1991) “Transfer Pricing and Organizational Form,” Journal of Law, Economics, and Organizations 7:2 (Autumn), 201-228. Horton, Marshall J., and Kent S. Faught (2008) “Transfer Pricing and the American University,” Southwestern Economic Review 35:1 (Spring), 227-242. James, Estelle (1990) “Decision Processes and Priorities in Higher Education,” Chapter 4 in The Economics of American Universities: Management, Operations, and Fiscal Environment, Stephen A. Hoenack, Editor, pp. 77-106. Albany: State University of New York Press. Johnson, William R., and Sarah Turner (2009) “Faculty without Students: Resource Allocation in Higher Education,” Journal of Economic Perspectives 23:2 (Spring), 169-189.
7 Kaplan, Robert S. and Steven R. Anderson (2004) “Time-Driven Activity-Based Costing,” Harvard Business Review 82:11, 131-138. Kaplan, Robert S., and Robin Cooper (1998) Cost and Effect: Using Integrated Cost Systems to Drive Profitability and Performance. Cambridge, Massachusetts: Harvard Business School Publishing. Pope, Justin (2013) “From the Recession’s Wake, Education Innovation Blooms,” Hot Springs Sentinel Record (Sunday) August 4, p. 6B Pugliese, Alfred (1970) “Tuition, Fees, Deposits, and Other Charges,” Chapter 2 in Handbook of College and University Administration, General, Asa S. Knowles, Editor-in-Chief, pp. 8-13 to 8-26. New York: McGraw-Hill Book Company. Strange, Kevin (2013) “Differential Pricing in Undergraduate Education: Effects on Degree Production by Field”, NBER Working Paper No. 19183 (June) available at http://nber. org/papers/w19183 Trussel, John M., and Larry N. Bitner (1996) “As Easy as ABC: Re-engineering the Cost Accounting System,” NACUBO Business Officer (June), 34-39. Williamson, Oliver E. (1975) Markets and Hierarchies: Analysis and Antitrust Implications. New York: The Free Press.
8 9 TABLE 1. SUPPLY SIDE ANALYSIS USING ACTIVITY BASED COST TEMPLATE
Transfer Pricing Cost Template Direct Activity-Based Instructional Costs by division (functional area) SUPPLY SIDE Major: Accounting Proportional Flow Performance by Area
Weighted % of Academic Average Academic Phil/ Psych/ Fine Phys Life Areas Hrs Cost Cost Educ Eng Lang Rel Soc His t Arts Bus Ec on Acct Math Sci Sci Total Core 46 $6,769.58 0.2719 0.079 0.185 0.000 0.126 0.000 0.216 0.108 0.031 0.000 0.000 0.066 0.093 0.098 1 Major core 39 $9,422.08 0.3785 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.734 0.125 0.141 0.000 0.000 0.000 1 Major emphasis 27 $5,658.75 0.2273 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 1.000 0.000 0.000 0.000 1 Basic School 12 $2,246.25 0.0902 0.000 0.000 0.000 0.000 0.202 0.000 0.000 0.798 0.000 0.000 0.000 0.000 0.000 1 Language 6 $799.17 0.0321 0.000 0.000 1.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 1 0 0.1 0.1 0.1 0.5 0.2 1 2 3.9 7.7 15.3 30.1 60
TOTAL 120 $24,895.83 1.0000
Departmental Cost 0.021 0.05 0.032 0.034 0.018 0.059 0.0293 0.3582 0.0473 0.2807 0.0179 0.0252 0.0265
Non-instructional Costs may be added as needed below in format illustrated above
TABLE 2. DEMAND SIDE ANALYSIS USING REVENUES AND MARKETING COSTS
Transfer Pricing Revenue Template Revenue Generated and Marketing Costs by division (functional area) DEMAND SIDE Major: Accounting Proportional Flow Performance by Area
Weights contributed by each functional area Benefit to each component of marketing cost: Potential of Proportion of each channel flow performed by each functional area Proportionate Channel ACADEMIC------Non-academic departments------Marketing Channel Flow FINAL DEP T: Financial Fl ows Cost of Flow Value WEIGHT Accounting Development Admissions Services Admin Student TOTAL Promotion 0.6 HIGH 0.75 0.2 0.7 0.05 0 0 0.05 1 Financing 0.15 MEDIUM 0.15 0 0 0 0.7 0 0.3 1 Negotiation 0.1 HIGH 0.03 0.2 0.5 0.05 0.1 0 0.15 1 Ordering 0.05 LOW 0.02 0.1 0.1 0.15 0.1 0.5 0.05 1 Risking 0.05 LOW 0.02 0 0 0 0 0.25 0.75 1 Payment 0.05 LOW 0.02 0 0.1 0.1 0.1 0.5 0.2 1
TOTAL 1 0.99 0.5 1.4 0.35 1 1.25 1.5
Normative Channel Revenue Share 0.158 0.544 0.044 0.112 0.025 0.107 0.99
10 11 12 13 14 THE DETERMINANTS OF BOX OFFICE REVENUE FOR DOCUMENTARY MOVIES Neil Terry, West Texas A&M University Leigh Browning, West Texas A&M University Lisa Mitchell, West Texas A&M University
ABSTRACT
This paper examines the domestic box office revenue determinants of movies from the documentary genre. The sample consists of the top 100 gross box office revenue documentary films released during 1978-2014. Regression results indicate the primary determinants of box office revenue for documentaries are the ability of a movie to achieve wide release, revenue generation from opening weekend, receiving a restricted rating, and award nominations. One of the more interesting results include the observation that within the documentary movie genre critical ratings do not significantly impact box office revenue based on the observation that documentaries receive relatively high critical ratings for most films, which negates the uniqueness of positive critical reviews. In contrast, acclaim via award nominations is a positive and significant determinant of box office revenue. The results provide evidence that critical acclaim via positive reviews does not help distinguish a documentary but award nominations can play a major role in helping a film achieve commercial success. JEL Classification: L8
INTRODUCTION
The documentary film is unlike any other genre. In fact, the documentary film is the blank canvas that allows filmmakers to shape and polish story arcs as the plot unfolds. This process leads to films that take the viewer inside the minds and lives of those who have experienced what most of us have not. However, this naïve view of the power of the documentary film has not always been the norm. A surprising large number of people, including documentary filmmakers, will strive to differentiate the nonfiction films they enjoy from something they have stereotyped as documentaries. Documentaries, from the reputation they seem to hold, are the films some of us had to watch in the fifth grade. They tended to be filled with facts and can be painful to watch (Bernard, 2011). Fast-forward to a more contemporary view of this genre, the documentary film is no longer considered second class. In the modern world of the movie industry the documentary genre is uniquely positioned to play multiple roles, which include serving as forms of entertainment, art, an information source, and outlet for investment or grants. A single movie can be the difference between millions of dollars of profits or losses for a studio and filmmaker in a given year (Simonoff & Sparrow, 2000). The purpose of
15 this research is to analyze the motion picture industry with a focus on the determinants of domestic box office revenues of documentaries. This manuscript is divided into five sections. First, a survey of the related literature on the movie industry is discussed. The second section offers background information specific to the documentary genre. The third provides the model specification. The next section puts forth an empirical evaluation of the determinants of domestic box office revenues for the top grossing documentary movies released during the years 1978-2014. The final section offers concluding remarks.
SURVEY OF THE LITERATURE
Although no research has focused exclusively on the documentary movie genre, many researchers have developed models that explore the potential determinants of motion picture box office performance and related issues. Litman (1983) was the first to develop a multiple regression model in an attempt to predict the financial success of films. The original independent variables in the landmark work include movie genre (science fiction, drama, action-adventure, comedy, and musical), Motion Picture Association of America rating (G, PG, R and X), presence of a superstar in the cast, production costs, release company (major or independent), Academy Awards (nominations and winning in a major category), and release date (Christmas, Memorial Day, summer). Litman’s model provides evidence that the independent variables of production costs, critics’ ratings, science fiction genre, major distributor, Christmas release, Academy Award nomination, and winning an Academy Award are all significant determinants of the success of a theatrical movie. Litman and Kohl (1989), Litman and Ahn (1998), and Terry, Butler, and De’Armond (2004) have replicated and expanded the initial work of Litman. None of the extensions of Litman’s work has explicitly focused on the determinants of domestic box office revenue of the documentary movie genre. One strong area of interest in the movies literature has been the role of the critic (Weiman, 1991). The majority of studies find that critics play a significant role on the success or failure of a film. Eliashberg and Shugan (1997) divide the critic into two roles, the influencer and the predictor. The influencer is a role where the critic will influence the box office results of a movie based on his or her review of the movie. Eliashberg and Shugan’s results suggest that critics have the ability to manipulate box office revenues based on their review of a movie. The predictor is a role where the critic, based on the review, predicts the success of a movie but the review will not necessarily have an impact on how well the movie performs at the box office. Eliashberg and Shugan show that the predictor role is possible, but does not have the same level of statistical evidence as the influencer role. King (2007) explores the theoretical power and weakness of critics on the box office performance of movies. The substantial market power of critics is derived from the following: (1) Film reviews are widely available in newspapers, magazines, and websites. The ubiquitous availability of critical reviews in advance of a movie release creates positive or negative energy in the critical opening weeks; (2) Film critics regard themselves as advisors to their readers. They are often as explicit in their recommendations as Consumer Reports is about other consumer purchases; and (3) Film critics are likely to be considered objective. There are too many critics and too many films for serious critical bias to develop. Those who are skeptical about the influence of film critics point to the following counter arguments: (1) It is possible that the effects of aggressive marketing at the time of a film’s
16 release might dominate critical evaluations in determining opening attendance; (2) Critics may raise issues that do not concern most audiences. They are more likely to notice and comment on technical issues, like cinematographic technique, than the average member of the audience; and (3) Critics may write for a readership that has different tastes from the average cinemagoer. The most obvious potential reason for this is demographic. Cinema audiences are younger than the general population and less likely to pay attention to print reviews. Critics might therefore, be expected to aim their reviews at the older demographic audience and give relatively negative reviews to certain film genres. The empirical results put forth by King (2007) are mixed with respect to the impact of critics on box office earnings for the U.S. box office in 2003. He finds zero correlation between critical ratings for films and gross box office earnings when all releases are considered because of the affinity critics have for foreign movies and documentaries relative to the general public. For movies released on more than 1,000 screens, critical ratings have a positive impact on gross earnings. Reinstein and Snyder (2000) focus on the critics Siskel and Ebert and how their reviews impact box office success. The authors report that the correlation between good movie reviews and high demand might be false due to unknown quality measurements. In order to circumvent the proposed false correlation Reinstein and Snyder apply a “differences in differences” approach that yields a conclusion that positive reviews have a surprisingly large and positive impact on box office revenue. Reinstein and Snyder also report that their results show that the power to influence consumer demand does not necessarily lie in the entire critic population, but may lie in the hands of a few critics. Wallace, Seigerman, and Holbrook (1993) employ a sample of 1,687 movies released from 1956 through 1988 to investigate the relationships between movies box office success and critic ratings. They find a poorly rated movie will actually lose money for every positive review it receives while a highly rated movie will continue to gain money for every positive review it receives. They conclude that a bad movie has something to gain by being as trashy as possible, while it pays for a good movie to strive for excellence. Ravid (1999) has also looked at movie reviews as a source of projecting higher revenues. He concludes that the more reviews a film receives, positive or negative, the higher revenues it will obtain. Although much research has supported the critic as a positive indicator of box office success, others have shown that the critic plays a much less important role. Levene (1992) surveyed students at the University of Pennsylvania and concludes from her 208 useable surveys that positive critic reviews ranked tenth, behind plot, subject, and word-of-mouth on a list of factors that influence the decision to watch a film. Levene’s study reveals that theatre trailers and television advertising were the two most important determinants. Faber and O’Guinn (1984) conclude that film advertising, word-of-mouth and critics’ reviews are not important compared to the effect that movie previews and movie excerpts have on the movie going public. Wyatt and Badger (1984) find that negative or positive reviews have little effect on the interest of an individual to see a movie over a mixed review or seeing no review. Further research by Wyatt and Badger (1987) conclude that positive reviews and reviews that contain no evaluative adjectives, which they called non-reviews, are deemed more interesting than a review that was negative or mixed. More recently, Wyatt and Badger (1990) report that reviews containing high information content about a movie raise more interest in a film than a positive review. Research has shown a seasonal pattern in movie releases and box office performance. Litman (1983) reports that the most important time for a movie release is during the
17 Christmas season. Sochay (1994) counters this with evidence that the summer months are the optimal time of year to release a motion picture. Sochay, referencing Litman (1983), explains his conflicting results are due to competition during the peak times. Sochay adds that the successful season will shift from the summer to Christmas in different years due to film distributors avoiding strong competition. Radas and Shugan (1998) developed a model that captures the seasonality of the motion picture industry and apply it to the release of thirty-one movies. The authors find that the length of a movie release on average is not longer during the peak season but peak season movies typically perform better at the box office. Einav (2001) investigates seasonality in underlying demand for movies and seasonal variation in the quality of movies. He finds that peak periods are in the summer months and the Christmas season because distributors think that is when the public wants to see movies and when the best movies are released. He comments that distributors could make more money by releasing “higher quality” movies during non-peak times because the movie quality will build the audience and there will be less competition than at peak times. Film ratings passed down from the Motion Picture Association of America (MPAA) may also influence box office performance. Many film companies fight for a better rating, often re-shooting or re-editing scenes multiple times in order to get their preferred ratings, most often being PG or PG-13 because these ratings exclude virtually no one from seeing the movie. Sawhney and Eliashberg (1996) develop a model where the customer’s decision- making process on whether to see a movie can be broken into a two-step approach, time-to- decide and time-to-act. The results of their study show that movies with an MPAA rating of restricted (rated R) perform worse at the box office than movies without a restricted rating. The analysis shows that restricted rated movies have a higher time-to-act but have longer time-to-decide periods than family movies. Terry, Butler, and De’Armond (2004) verify the negative impact of the restricted rating on box office performance, providing evidence of a penalty in excess of $10 million. Ravid (1999) provides evidence from a linear regression model that G and PG rated films have a positive impact on the financial success of a film. Litman (1983) on the other hand, finds that film ratings are not a significant predictor of financial success. Austin (1984) looks at film ratings in an attempt to find a correlation between ratings and movie attendance but find no significant relationship. Anast (1967) was the first to look at how film genre relates to movie attendance. His results show that action-adventure films produce a negative correlation with film attendance while films containing violence and eroticism had a positive correlation. Litman (1983) shows that the only significant movie genre is science fiction. Sawnhey and Eliashberg (1996) use their two-step approach and find that the drama genre has a slower time-to-act parameter while action movies result in a faster time-to-decide than other movie genres. Neelamegham and Chinatagunta (1999) employ a Bayesian model to predict movie attendance domestically and internationally. They find that across countries the thriller and action themes are the most popular, while romance genre was the least popular. Terry, King, and Patterson (2011) examine the determinants of horror movie box office revenue for the years 2006-2008. The most interesting result of the study is the observation that slasher movies are the most profitable theme and zombie movies are the least profitable theme in the horror movie genre. Fictional characters have always driven the horror movie genre. The early years of the horror movies focused on characters like Dracula, Wolfman, and Frankenstein. The modern star fictional characters with box office draw are slasher killers like Freddy, Jason, and Jigsaw. Number of theatres featuring a movie during opening weekend is revealed to have a significant impact on box office revenue but holiday release is not a significant determinant. One of the most influential determinants
18 of domestic box office performance of horror movies is critical acclaim. Horror movies are one of the most harshly reviewed movie genres. The fact that the majority of horror movies receive poor critical reviews creates a box office opportunity for the relatively rare horror movies that receives critical acclaim. Movie sequels are shown to have a positive and statistically significant impact on domestic box office performance of horror movies. The built in audience associated with a sequel is worth approximately $7 million in domestic box office revenue. Horror movies earning a restricted rating pay a significant financial box office penalty of $15 million to $29 million. Production budget is also identified as a positive and significant determinant of domestic box office performance of horror movies. Awards are important to every industry but few industries experience financial compensation from an award more than the motion picture industry (Lee, 2009). Litman (1983) shows that an Academy Award nomination in the categories of best actor, best actress, and best picture is worth $7.34 million, while winning a major category Academy Award is worth over $16 million to a motion picture. Smith and Smith (1986) point out that the power of the Academy Award explanatory variable in models explaining patterns in movie rentals will change over time as the effects of different Academy Awards could cause both positive and negative financial results to a movie in different time periods. Nelson, Donihue, Waldman, and Wheaton (2001) estimate that an Academy Award nomination in a major category could add as much as $4.8 million to box office revenue, while a victory can add up to $12 million. The authors find strong evidence toward the industry practice of delaying film releases until late in the year as it improves the chances of receiving nominations and monetary rewards. Dodds and Holbrook (1988) look at the impact of an Academy Award after the nominations have been announced and after the award ceremony. The authors find that a nomination for best actor is worth about $6.5 million, best actress is worth $7 million and best picture is worth $7.9 million. After the award ceremony the best actor award is worth $8.3 million, best picture is worth $27 million, and best actress award is not statistically significant. Simonoff and Sparrow (2000) find that for a movie opening on less than ten screens, an Academy Award nomination will increase the movies expected gross close to 250 percent more than it would have grossed if it had not received the nomination. For movies opening on more than ten screens, an Academy Award nomination will increase the movies gross by nearly 30 percent. Opening weekend box office performance of a movie is often a critical determinant of the overall financial success of a film. In the case of motion pictures, decay effects means the diminishing attractiveness of movies as time goes on. The sales pattern of widely released movies shows an exponential distribution where the early period sales are the highest and drop throughout the life cycle (Liu, 2006). Therefore, the box office revenue at some later week is usually less than that of previous weeks. According to Einav (2007), in his reduced form of individual utility, there could be two possible effects that are captured by the decay effect: First, potential markets shrink over time because most people go to a movie only once. Second, consumers prefer watching a movie earlier. The result of the study shows the estimated decay of revenues of almost 40 percent per week. Literature investigating movie revenue streams beyond the box office are limited. Chiou (2008) explores the timing of a theatrical release as it relates to the home video industry and finds the highest demand season for the video market is between Thanksgiving and Christmas. Terry and De’Armond (2008) employ regression analysis to investigate the determinants of movie video rental revenue. They find domestic box office, Academy Award nominations, and domestic release exposure to be positive and significant determinants of movie video rental revenue. Time to video, sequels, and children’s movies are shown to
19 have a negative and statistically significant impact on video rental revenue.
THE DOCUMENTARY AS A COMMERCIAL PRODUCT
Bernard (2011) notes that documentaries bring viewers into new worlds and experiences through the presentation of factual information about real people, places, and events generally - but not always – portrayed through the use of actual images and artifacts. It is those gritty stories we tend to love the most. Stories we tell are a part of the fabric of our culture. Whether documentary or narrative, they are the foundation of all things in the mass media world (Campbell, Martin, Fabos, 2011). Documentaries are based on facts, as seen by the filmmaker, which bring us in touch with a world we would not otherwise have the ability to see. Although a separate genre from the narrative film, the same process applies to the documentary. A filmmaker who sets out to get a documentary film produced must deal with issues at all five stages of the process. Development, pre-production, production, post-production, and distribution are all complicated stages of filmmaking, regardless of the type. Where documentary filmmaking takes a serious turn from other genres, within the narrative umbrella, is the business model surrounding profit. Scholars and industry leaders differ in their opinion as to the profit viability of the documentary film today. Chris Anderson (2006) captures the paradigm shift in business today in his book, The Long Tail. Companies are surviving at the end of a long tail, finding very niche audiences. Many believe this is the place for the documentary genre. For goods like music, video, and information, which can be digitized, distribution costs approach zero, so the tail can be extremely long. Anita Elberse (2008) advises documentary filmmakers to keep costs low and take advantage of digital distribution formats in order to enjoy larger profit margins that exist at the end of the tail. Alex C. (2012) explains that the increase in documentary filmmaking can be attributed to many factors. Perhaps we are tired of reality television, sick of Hollywood, and actually care about our fellow human beings. The rise in the number of documentaries being produced started in 2004, particularly those focused on social issues. For committed filmmakers, new technology has made it easier and cheaper to create beautiful stories. For social activists, documentaries add a compelling narrative element to their struggle for social change. Documentary film is undergoing major shifts due to technology and distribution portals. Alex C. (2012) unpacks the reason so many documentaries are produced with such a low possibility of a big profit margin. The research indicates that most documentaries operate outside of the market forces of major Hollywood blockbusters, which allows them to seek a different audience. Because of this, they often garner an alternative revenue stream from grants, crowd funding, non-governmental organizations (NGO), and private donations. Documentary film projects received over $70 million in grants for documentary film projects from 2002-2011. Distribution portals have enhanced the indie filmmaker’s ability to get a documentary noticed. Some of the more substantial platforms are Hulu, Indieflix, the iTunes Store, YouTube, Vimeo, Distrify, Reelhouse, Pivotshare, VHX, and Yekra (Valentini 2014). Making money from—let alone finishing—a documentary remains a great challenge, but their evident proliferation over the past ten years should encourage cautious optimism for the future of the genre. The long tail may not hold large profits, but it contains the seeds for influence and impact. The most comprehensive lens put to the film industry is the book, Sleepless in
20 Hollywood: Tales from the New Abnormal in the Movie Business, by Lynda Obst (2013). To understand the fate of the documentary today, an examination of the new abnormal is essential. Narrative filmmaking today is about the international box office appeal and the mega deal. The documentary film has been immune to this change, so far. As noted, this is due to technology, distribution, and the rise in audience in the past few years. Obst (2013) breaks down a landscape for the narrative film business that is a bit bleak for the industry enthusiasts. Original narrative film scripts produced and released in 1981 was a meager seven. In 2011 there were none. Narrative film today is all about the deal and not the story. Documentary fills a gap today, that earlier did not exist. Technology, digital distribution portals, the desire for the truth, and the fundamental concern for humanity, are all viable reasons to embark on the process of making a documentary film today. Big profit is purposely left off that list. But, that is not to say that one cannot make a profit in this genre. Today, the path to profit is an ever-changing road.
DATA AND MODEL
Predicting the financial performance of films is widely regarded as a difficult endeavor. Each movie has a dual nature, in that it is both an artistic statement and a commercial product (Sochay, 1994). Many studies have attempted to estimate the determinants of box office performance by employing empirical models to high profile features. The approach of this study provides a unique focus on the determinants of box office revenue for documentary movies. The sample includes the top 100 grossing motion pictures released during 1978-2014 that are classified by boxofficemojo.com as documentary movies. The movies selected for the research cohort are the 100 documentaries with domestic box office gross revenue over $1.5 million. The primary source of data for this study is the Rotten Tomatoes website (rottentomatoes.com). The website utilizes a unique rating system that summarizes positive or negative reviews of accredited film critics into an easy to use total percentage that is aggregated for each motion picture. In addition to providing a system of aggregate reviews, the website also contains information pertaining to revenue, release date, movie rating, and genre. Movies.com, Oscars.org, WorldwideBoxoffice.com, boxofficemojo. com, imdb.com, and the-numbers.com are additional sources of data and information. The empirical model employed to investigate the determinants of documentary box office performance for this study is specified as:
(1) BOXOFFICEi = B0 + B1OPENINGi + B2THEATRESi + B3CRITICi + B4AWARDSi +
B5RESTRICTEDi + B6BUDGETi + ui,
Where BOXOFFICE is domestic box office earnings adjusted for inflation and presented in real 2014 dollars, OPENING is domestic box office revenue generated from the opening weekend of a documentary release, THEATRES is number of theaters featuring the movie during the weekend of widest release, CRITIC is the percent of composite approval rating for a movie by a group of leading film critics, AWARDS is a categorical variable for movies that receive five or more nominations for major awards (e.g., Academy Award, Los Angeles Film Critics, New York Film Critics), RESTRICTED is a categorical variable for movies with a restricted rating (Rated R), and BUDGET controls for the estimated production for each movie after being adjusted for inflation by converting all nominal values to 2014 real
21 values. Several alternative model specifications were considered but the limited box office revenue for most documentaries restricts the ability to put forth a more robust model. Descriptive statistics for the model variables are presented in Table 1. The average box office revenue in the sample is $10.7 million, with a maximum of $149.6 million (Fahrenheit 9/11). The five films in the research cohort grossing more than $34 million in 2014 real dollars are Fahrenheit 9/11 ($149.6 million), March of the Penguins ($94 million), Justin Bieber: Never Say Never ($77 million), Earth ($35.3 million), and 2016: Obama’s America ($34.5 million). The explicit marketing from production studios and implicit marketing from movie reviews tend to make the opening weekend the most profitable single weekend for most films. The average opening weekend for the research cohort is $1.4 million, which includes 19 movies with an opening weekend that are above the mean and 81 that are below the mean. The relatively modest opening weekend numbers implies that documentaries do not tend to rely on opening weekend as much as most other movie genres. The 19 movies that have an opening weekend box office above the mean include six focusing on various musical artists, which includes Justin Bieber, One Direction, Katy Perry, Tupac Shakur, John Lennon, and Madonna. Number of theaters showing a film during the weekend of widest release is expected to have a positive impact on domestic box office. Average wide release for a documentary in the sample is 423 theatres. Justin Bieber: Never Say Never is the cohort leader for number of theatres at 3,118. Only 15 movies in the research cohort ever reach a wide release of more than 1,000 theatres. Average critical rating of the movies in the research cohort is approximately 84 percent positive with a standard deviation of 16.4. Five movies in the sample earn 100 percent positive rating from the critics. The five movies are Roger and Me ($15.9 million), Paris is Burning ($6.6 million), Flying Monsters ($6 million), Jerusalem ($5.2 million), and Man on Wire ($3.3 million). The expectation is for critical acclaim to have a positive impact on box office success. An alternative form of critical acclaim is formal recognition for industry awards such as an Academy Award nomination. Sixty percent of the movies in the research sample received award nominations. Rated R movies are expected to have a negative impact on box office revenue because the rating restricts attendance to individuals 17 years of age and older. The restricted rating is applied to nineteen percent of the movies in the research sample. Production budget for a documentary is normally very modest compared to other movie genres. Sicko by Michael Moore leads the way with a budget of $9 million but the average production budget for a documentary in the sample is only $866,025.
DETERMINANTS OF BOX OFFICE REVENUE OF DOCUMENTARIES
The estimated empirical relationship between the explanatory variables and box office revenue for documentaries is presented in Table 2. Two model specifications are put forth based on incomplete data for the BUDGET variable. The first is the full model, which includes all 100 movies in the sample. The second specification only includes the 54 movies that have BUDGET information available. The full and reduced model specifications are extremely consistent. Both models explain over 65 percent of the variance in documentary movie box office revenue. None of the independent variables have a correlation higher than 0.75 (THEATRES and OPENING have the highest correlation at 0.69), suggesting that excessive multicollinearity is not a problem with the analysis. Four out of the six
22 model independent variables are statistically significant. The first two independent variables in the model relate to the ability of documentaries to access exposure. The OPENING variable has positive and statistically significant impact on box office revenue for the research sample. The results are not surprising given that musical artist based documentaries dominate opening weekend numbers and tend to be major studio supported efforts that treat the opening weekend as a nationwide concert event. Justin Bieber, One Direction, Katy Perry are a few of the recent musical acts opening with a weekend box office revenue ranging between $7 million to $30 million. Of course, most documentaries do not enjoy an organized wide release effort on the opening weekend. March of the Penguins opened with revenue of only $137,000 on the first weekend but built momentum over time to expand to a peak of 2,506 theatres and total box office revenue of $94 million. Hence, opening weekend revenue is a significant predictor of box office revenue but total number of theatres during the widest week of release can also be an important determinant for documentaries that build momentum over time. The empirical results from this study indicate that number of theatres has a positive and statistically significant impact on box office revenue for documentaries. The two regression models imply an additional $143,000 to $150,000 increase in box office revenue per 100 screens of extra theatre exposure. Regardless to if the source of number of theatres is a push from a major studio or simple momentum from an independent following, achieving some degree of wide release is important for a documentary to find box office success. The variables CRITIC and AWARDS explore the impact of critical acclaim from both a micro and macro perspective. Good reviews and award recognition are expected to stir curiosity and identify quality. The CRITIC variable is positive but not statistically significant. As mentioned in the previous section, a composite positive rating of 84 percent makes it very difficult for documentaries to distinguish one from another. Horror movies represent the opposite side of the of the critical review spectrum with less than 38 percent of horror movies receiving a positive rating (Terry, King, and Patterson, 2011). Horror movies with acclaim across multiple critics provide an opportunity for distinction from the crowd but documentaries have little room to standout as even a film with 100 percent rating is only one standard deviation from the mean. Critical acclaim in the form of multiple major award nominations has a positive and statistically significant impact on box office revenue of documentaries. Holding other model variables constant, the empirical results imply the AWARD variable has a coefficient of approximately $5 million. One possible explanation for the result is that most documentaries do not have a large marketing budget so recognition from award nominations provides indirect promotional support. In addition, award recognition may serve as a signal to casual fans of the genre that a film merits priority viewing. As a rule of thumb, most documentaries are popular with critics but those that achieve award nominations are considered extraordinary. Another element that can affect the financial performance of a film is the rating assigned by the Motion Picture Association of America. The motion picture industry established the code as a means of giving advance information to parents and others about the theme and treatment of films. This voluntary code was adopted to prevent stringent forms of governmental controls. There are four possible ratings given to films in the research sample—G (general audiences), PG (parental guidance suggested), PG-13 (possibly unsuitable for children less than 13 years of age), and R (restricted; children not admitted unless accompanied by an adult). The conventional wisdom is that the family product sells, while an adult theme or treatment has a limited customer base because of age restrictions limiting access to the lucrative teenage market. This hypothesis does not appear to be
23 valid for the documentary genre. Empirical results from this study reveal a RESTRICTED variable that is positive and statistically significant in both model specifications. The box office gain associated with restricted movies is approximately $4 million in both model specifications. One possible explanation is that documentary fans prefer authentic films that illustrate some of the more graphic story details that are associated with a restricted rating. A second explanation is to recognize that youth audiences are not the target for the human interest and political themes associated with many documentaries with restricted ratings. The last variable in the model is BUDGET. Model results indicate the BUDGET variable is a positive but not a statistically significant determinant of domestic box office revenue for documentaries. Big budget movies with high profile movie stars, brand name directors, expensive special effects, and large advertising budgets have an obvious advantage drawing crowds at the box office but are not the normal domain of movies that fall in the documentary category. With an average budget of only $866,000 in the research sample, documentaries are the antithesis of the Hollywood summer blockbuster. In the extreme, most documentaries are more of a form of art and history than a viable commercial product that attracts significant investors. Another possible explanation is that documentaries are simply not comparable to most other motion picture genres with respect to revenue and expense considerations.
CONCLUSION
The documentary movie genre is a unique form of reality entertainment. Scholars and industry leaders differ in their opinion as to the profit viability of the documentary film today. This study examines the determinants of documentary movie box office revenue for the years 1978-2014. This study provides evidence that box office success for most documentaries is driven by exposure, award recognition from critics, and overall quality instead of studio production budgets. Opening weekend box office revenue and number of theatres during the week of widest release exposure variables are both positive and statistically significant determinants of documentary box office revenue. The positive or negative review of aggregated measures of critical acclaim captured by a rating database like the Rotten Tomatoes website is not statistically significant in this study. The documentary genre receives relatively high positive ratings across the board making it hard to employ the metrics as a form of distinction. That being said, critical acclaim via multiple award nominations is found to be positively and statistically significant with a coefficient estimate value of approximately $5 million. Unlike most other genres, documentaries appear to benefit from the graphic exposition associated with a restricted rating. Specifically, the positive and statistically significant coefficient for the restricted rating variable is approximately $4 million. Finally, production budget has a positive impact on the box office revenue of documentaries but the variable is not a statistically significant determinant. One avenue for future research into the documentary movie genre profitability is to extend the research focus to include alternative forms of revenue streams such as the various home video markets and foreign box office. A second avenue for future research is focusing on other specialty genres, which include music/concert, political, nature, and human-interest themes.
24 REFERENCES
Anast, Philip (1967). Differential movie appeals as correlates of attendance. Journalism Quarterly, 44, 86-90. Anderson, Chris. (2006). Long tail: Why the future of business is selling less for more. New York, NY: Hyperion. Austin, Bruce A. (1984). Portrait of an art film audience. Journal of Communication, 34 (winter), 74-87. Bernard, Sheila C. (2011). Documentary storytelling: Creating nonfiction on screen. Burlington, MA: Focal Press. C. Alex (2012). The Ripple Effect Living in the Long Tail: On the state of documentary film [Web log comment]. Retrieved from http://harmony-institute.org/ therippleeffect/2012/07/05/living-in-the-long-tail-on-the-state-of-documentary- film/ Campbell, Richard; Martin, Christopher R.; Fabos, Bettina G. (2011). Media and culture: An introduction to mass communication. New York, NY: Bedford/St. Martins. Chiou, Lesley (2008). The timing of movie releases: Evidence from the home video industry. International Journal of Industrial Organization, 26 (5), 1059-1073. Dodds, John C. & Morris B. Holbrook (1988). What’s an oscar worth? An empirical estimation of the effect of nominations and awards on movie distribution and revenues. Current Research in Film: Audiences, Economics and the Law, Vol. 4. Einav, Liran (2001). Seasonality and competition in time: An empirical analysis of release date decisions in the U.S. motion picture industry. Working Paper, Harvard University. Einav, Liran (2007). Seasonality in the U.S. motion picture industry. The RAND Journal of Economics, 38 (1), 127. Elberse, Anita (July, 2008). Should you invest in the Long Tail? Retrieved from http://hbr. org/2008/07/should-you-invest-in-the-long-tail/ar/1 Eliashberg, Jehoshua & Steven M. Shugan (1997). Film critics: Influencers or predictors? Journal of Marketing, 61, 68-78. Faber, Ronald & Thomas O’Guinn (1984). Effect of media advertising and other sources on movie selection. Journalism Quarterly, 61 (summer), 371-377. King, Timothy (2007). Does film criticism affect box office earnings? Evidence from movies released in the U.S. in 2003. Journal of Cultural Economics, 31, 171-186. Lee, F. L. (2009). Cultural discount of cinematic achievement: The Academy Awards and U.S. movies’ East Asian box office. Journal of Cultural Economics, 33(4), 239- 263. Levene, Catherine (1992). Marketing art-films to college students. Working Paper, The Wharton School, University of Pennsylvania. Litman, Barry R. (1983). Predicting success of theatrical movies: An empirical study. Journal of Popular Culture, 16 (spring), 159-175. Litman, Barry R. & H. Ahn (1998). Predicting financial success of motion pictures. In B.R. Litman, The Motion Picture Mega-Industry, Allyn & Bacon Publishing, Inc.: Boston, MA. Litman, Barry R. & A. Kohl (1989). Predicting financial success of motion pictures: The 80’s experience. The Journal of Media Economics, 2 (1), 35-50. Liu, Yong (2006). Word of mouth for movies: Its dynamics and impact on box office
25 revenue. Journal of Marketing, 70 (3), 74-89. Neelamegham, R. & P. Chinatagunta (1999). A Bayesian model to forecast new product Performance in domestic and international markets. Marketing Science, 18 (2), 115-136. Nelson, Randy A., Michael R. Donihue, Donald M. Waldman & Calbraith Wheaton (2001). What’s an Oscar worth? Economic Inquiry, 39 (1), 1-16. Obst, Lynda (2014). Sleepless in Hollywood: Tales from the new abnormal in the movie business. New York, NY: Simon & Schuster, Inc. Radas, Sonja & Steven M. Shugan (1998). Seasonal marketing and timing introductions. Journal of Marketing Research, 35 (3), 296-315. Ravid, S. Abraham (1999). Information, blockbusters, and stars: A study of the film industry. Journal of Business, 72 (4), 463-492. Reinstein, David A. & Christopher M. Snyder (2000). The influence of expert reviews on consumer demand for experience goods: A case study of movie critics. Working Paper, University of California-Berkeley and George Washington University. Rosen, S. (1981). The economics of superstars. American Economic Review, 71 (5), 845- 858. Sawhney, Mohanbir S. & Jehosua Eliashberg (1996). A parsimonious model for forecasting gross box office revenues of motion pictures. Marketing Science, 15 (2), 113-131. Simonoff, Jeffrey S. & Ilana R. Sparrow (2000). Predicting movie grosses: Winners and losers, blockbusters and sleepers. Chance Magazine, 13 (3), 15-24. Smith, Sharon P. & V. Kerry Smith (1986). Successful movies: A preliminary empirical analysis. Applied Economics, 18, 501-507. Sochay, Scott (1994). Predicting the performance of motion pictures. The Journal of Media Economics, 7 (4), 1-20. Terry, Neil, Michael Butler & De’Arno De’Armond (2004). The economic impact of movie critics on box office performance. Academy of Marketing Studies Journal, 8 (1), page 61-73. Terry, Neil & De’Arno De’Armond (2008). The determinants of movie rental revenue earnings. Academy of Marketing Studies Journal, 12 (2), page 35-47. Terry, N., King, R., & Patterson, R. (2011). Vampires, slashers, or zombies: Opening weekend’s favorite box office monster. Journal of Business & Economics Research, 9 (2), 95-105. Valentini, Valentina I. (2014). The 2015 complete guide to making movies: Festival Strategy 101. Movie Maker, 84-91. Wallace, W. Timothy, Alan Seigerman & Morris B. Holbrook (1993). The role of actors and actresses in the success of films: How much is a movie star worth? Journal of Cultural Economics, 17 (1), 1-27. Weiman, Gabriel (1991). The influentials: Back to the concept of opinion leaders. Public Opinion Quarterly, 55, 267-279. Wyatt, Robert O. & David P. Badger. 1984. How reviews affect interest in and evaluation of films. Journalism Quarterly, 61 (winter), 874-878. Wyatt, Robert O. & David P. Badger (1987). To toast, pan or waffle: How film reviews affect reader interest and credibility perception,” Newspaper Research Journal, 8 (summer), 19-30. Wyatt, Robert O. & David P. Badger (1990). Effects of information and evaluation in film criticism. Journalism Quarterly, 67 (summer), 359-368.
26 27 28 A CANONICAL CORRELATION APPROACH TO INVESTIGATE THE DETERMINANTS OF INVESTMENT SAFETY
Roman Wong, Andreas School of Business Barry University Nichole Castater, Andreas School of Business Barry University Bruce Payne, Andreas School of Business Barry University
ABSTRACT
In this study, we use a canonical correlation approach to investigate the constituencies of the concept of investment safety related to investments in common stocks. We identify a parsimonious set of financial indicators that, collectively, predict the safety level of a given firm in an investment context. The results from the canonical analysis show that there is a significant correlation between the multi-faceted concept of investment safety and a set of three financial measures – cash flow (measured by cash flow per share), earnings (measured by earning growth), and liquidity (measured by the current ratio). JEL Classifications: C38; E22; L25
INTRODUCTION
The equities markets have been experiencing increasing volatility since 2009 (Market Watch 2013; Murphy 2014). Such increases can be illustrated by discussing the VIX Index. The VIX Index Fund is a fund that measures the expectation of volatility on the S&P 500 market index over a one month period. The VIX is traded on the Chicago Board Options Exchange and is constructed using the implied volatilities of a number of puts and calls on the S&P 500 Index. When the VIX trades above 30, a large amount of volatility is said to exist in the market, and it is often called the “investor fear gauge.” When the VIX trades below 20, these times are considered less stressful in the stock market. Safe assets have been traditionally identified as fixed income assets that include US Treasuries, US Agency Debt and their Asset Backed Securities (ABS) and Mortgage Backed Securities (MBS) as well as municipal debt. These instruments are considered safest, on both a domestic and global scale, especially when issued by a government with a stable monetary policy. In addition, Many safe assets are also provided by the private sector and “shadow banking system.” (Gorton and Ordoñez 2013) These assets include investment grade bonds, high-yield bonds, and private sector mortgage-backed securities. Although debt investments would be a natural alternative to stock investments during volatile times, and have traditionally been “safe assets,” bond yields are at historically low levels and cannot provide a sufficient return for most investors.
29 The Federal Reserve’s third quantitative easing initiative (QE3) focused on government purchases of its own treasuries and agency debt and crowded out private investors to seek safety in corporate bonds and private sector MBS (Gorton, Lewellen et al. 2012). In addition, the Fed’s quantitative easing policies have caused record low yields along with record high prices for debt. Martin Feldman, in an interview with Goldman Sachs, states that the Fed’s policy is the primary cause of a “bond bubble” that may “pop” with the tapering of Fed funds (Nathan 2013). With the Fed deciding to “taper” QE3 from $85 billion to $75 billion per month in December 2013, and another reduction of QE3 to $65 billion in January 2014, the price of safe assets will decrease (Wearden 2013). In June 2013, Ben Bernanke’s mere mention of tapering caused a worldwide decrease in equity markets (Hargreaves 2013). However, the moderate taperin announcement of just $10 billion in December 18, 2013 signaled a more moderate policy than originally predicted and global stock indices rose on December 19, 2013 (Wearden 2013). Destabilizing political events in Europe and the United States have also affected global bond volatility. The European Debt Crisis in the PIGS (Portugal, Ireland, Greece, and Spain) has created doubt in European sovereign bond markets. The refusal of Congress to increase the debt ceiling in July 2011 and again in October 2013 resulted in the US losing its AAA credit rating in sovereign bond markets, as well as caused volatility and doubt in US sovereign bond markets. In order for investors to earn higher returns in bonds now requires investment in riskier investment grade or high-yield bonds. The risks associated with earning higher returns in bonds may not be appropriate for many investor profiles, especially for those who are approaching retirement. So this leaves investors searching for not just safe assets, but safe assets that yield higher than the risk-free rate of return. When the reality of more volatile stock and bond markets collides with the rapidly aging populations in the US and other developed countries, the need to identify safe assets which earn appropriate returns becomes paramount. This situation highlights the importance of this study. Part of our objective is to create a means to better identify safe equity investments using easily accessible financial indicators. These financial indicators are closely correlated with the Value Line ranking system, whose highly-ranked stocks have exhibited positive abnormal returns.
VALUE LINE RANKINGS
Value Line has created rankings for its universe of 1,700 companies. The “Safety” rank is an overall measure of risk of the stocks analyzed by Value Line. It is derived from the “Stock Price Stability” the “Financial Strength” ratings of a company. Those stocks with a ranking of 1 are the “safest, most stable, and least risky investments” relative to the other stocks in the Value Line universe (2013). Those stocks with high “Safety” ratings are usually large, financially sound companies, which pay regular cash dividends, and can have less than average growth prospects.(2013) These “safe” stocks often provide two sources of income for the investor in the form of dividends and slow capital gains growth. In volatile markets, these stocks can become invaluable sources of stability and income for those investors with certain risk profiles. According to Value Line research, those stocks with high “Safety” rankings often fall less during times of market downturn, while still issuing dividends Value Line universe (2013). The Value Line scores that measure “Stock Price Stability” and “Price Growth
30 Persistence” for the 1,700 stocks it follows are the basis of research in this paper. These measures can range from a low score of 5 to a high of 95, and are comparisons between the stock in question and the rest of the stocks in Value Line’s universe (2013). “Stock Price Stability” is basically a measure of the stock’s price volatility. It is “a relative ranking of the standard deviation of weekly percent changes in the price of [the stock] over the last five years.”(Greene 2010) If there is not five years of data available, then there is no “Stock Price Stability” ranking for that stock. “Price Growth Persistence” is a “measurement of the historical tendency of a stock to show persistent growth over the past 10 years compared to the average stock.” (Greene 2010) Again, if there is not 10 years of data available for the stock in question, then there is no “Price Growth Persistence” ranking for that stock. The need for further research on the “Safety” rank and its components is the foundation for this research. We notice that Value Line uses market prices, growth rates, and standard deviations for “Price Growth Persistence” and “Stock Price Stability,” respectively. These rankings are both good indicators of equity investment safety.(Waggle 2001) However, this research seeks to further investigate these rankings by using canonical correlation to create a financial profile to explain financial safety. In addition, instead of using market prices, this research uses fundamental financial indicators, including cash flow, earnings, and liquidity and further identifies a financially safe equity investment. This expands on existing research by O’Hara, Lazdowski, et al. which finds that companies that exhibit “17 years of consistent growth in dividends, cash flow and earnings can outperform the market on a constant, long-term basis.” The indicators used to measure dividends, cash flow, and earnings include dividends per share, and earnings per share (O’Hara, Lazdowski et al. 2000). These fundamental indicators are available to all investors and researchers, not just those who subscribe to Value Line, and for firms other than those contained in the 1,700-member Value Line universe.
OBJECTIVE
In an age of volatile markets, safe investments can help preserve principal and compound investments at a steady rate of return. Our objective in the current study is two-fold. First, we provide a discussion on the two different dimensions, price stability and growth persistence of a stock that define the concept of financial safety in the context of investment. Second, we then investigate the finer contents of the multi-dimensional concept of financial safety. Using the “stability” and “persistence” rankings as well as multiple financial measures from Value Line, we use a canonical correlation analysis approach as a tool to investigate the relationships between the concept of investment safety and its underlying predicting factors. In the following sections, we first provide a discussion on the concept of investment safety, its underlying dimensions as conceptualized in this study, and a discussion on the set of financial indicators selected to be used as predicting variables for the concept. We then briefly discuss the analysis methodology and the canonical correlation analysis (CCA) technique before describing the data and their analysis. The results and their interpretation will be discussed before our conclusion and recommendations for future studies.
31 INVESTMENT SAFETY
Investment safety is not a single-dimension concept. Primarily, safe investments are investments that are subject to little or no risk of loss. These investments may or may not be guaranteed by a government or quasi-government agency. Secondarily, safe investments typically pay modest and continuous returns to investors. The capability to generate continuous stream of income should reflect in its stock price. As such, we posit in this study that the concept of investment safety should comprise of two underlying dimensions: The stock price stability and the persistence of growth in the value of common stocks. The first dimension, price stability, of the safety construct is the Value Line rating for stock price stability. This measurement is based on the ranking of the standard deviation of weekly percent changes in the price of a stock over the past five years. The lack of price level volatility may be used as a measure of the absence of risk, or safety of investment. In a recessionary market, stocks with price stability allow investors’ capital to stay in a safe haven. While the investment with price stability may still have some potential for growth, its investor does not run the risk of losing vast amounts of wealth. The persistence of growth in the value of common stocks has long been of interest to investors, investment counselors, financial managers, and academicians. It seems to be a common belief that a firm that has grown rapidly for the past for several years is highly likely to repeat this performance in the future. Conversely, stocks that have done poorly over prolonged periods are shunned and trade at low multiples. In a trend of rising prices, the price of an investment is expected to grow at least as much as inflation in order to maintain its value over time. Indeed, the persistence rather than the magnitude of growth has become of primary importance to the selection of securities by both institutional investors and inside traders (Meisheri 2006, Damodaran 2002, and Payne 2004). The Value Line proprietary measure of price growth persistence rewards a firm for the consistency with which it outperforms the broader universe of equity offerings over an extended period of time. It is used here as a proxy measure of the second dimension, persistence of growth.
PREDICTING VARIABLES
Previous studies have chosen predictor (explanatory) variables by various methods and logical arguments. Despite the variation in the selection methods, the safety of an investment can be effectively reflected in three aspects: the firm’s financial structure (the long term debt to total capital is a measure of financial risk), its potential of sustaining earnings, and the firm’s liquidity. Accordingly, the group of predictor variables chosen in this study for analysis includes a measure of the overall finance risk, two measures of return or potential return to investors, and two measures of liquidity. Table 1 below summarizes the criterion variables and predictor variables used in the current study. The first measure of return is return to total capital (ROTC). Return to total capital includes a return to creditors as well as owners, and recognizes that value is affected by the cost of debt. A measure of return to equity could be used, but it would ignore the cost of debt and the fact that debt as well as equity is used to finance assets. This is consistent with the use of the debt to total capital ratio as a measure of financial leverage (financial risk). It is an understatement that growth, whether it be measured as positive changes in sales, share price, or the present value of invested dollars, in a period of economic recession
32 and financial market turmoil is unusual. Financial literature is replete with methods and theories on how to achieve growth, how to determine the optimum rate of growth, and even if an optimum growth rate exists (Miller and Modigliani 1961). On one hand, one can look at growth with a euphoric view that it would bring in new cash flows and incomes; hence, more safety. On the other hand, in situations where a company has to maintain growth by cutting prices may not be able to contribute as much in terms of income. In addition, growth usually requires extra funding for increases in the working capital. During a time when overall liquidity is sluggish, such demands for additional working capital can exert financial stresses on the company. In this paper, we posit there is no a priori expectation on how positive growth would affect the safety of a company. It is simply not known. The current ratio, to a certain extent, can be used to indicate a given company’s ability to pay its short-term liabilities (debt and payables) with its short-term assets (cash, inventory, receivables). The higher the current ratio, the more capable the company is of paying its obligations. Generally, a ratio under 1 suggests that the company is unable to pay off its obligations if they were to become due at that point. While a current ratio less than 1 shows a lack of liquidity, it does not necessarily mean that the company will go bankrupt; as there are many ways to access financing. Furthermore, a low current ratio can indicate a highly-efficient operating cycle or an uncanny ability to turn its product into cash (i.e. Wal-Mart). Companies that have trouble getting paid on their receivables or have long inventory turnover can run into liquidity problems while maintaining a high current ratio. The high current ratio can be the result of a high level of accounts receivable and inventory due to inefficient turnover. Because business operations differ in each industry, it is always more useful to compare companies within the same industry. Cash-flow-per-share (CFPS) is considered another measure for liquidity. The capability of business to meet claims on a timely basis becomes difficult if the business is low on cash. Furthermore, lower operational cash flows may compel a company to cut back on planned profitable projects. Conversely, if there are adequate cash flows, the chance that the company has to forego earning opportunities due to projects being underfunded will be minimized. Thus, cash flow per share is included in the profile of explanatory variables. There is an a priori expectation that firms with higher levels of safety will have significantly greater cash flows per share that firms selected at random. In summary, there are five explanatory variables in this canonical correlation model: ROTC, GROWTH, CURRATIO, DTC, and CFPS are compared against the concept of investment safety consisting of the aforesaid two variables. These five variables measure the firm’s profitability (ROTC), financial risk (DTC), its ability for sustained growth (GROWTH), its liquidity (CURRATIO and CFPS). Thus, the study contains measures of both risk and return that determine the value of the firm. A basic tenet of this study is that investors at the margin evaluate the degree of risk in an investment and compare it to the investment’s potential rate of return. In modern textbooks this is a fundamental principle referred to as the “risk-return tradeoff.” (Brigham and Daves, 2012) Investors at the margin “trade off” proxies for risk and return in buying and selling securities to establish demand and thus, price or market value. Safety is simply one side of that tradeoff indicating a lack of risk, and when investors become more risk adverse, as in a recent period of recession and slow recovery, they must have a greater potential return to assume marginal risks.
33 DATA ANALYSIS
The objective for the data analysis in the current study is to obtain insights into the interrelationships between the two investment safety dimensions and the collection of predicting variables. In the current study, we have two metric scales, STAB and PERSIS, as the criterion variables and a set of five metric predictors as the independent variables. Ordinary regression analysis or stepwise regression analysis handles only a single criterion variable; therefore, these statistical techniques do not provide an accommodation for the multiple criterion variables in our study. As such, we use the canonical correlation analysis method to analyze the data we collected for this study.
Canonical Correlation Analysis
Canonical correlation analysis (CCA) is a multivariate statistical model aimed at studying the interrelationships among sets of multiple dependent variables and multiple independent variables. It is generally accepted to be the most appropriate and powerful multivariate approach (Hair et. al 1998). CCA develops a number of orthogonal canonical functions that maximize the correlation between the linear composites, also known as canonical variates, which are sets of dependent and independent variables. Each canonical function is actually based on the correlation between two canonical variates, one variate for the dependent variables and another for the independent variables. Those variates are derived to maximize the correlation between them. In addition, CCA generates a number of canonical functions (pairs of canonical variates). The number of functions generated each time is equal to the number of variables in the smaller set of variables.
The CCA model
In the current study, our objective is to evaluate the strength of the associations between the canonical variates which include the five metric scales: ROTC, GROWTH, CURRATIO, DTC, and CFPS and the concept of investment safety. The set of dependent variables includes the two underlying dimensions of STAB and PERSIS. The five independent variables resulted in a 40-to-1 ratio of observations to variables, exceeding the guideline of 10 observations per variable. The CCA in our study was restricted to deriving two canonical functions as the dependent variable set contained only two variables, STAB and PERSIS. To determine if both canonical functions should be included in the interpretation stage, we focused on the level of statistical significance, the practical significance of the canonical correlation, and the redundancy indices for each variate. The first statistical significance test is for the canonical correlations of each of the two canonical functions. Both of the two canonical functions were tested to be statistically significant at the 0.01 level (see Table 2). In addition to tests of each canonical function separately, multivariate tests of both functions simultaneously are also performed. The test statistics employed are Pillai’s criterion, Hotelling’s trace, Wilks’ lambda, and Roy’s largest root. Table 2 also details the multivariate test statistics, which all indicate that the canonical functions, taken collectively, are statistically significant at the 0.01 level. In terms of the magnitude of the canonical relationships, the practical significance of the canonical functions as represented by the size of the canonical correlations cannot be said to be strong. The canonical correlation coefficients that indicate the variates’ linear relationship are 0.3497 and 0.2661 for Function 1 and Function 2, respectively. In
34 other words, the proportions of variances in the respective criterion variates that can be explained by the canonical functions are respectively 12.23% and 7.1%. Nonetheless, when we consider that the set of predictor variables selected for this study are relatively small in number and constitute only a small subset of all key financial indicators, having the canonical correlations to be in such a size is not unexpected.
Redundancy analysis
The squared canonical correlations provide an estimate of the shared variance between the canonical variates rather than that from the sets of dependent and independent variables (Alpert and Peterson 1972). The interpretation of the canonical correlations can be misleading. This is particularly true when the roots are considerably larger than previously reported bivariate and multiple correlation coefficients. The researcher may be tempted to assume that the canonical analysis has uncovered substantial relationships of conceptual and practical significance (Hair et. al 1998). To overcome this issue, Steward and Love (1968) proposed the calculation of the redundancy index as a summary measure of the ability of a set of independent variables (taken as a set) to explain variation in the dependent variables (taken one at a time). In the earlier statistical significance tests, although both of the functions are statistically significant, we focus on interpreting the first function as the squared correlation for the second function is relatively insignificant. In Table 3 we summarize the computation of the redundancy indices for the predictor and criterion variates in the first function. The redundancy index for the criterion variables is 0.0785. That is, approximately 8% of the variances in the criterion variables can be explained by the predictor variate. As there have not been any generally accepted guidelines to judge what level, above which, a redundancy index is supposed to be acceptable, one needs to make his judgment in accordance to the context of the study. In the current study, as the input and output scales we use are only small subsets of the whole repertoire of indicators, it is not particularly unexpected that the redundancy indices are not much higher (0.08 and 0.03 respectively).
Interpretation of the canonical variates
The CCA results are summarized in the Canonical Structure Matrix (Table 4). The canonical structure matrix reveals the correlations between each variable and its own variate in the canonical functions. It can be said that these correlations are like the factor loadings of the variables on each discriminant function and can be interpreted as they are standardized. It allows the comparison of the variables in terms of their correlations and how closely a variable is related to each function. Generally, any variable with a correlation of 0.3 or more is considered to be significant. Recall the results of the significance tests in the earlier paragraphs, our interpretation focuses only on the first canonical function and not the second one. Adopting the traditional 0.3 as the cutoff point, three predictor variables; GROWTH (0.5939), CURRATIO (-0.5226), and CFPS (0.7355); are significant contributors to the composite independent variate. Both of the variables in the criterion variate have significant contributions at 0.7806 and 0.8215 for STAB and PERSIS, respectively. Based on their loading values, we can rank the importance of the independent variables in the following orders: CFPS, GROWTH and CURRATIO. That is to say that CFPS, the variable indicating cash flow, contributes most highly in explaining the variances in the dependent variate in the canonical function.
35 In terms of the dependent variables, it appears that the variances in the stock value growth consistence variable (PERSIS) are slightly more ready to be explained by the independent variate than the stock price stability variable (STAB). The positive signs of CFPS and GROWTH indicate that the higher the loading are on these variables the higher the levels of STAB and PERSIS. However, the negative loading on CURRATIO, while maximized for correlation with the dependent variate, renders the interpretation to be a complicated task. Figure 1 graphically depicts the relationships between the independent and dependent variables and their variates.
DISCUSSION
Our independent variate is composed of stock price growth consistency PERSIS and stock price stability STAB, with most explanatory power expressed through the PERSIS component. This represents the fact that our canonical correlation model is a better suited to those stocks that have a consistent growth rate over at least the last ten years. Since stock price growth is a desirable trait for any long term investment strategy, these firms would generate returns that are at least approximating the inflation rate, if not higher. These returns would then grow investment funds over the long term to insure future financial goals were met. This is a prime component of what we have explained as safety in equity assets. Consistent with an emphasis on price growth consistency, the highest positive loading components of our dependent variate are cash flow per share (CFPS) and earnings growth (GROWTH). This indicates that those firms with the highest cash flow per share and the highest earnings growth are those that will have the most consistent stock price growth over time. This stands to reason due to the fact that firms with higher CFPS and GROWTH will generate more internal equity, and will be less dependent on outside sources of financing during tough economic times. With higher amounts of internal equity, these firms can take advantage of growth opportunities during times when outside financing may be scarcer. The highest negative loading component of our dependent variate is the current ratio, CURRATIO, one of the primary measures of liquidity. The negative sign suggests that the relationship between CURRATIO and the predictor variate is inverse. This is a counterintuitive result. To obtain plausible explanations for such an unexpected result, we performed several procedures. The first procedure was a sensitivity analysis to test how the liquidity variable, CURRATIO, and the financial structure variable, DTC, individually contribute to the predictor variate. The underlying question we had was: Would the inclusion of DTC, which already includes the constituencies of CURRATIO, be a disturbing factor to have caused the unexpected sign of CURRATIO? In the sensitivity analysis, the CCA model was run again twice, one without CURRATIO and another without DTC, to verify whether the canonical loadings would change significantly when the model changes. Compared to the full model (contains all five predictor variables), neither of the revised models resulted in any significant change in the canonical correlation coefficient. In the model that CURRATIO was left off, the variation in the distribution of the loadings mainly occurred to the variable GROWTH (increased from 0.5939 to 0.7314). As the loadings were more evenly distributed between the growth and cash flow variables, the criterion variate was leaning more towards the variable PERSIS. In the other revised model in which DTC was left off it instead of CURRATIO, the
36 loading distributions remained almost unchanged in both of the predictor and criterion variate. The second revised model suggests the financial structure, represented by DTC, does not have much correlation with the safety construct we defined earlier in this paper. However, our sensitivity analysis of the components in the model does not seem to have offered too much of an explanation to the opposite sign of the variable CURRATIO in the CCA model. Another possibility for the negative loading of CURRATIO may be attributed to the inherent contradictory nature between STAB and PERSIS. The variable STAB is the measurement of how stable the stock prices of a given company has been: A stock with its price unchanged over time will rank higher than one that appreciates gradually. On the other hand, the variable PERSIS rates more highly the stocks of those companies whose values grow over time. To verify this assertion, we conducted another sensitivity analysis on the alternate canonical correlation analysis function. In CCA, usually the first function, which has the highest canonical correlation, is chosen for interpretation. So did us in this current paper. However we also looked into an alternate combination of the variables in the second CCA function with a view to uncover additional insights into the relationships between the predictor and criterion variables. Table 6 summarizes our analysis of swapping the variables of CURRATIO and DTC in and out of the CCA model. In the model where all the five predictor variables were present, the predictors GROWTH and CURRATIO have loadings that are higher than 0.3, and with a positive sign, which are in the same direction as the criterion variable PERSIS. The positive sign of the two predictor variables and the criterion variable shows that the relationship between them is a direct one. Another predictor variable that has a loading greater than 0.3 is DTC, that has a negative direction, the same as that of STAB. This loading distribution provides us with some initial evidence to show that the two dimensions underlying the investment safety construct, stability in stock price (STAB) and persistence of growth (PERSIS), are orthogonal and predicted by different sets of predictors. In that, growth and liquidity seem to support a company’s stock price sustaining a steady growth; while the long term debt to capital ratio is directly related to the stability of stock price of a company. When the liquidity variable was removed from the model, the loading on the cash flow variable became much more significant to replace the role of CURRATIO in supporting STAB. As DTC was swapped with CURRATIO, the shift in the loading distributions further strengthened the pattern that GROWTH and CURRATIO tend to be directly and significantly connected with the dimension of stock price growth persistence. Yet, another possible explanation as we discussed earlier in this paper is that a small current ratio might be a primary indicator of efficiencies in receivable collections and inventory management. Warren Buffet states in his investment philosophy that companies with a current ratio less than one over a long period of time are usually companies that have what is called a “sustainable competitive advantage,” (Livy 2013) These companies can convert inventory to revenues so quickly that this economic power far outweighs any lack of liquidity over time. The negative loading on CURRATIO substantiates this philosophy by showing that investors are investing in companies with sustainable competitive advantages. It is noteworthy that the correlation between the two canonical variates in the primary canonical correlation function is relatively low. One direction for further research is that the low correlation could have been caused by the incompatible scales of the variables used in the study. Among the five predictors and two criterion variables used in the CCA
37 model, three (STAB, PERSIS, and GROWTH) of the seven variables measure each an aspect of a company across a time span; while the remaining four variables are relatively ‘snapshot’ indicators measuring a company at a specific point in time. Note that both of the dimensions in the criterion variate are measured by variable covering a time span and only one of the five predictor variables measures a time interval. Some important information relevant to earlier periods might therefore be missing. This could be a major issue that explains the low canonical correlation. The data analysis procedure, canonical correlation analysis, seeks to maximize the correlation between the linear combination of the criterion variate and its predictor variate. However, such combination may not always make much theoretical sense. In addition, nonlinearity can also pose the same problem as it does in simple correlation i.e. if there is a nonlinear relationship between the sets of variables, the CCA technique is not designed to detect that. While CCA is an appropriate data analysis procedure that facilitates the description of the relationship between the set of predictor variables with the construct of investment safety, it is very sensitive to the data involved, i.e. adding or leaving out influential cases and/or individual variables of the analysis can change the outcome dramatically.
CONCLUSION
In summary, the results of this analysis indicate that when investors are primarily concerned with safety, consideration should be given to consistent growth, followed by low variance in stock price. The three financial variables to consider should be earnings growth, cash flow per share, and the current ratio. Over the long term, consistent earnings growth and positive cash flow per share contribute to a safe equity investment. Conversely, over the same long time horizon, firms with current ratios consistently less than 1 show an operational efficiency and earnings power that also contribute to the stock’s safety. Further analyses suggest the somewhat unexpected sign of loading on the liquidity proxy, CURRATIO, is likely due to the opposing natures of the two dimensions, capital growth and stability of investment value, which underlie the construct of investment safety. Of course it may be argued that such phenomena are not inconsistent with the aforementioned Buffet philosophy of sustainable competitive advantage. Although these results are promising, the explanatory power of this canonical correlation analysis in its current form is relatively low. Future studies could involve different financial variables, as well as more strategic and operational variables. These could further enhance this model and method of analysis. Interestingly enough, the low explanatory power of the model, but high correlation of components within the independent and dependent variates show that there are a few but very key financial characteristics that must be addressed in order for a stock to be considered safe. However, the lack of explanatory power of the overall model points to more as of yet undefined characteristics that must be analyzed to make the model more robust. These undefined characteristics could range from management styles to operational efficiencies to innovation at the most basic level. This study has resulted in a contribution to the construction of a theory. Further research in this area may be rich in potential contributions for constructing a complete theory of the safety of invested capital. Such a theory would be invaluable to investors, investment managers, companies, and academicians.
38 REFERENCES
Alpert, Mark I. and Peterson, Robert A. Peterson. (1972). On the Interpretation of Canonical Analysis. Journal of Marketing Research, 9 (May), 187-192. Brigham, Eugene and Philip Daves (2012). Intermediate Financial Management, 11th edition, Boston, MA: Centage Learning. Damodaran, Aswath. (2002). Investment Valuation: Tools and Techniques for Determining the Value of Any Asset. New York, NY: John Wiley and Sons. Gorton, G. B. and Ordoñez, G.L. (2013). The Supply and Demand for Safe Assets. The American Economic Review,102, (3), 101-106. Greene, R. (2010). Earnings Predictability, Price Growth Persistence and Stock Price Stability. Value Line. Hair, Joseph F., Anderson, Rolph E., Tatham, Ronald L. & Black, William C. (1998). Multivariate Data Analysis, 5th edition. Englewood Cliffs, NJ: Prentice Hall. Hargreaves, S. (2013). Fed Sets Road Map for End of Stimulus. CNNMoney.com Livy, Julian. (January, 2013). Warren Buffett Does Not Like Debt: http://www. buffettsecrets.com/warren-buffett debt.htm Meisheri, Kaush. (2006). One Page Quick Analysis of a Stock. https://home.comcast. net~lowellherr/quickvl.htm Modigliani, Franco. & Miller Merton H. (Oct., 1961) Dividend Policy, Growth, and the Valuation of Shares: The Journal of Business, Vol. 34, No. 4, 411-433. Nathan, A. (April, 2013). Bond Bubble Breakdown. Top of Mind 11. O'Hara, H. T. (2000). Financial indicators of stock price performance. American Business Review 18, 1, 90-100. Payne, Bruce C. (1993). A Multiple Discriminant Investigation Into the Financial Characteristics of High Growth Firms: Advances In Quantitative Analysis of Finance and Accounting. 2, 19-33. Stewart, Douglas. & Love, William. (1968). A General Canonical Correlation Index. Psychological Bulletin, 70, 60 163. Value Line. (2013). Definitive Guide: The Value Line Ranking System. Wearden, G. (December, 2013). Markets Rally as Traders Take Taper in Their Stride. The Guardian.
39 TABLE 1
PREDICTOR AND CRITERION VARIABLES
Criterion Variables STAB Stock Price Stability PERSIS Persistence of Growth in Value of Common Stocks
Predictor Variables ROTC Return to Total Capital GROWTH The Five Year Growth Rate in Earnings CURRATIO Current Ratio DTC Long Term Debt to Total Capital Ratio (Financial Risk) CFPS Cash Flow Per Share
TABLE 2
MEASURES OF OVERALL MODEL FIT AND
MULTIVARIATE TEST OF SIGNIFICANCE
Canon. Can. Function F Corr. Sq. Corr p 1 4.14 0.35 0.12 0.000 2 3.70 0.27 0.07 0.006
Test Name Value Approx. F Sig. of F Pillais 0.1931 4.15 0.000 Hotellings 0.2155 4.14 0.000 Wilks 0.8156 4.14 0.000 Roys 0.1223
40 41 TABLE 5: SENSITIVITY ANALYSIS FOR THE PRIMARY
CANONICAL CORRELATION FUNCTION
Canonical Canonical Canonical Predictor Variable Loading Loading Loading ROTC -0.0545 -0.0189 -0.0572 GROWTH 0.5939 0.7314 0.5839 CURRATIO -0.5226 -0.5366 DTC 0.0369 -0.0571 CFPS 0.7355 0.7089 0.7378
Criterion Variable STAB 0.7806 0.5459 0.7956 PERSIS 0.8215 0.9587 0.8074
Canonical Correlation 0.3497 0.3353 0.3491
TABLE 6: SENSITIVITY ANALYSIS FOR THE ALTERNATE CANNONICAL
CORRELATION ANALYSIS FUNCTION
Predictor Variable Canonical Canonical Canonical Loading Loading Loading ROTC 0.1416 -0.2545 0.1446 GROWTH 0.577 -0.4929 0.6163 CURRATIO 0.7181 0.7251 DTC -0.3773 0.6057 CFPS -0.0788 0.6116 -0.057
Criterion Variable STAB -0.6251 0.8379 -0.6058 PERSIS 0.5702 -0.2846 0.5901
Canonical Correlation 0.2661 0.1641 0.2572
42 FIGURE 1
CANONICAL CORRELATION MODEL FOR INVESTMENT SAFETY AND ITS FINANCIAL INDICATORS
ROTC
-0.06
0.59 GROWTH 0.78 STAB FINANCIAL 0.35 INVESTMENT -0.52 INDICATORS SAFETY CURRATIO 0.82 PERSIS 0.04
DTC 0.74
CFPS
43 44 HOW PROGRESSIVE IS THE U.S. TAX SYSTEM? Syed Shahabuddin, University of Central Michigan
ABSTRACT
The U.S. income tax system is used to collect revenue, and influences the economy. Tax rates have changed many times and have now decreased to a point where many believe the tax system is no longer progressive. The issue of progressivity or lack thereof has become more pronounced since the election to Congress of Tea Party candidates, who insist on cutting spending and lowering taxes. This paper shows that the income tax system has, in fact, become less progressive, and that if nothing is done to balance revenue with expenditures, a long-term budget deficit will result. JEL Classification: E6
INTRODUCTION
The U.S. taxation laws have gone through many changes, as reflected in the U.S. Treasury Fact Sheet (2003). Initially, the governments (states and federal) collected money by imposing taxes on a variety of sources, i.e., excise taxes, tariffs, and custom duties. Later, the federal government was allowed to impose income taxes with the adoption of the Sixteenth Amendment, ratified on February 3, 1913. Since then the tax rates have been changed many times to adjust rates to meet the growing need for revenue. Tax rates also have been adjusted to manipulate the economy by providing incentive to encourage or discourage economic activity. Since the 1950s, the government has manipulated income tax rates to control economic fluctuations, i.e. business cycles. In 1951, the House passed the biggest across-the-board tax increase (12.5 percent) in U.S. history. Then later, in 1954, the taxes were reduced to spur private investment. Other reductions occurred in 1962, 1966, 1969, 1971, 1975, and 1981 and were also implemented mainly to spur investment as unemployment had risen. In 1975, President Ford signed into law a tax refund of $22.8 billion. In 1977, President Ford proposed a permanent tax cut, and in 1979, he suggested automatic adjustments that would keep the tax rate constant for families. Other tax changes include the Economic Recovery Tax Act of 1981, the Tax Reform Act of 1986, the adjustment of 1993, The Tax Payer Relief Act of 1997, and the Economic Growth and Tax Relief and Reconciliation Act of 2001. All these actions were taken to either lower or raise marginal tax rates, to spur economic activities, and/or to solve budgetary problems. GDP of the last 50 years indicates that the economy has gone through many business cycles and has required intervention from time to time. Taxation has been one of the tools used by the federal and state governments to manipulate economic activity. As tax changes were made, the federal deficit (Figures 1) and national debt has skyrocketed. On December 31, 2013, the U.S. debt stood at almost 17 trillion dollars or 108 percent of the GDP. Obviously, the new Congress elected in November 2010 and 2012 are trying to control spending while advocating lower taxes. The question most people may ask is, “Who pays
45 taxes and has the tax system become progressive or regressive?” Research consists of analyzing economic and tax data collected from many sources to determine whether a case can be made for progressivity of the tax structure of the United States. Statistical methods were used for analysis as well as other published research on the topic was evaluated to determine the progressivity or regressivity of the tax structure. Based on the analyses and evaluation, the conclusions were reached. The purpose of the paper is to statistically analyze economic and tax data to determine whether the tax system is progressive or regressive.
GDP AND POLICIES AND EFFECTS
The U.S. economy has gone through many business cycles and has needed a shot in the arm to revive it. The two methods of revival are fiscal policy and/or monetary policy. Fiscal policy has been used by governments in times of crisis, e.g., high unemployment, inflation, budget deficit, or other social and economic problems of the time. To be effective, fiscal policy changes must be timely. The purposes of changing fiscal policy are to temporarily affect economic activity by lowering corporate tax rates or allowing deduction for business investment. Fiscal policy can also be used to boost consumer demand by lowering individual tax rates or giving tax “holidays” for a certain period of time. According to Myles (2000),”through its effect on the return to investment or the expected profitability of research and development, taxation can affect what choices are made and, ultimately, the rate of growth” (p. 145). In addition to using fiscal policy to influence the economy, fiscal policy is also used by politicians to prove that they are doing something, but monetary policy can also be effective as countercyclical , because “implementation lags are much shorter for monetary policy than for fiscal policy” Taylor (2000, p. 27). Further, monetary policy can be easily rescinded when it achieves its purpose. In addition, tinkering with discretionary fiscal policy makes it harder for the Federal Reserve to determine the need and scope of the monetary policy. Therefore, Taylor (2000) concludes that it is better for fiscal policy to function as an automatic stabilizer while allowing monetary policy to try and keep “the aggregate demand close to potential GDP. […] Empirical evidence suggests that monetary policy has become more responsive to the real economy, suggesting that fiscal policy could afford to become less responsive” (p. 34). Fiscal policy can shift aggregate demand and can change real GDP; therefore, fiscal policy should be carefully managed to keep the real GDP close to potential GDP under inflationary conditions, and fiscal policy “reduces deadweight loss and creates greater efficiency.” As Taylor (2000) states, “Running a budget surplus to keep real interest rate down provides for more private investment and higher economic growth. […the ] unemployment component, payroll tax policy and other laws affecting the labor market can change the natural rate of unemployment” (p. 26). Regardless of the purpose, sometimes fiscal policy may involves changing the tax structure, which could favor some sectors of the economy more than others or benefit some people more than others. Discretionary changes in taxes and spending are automatic stabilizers (Taylor, 2000), however. The effect of automatic stabilizers on spending and taxes is much larger “than even the proposed discretionary changes. […] Both types of changes in taxes and spending impact aggregate demand, but the automatic ones are more predictable and work more quickly than the discretionary ones” (p. 26).
46 However, the federal government outlays have increased from less than 2.78 percent of the GDP in 1954 to almost 26.29 percent of the GDP in 2010 (Figure 1). But the receipts have increased from 2.62 percent of the GDP in 1954 to 16.65 percent of the GDP in 2010. Figure 1 shows that receipts remained equal to outlays until 1982, but dropped below the outlays since then except in 2001 and 2002 when receipts exceeded outlays resulting in surplus in 46 years. Since then, the outlays have exceeded the receipts causing large deficits. These deficits have become a basis of political argument over how to cut spending to balance the budget. The questions is whether balancing should be via tax increases, spending reductions, or both. Further, if taxes have to be raised, who should pay more? The overarching question is, “What is the fairest way to increase or decrease taxes?” To answer this question requires knowing who is paying the most taxes now as well as evaluating the history of taxation in the United States.
TAXATION IN THE UNITED STATES
Governments tax people, businesses, or other sources of income to generate revenue. However, governments should be careful not to impose taxes so high that they result in loss of revenue by discouraging economic activities (Fullerton, 1982). Further, even Smith (1776) argued about the inverse relationship between marginal tax rate and tax revenue. In other words, tax receipts and tax rates have a concave relationship. That is, after a certain point, higher tax rates result in decreasing revenue, a point called the “prohibitive” range. Saez (2003) states, “The elasticities of taxable income and adjusted gross income are around 0.4 and significant but the elasticities of wage income are in general insignificant and close to zero” (p. 1231). Goolsbee (2000) study confirms that “the short-term elasticity of taxable income with respect to the net-the-tax should exceeded one, but taking out the temporary component yield longer-run elasticities between zero and .4” (p. 375). Therefore, Goolsbee (2000) concludes “that taxing the rich can lead to dramatic shifting of taxable income in the years immediately surrounding a tax change. […] But after the shifting is done, the total changes in taxable income, […], seem to be more limited” (p. 375). Feldstein (1995) found that “there is very substantial response of taxable income to change in marginal tax rates” (p. 552). However, Feldstein (1995) found that changes in tax rates have less impact on tax revenue. Further, the high marginal tax rate creates significant deadweight losses due to changes in the behavior. Therefore, Feldstein (1995) suggests that governments should keep these behavior and effect in mind when developing tax policies and spending levels. Tanzi (1969) found that the long-run elasticity is estimated at 1.42, and Blackburn (1967) concluded that for every one percent increase in income of the tax payers, the federal personal income tax revenue will rise about 1.4 percent (p. 168). Tanzi (1969) found the same relationship between taxes and revenue (p. 209). Table 1 shows that from 1955 to 1960, GDP increased by 13.22 percent while receipts increased by 25.51 percent and outlays increased by 17.33 percent. Change in receipt to change in GDP was 4.89 percent from 1955 to 1960, while change in outlays to change in GDP was 3.46 percent during the same period. The change in outlays to change in receipts was 70.89 percent during the same period. That is, during the five years (1955 - 60), receipts increased by $16,160 billion compared to outlays which increased by $11,456 billion. The change in receipts to the change in GDP was 1.55 percent, and the change in outlays to change in GDP was 158.86 percent during 2005 - 2010. The change in outlays compared to
47 change in receipt was 10,265 percent. That is, receipt in 2010 was $2,205 billion in 2010 compared to receipt in 2005 of $2,196, an increase of $9 billion. The outlay, on the other hand, was $3,483 billion in 2010 compared to outlays in 2005 of $2,518 billion, an increase of $965 billion, thus resulting in an increase of 10,265 percent of outlays in relation to receipts. The budget deficit has increased to almost 1.46 trillion dollars in 2011 and accounts for almost 9 percent of the GDP (Figure 1). Figure 1 also indicates the outlays have increased from 2.78 percent of the GDP in 1954 to almost 26.29 percent of the GDP in 2010. The receipts have increased from 2.62 percent of the GDP in 1954 to almost 16.65 percent of the GDP in 2010. As shown in Table 1, outlays have outpaced receipts almost every five years except in 1955-60, 1975-80, 1990-95 and 1995-2000. In most years, tax rate or taxes have been reduced while expenditures have continued to increase every year. As a result, the federal budget has always been in the red. Figure 1 shows the flow of outlays and receipts of the U.S. Federal government. It indicates either the budget is barely balanced or in deficits. This trend has been in existence since 1945, and shows no end in sight. Under Keynesian philosophy, it is economically justifiable during recessionary periods to deficit spend, and automatic stabilization philosophy suggests that revenue should rise during boom periods. Deficits have persisted almost every year, however, regardless of whether the economy is in recession or in a boom. The discussion in Congress in 2011 was to reduce spending while allowing taxes to remain at the same level as approved by President Bush, even though the budget deficit is at its highest level in history ($1.45 trillion). The Republicans insist on lower taxes while knowing that during the weak economy, revenue will either remain the same or rise slight during the 2010 - 2011 recovery years. What is the justification for keeping taxes low without cutting spending to match receipts? Should taxes be raised to cover spending, or should spending be cut to match receipts? The current Congress obviously is insisting on cutting spending, leading some to claim that we are mortgaging the future of our children. They blame spending as the cause of large deficit; however, Manage and Marlow (1986) suggested that “it seems incorrect to focus on hypothesized symptoms of deficit rather than causes of deficits” (p. 620). Therefore, the continued failure to attack causes will have no or little effect on the deficit. Normally, spending in business should be constrained by income (receipts) and relative cost (price). The same should apply to governments. Government income (receipts) comes from taxes, direct and indirect. However, governments have the ability to overspend by borrowing. Thus, the governments’ receipts consist of taxes (direct and indirect) and debt. The urge to overspend is the result of relative costs. These costs are based on the need among politicians to get re-elected. Manage and Marlow (1986) suggested that when the cost of debt increases relative to the cost of re-election, the outcry over deficit becomes loud. The U.S. government has run deficits through most of its history without any real attempt to reduce or eliminate it. However, since the November 2010 and 2012 elections, when the Tea Party showed its power by electing members to Congress, the Republicans are now advocating for the control of spending and Democrats are calling for higher taxes on wealthy. This outcry is based on the assumption that reducing spending will reduce the deficit, thus helping them get re-elected. As Manage and Marlow (1986) stated, “The key, however, is the total funding level which must balance out aggregate spending” (p. 625). Therefore, neither higher taxes nor cutting spending will necessarily solve the deficit problem. Progressive tax is defined as the marginal rate of tax on the additional dollars and
48 which must be higher than the average rate on all income. Thus, “progressive” implies that if revenue increases faster than the total income, then elasticity will be greater than 1. Automatic stabilizers result in increased revenue due to increased income. This increase can be calculated by the average rate of responsiveness (ARR) (Johnson and Lambert, 1989, p. 3), which is the difference between the effective marginal rate (EMR) and the average tax rate (ATR), thus, ARR = EMR - ATR. A change in marginal tax rate can cause tax payers to change the way they collect their income. For example, they can change investment strategies, form of compensation, expenses and itemized deduction, and compliance (Feldstein, 1995). It is more likely for people in higher income groups to change the way they collect their income, because they have the wider discretion to bring about a change in their income (Feldstein, 1995). Further, a higher marginal tax rate could encourage taxpayers to take defensive legal or illegal steps to reduce their taxes. Feldstein (1995) found a substantial response of taxable income to changes in marginal tax rates. He calculated an elasticity of taxable income with respect to the marginal tax rate of at lease 1.0 and stated that it “could be substantially higher” (Feldstein, 1999, p. 570). Feldstein and Feenberg (1996) state that “high-income taxpayers would have reported 7.8 percent more taxable income in 1993 than they did if their tax rate had not increased” (p. 90). This conclusion was supported by Long (1999), who noted that “it is not surprising that upper-income tax rate increases legislated in 1993 raised far less revenue than would have been generated had there been no behavioral responses by taxpayers to the higher rates” (p. 686). Taxes and tax rates are controversial subjects and are constantly debated. As they represent deep, abiding, and contradictory attitudes in this country toward wealth (Kornhauser, 1993, p. 119). As Kornhauser (1993) has stated,” Today it [the problem] is an amalgam of consumption and income provisions, of progressive rates and tax expenditures which undermine progressivity” (p. 116). Progressivity is a contentious issue. Questions are always raised about appropriate tax rates, exemption amounts, and the taxability of certain types of income. Therefore, as Kornhauser (1993) has stated, “the debates show great partiality for preferences, and the tax laws reflect this” (p. 167). The current rates are based on taxing income as well as consumption, earned and unearned income, and progressivity.
HISTORY OF TAXES
Marginal tax rates and the associated income brackets have changed over the years. The IRS data indicate that in 1941, the minimum marginal tax for “married filing jointly” was 10 percent for income between $0 - $2,000, 59 percent for income between $44,000 and $50,000 and 81 percent for income over $5,000,000. The rates and income groups were changed almost every year starting in 1941. These changes continued until 2003. Since 2003, the tax rates remained the same until 2011, but income levels in each bracket were increased every year during those years. The tax rates and income brackets for 2011 are shown in Table 2. Countries need to tax their citizens to cover the cost of the services provided to them. However, some may resist paying their full share of taxes or may pay none at all. Cummings, Martinez-Vazquez, McKee, & Toggle (2004) found that “compliance behavior and tax morale can be explained by differences in the fairness of tax administration, in the
49 perceived equity of the fiscal exchange, and in the overall attitude towards the respective governments across the countries” (p. 2). “Tax morale” is defined as the intrinsic motivation to pay taxes (Alm and Togler, 2004, p. 5). There is considerable evidence that enforcement efforts can increase compliance. However, compliance has been shown by many studies to be related to the perceived fairness of tax administration, fiscal exchange, and the overall attitude towards the respective government (Cummings, Martinez-Vazquez, McKee, & Toggle, 2006, p. 2). Tax compliance is a human behavior affected by many factors. As usual, the threat of punishment and increased enforcement are likely to affect compliance. However, some theories (prospect theory and rank-dependent expected utility) suggest that social norms may encourage better compliance than the threat of punishment. But Cummings, Martinez- Vazquez, McKee, & Torgler (2009) found that the perceived fairness of the tax code and government behavior is a major factor in determining the level of compliance with tax laws. Taxpayers in the United States have higher compliance than European countries (Alm, Sanchez, and De Juan, 1995, p. 15). A tax law is framed to collect taxes from income (e.g., wages, dividends, interest, capital gains, etc.). Legislators pass tax laws, called statutory tax functions, as a basis for income earners to pay a certain amount of taxes. What taxpayers pay, however, is subject to adjusted gross income (effective tax rate), which may result in a different amount of tax than stipulated by the statutory rate. There are two reasons for the difference between the effective tax rate and the statutory rate: 1) Taxable income and economic income are never the same due to the prevailing tax laws. That is, income is reduced by many deductions or loopholes. 2) Tax law influences tax payers’ behavior by changing either the timing of income received or the form in which it is received (Gouveia and Strauss, 1994, p. 318). Therefore, several questions need to be answered: How close are the statutory and effective rates? What percent of the income in each income group is paid in taxes? What percent of GDP is received by each income group, and what percent of total taxes is paid by each income group? What difference, if any, is there between statutory and effective tax rates in each income group? Taxes are collected from different sources (Figure 2). The corporate share was 13.94 percent of the total taxes in 1980 and 12.91 percent in 2008. The corporate taxes went down, however, to as low as 8.77 percent in 2001. Individual share was 55.36 percent of the total taxes in 1980 and 51.95 percent of the total taxes in 2008. Employment-tax share was 24.71 percent of the total taxes, and it went up to 32.17 percent of the total taxes in 2008. The share of estate taxes was 1.21 percent in 1980 and .97 percent in 2008. The share of gift taxes was .04 percent in 1980 and .12 percent in 2008. The share of excise taxes was 4.74 percent in 1980 and 1.88 percent in 2008. Interestingly, excise taxes have gone down almost each year. Table 3 shows the share of different taxes as a percent of gross national income (GNI). The total taxes collected as a percent of GNI was 18.67 in 1980, was 19.25 percent in 2008 and was 18.14 percent in 2009. The share of corporate tax was 2.60 percent in 1980, was 2.48 percent in 2008 and was 7.74 percent in 2009. The share of individual tax as a percent of GNI was 10.34 percent, was 10.00 percent in 2008, and was 9.21 percent in 2009. The share of employment tax as a percent of GNI was 4.61, was 6.19 percent in 2008, and was 6.62 percent in 2009. The share of estate tax as a percent of GNI was .23 in 1980, was .19 percent in 2008, and was .17 in 2009. The share of excise tax as a percent of GNI was .88 percent in 1980, was .36 percent in 2008, and was .36 in 2009. As Table 3 shows, the share of taxes collected in each taxpayer group has gone down
50 except the employment tax which went up in 2008 and then decline in 2009. This obviously indicates that less taxes were collected as a share of GNI than in 1980. Table 4 shows a breakdown of income groups and taxes as a percent of taxable income. The first row in the table shows the total taxes for all income groups combined as a percent of total taxable income. Obviously, it shows that total taxes (row 1) as a percent of taxable income have been decreasing. It went down from 20.74 percent in 1993 to 19.18 percent in 2009. The table further indicates that taxes as a percent of taxable income have gone down of all income groups. Figure 3 graphically represents the data in Table 4. However, some income brackets were combined to create four income groups to reduce the clutter of lines. Obviously, taxes as a percent of taxable income have down for all income groups. The difference in percent from 1993 to 2009 indicates that even though taxes for almost every income group have declined, the income group between $25,000 and $100,000 has declined the least (4 percent) and the income group between $500,000 and $2,000,000 has declined the most (7 percent). Therefore, higher income earners are saving more in taxes than lower income earners. Figure 4 shows percent of taxes paid as per dollar of taxable income. Taxpayers in the income group between $500,000 and $1,000,000 and $1,000,000 and $1,500,000 pay more as a percent per dollar of taxable income than the other income groups. Taxpayers in income group of $100,000 and $200,000 pay more as a percent of taxable income than the other income groups. The percent of taxes per dollar for all income groups combined varies between 21 percent per dollar in 1993 to 17 percent per dollar in 2009. However, as the figure indicates, percent per dollar dropped in 2002 and rose again in 2007. IRS publishes data on the number of filers in each income group. The data for years 1990-2009 indicate that the number of filers since 1990 has decreased in lower income brackets ($500-$35,000). In 1990, there were 54 percent filers in this bracket, but it dropped to 20 percent of filers in 2009. The number of filers in the income bracket at or below $55,000 has decreased from 75 percent since 1990 to 52 percent in 2009. Obviously, this change is complemented by change in the number of filers above $55,000, which has increased from 25 percent to 48 percent. The change indicates a positive economic trend in that more tax payers are moving into the higher income group. In 1993, the number of returns for taxable income was 91 million, while the total number of people employed was 119 million, the total number of registered businesses was 2.13 million, and the total number potential tax filers was 121.3. But only 76.4 percent of potential taxpayers filed returns in 1993. In contrast, in 2009, the number people employed was 141.2 million, the number of registered business was 3.21 million, and the total number of potential tax filers was 144.4 million. In 2009, 104 million potential taxpayers filed returns, which is only 72 percent of the total number of potential taxpayers. In 1993, a total of $ 2.45 trillion of taxable income was reported when the GDP was $ 8.5 trillion or 28.25 percent of the GDP. In 2009, a total of $5.9 trillion of taxable income was reported when the GDP was $12.9 trillion or 39.51 percent of the GDP. The percent change in taxable income reported for each group in Table 6. Table 5 shows change in taxable income. The amount of reported taxable income between $1- $5,000 went up 110.81 percent from 1993 to 2000. During the same period, the amount of reported taxable income more than $1,000,000 went up 489.64 percent. From 1993 to 2009, the amount of reported taxable income between $1 and $5,000 went up 20.34 percent, and the amount of reported taxable income of more than $1,000,000 went up 415.13 percent. However, the amount of reported taxable income between $1 and $5,000
51 went up 18.36 percent from 2000 to 2009. During the same period, the amount of reported taxable income of more than $1,000,000 went up 84.78 percent. This trend indicates that more tax payers have moved into higher income bracket and that the bad economy in 2009 decreased the percent change from 2000 to 2009 compared to 1993 to 2000. Figure 5 shows reported taxable income as percent of GDP for the years 1993, 2000, and 2009. It shows that the share of income for the income group of more than $1,000,000 as percent of GDP has increased from 1.77 in 1993 to 6.58 percent in 2000 and to 5.60 percent in 2009. Obviously, taxpayers in higher income groups are getting a larger share of the GDP. Unfortunately, taxable income as a share of GDP in all income groups except in income group between $100,000 and $200,000 dropped in 2009 due to the country’s economic problems. As Figure 5 indicates, the taxable income as a percent of GDP for income groups below $50,000 has declined since 1993. But taxable income as a percent of GDP among the $50,000 and $75,000 income taxpayers increased (from 6.06 percent to 6.45 percent) from 1993 to 2000, but it has declined from 2000 to 2009 (from 6.45 percent to 5.60 percent). Further, taxable income as a percent of GDP increased in the income group $75,000 and $200,000 from 1993 to 2000 and from 2000 to 2009. However, taxable income as a percent of GDP for the income group $200,000 and higher has gone down from 2000 to 2009 (from 6.58 percent to 4.85 percent). Figure 7 shows the share of income for the years 1967 to 2009. Share of income in the highest range (top 20 percent, 5th fifth) has been increasing every year since 1967. For example, in 1967, the top 20 percent received 44 percent of the total income and they received 50 percent of the total income in 2009. It seems that their share has remained within 44 to 50 percent of the income. In addition, the share of income received by the top 5 percent has increased from 16 percent in 1967 to 21 percent in 2009. On the other hand, the share of income for the bottom fifth has gone down from 4 percent to 3.27 percent. The share of income received by 1st fifth, 2nd fifth, 3rd fifth, and 4th fifth has remained the same or has gone down. Figure 8 shows how much taxpayers pay in taxes for each dollar increase in income. The change was calculated by calculating a change in taxes paid to change in taxable income. Taxpayers paid more in taxes as they moved from lower income groups to higher income groups. For example, taxpayers with taxable income between $2,000,000 and $5,000,000 paid $.32 per dollar in 2000; subsequently, the rate dropped and then increased to $.299 per dollar of taxable income in 2009. The same trend can be seen for the taxable income between $1,000,000 and $2,000,000. However, the tax paid per dollar dropped for all other taxable incomes groups except for the group $200,000 to $1,500,000. As can be seen, the rate dropped but then went up in 2008 and 2009. As all the rates indicate, most of the taxes paid per dollar of taxable income have dropped since 1993 except for those in the income group of $200,000 to $1,500,000. Figure 9 how much tax was paid per taxpayer in each income group. Obviously, taxpayers in higher income groups ($1,000,000 to $5,000,000) paid higher taxes. That is, they paid $810 per taxpayer in1993 and paid $213 per taxpayer in 2009. Taxpayers in income group $500,000 to $1,000,000 paid $202 in 1993 and paid $71 in 2009. Taxpayers in income group $100,000 to $500,000 paid $49 in 1993 and paid $28 in 2009. Taxpayers in income group between $30,000 and $100,000 paid $7 in 1993 and paid $5 in 2009. Overall, all taxpayers paid an average tax of $6 in 1993 and paid $8 in 2009. The analyses of taxes presented in this section indicate that the rates of taxes paid and amount of taxes paid by each income group have been going down. However, higher
52 income groups have benefitted more that the other income groups. The trend may indicate regressivety. But to prove whether the tax system is progressive or regressive, one needs to analyze the data using many measures commonly used to identify progressivity.
IS THE U.S. TAX REALY PROGRESSIVE?
Many tax reforms have taken place in the United States over the years. The reforms have resulted in changes in the rate schedule, allowable deductions and exemptions, and exemption of income subject to tax. As a result, progressivity may have been reduced or eliminated (Hayes, Lambert, and Slottje, 1995). Hayes et al. (1995) found these reforms have generally reduced progressivity, and Piketty and Saez (2007) concluded that tax reform “seems to have the federal tax system in the direction of less progressivity” (p. 1). Major changes in the tax rate occurred in 1964, when the marginal rate of 91% for highest income ($400,000 or more) was reduced to 77%. It was then reduced to 70% for the highest income level of $200,000 or more. The marginal rate was changed again to 50% on income over $85,600 in 1983. In 1983, that rate (50%) was applied to income over $100,000. Other changes in income were made in 1984, 1985, and 1986. However, drastic changes in the marginal rates occurred in 1987 that finally settled the highest rates at 35% for income over $379,150. Concurrent with these changes, the tax rates were reduced and income levels were increased, but Social Security and Medicare contributions were increased. Have these tax reforms affected progressivity? Piketty and Saez (2007) have defined a progressive tax as “one in which the share of income paid in taxes rises with income, [and] a regressive tax is one in which the share of income paid in taxes falls with income” (p. 4). They modified the definition somewhat: “a tax system can be defined as progressive if after-tax income is more equally distributed than before-tax income, and regressive if after-tax income is less equally distributed than before-tax income” (p. 5). Duncan and Peter (2010) define progressivity as taxes that are designed to collect a greater proportion of income from the rich relative to the poor. Tax progressivity is often misunderstood. There are many suggested indexes for measuring progressivity, but there is no single commonly used method. That is, there are methods for measuring the distribution of tax burden, methods for measuring the effect of tax burden, and methods for measuring the effect of taxes on the distribution of income. Some people may not consider the latter as a measure of progressivity. Many researchers have studied the progressivity of taxes (Silber, 1994; Dunbar and Groff, 2000; Thorensen, 2004; Alm, Lee, and Wallace, 2005; Iyer, Schmidt, and Seetharaman, 2008; and Stroup, 2005) and using a variety of measures, e.g., Kakwani progressivity index, Kakwani distribution index, Standard Tax Rate (STR), Suits S Index, GINI Index, and Lorenz curve. Stroup (2005) has argued that progressivity is often measured by how much taxes are paid by certain income groups but that this measure does not relate tax share to income share. He has proposed that tax progressivity should be determined by calculating the share of income tax paid in relation to income share earned, i.e., marginal taxes. Table 6 shows the average tax paid by taxpayer in each income group. Taxpayers in the income group between $15,000 and $30,000 paid an average of $2 per taxpayer in 1993 and $1 in 2008, a drop of 50 percent. Taxpayers in the income group between $30,000 and $100,000 paid an average of $7 per taxpayer in 1993 and $5 per taxpayer in 2009, a drop of 29 percent. Taxpayers in income group between $100,000 and $500,000 paid an average of
53 $49 per taxpayer in 1993 and they paid an average of $28 per taxpayer in 2009, a drop of 43 percent. Taxpayers making more than a million dollars in 1993 paid an average of $810 per taxpayer in 1993 and $169 in 2009, a drop of 79 percent. Obviously, the higher their income, the more people saved in taxes. This indicates regressivety as the rich taxpayers have benefitted more from tax reduction and they pay less in taxes as a share oftheir income. If one uses the Piketty and Saez (2007) definition of progressivity, one may conclude that the U.S. tax system is regressive, because, as Table 6 shows, as income increased, taxpayers paid less in taxes. In 1993, the lowest income group was paying $2 in taxes per tax payer, and highest income group was paying $810 in taxes per tax payer. In 2009, the lowest income group paid $1 in taxes per tax payer and the highest income group paid $213 in taxes per tax payer. Obviously, the higher income taxpayers are paying less in taxes. Figure 10 shows the average marginal tax paid by each income group for the years 1993 to 2009. Additional dollars paid in taxes for each additional dollar of income is negative for income between $75,000 and $100,000. The same trend is true for all income above $75,000 except those in income groups between $100,000 and $200,000 and between $5,000,000 and $10,000,000. Filers in income groups between $5,000 and $75,000 paid more in taxes for each dollar of income. It indicates that some income groups are paying less in taxes for each dollar of income earned. That means they have benefited from tax reduction. This also indicates regressivety. The Gini-coefficient is most often used as measure of inequality. The coefficient varies between a zero and one. A one means complete inequality and zero means complete equality. According to Farris (2010),”the Gini index offers […] a single number that measures how equitably a resource is distributed in population. […] It allows us to illustrate how equity has changed in a given situation over time” (p. 851). Figure 11 shows the Gini index of reported adjusted gross income. Income distribution has become increasingly unequal. In 1993, the Gini index was .189, close to zero indicating more equality of reported gross income. However, in 2009, the Gini index was .3565, indicating a greater inequality of income. Figure 12 shows the Gini index of the reported average income tax (before credit) declared by tax filers for 1993 to 2009. In 1993, the index was .32 indicating inequality of taxes paid, it went up as high as .4532 in 1999 and then dropped to .38 in 2007. The indices indicate inequality of taxes paid. However, they show that inequality decreased after 2004, i.e., the average reported income tax before credit each year became more equal until 2007 but has edged back up since then. This indicates inequality in taxes. Other methods of measuring inequality have been proposed. According to Conceicao and Ferreira (2000), “a measure of economic inequality provides, ideally, a number summarizing the dispersion of the distribution of income among individuals. Such a measure is an indication of the level of inequality of a society” (p. 2). Hale (2001) suggested using skewness, dispersion, variance, and coefficient of variation to measure inequality, and these are common statistical measures used to describe the behavior of a distribution, including the spread, and the bigger the spread, the larger the inequality. In contrast, coefficient of variation describes the peakedness of distribution. Therefore, a smaller coefficient of variation indicates more equality. Figure 13 shows the mean, median, and standard deviation of average reported taxable income from 1993 to 2009. The standard deviation has increased each year, indicating a broader spread of reported taxable income. The average median reported taxable income
54 has dropped in 1999 and has continued its decline until 2003, but continued its rise after that. The median taxable income has dropped in 20000 and continued its decline in 2004 and then started income higher. This indicates higher taxes. Figure 14 shows mean, median, and standard deviation of average taxes paid before credit. The standard deviation has increased each year except it dropped in 1999 and then started its rise in 2003 indicating a broader spread of taxes paid. The median and the average indicate same pattern. All these indicate that share of taxes paid has dropped. These measures perform well if they are calculated using complete, individual point data (Hale, 2001). However, most available data is aggregate data, which is true of income tax data also. For this type of data, the Theil (T) method (Pedro and Pedro, 2000, p. 3) is more appropriate for measuring inequality. Like the Gini index, the Theil measure goes from zero, which means complete equality, to one, which means complete inequality. Also, the T can be calculated using group rather than individual data (Hale, 2001). Suits (S) measures progressivity by relating cumulative percent of tax payment to cumulative percent of income (Formby, Seeks, and Smith, 1982). Kakwani (1977) has also developed a method for measuring tax progression. Kakwani’s (K) index utilizes a measure of tax concentration. Both Suits (S) and Kakwani (K) measure progressivity. “Both measures are based on the difference between income and taxes, but Suits integrates this difference with respect to income and Kakwani with respect to return (R)” (Fromby, Seeks, and Smith, 1982, p. 1018). According to Fromby, Seeks, and Smith (1982) “the Suits and Kakwani indices, although identical in intent, are fundamentally different measures of tax progression” (p 1019). According to Formby, Seeks, and Smith (1982), “the only difference is […] simply the slope of the Lorenz curve” (p 3). Thus, it can result in “different estimates of the degree of progressivity,” and there is no reason to believe that one gives better results than the other (p. 3). Table 7 summarizes all the common measures of inequality for readers to compare the outcomes. All the measures indicate inequality of taxes. All of them have shown a constant increase since 1993. Interestingly, all of them, except Gini, show a drop during economic recession (1997 through 1999), but Gini dropped in 2001 and 2002. All measures dropped again in 2006 and 2007, again except Gini, which dropped in 2008 and 2009. Regardless, all the measures indicate increasing inequality of taxes. Another way to measure inequality is to compare changes in share of income and taxes paid over time (Table 8). In 1993, 77 percent of the tax filers reported taxable income under $50,000, received 38.04 percent of the total taxable income and paid 28.81 percent of all the taxes; 18.10 percent of the filers reported income between $50,000 and $100,000, received 31.78 percent of the total taxable income and paid in taxes 28.74 of the total taxes; 4.44 percent of the filers reported income between $100,000 and $1,000,000, received 24.04 percent of the total taxable income and paid 31.88 of the total taxes. Obviously, more filers have moved into higher income brackets i.e. income between $50,000 and $100,000 (18.10 (1993) vs. 28.78 (2009) percent), income between $100,000 and 1,000,000 ((4.44 in 1993) vs. (16.48 in 2009)) and income of more than a $1,000,000 (.07 in 1993 vs. .23 percent in 2009). The share of taxes paid by these income groups has also changed. Comparing 1993 with 2009, filers in income group between $50,000 and $100,000 received 4.45 percent less in taxable income and paid 8.10 percent less in taxes. Filers in income group between $100,000 and $1,000,000 received 21.24 percent more in taxable income and paid 18.95 percent more in taxes. Filers in income group of more than a million received 6.17 more in taxable income and paid 8.46 more in taxes. Comparing 2002 with 20009 (in 2001, the Bush tax took effect), filers in income group between $50,000 and
55 $100,000 received 3.87 percent less in taxable income and paid 4.42 percent less in taxes. Filers in income group between $100,000 and $1,000,000 received 8.94 percent more in taxable income and paid 6.72 percent more in taxes. Filers in income group of more than $1,000,000 received 1.97 percent more in taxable income and paid 2.37 percent more in taxes. Comparing change in taxes from 1993 to 1996, income group of $ 0 - $ 50000 paid .88 less in taxes for one percent drop in their share of income. Income group $ 50000 - $ 100000 paid 7.54 percent less in tax for every one percent drop in their share of income. Income group $100000 - $1000000 paid .91 percent more in taxes for one percent increase in their share of income. Income group over $1,000,000 paid 1.38 more in taxes for one percent increase in their share of income. Comparing change in taxes from 2004 to 2009, income group of $ 0 - $ 50000 paid .84 less in taxes for one percent drop in their share of income. Income group $ 50000 - $ 100000 paid 1.46 percent less in tax for every one percent drop in their share of income. Income group $100000 - $1000000 paid .95 percent more in taxes for one percent increase in their share of income. Income group over $1,000,000 paid .81 less in taxes for one percent drop in their share of income. These results may indicate that taxpayers in higher income groups are paying more in taxes. This could be explained that in those years when the higher income groups paid more in taxes is they received a larger share of income, thus putting them in much higher tax rate. However, as Figure 10 indicates, the average marginal tax paid per year has dropped since 2003, and Figure 11 indicates that the average marginal tax by higher income groups above $75,000 has been either negative or smaller compared to taxpayers with income below $75,000.
CONCLUSION
Income tax is a major share of budgeted revenues (Federal or states) that covers many vital services of the society. Obviously, the cost of these services does go up and needs to be covered. To cover the cost, governments must either increase the sources of taxes or increase the tax rate. Failing to generate enough revenue means either cutting the services or running a deficit. In the U.S., a budget deficit has occurred for more than fifty years, and there is no end in sight. Many politicians are insisting on reducing expenses as well as insisting on reducing tax rates. Unfortunately, the emphasis is on cutting taxes regardless of whether expenses can be reduced. Most of those who advocate cutting taxes argue that the tax rates are unfair to the rich and, if taxes are reduced, the economy will improve. This paper has not dealt with the issue of whether reducing taxes will spur economic growth. However, this paper has dealt with the issue of tax fairness as defined in terms of specific statistical models. The analyses presented here using four progressivity measures of inequality—i.e. Suits, Kakwani, Gini, and Theil—indicate that taxes are unequal. Other measures such as average marginal tax rate, coefficient of variation, dispersion, median, mean, and other measures discussed in the paper all indicate inequality. Therefore, the results presented here do not support the position of those who say that the rich are paying a higher share than everyone else and that their tax burden should be reduced. However, the claim that spurring economic growth requires cutting taxes for the rich increases economic growth still needs to be proven. That is beyond the scope of this paper but may be pursued in further research.
56 REFERENCES
Alm, J., Lee, F., & Wallace, S. (2005). How fair? Changes in federal income taxation and the distribution of income 1997-1998. Journal of Policy Analysis and Management, 24, pp. 5-22. Alm, J., Sanchez, I. & De Juan, A. (1995). Economic and noneconomic factors in tax compliance. KYKLE, 48,p. 218. Alm, J., & Torgler, B. (2006). Cultural difference and tax morale in the United States and in Europe. Journal of Economic Psychology, 27, 2, pp. 224-246. Blackburn, J. O. (1967). Implicit tax rate reduction with growth, progressive taxes, constant progressivity, and a fixed public share. America Economic Review, 57, 1, pp. 162- 169. Conceicao, P., & Ferreira, P. (2000). The young person’s guide to the Theil Index: Suggesting intuitive interpretation and exploring analytical application. Working Paper 14, University of Texas, Austin, Texas. Cummings, R., Martinez-Vazquez, G. J., McKee, M., & Torgler, B. (2004). Fiscal exchange and social capital: Effects of culture on tax compliance, Unpublished manuscript. Retrieved www.CREM-Research.ch. Cummings, R., Martinez-Vazquez, G. J., McKee, M., & Torgler, B. (2006). Effects of tax morale on tax compliance: Experimental and survey evidence. Berkley Program in Law and Economics Working Paper Series. Retrieved from http://escholarship.org/ uc/item/8sch2wafp Cummings, R., Martinez-Vazquez, G. J., McKee, M., & Torgler, B. (2009). Tax morale affects tax compliance: Evidence from surveys and an artefactual field experiment. Journal of Economic Behavior and Organization, 70, 3, pp. 447-457. Dunbar, A., & Groff, J.E. (2000). Determination of income mobility using tax return data. Public Finance Review, 28, pp. 511-539. Duncan, D., & Peter, K. S. (2010). Tax progressivity and income inequality. Working paper 2008-6-5, June 2008, Department of Economic, Georgia State University, Atlanta, GA. Easterly, W., & Rebelo, S. (1993). Fiscal policy and economic growth: An empirical investigation. Working Paper No. 4497, National Bureau of Economic Research, Cambridge, MA. Farris, F. A. (2010). The Gini Index and measure of inequality. American Mathematical Monthly, 117, 10, pp. 851-864. Feldstein, M. (1995). The effect of marginal tax rates on taxable income: A panel study of the 1986 Tax Reform Act. The Journal of Political Economy, 103, 3, pp. 551-572. Feldstein, M., & Feenberg, D. (1996). The effect of increased tax rates on taxable income and economic efficiency: A preliminary analysis of the 1993 tax rate increases. Tax Policy and the Economy, 10, p. 89-117. Formby, J. P., Seaks, T. G., & Smith, W. J. (1982). Measuring the net progressivity of government expenditures and taxes. Working Paper E820401, University of North Carolina at Greensboro, Center for Applied Research. Fullerton, D. (1982). On the possibility of an inverse relationship between tax rates and government revenues. Journal of Public Economics, 19, 1, pp. 3-22. Gouveia, M., & Strauss, R. P. (1994). Effective federal individual tax function: An exploratory empirical analysis. National Tax Journal, 47, 2, pp. 317-319.
57 Goolsbee, A. (2000). What happens when you tax the rich? Evidence from executive compensation. The Journal of Political Economy, 108, 2, pp. 352-378. Hale, T. (2003). The theoretical basics of popular inequality measures: Online computation of examples. Working Paper, University of Texas Inequality Project. Hayes, K., Lambert, P. J., & Slottje, D. (1995). Evaluating effective income tax progression. Journal of public Economics, 56, 461-474. Internal Revenue Services (IRS), U. S. Federal Government, Washington, D.C. Iyer, S. G., Schmidt, A., & Seetharaman, A. (2008). The effects of standardized tax rates, average tax rates, and the distribution of income on tax progressivity. Journal of Accounting and Public Policy, 27, pp. 88-96. Johnson, P., & Lambert, P. (1989). Measuring the responsiveness of income tax revenue to income growth: A review and some UK values. Fiscal Studies, 10, 4, pp. 1-18. Kakwani, N. C. (1977). Measurement of tax progressivity: An international comparison. Economic Journal, 87, pp. 71-80. Kornhauser, M. E. (1993). The morality of money: American attitude toward wealth and the income tax. Indiana Law Journal, 70, 1, pp. 119-69. Long, J. E. (1999). The impact of marginal tax rates on taxable income: Evidence from state income tax differentials. Southern Economic Journal, 65, 4, p. 855-869. Manage, N., & Marlow, M. L. (1986). The causal relation between federal expenditure and receipts. Southern Economic Journal, 52, 3, pp. 617-629. Myles, G. D. (2000). Taxation and economic growth. Fiscal Studies, 2, 1, pp. 141-168. Pedro C., & Pedro F. (2000). The young person’s guide to the Theil Index: Suggesting intuitive interpretations and exploring analytical applications. Working Paper Number 14, The University of Texas at Austin. Piketty, T., & Saez, E. (2007). How progressive is the U.S. federal tax system? A historical and international perspective. Journal of Economic Perspective, 21, 1, pp. 3-24. Saez, E. (2003). The effect of marginal tax rates on income: A panel study of ‘bracket creep’. Journal of Public Economics, 87, 5-6, pp. 1231-1255. Silber, J. (1994). Income distribution, tax structure and the measurement of tax progressivity. Public Finance Quarterly, 22, 1, pp. 86-102. Smith, A. (1776). The wealth of nations. London: J. M. Dent & Sons. Stroup, M. D. (2005). An index for measuring tax progressivity. Economic Letters, 86, pp. 205-213. Sutherland, A. (1997). Fiscal crises and aggregate demand: Can high public debt reverse the effects of fiscal policy? Journal of Public Economics, 65, pp. 147-162. Tanzi, V. (1969). Measuring the sensitivity of the federal income tax from cross-section data: A new approach. The Review of Economics and Statistics, 5, 2, pp. 206-209. Taylor, J. B. (2000). Reassessing discretionary fiscal policy. Journal of Economic Perspectives, 14, 3, pp. 21-36. U. S. Dept. of the Treasury. (2003, Aug.). Fact sheet on the history of the U.S. tax system. (Retrieved May, 20, 2011) http://www.policyalmanac.org/economic/archive/tax_ history.shtml “Tax Revenue as a Fraction of GDP” Department of Numbers, http://www.deptofnumbers. com/blog/2010/08/tax-revenue-as-a-fraction-of-gdp, (accessed December 22, 2011) “Income” U.S. Census Bureau, Washington, D.C. http://www.census.gov/hhes/www/ income/income.html (Accessed December 22, 2011)] Source: http://www.deptofnumbers.com/blog/2010/08/tax-revenue-as-a-fraction-of-gdp/
58 FIGURE 1: USA FEDERAL GOVERNMENT RECEIPTS, OUTLAYS, AND DIFFERENCE AS PERCENT OF GDP
USA Budget and Difference as Percent of GDP
30.00 25.00 20.00 15.00 Receipt 10.00 Outlays 5.00 Difference 0.00
Percent of GDP -5.00 -10.001954 1958 1962 1966 1970 1974 1978 1982 1986 1990 1994 1998 2002 2006 2010 -15.00 Year
TABLE 2: TAX TABLE FOR 2011 Percent tax Rate Income bracket more than Income bracket less than $ 10 0 17,000 15 17,000 69,000 25 69,000 139,350 28 139,350 212,300 33 212,300 375,150 35 375,150
59 FIGURE 2: SOURCES OF TAXES
Percent of tax collected from difference sources
60.00
50.00
40.00
Corp Individual 30.00 Emplyment Estate Gift Excise
20.00
10.00
0.00
Year
TABLE 3: SOURCES OF TAXES AS A PERCENT OF TOTAL TAXES AND GNI % of Total Taxes % of GNI Source 1980 2009 1980 2008 2009 Corporate 13.94 9.61 2.60 2.48 7.74 Individuals 55.34 50.12 10.34 10.00 9.21 Employment 24.71 36.59 4.61 6.91 6.64 Estate 1.21 .92 .23 .19 .17 Gift .04 .13 - .02 .02 Excise 4.74 1.99 .88 .36 .36 Taxes as share of 18.66 19.25 18.14 GNI
FIGURE 3: TAXES AS A PERCENT OF TAXABLE INCOME
Share of Taxes as a Percent of Taxable Income
40.00 $1 under $25,000 $25,000 under $100,000 $100,000 under $500,000 $500,000 under $2,000,000
35.00
30.00
25.00
20.00
15.00
10.00 Taxes as a Percent of Taxable Incom
5.00
0.00 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year
60 FIGURE 4: TAXES PAID PER DOLLAR OF TAXABLE INCOME
Percent of Taxes paid Per Dollar of Taxable Income
40.00 $1 under $5,000 $15,000 under $20,000 $50,000 under $75,000 $100,000 under $200,000 $500,000 under $1,000,000 $1,000,000 under $1,500,000 $1,500,000 under $2,000,000 T axable returns , total
35.00
30.00
25.00
20.00
15.00
10.00 Percent of Taxes paid Per Dollar of Taxable Income Taxable of Dollar Per paid Taxes of Percent
5.00
0.00 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year
TABLE 5: PERCENT CHANGE IN INCOME GROUP REPORTING TAXABLE INCOME Taxable Income Percent Percent Percent Size of adjusted gross income change change change 1993 - 2000 1993 - 2009 2000-2009 All returns, total 185.21 207.39 111.97 No adjusted gross income 0.00 0.00 0.00 $1 under $5,000 110.81 20.34 18.36 $5,000 under $10,000 84.43 24.89 29.48 $10,000 under $15,000 77.29 41.42 53.59 $15,000 under $20,000 89.40 55.83 62.45 $20,000 under $25,000 94.78 63.47 66.96 $25,000 under $30,000 102.51 77.85 75.94 $30,000 under $40,000 106.10 91.39 86.13 $40,000 under $50,000 113.11 102.02 90.20 $50,000 under $75,000 140.20 139.46 99.47 $75,000 under $100,000 204.62 254.81 124.53 $100,000 under $200,000 262.18 416.81 158.98 $200,000 under $500,000 269.88 125.48 46.49 $500,000 under $1,000,000 288.67 338.34 117.21 $1,000,000 or more 489.64 415.13 84.78
61 FIGURE 5: REPORTED TAXABLE INCOME AS PERCENT OF GDP – 1993, 2000, AND 2009
Reported Taxable Income as Percent of GDP by Income Group
12.000
10.000
8.000
1993 GDP 6.000 2000 GDP 2009 GDP 4.000
Taxable Income as Percent of GDP 2.000
0.000
$1 under $5,000 $1,000,000 or more $5,000 under $10,000 $10,000 under$15,000 $15,000 under$20,000 $20,000 under$25,000 $25,000 under$30,000 $30,000 under$40,000 $40,000 under$50,000 $50,000 under $75,000 $75,000 under $100,000 $100,000 under$200,000 $200,000 under $500,000 $500,000 under $1,000,000 Income Group
FIGURE 6: TAXES PAID PER TAXPAYER IN EACH INCOME GROUP
Taxes per Capita in each Income Group
1000.00
900.00
800.00
700.00
600.00 $1 under $15,000 $15,000 under $30,000 $30,000 under $100,000 500.00 $100,000 under $500,000 $500,000 under $1,000,000 $1,000,000 under 5000,000 400.00 T axable returns , total Taxes Paid Per Capita Per Paid Taxes 300.00
200.00
100.00
0.00
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year
62 FIGURE 7: DISTRIBUTION OF INCOME
Share of Income Received by each 5th and the Top 5 Precent 60.00
50.00
40.00
1st F ifth
30.00 2nd F ifth
3rd F ifth
Percent of Income 4th F ifth
5th F ifth 20.00 T o p 5 percent
10.00
0.00 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Year
Source: http://www.census.gov/hhes/www/income/income.html
FIGURE 8: CHANGE IN TAXES PAID PER DOLLAR OF CHANGE IN INCOME
Change in Taxes due to Change in Taxable Income in dollars
0.400
0.350
0.300
0.250 $5,000 under $25,000 $30,000 under $50,000 $50,000 under $200,000 0.200 $200,000 under $1,500,000 $1,500,000 under $2,000,000 $2,000,000 under $5,000,000 0.150
0.100
0.050 Change in Taxes and Change in Taxable Income in Dollars in Income Taxable in Change and Taxes in Change
0.000 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year
63 FIGURE 9: TAXES PAID PER TAXPAYER IN EACH INCOME GROUP
Taxes per Capita in each Income Group
1000.00
900.00
800.00
700.00
600.00 $1 under $15,000 $15,000 under $30,000 $30,000 under $100,000 500.00 $100,000 under $500,000 $500,000 under $1,000,000 $1,000,000 under 5000,000 400.00 T axable returns , total Taxes Paid Per Capita Per Paid Taxes 300.00
200.00
100.00
0.00
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year
FIGURE 10: AVERAGE MARGINAL TAX PAID FOR THE YEARS 1993-2009
Average Marginal Tax Paid by Income Group
2.0000
1.5000
1.0000
0.5000
0.0000 Average Marginal Tax Paid
-0.5000
-1.0000 Income group
FIGURE 11: GINI COEFFICIENT OF REPORTED ADJUSTED INCOME
Gini Coefficient for Adjusted Gross Income Reported 0.4000
0.3500
0.3000
0.2500
0.2000
0.1500 Gini Goefficient 0.1000
0.0500
0.0000 1993 1884 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year
64 FIGURE 12: GINI COEFFICIENT OF REPORTED INCOME TAX BEFORE CREDIT
Gini Coefficient of Income Tax Before Credit
0.5000
0.4500
0.4000
0.3500
0.3000
0.2500
Gini Coeffient 0.2000
0.1500
0.1000
0.0500
0.0000 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year
FIGURE 13: DESCRIPTIVE STATISTICS OF AVERAGE TAXABLE INCOME
Descriptive Statistics of Taxable Income
400,000,000
350,000,000
300,000,000
250,000,000
Mean 200,000,000 Median Standard Deviation
150,000,000
Mean, Standard Median, Error, Standard Deviation 100,000,000
50,000,000
0 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year
65 FIGURE 14: DESCRIPTIVE STATISTICS OF INCOME TAX PAID BEFORE CREDIT
Descriptive Statistics of Income Tax Paid
80,000,000
70,000,000
60,000,000
50,000,000 Mean Median Standard Deviation 40,000,000
30,000,000 Mean, Median, Standard Error Standard Median, Mean,
20,000,000
10,000,000
0 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year
66 TABLE 7: SUITES (S), GINI (G), KAKAWANI (K), and THEIL (T) MEASURES Year S G K T 1993 0.157 0.560 0.096 0.1651 1994 0.157 0.565 0.000 0.1647 1995 0.161 0.574 0.097 0.1622 1996 0.164 0.586 0.099 0.1587 1997 0.156 0.599 0.096 0.1565 1998 0.149 0.603 0.094 0.1606 1999 0.151 0.611 0.095 0.1591 2000 0.151 0.620 0.097 0.1556 2001 0.157 0.590 0.100 0.1687 2002 0.184 0.579 0.121 0.1727 2003 0.166 0.586 0.109 0.1771 2004 0.159 0.603 0.106 0.1757 2005 0.155 0.623 0.104 0.1693 2006 0.152 0.628 0.103 0.1688 2007 0.149 0.635 0.102 0.1676 2008 0.165 0.609 0.110 0.1822 2009 0.177 0.587 0.115 0.2004
67 68 DETERMINANTS OF THE SOARING GROWTH OF U.S. NON-MARITAL BIRTHS Frederic L. Pryor, Swarthmore College
ABSTRACT
Between 1970 and 2006 the number of non-marital births in the U.S. almost quintupled and the percentage of non-marital to total births almost quadrupled rose from 10.7 percent to 38.5 percent. In early 1973 when the Supreme Court announced the Roe v Wade decision, few would have expected such a development.1 Discussions in both the popular press and technical literature have pointed to several possible factors: the weakening of the stigma attached to children born out of wedlock, the fall in the proportion of “shotgun marriages,” the rising divorce rate, the diminishing ratio of suitable men to marriageable women in certain minority groups, and the growing financial independence of women as more entered the work force. As noted below, non-marital births have been increasing in most industrialized nations, so similar discussions on the issue can be found in many languages. This analysis focuses on the issue of whether the increase in non-marital births in the U.S. is due to a rising number of women entering cohabitating unions and then bearing children or, alternatively, to the increased fertility of unmarried women. For this purpose, women fifteen and over are separated into three groups: married, single and cohabiting, and single, non-cohabiting. Then it is shown that it is not a rising fertility rates of any of these groups that primarily accounts for the rising percentage of non-marital births (in fact, quite the reverse), but rather the rising incidence of non-marital cohabitation. Factors adduced by others, such as a changing ethnic/racial composition of the population, play only a very minor role in the increasing rate of non-marital births. The analysis is straightforward. After outlining the basic trends and discussing the possible underlying causes of the changes, the discussion turns to two basic questions: Why are now so many couples cohabiting? And why has the birth rate of cohabiting women declined? JEL Classification: J1
THE BASIC TRENDS
Table 1 shows that the rate of live babies born per 1000 women between fifteen and forty-four fell between 1950 and 1980 and has remained relatively constant since then. Among African-Americans, it has slightly fallen. The share of non-marital births has been rising, but at a decelerating rate until 2000 when it accelerated again.2 Between 1970 and 1980, the increase of this ratio was 72 percent; between 1980 and 1990, 52 percent; and between 1990 and 2000, 19 percent. Non-marital birth ratios for African-American as well as Hispanic and Latina women were considerably higher than the rate for white women
69 (of all ethnicities), but these gaps were slowly narrowing. Surprisingly, the percentage of non-marital births to women under twenty in all groups has declined dramatically, especially before 1990, while the non-marital birth rates of women twenty-five and over have considerably increased, especially before 1990 as well. This suggest an important ideological shift that deserves greater attention. The legalization of abortions might be expected to reduce the rate of non-marital births. Table 2 shows that the rate of abortions has unexpectedly fallen since 1980 for all ages, races, and ethnic groups, even though the rate of non-marital births has risen. Of importance to this analysis, the abortion rate has particularly declined among unmarried women. Such data might mean that most non-marital babies were wanted, a conclusion supported in part by other findings on the relatively low share of unwanted pregnancies of this group.3 In turn, this suggests that measures to reduce non-marital births may not prove to be very effective. The relative constancy of the birth rate between 1900 and 2006, combined with the falling abortion rate, could have occurred as a result of more effective contraception.
POSSIBLE CAUSAL FACTORS
As shown in Table 1, some minority groups have a higher rate of non-marital babies than the white population and some have suggested that the changing racial composition of the U.S. population has had an impact on the overall share of non-marital births. If the racial composition had not changed between 1980 and 2000, the overall rate of non-marital births would have been only 6 percent less. Clearly the changing racial composition does little to explain the rising share of non-marital births. We learn much more by looking at Sweden or Iceland, which recently have had non- marital birth ratios of over 50 percent (Kiernan 2001, 2004).4 In these countries, most of these couples accounting for the non-marital births have been cohabiting for a relatively long term. As argued below, ideas about marriage and non-marital births in the U.S. are moving in the same direction. Although exploring non-marital births by examining trends in cohabitation in the U.S. is crucial, some serious data problems arise. When the Census Bureau began to estimate cohabitation, they labeled this statistic “POSSLQ” (persons of opposite sex sharing living quarters), measuring it inferentially from information on household composition: any household containing just two unrelated adults over fifteen years and of the opposite sex were classified as POSSLQ. This definition, however, missed those couples in group living arrangements, as well as cohabiting couples with children over fifteen; and, in addition, may have erroneously included simple roommates, roomers, and live-in servants. Casper and Cohen (2000) developed an improved measure (adjusted POSSLQ) to take account of some of these failings, and by the mid-1990s this measure was about 17 percent higher than the Census Bureau’s POSSLQ, The most accurate estimates of cohabitation have recently been derived from the Integrated Public Use Microdata Series (IPUMS) from various censuses by Fitch, Goeken, and Ruggles (2005). Since these estimates cover only census years, it is necessary to interpolate or extrapolate (beyond 2000) for other years, using the “adjusted POSSLQ” estimates for this purpose. In the mid-1990s the Census Bureau took another approach and began to directly ask unmarried householders if they were partners, but this measure appears to have some serious flaws.5 For 2002 we have five estimates of U.S. cohabiting couples (in millions): 4,898
70 POSSLQ), 6,731 (adj. POSSLQ), 4,193 (“unmarried partners”), 5,578 (National Study of Family Growth), and my estimate of 4,720 derived from Fitch et al. (2005). Although the Fitch estimates are the second lowest, they also show an increase between 1980 and 2006 of 310 percent, in contrast to 238 percent and 202 percent respectively for the first two. Table 3 allows us to determine what part of the rise in non-marital births has been due to fertility changes and what part to changes in the relative number of married women, cohabiting women, and single, non-cohabiting women. Part A of the table shows a declining share of married women and a rising share of cohabiting women among the female population, although the share of the latter was still small. Part B shows the percentage of births attributable to the three groups of women. The share of births to married women declined, as we would expect, while those to cohabiting women dramatically increased and was responsible for the major share of increase in non-marital births. Part C of Table 3 shows that the birth rate of cohabiting women showed an uneven pattern but, nevertheless, declined in the two decades since 1980, while that of single, non-cohabiting women reveals an uneven increase. The major conclusion to be drawn from Table 3 is that the rising share of cohabiting women is primarily responsible for the explosion in non-marital births, not their rising fertility. Births to single, non-cohabiting women have also increased considerably. Also noteworthy is that the birth rate per 1,000 women is much higher among women in cohabiting than married unions. Partly this is because cohabiting couples are much younger on average than married couples. Furthermore, their coital frequency is much higher, other factors such as age held constant (Rao and Demaris 1995; Yabiku and Gager 2009). But such information tells us little about trends in the key variables under examination. Answers to two crucial questions examined below are necessary to understand the rising number and ratio of non-marital births: Why has cohabitation increased? And why have the birth rates of married and cohabiting women declined?
WHY ARE MORE COUPLES COHABITTING?
The literature on factors underlying a couple’s choice to cohabit rather than marry is large,6 but it usually focuses on these decisions at a single point in time. This brief discussion explore those factors that might allow us to make predictions about future trends.
Cultural Influences A common explanation of the rising rate of cohabitation is cultural change: it is more acceptable now for couples to live together outside of marriage. For instance, from 1980- 81 to 1997-98, the percentage of U.S. women who agreed with the statement that it is “all right” for an unmarried couple to live together as long as they plan to eventually marry rose from 33 to 59 percent; for men, the percentage rose from 47 to 67 (Thornton, and Young-DeMarco). Such an attitude may underlie explanations cohabiting couples give for their living arrangement, presenting their cohabitation as a screening device for eventual marriage. Since “eventual marriage” is vague, this excuse might also cover couples who are cohabiting simply for convenience. In his very useful history of U.S. attitudes toward marriage, Cherlin (2009: 137) takes a more radical view, noting that a critical puzzle is not “why there is so little
71 marriage in the U.S., but why there is so much of it...[W]hy does anyone bother to marry anymore?” Cohabitation in the U..S. is, however, somewhat different than in other countries. Comparing nine OECD nations, we see that the U.S. had a higher percentage of cohabiting unions that dissolved within five years than in any of the others countries; but it also had the highest percentage of surviving unions that survived and converted to marriage within five years (Kiernan 2004, Table 1). It might appear that cohabitation may have become more common because it is increasingly viewed as “preparation for marriage,” rather than as an alternative life style. Strong evidence, however, also suggests that social class and rising class divisions also play a crucial causal role. Murray (2012) defines “social class” in terms of the education of the mother and, if she or the household head are employed, by the prestige rating of the occupation. He shows that in the forty years since 1970, the rate of non-marital births of white women falls as the social class of the mother rises; and, moreover, the increase of the non-marital birth rate follows the same pattern. Regarding cohabitation, tor the entire population between 15 and 44 Goodwin et al., (2010) show that in 2002 as the educational level of the mother rose, the share of women who were married increased and the share women who were cohabiting fell, a trend also apparent in previous decades as well (Bumpass and Lu 2000). It has been hypothesized that for women, low income men with poor education do not make potentially good marriage partners. But education and income have other effects as well. One factor is the rising share of families in which the woman has more education and more income than the man (Fry and Cohn 2010). In the mid-1990s women became the majority of college graduates and by 2007, 28 percent of women had more education than their husbands, in contrast to the 19 percent of husbands who had more education than their wives. In the same year 22 percent of wives earned more than their husbands. Such trends mean that in recent years marriage offers relatively fewer economic advantages to woman than formerly. Moreover, despite the easing of divorce, if the relationship sours, a cohabiting union remains easier to escape from than a marriage. Both the declining economic advantage of marriage to women and the relative ease of dissolution encourage cohabitation.7 Some have suggested that other cultural factors have also played a causal role. For instance, simple cross-section regressions of data from U.S. states show that church adherence (depending on the denomination) may partly explain different cohabitation rates (Wydick 2007). In contrast, panel regressions to explain cohabitation show no influence of religion (ibid.). Since attitudes toward religion change very slowly, it does not seem likely that religion, if it has any impact, will have much influence on cohabitation rates in the future. Micro-cultural factors also influence divorce. For instance, according to some international studies, a woman who experienced her parent’s divorce as a child is more likely to cohabit and also to bear children at an earlier age (Haveman, et al. 2001; Kiernan 2001, 2004). Although divorces per 1000 marriages in the U.S. rose until 1980, the rate has declined since then. If divorce is indeed a factor in the rising rate of cohabitation in past years in the U.S., it should play an ever smaller role in the future. On the other hand, children of cohabiting couples are themselves more likely to cohabit (Hetherington and Elmore, 2004) and the currently rising rate of cohabitation should lead to an even higher cohabitation rate in the future, other things equal. Other micro-cultural influences include the rising age of marriage for both men and women. For men in 1980, the average age was 24.7 and by 2006, it was 27.5; for women
72 these averages were respectively 22.0 and 25.5 (U.S. Bureau of the Census 2010b). Since younger people have less knowledge of their potential marriage partner and of marriage itself and, at the same time, have a stronger sex drive, cohabitation seems to provide a solution to both problems. Finally, the delay of marriage might have influenced some cohabiting couples wishing children to have their first child before wedlock. Other studies of cohabitation rates (summarized Lillard, Brian and Waite 2003) reveal still other cultural influences. A serious problem of most of these studies is that cohabitation, marriage, and divorce are endogenous to each other and it is difficult to separate the various causal strands (ibid.; Walters and Ressler 1999). And from the discussion above, we see that macro- and micro-cultural forces are sometimes operating in opposite directions, so that firm conclusions about their combined influences are often difficult to draw.
Exogenous Influences
Three exogenous factors deserve brief mention: the legalization of abortion in 1973, the introduction of the birth-control pills in the late1950s, and easing of divorce laws. Legalization of abortion. As shown in Table 2, the abortion rate has declined since 1980 at the same time as non-marital births have soared. No believable causal connection is apparent, so the impact of legalized abortions on future rates of non-marital births should not be significant.8 The Pill. The diffusion of oral contraceptive pills around 1960 made both cohabitation and marriage more attractive by allowing closer planning (and prevention) of births. But the absence or the planned lack of children also makes dissolution of both types of living arrangements easier as well. Table 2 shows that unwanted pregnancies (as measured by the abortion rate) are more common among unmarried women than married women and somewhat greater among single, non-cohabiting women than cohabiting women, so the birth control pill probably led to more non-marital sex and cohabitation. Akerlof and his colleagues (1996) raise the interesting argument that the Pill and the easing of abortion might in some cases actually increase non-marital births since the increased availability of abortion “decreases the incentives to obtain a promise of marriage if premarital sexual activity results in pregnancy .” (ibid.: 280). Since a woman who does not want an abortion is placed at a competitive disadvantage with other women in snaring a husband, any accidental and unwanted pregnancy is more likely to be carried to term. Moreover, they also argue that that the pill (and the easier availability of abortion) reduces the number of “shotgun marriages,” a controversial claim that is supported by data from Ventura and Bachrach (2000) showing that the share of marriages involving pregnant women declined considerably after 1960. Both these lines of argument suggest that the Pill leads to an increase in non-marital births and cohabitation. Nevertheless, since the Pill has been around long enough to be taken for granted, it should not play a significant role in any change in the rate of cohabitation in the years to come. Divorce. Wydick (2007) argues that liberalization of divorce law in the last half of the twentieth century makes marriage commitment more difficult and the consequences of marital failure less harsh. That is, couples enter into marriage more casually and have less incentive to make the marriage last. Although he argues that easier divorces would have a favorable impact on cohabitation, cross-section empirical studies of both U.S. states and counties show few significant relations between these variables. Moreover, his own regressions over time to explain changes in cohabitation in U.S. states also show no impact of divorce law on cohabitation, so this factor does not seem to have an important impact on
73 future non-marital unions.
Economic Influences
Almost all studies of the rising rate of cohabitation point to the increasing role of women in the labor force and their growing non-household incomes, a phenomenon reducing the economic advantage of the traditional division of labor between men and women.9 Many also mention the incomes of poorly educated males are falling ever behind the median income and their employment record is more sporadic, both of which do not make them desirable marriage partners for higher earning females. Wydick’s (2007) cross-country and cross-state regressions explaining cohabitation, as well as his regressions explaining changes in cohabitation by state between 1990 and 2000, provide important evidence for a positive relationship between female labor force participation and cohabitation. In contrast, Sayer and Bianchi (2000) find only weak empirical support for this hypothesis, once the quality of the cohabiting relationship is held constant. That is, if the relationship between the partners deteriorates, the woman is more likely to start working outside the household. Although serious questions of two- way causation arise, what is important for this analysis is the positive correlation between female labor force participation and cohabitation, whatever the cause of this correlation may be. Several mechanisms have been proposed to explain the linkage between female labor force participation and cohabitation. Most importantly, marriage-age women are economically less dependent on men, so that for an increasing number, marriage is no longer necessary for their economic security. Using a game-theoretic approach, Wydick (2007) argues that women’s labor force participation lowers the gains from household specialization (that is, the dollar value of a woman’s time doing household chores versus a man’s income) and thus lessens the benefits of mutual dependency. It also increases the payoff from non-cooperation in marriage (lack of economic dependency) and increases the independence that a woman can achieve in non-household activities. The impact of female employment on future rates of cohabitation is probably slight. Although the participation of women in the labor force increased steadily over the second half of the twentieth century, it remained relatively constant after 2000. The U.S. Department of Labor projects that by 2016 this participation rate should be little different from the rate in 2000. (U.S. Bureau of the Census 2010, Table 575). This suggests that the rate of cohabitation should have risen up to 2000 and very slowly tapered off thereafter. Finally, although the tax structure penalizes marriage, investigations by Alm and Whittington (2003) suggest that this tax issue played a very minor role in the initial decision to marry or cohabit. They do suggest, however, that tax consequences have a more important influence in the decision to move from cohabitation to marriage.
WHY HAVE BIRTH RATES OF COHABITATING WOMEN DECLINED?
Although a rising fertility rate of cohabiting women was not the cause of the increase in non-married births, the uneven downward trend in the birth rate of married and cohabiting women shown in Table 3 raises some puzzles whose solution might have some impact on the future of such births. Both psychological studies of why single women want babies (e.g., Hertz, 2006) and several theoretical studies (e.g., Akerlof et al. 1996; Wu et al.
74 1999; Willis 1999) look at the question for a single year. But none of these explains why the fertility rate would change over time. As previously mentioned, the rising age of marriage might have some impact on cohabitation and non-marital births, but this should increase, not decrease the fertility rate of cohabiting women. Similarly, the growing social acceptance of non-marital births would explain why the fertility rates of unmarried women, whether cohabiting or single, might rise, but in the case of cohabiting women we actually see a fall between 1980 and 2001. It is possible that the increasing cost of raising children might account for the lower fertility of cohabiting (and married) women, but this remains to be proven. In short, this fertility puzzle remains unsolved.
A PERSPECTIVE
The soaring ratio of non-marital births to total births can be primarily explained by the growth in the number of cohabiting couples and, at the same time, the decline in both the relative number of married couples and their fertility. Changing fertility rates of married and cohabiting women do not account for the growing proportion and number of non-marital births. Given the strong relation between female labor force participation and cohabitation, it seems likely that if current labor force trends continue, the rising non- marital birth ratio will finally taper off. It seems unlikely that incremental public policy will reduce the rate of non-marital births. For instance, the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 not only eliminated Aid for Dependent Children payments to unwed mothers but offered $100 million to the five states with the largest reduction of non-marital births. Non-marital births went up anyway, in part to the relatively small incentives offered in the program.10 Moreover, neither public officials nor the public has shown the will to carry out more drastic incentive programs to curtail non-marital births.11 In brief, the potential for the government to carry out effective programs to reduce non-marital births seems low. In sum, it is argued that the rise in non-marital births can be traced primarily to a shift in the proportion of married and cohabitating couples, and to the greater fertility of the latter. In turn, the rise in cohabitation can be explained in part by the greater education and labor force participation of women and by a change in attitudes toward non-marital births. In the near future, non-marital births should continue to increase.
75 REFERENCES
Akerlof, George, Janet L. Yellen, and Michael L. Katz. 1996. “An Analysis of Out-of- Wedlock Childbearing in the United States,” Quarterly Journal of Economics 111, no. 2: 277-317. Alm, James, and Leslie A. Whittington. 2003. “Shaking Up or Shelling Out: Income Taxes, Marriage, and Cohabitation.” Review of Economics of the Household 1: 169-186. Bumpass, Larry, and Hsien-Hen Lu. 2000. “Trends in Cohabitation and Implications for Children’s Family Contexts in the United States,” Population Studies 54, no. 1: 29- 41. Casper, Lynne M., and Philip N. Cohen. 2000. “How Does POSSLQ Measure Up? Historical Estimates of Cohabitation,” Demography 37, no. 2: 237-46. Chase-Lansdale, P. Lindsay, Kathleen Kiernan, and Ruth J. Friedman, eds. 2004. Human Development across Lives and Generations: The Potential for Change. New York: Cambridge University Press. Cherlin, Andrew. 2009. The Marriage-Go-Round: The State of Marriage and the Family in America Today. New York: Knopf. Deparle, Jason, and Sabrina Tavernise/ 2012. “Unwed Mothers Now a Majority in Births in 20’s, New York Times 2/18/2012, p. 1 Fitch, Catherine, Ron Goeken, and Steven Ruggles. 2005. “The Rise of Cohabitation in the United States: New Historical Estimates.” Minnesota Population Center, University of Minnesota
76 Vital Health Statistics 23(25)
77 78 79 80 81 82 BEHAVIOR OF THE VIETNAMESE EQUITY PREMIUM Chu V. Nguyen, University of Houston-Downtown
ABSTRACT
The Vietnamese equity premium over the period 2000:07 to 2013:10: (i) followed a stationary trend process with a break date of August 2007, (ii) adjusted around its estimated threshold value symmetrically in the long run. When the short-run dynamic components are introduced to the model: (i) the return on the market portfolio asymmetrically responded to both the widening and the narrowing of the equity premium, (ii) the T-bill rate did not respond to either the widening or the narrowing of the equity premium. Finally, the GARCH (3, 3) effect is present on the Vietnamese monthly equity returns and their variance. JEL classifications: C22, F36, and G14
INTRODUCTION
Equity premium, the difference between the return on the market portfolio and the risk- free interest rate has been a topic of considerable debate. From the theoretical perspective, the equity premium is the difference between the expected real return on market portfolio of common stocks and the real risk free interest rate. As initially recognized by Mehra and Prescott (1985), the historic U.S. equity premium, which is in the world’s largest economy, appears to be much greater than what can be rationalized in the context of the standard neoclassical paradigm of financial economics. Mehra (2003) articulated that for the 1889- 2000 period, the average annual real return on the US equity market has been about 7.9%, as compared to the real return on a relatively riskless security was 1.00%. This irrationally high average, dubbed “the equity premium puzzle” is not unique to U.S. capital market. Internationally, as reported by Dimson et al., (2006) over the 1900-2005 period, the equity premium measure relative to T-bills was 7.08% in Australia, 6.67% in Japan, 6.20% in South Africa, 3.83% in Germany, 5.73% in Sweden, 5.51% in the US, 4.43% in the UK, 6.55% in Italy, 4.54% in Canada, 6.79% in France, 4.55% in Netherlands, 4.09% in Ireland, 2.80% in Belgium, 3.07% in Norway, 3.40% in Spain, 2.87% in Denmark and 3.63% in Switzerland. The average equity premium for these 17 countries over this period of 106 years is 4.81%. In the late 2011, Dimson et al., (2011) updated the global evidence on the long-term realized equity risk premium, relative to both bills and bonds, in 19 different countries. Their sample was from 1900 to the start of 2011. They found that while there was considerable variation across countries, the realized equity risk premium was substantial everywhere. They reported that for a sample of 19-country World index, over the entire 111 years, geometric mean real returns were an annualized 5.5%; the equity premium relative to Treasury bills was an annualized 4.5%; and the equity premium relative to long-term government bonds was an annualized 3.8%. The expected equity premium is lower, around
83 3% to 3½% on an annualized basis. Since its introduction to the literature in 1985, the equity premium puzzle has spawned many efforts by a number of researchers to explain this anomaly away. With the exception of the following investigations, the majority of the studies concentrated on theoretically and empirically explaining the implausible equity premium puzzle. Buranavityawut and Freeman (2006) examined consumption risk and the equity premium. Blanchard (1993) studied the variation of the equity premium for a 50 year period. Fama and French (2002) compared the estimated unconditional equity premium to the realized market gains. Siegel (1999) investigated the variations of the size of the equity premium. Welch (2000) surveyed financial economists on their expectations on the future equity premium. While the theoretical and empirical debates are still unsettled, equity is the major instrument to channel the financial resources from the capital surplus economic units (the savers) to the financial deficit units (the borrowers) in the direct financing mode of the market economies. In the capital market, the realized equity premium is the premium that corporations have to pay to obtain their financial resources, when they issue new equities or to acquire their treasury stocks, just like the difference between the loan rate and the risk free interest rate that financial institutions charge for loans to corporations. Therefore, the time path on which the equity premium adjusts towards its “normal” or equilibrium level following a shock has a major consequence on the cost of capital to corporations. Thus, policymakers should have accurate knowledge of the adjustment process of the equity premium when being disturbed by economic shocks or countercyclical monetary policy action in the equity market. The remainder of the study is organized as follows: Section 2 briefly describes the nature of the equity premium; Section 3 summarizes the Vietnamese equity market; Section 4 describes the data set and its descriptive statistics; Section 5 discusses the methodology and model’s specification; Section 6 reports and discusses the empirical results; Section 7 provides some concluding remarks and recommendations.
NATURE OF EQUITY PREMIUM
Brealey and Myers (2003) articulated that an integral part of the economic and financial literature on equity premium is the assumption that “there is a normal, stable, risk premium on the market portfolio.” Therefore, to estimate the ex-ante equity premium, the most popular method is to extrapolate the historically realized equity premium into the future (Welch, 2000). For example, Brealey and Myers (2000), described how to estimate a return for a diversified stock market portfolio. They do this by taking the current interest rate on U.S. Treasury bills plus the average equity premium over some historical time period. In other words, they simply extrapolated past returns forward. Brealey and Myers (2000) noted that their result is consistent with security analysts’ forecasts of earnings growth. This assumption requires that the equity premium time series be mean-reverting. In addition, the capital asset pricing model (CAPM) conceptually postulates that investors set their required real earning yields as some markup relative to real risk free interest rates. In the equity market, this mark-up is the equity premium. If this equity premium becomes too high or low, the marketplace will put pressure on the investors to adjust it back to some “normal” or equilibrium equity premium. Specifically, the above assumption implies that the equity premium returns back to its long run equilibrium position following any shock. Perhaps the state of the equity premium puzzle today still can be described best by
84 one of the two researchers who originally recognized the anomaly: “After detailing the research efforts to enhance the model’s ability to replicate the empirical data, I argue that the proposed resolutions fail along crucial dimensions.” Mehra (2003). Also, Damodaran (2014) articulated that Equity risk premiums are a central component of every risk and return model in finance and are a key input into estimating costs of equity and capital in both corporate finance and valuation. Given their importance, it is surprising how haphazard the estimation of equity risk premiums remains in practice.
VIETNAMESE EQUITY MARKET
To this end, the recent inaugural equity market of the transitional economy of Vietnam is of particular interest. To describe what has transpired in the Vietnamese economy, as cited by Currie (2008), Klaus Rohland, the World Bank’s Vietnam country director from 2002 to 2007 stated “There is probably no other country in the world that, over the last 15 years, has moved its development so far and so fast.” This characterization is in diametric contrast with Vietnam in the1990s, which was mostly mentioned in policy circles for having some of the most inappropriate reforms among the transitional economies, Kim (2008). The equity market is the supposed poster child of the Vietnamese financial sector, which has benefitted greatly from market liberalization and demonstrated impressive economic growth in spite of the considerable international, political, and social turmoil of the 2000s. Historically, Vietnam’s first stock exchange, known as the Ho Chi Minh City Securities Trading Center, was established in July 2000. In early 2005, the exchange had 28 stocks listed and a total market capitalization of only U.S. $270 million. In March 2005, Vietnam opened an over-the-counter exchange known as the Hanoi Securities Trading Center to expedite the privatization of state-owned enterprises. Additionally, Vietnamese officials set the goal of expanding their combined market capitalization to 10 percent of gross domestic product by 2010 and gradually phased out restrictions on foreign ownership o f s h a r e s . In September 2005, Vietnam’s Prime Minister announced that the limit on foreign share ownership would rise from 30 percent to 49 percent. Actually, by the end of 2012, the number of companies listed was 183 and, as indicated by Table 1, the total market capitalization accounted for only 23.2 percent of gross domestic product. The Vietnam Stock Index or VN- Index is a capitalization-weighted index of all the companies listed on the Ho Chi Minh City Stock Exchange. The index was created with a base index value of 100 as of July 28, 2000. Prior to March 1, 2002, the market only traded on alternate days. Additionally, equity market regulations, issued in 2006, and the attendant consequences of becoming the WTO membership have changed the landscape of the Vietnamese equity market significantly. On May 03, 2006 the Vietnam Securities Depository officially went into operation under Decision No. 189/2005/QD-TTg signed on July 27, 2005 to increase the market performance in general, and the clearing and settlement system in particular. Moreover, on June 01, 2006 the Hanoi Securities Trading Center increased the number of trading days from 3 to 5 days a week, in order to increase the market liquidity. Starting on June 14, 2006, the Ho Chi Minh City Securities Trading Center increased the number of order matching phases from 2 to 3 phases a day (1st phase from 8h40 to 9h10, 2nd phase from 9h20 to 9h50, 3rd phase from 10h to 10h30) in order to meet investors’ trading demand. Finally, on November 07, 2006 Vietnam gained membership to become 150th member of the World Trade Organization, and would
85 officially join on January 11, 2007. That event created new opportunities and challenges for the course of the country’s cultural and economic integration into the globe, especially with respect to the financial aspects of the economy. As Table 1 indicates, the Vietnamese equity market is still relatively undeveloped.
DATA
This study utilizes monthly stock price indices in Vietnam and the T-bill rate as the proxy measure for risk-free rate. The data set, used in this investigation, covers the period from its inaugural month of July 28, 2000 to October 2013 where the data on risk-free rate is available. The time-series data is obtained from the Vietnamese equity market: http://www.cophieu68. vn/historyprice.php?id=^vnindex, (retrieved on April 18, 2014.) The data on the T-bill rate, used as the proxy for the risk free rate, is obtained from the International Financial statistics, published by the International Monetary Fund. In this analysis, let ER and RF denote the annualized monthly return on the Vietnamese monthly equity market portfolio and the risk free rate, respectively. The monthly return on the market portfolio is annualized to be comparable to the risk-free rate which is stated in the annual basis. The difference between ER and RF is defined as equity premium and is denoted by EP. Figure 1 illustrates the behaviors of ER, RF and EP over the sample period. As to the descriptive statistics of the time series of the Vietnamese equity annualized monthly return, its mean is 20.31 percent, ranging from -452.63 percent to 417.99 percent with standard error being 139.46 percent. The corresponding figures for the T-bill rate were 7.38 percent, 3.34 percent, 15.60 percent and 2.76 percent, respectively. The average Vietnamese equity premium since its inauguration is 12.93 percent. Additionally, in their 2014 survey of market premium used in 2014 in 88 countries, Pablo (2014) reported, in Table 2, the average premium of the following selected countries which show that Vietnamese equity premium is the among the highest premia in its neighboring Asian countries and much higher than the corresponding figures in the advanced economies.
METHODOLOGICA ISSUES AND ANALYTICAL FRAMEWORK
Structural Break
Historically, every economy would experience many business cycles caused by internal and external shocks; therefore, countercyclical monetary policy measures would be used to bring the economy back to its long-run path. Vietnam is no exception! Consequently, the spread between return on market equity index and the risk free rate, the equity premium, is most likely to suffer some structure breaks. To search endogenously for the possibility of any structural break in the basis, this investigation utilized Perron’s (1997) endogenous unit root test function with the intercept, slope, and the trend dummy, as specified by equation (1), to test the hypothesis that the spread between stock price index and the money supply has a unit root.
k EP = µ +θDU + ςt + ξDT + δD(T ) + ζEP + ψ ΔEP +υ t b t−1 ∑i=1 i t−i t
(1)
86 DU = 1(t > Tb ) where is a post-break constant dummy variable; t is a linear time trend; D(T ) = 1(t = T +1) b b is a post-break slope dummy variable;is the break dummy
variable; andet are white-noise error terms. The null hypothesis of a unit root is stated as
. The break date,Tb , is selected based on the minimum t-statistic for testing z = 1 (see Perron, 1997).
Threshold Autoregressive (TAR) model
To further investigate the nature of the Granger causality between the equity premium and the risk-free rate, this study uses the threshold autoregressive (TAR) model, developed by Enders-Siklos (2001) that allows the degree of autoregressive decay to depend on the state of the equity premium, i.e. the “deepness” of cycles. The estimated TAR model would empirically reveal if the premium tends to revert back to the long-run position faster when the premium is above or below the threshold. Therefore, the TAR model indicates whether troughs or peaks persist more when shocks or countercyclical monetary policy actions push the equity premium out of its long-run equilibrium path. In this model’s specification, the null hypothesis that the basis contains a unit root can be expressed as r 1 = r 2 = 0 , while the hypothesis that the basis is stationary with symmetric adjustments can be stated as r 1 = r 2 . The first step in the Enders-Siklos’ (2001) procedure is to regress the equity premium,
EP t , on a constant and an intercept dummy (with values of zero prior to the structural break date and values of one for the structural break date and thereafter), as specified by equation (2).
EP t = p 0 + p1Dummyt + et (2)
The saved residuals, et from the estimation of equation (2), denoted byeˆt , are then used to estimate the following TAR model:
(3)
2 where uˆt ~ i.i.d.( 0,s ) , and the lagged values of �eˆt are meant to yield uncorrelated residuals. As defined by Enders and Granger (1998), the Heaviside indicator function for the TAR specification is given as:
#1 i f εˆt−1 ≥ τ I t = " (4) !0 if εˆt−1 < τ
The threshold value,t , is endogenously determined using the Chan (1993) procedure which obtains t by minimizing the sum of squared residuals after sorting the estimated residuals in an ascending order, and eliminating 15 percent of the largest and smallest values. The elimination of the largest and the smallest values is to assure that the eˆ series t
87 crosses through the threshold in the sample period. Throughout this study, the included lags are selected by the statistical significances of their estimated coefficients as determined by the t-statistics.
The Asymmetric Error-Correction Models
Moreover, to investigate the short-run asymmetric dynamic behavior between the return on the market equity portfolio and the risk–free interest rate, this study specifies and estimates the following asymmetric error-correction model. The estimation results of this model can be used to discern the nature of the Granger causality between the return on the market stock portfolio and the risk-free rate. Additionally, the following TAR-VEC model differs from the conventional error-correction models by allowing asymmetric adjustments toward the long-run equilibrium. n q ΔER = α + ρ I εˆ + ρ (1 − I )εˆ + α ΔER + γ ΔRF + u t 0 1 t t−1 2 t t−1 ∑i=1 i t−i ∑i=1 i t−i 1t (5)
n q ΔRF = α~ + ρ~ I εˆ + ρ~ (1 − I )εˆ + α~ ΔER + γ~ ΔRF + u t 0 1 t t−1 2 t t−1 ∑i=1 i t−i ∑i=1 i t−i 2t (6)
2 where u1,2t ~ i.i.d.( 0,s ) and the Heaviside indicator function is set in accord with (4). This model specification recognizes the fact that the stock price index responds differently depending on whether the equity premium is widening or narrowing, due to the nature of economic shock or countercyclical policy.
GARCH(s, r)-M MODEL
As to the equity premium in relation to market volatility and economic condition, Graham and Harvey (2009) analyzed the history of the equity premium from surveys of U.S. Chief Financial Officers conducted every quarter from June 2000 to March 2009. They defined equity premium as the expected 10-year S&P 500 return relative to a 10-year U.S. Treasury bond yield. They noted that these surveys were conducted during the darkest parts of a global financial crisis. They further indicated that the equity premium sharply increased during the crisis. The authors also found that the level of the equity premium closely tracks the market volatility as measured by the VIX. Additionally, from June 2000 to March 2012 surveys, Graham and Harvey (2012) found that while the equity premium sharply increased during the financial crisis peaking in February 2009, and then steadily fell until the second quarter 2010. These aforementioned results indicated that the equity premium is affected by market volatility and economic condition of the economy. The Vietnamese economy has become more and more internationalized and the international economic landscape over the existence of the Vietnamese stock market has been dotted with international political and social turmoil. These developments exacerbate the variance of equity premium and cause the variance to be different from some sub- periods to others over the sample period. Additionally, the graph of the Vietnamese equity premium in Figure 1 strongly supports the different variances in the Vietnamese equity premium from one sub-period to another period. Therefore, another important question
88 for investors, policy makers, and corporate executives is whether the fluctuations in the equity premia of the market portfolio and hence their variances from the one month affect the premia and the variances in the next month. To this end, this investigation specifies and estimates the following GARCH(s, r)-in-Mean (GARCH-M) model to discern this possibility. GARCH-M models have been very popular and effective for modeling the volatility dynamics in many asset markets.
EP = c + λω 2 + ε (7) t t t r s 2 2 2 (8) ωt = α + ∑ β lε t−l + ∑η mωt−m l=1 m=1 2 where EP is the equity premium, Ln is the natural Logarithm, and wt is the variance of the Vietnamese equity index at time t; et is a disturbance; c is a constant; l , a , b l , and h m are the parameters to be estimated of the model. The retentions of these estimated coefficients are determined by the calculated z-statistics at the 5 percent level of significance. Ther and s indices are the highest subscripts l and m of retained b l andh m .
EMPIRICAL RESULTS
Results of the Test for Structural Break
The estimation results of Perron’s endogenous unit root tests are summarized in Exhibit 1. An analysis of the empirical results reveals that the post-break intercept dummy variable, DU, and the post-break slope dummy variable, DT, are negative and insignificant at any conventional level. The time trend is positive and is significant at a 1 percent level. The empirical results of these tests suggest that the Vietnamese equity premium followed a stationary trend process with a break date of August 2007, which may be attributable to Vietnam officially becoming the 150th member of the World Trade Organization, and its attendant consequences.
Results of Cointegration Test with Asymmetric Adjustment
To examine whether or not the Vietnamese equity premium, EP, and the risk-free rate, RF, are co-integrated when allowing for possible asymmetric adjustments, the equity premium is regressed on a constant and an intercept dummy with values of zero prior to August 2007 and values of one for August 2007 and thereafter. The estimation results are reported in Exhibit 2. The residuals from them above estimation are used to estimate the TAR model specified by equations (3) and (4). The estimation results for the TAR model are reported in Exhibit 3. Over all, the empirical results reveal that the null hypothesis of symmetry, r 1 = r 2 , cannot be rejected at any significant level, based on the partial F = 0.1010, indicating statistically that adjustments around the threshold value of Vietnamese equity premium are symmetric. Additionally, the calculated statistic 21.5811 indicates that the null hypothesis of
89 no co-integration, r 1 = r 2 = 0 , should also be rejected at the 1 percent significance level, confirming that the equity premium is stationary. With regard to the sationarity of the premium, Ewing et al., (2007) pointed out that this simple finding of stationarity is consistent with the two underlying series comprising the premium (the monthly return on the Vietnamese market portfolio and the risk-free rate) being co-integrated in the conventional, linear combination sense.
More specifically, the estimation results reveal that both r 1 and r 2 are statistically significant at any conventional level. In fact, the point estimates suggest that the premium tends to decay at the rate of r 1 = 0.7664 for eˆt�1 above the threshold t=-2.8391 and at the rate of r 2 = 0.8391 for eˆt�1 below the threshold.
eˆt�1 > -2.8391 is indicative that an economic shock or a countercyclical monetary policy action causing a decline in the risk-free rate, such as an expansionary monetary policy, has widened equity premium. This widening of the premium initiates a downward adjustment in the equity premium. Similarly, eˆt�1 < -2.8391 is indicative that an economic shock or a countercyclical monetary policy action causing an increase in the risk-free rate, such as a contractionary monetary policy, has narrowed equity premium. This narrowing of the premium initiates an upward adjustment in the premium. Numerically, the estimation results reveal r 2 > r 1 which seems to indicate a slower convergence for positive disequilibrium than for negative disequilibrium, i.e., an asymmetric adjustment process. However, the aforementioned failure to reject the null hypothesis that r 1 = r 2 at any significant level, based on the partial F = 0.1010, indicates a symmetric adjustments of the equity premium about its threshold to negative and positive shocks in the long run.
Results of the Asymmetric Error-Correction Models
Exhibit 4 summarizes the estimation results for the TAR-VEC model specified by equations (4), (5) and (6) using the Vietnamese return on the market equity portfolio and the risk-free rate. In the summary of the estimation results, the partial Fij represents the calculated partial F-statistic with the p-value in square brackets testing the null hypothesis that all coefficients ij are equal to zero. “*” indicates the 1 percent significant level of the t-statistic. QLB (12) is the Ljung-Box statistic and its significance is in square brackets, testing for the first twelve of the residual autocorrelations to be jointly equal to zero. lnL is the log likelihood. The overall F-statistic with the p-value in square brackets tests the overall ~ ~ fitness of the model. The retained estimated coefficientsa i ,gi ,a i , and gi are based on the 5 percent level of significance of the calculated t-statistics. An analysis of the overall empirical results indicates that the estimated equations (5) and (6) are absent of serial correlation and have good predicting power as evident by the Ljung-Box statistics and the overall F-statistics, respectively. With regard to the short-run dynamic Granger causality between equity premium and the risk-free rate, the partial F-statistics in equation (5) reveal a bi-directional Granger- causality between the risk-free rate to the equity premium; i.e., the equity premium responds
90 to both its own lagged changes and the lagged changes of risk-free rate as well. Similarly, the empirical results for equation (6), the partial F-statistics suggest that the risk-free rate responds not only to its own lagged changes but also to lagged changes of the equity premium in the short run. Over all, the TAR-VEC estimation results seem to suggest that the Vietnamese equity market responds to monetary, fiscal policy and economic shocks which change the T-bill rates. This finding indicates that the Vietnamese economic policies matter in the short run. As to the long-run adjustments, the statistical significances of the error correction terms and r 2 > r 1 in equation (5) indicates that the equity premium asymmetrically responds to negative and positive shocks when short-run dynamic components are introduced to the model. Since r 1 and r 2 are significant at any conventional level, the estimation results of the TAR-VEC reveal that equity premium reverses to the long-run equilibrium faster when the equity premium is below the threshold than when it is above the threshold. With regard ~ ~ to the risk-free rate, the estimation results of equation (6) show | r 2 | >| r 1 | . However, ~ ~ both | r 1 | and | r 2 | are not statistically significant at any conventional level, indicating that the risk-free rate does not respond to either the widening or the narrowing of the equity premium in the long run.
GARCH(s, r)-M Model
As aforementioned, the retentions of the estimated coefficients of equations (7) and (8) are determined by the calculated z-statistics at the 5 percent level of significance. The r and s indices are the highest subscripts l and m of retained b l andh m which are l =3 and m=3, respectively. The values of l and m, in turn, suggest GARCH (3, 3) be the best model for this investigation. The estimation results of the GARCH (3, 3)-M model are reported in Exhibit 5. An analysis of the estimation results of the GARCH(r, s)-M model suggests the presence of GARCH (3, 3) effect on the Vietnamese monthly equity returns and their variance. Financially, the empirical results indicate that the fluctuations in the equity premia on the market portfolio and their variances from the one month affect the premia and the variances in the next month.
CONCLUDING REMARKS
While the theoretical debate on the anomalous equity premium is unsettled, equity has been an important instrument channeling the financial resources from the capital surplus economic units (the savers) to the financial deficit units (the borrowers) in the direct financing mode of the market economy. This study uses the well known TAR and the GARCH (3, 3)-M models to analyze the behavior of the Vietnamese equity premium. This study utilizes monthly stock price indices in Vietnam and the T-bill rate as the proxy measure for the risk-free rate. The equity premium is defined as the difference between the monthly change in the Vietnamese equity index and the T-bill rate. The data set used in this investigation, covers the period from its inaugural month of July 28, 2000 to October 2013 where the
91 data on risk-free rate is available. The descriptive statistics reveal that the equity premium over the sample period is 12.93 indicating that the Vietnamese equity premium is among the highest premia in its neighboring Asian countries and much higher than the corresponding figures in the advanced economies. Perron’s endogenous unit root test revealed that the equity premium is a stationary process with a structural break date of August 2007, which may be attributable to Vietnam officially becoming the 150th member of the World Trade Organization, and its attendant consequences. The threshold autoregressive TAR model reveals that the Vietnamese equity market symmetrically responds to monetary and fiscal policy, which is indicative that the policy makers use these instruments to effectively manage the equity market in the long run. With regard to the short-run dynamic Granger causality between the equity premium and the risk-free rate, the estimation of equation (5) revealed a bi-directional Granger- causality between the risk-free rate to the equity premium. Similarly, the empirical results for equation (6), suggest that the risk-free rate responds not only to its own lagged changes but also to lagged changes of the equity premium in the short run. Taken together, the empirical results of the TAR-VEC suggest that the Vietnamese equity market responds to monetary, fiscal policy and economic shocks which change the T-bill rates. This finding indicates that the Vietnamese economic policies matter in the short run. As to the long-run and when short-run dynamic components are introduced to the model, the TAR-VEC reveal that the equity premium reverses to the long-run equilibrium faster when the equity premium is below the threshold than when it is above the threshold. However, the risk-free rate does not respond to either the widening or the narrowing of the equity premium. Finally, the empirical investigations suggest GARCH (3, 3)-M be the best model for this investigation. The significance of the GARCH (3, 3)-M indicates the presence of GARCH (3, 3) effect on the Vietnamese monthly equity returns and their variance.
92 REFERENCES
Blanchard, O.J. (1993). Movements in the equity premium. Brookings Papers on Economic Activities Macroeconomics 2: 75–118. Brealey, R.E. and Myers, S. C. (2003). Principles of Corporate Finance, 7th Edition, McGraw-Hill Brealey, R.E. and Myers, S. C. (2000). Principles of corporate finance. 6th ed. Boston: McGraw-Hill. Buranavityawut, N., Freeman, M. C., and Freeman, N. (2006). “Has the Equity Premium Been Low for 40 Years?” North American Journal of Economics and Finance, 17(2) (August), 191-205. Chan, K.S. (1993). “Consistency and Limiting Distribution of the Least Squares Estimator of a Threshold Autoregressive Model”, Annals of Statistics, 21(2), 520-533. Currie, D. (2008). “Vietnamese Growing Pains”, The American online Magazine, American Enterprise Institute. Damodaran, A. (2014). “Equity Risk Premiums (ERP): Determinants, Estimation and Implications – The 2014 Edition. Available at SSRN: http://ssrn.com/abstract=2409198 or http://dx.doi.org/10.2139/ssrn.2409198 Dimson, E., Marsh, P., and Staunton, M. (2006). “The Worldwide Equity Premium: A Smaller Puzzle”, SSRN Working Paper, No. 891620. Dimson, E., Marsh, P. and Staunton, M. (2011). “Equity Premia Around the World. Available at SSRN: http://ssrn.com/abstract=1940165 or http://dx.doi.org/10.2139/ ssrn.1940165 Enders, W. and Granger, C. W. J. (1998). “Unit Root Tests and Asymmetric Adjustment with an Example Using the Term Structure of Interest Rates”, Journal of Business and Economic Statistics, 16(3), 304-311. Enders, W. and Siklos, P. (2001). “Cointegration and Threshold Adjustment”, Journal of Business & Economic Statistics, 19(2), 304-311. Ewing, B.T. and Kruse, J. (2007). “The Prime Rate-Deposit Rate Spread and Macroeconomic Shocks”, in Advances in Quantity Analysis of Finance and Accounting, Edited by Cheng-Few Lee, pp. 181-197, Rutgers University, USA Fama, E. F. and French, K. R. (2002). “The Equity Premium,” Journal of Finance, 57 (April), 637-659. Graham, J. R. and Harvey, C. R. (2009). “The Equity Risk Premium amid a Global Financial Crisis”. Available at SSRN: http://ssrn.com/abstract=1405459 or http:// dx.doi.org/10.2139/ssrn.1405459 Graham, J. R. and Harvey, C. R. (2012). “The Equity Risk Premium in 2012”. Available at SSRN: http://ssrn.com/abstract=2020091 or http://dx.doi.org/10.2139/ssrn.2020091 Kim, A. M. (2008). Learning to Be Capitalists in Vietnam’s Transition Economy, Oxford University Press, New-York, USA. Mehra, R. (2003). “Equity Premium: Why Is It a Puzzle?” Financial Analysts Journal, 59(1), (January/February), 54-69. Mehra, R. and Prescott, E. C. (1985). “The Equity Premium: A Puzzle.” Journal of Monetary Economics, 34(1), (March), 145-161. Perron, P. (1997). “Further Evidence on Breaking Trend Functions in Macroeconomic Variables”, Journal of Econometrics, 80(2), 355-385. Siegel, J. J. (1999). “The Shrinking Equity Premium.” Journal of Portfolio Management,
93 26(1) (Fall), 10-17. Welch, I. (2000). “Views of Financial Economists on the Equity Premium and on Professional Controversies.” The Journal of Business 73(4), (October), 501–537.
94 Figure 1
600 Vietnamese Monthly Data from August 2000 to October 2013 400
200
0
-200
Annualized Returns on Market Portfolio -400 Vietnamese T-bill Rates Equity Premium
-600 Sources: International Financial Statistics, IMF; and http://www.cophieu68.vn/ historyprice.php?id=^vnindex.
95 96 97 98 THE ECONOMIC IMPACT OF OKLAHOMA TOBACCO SETTLEMENT SPENDING ON RESEARCH Fritz Laux, Northeastern State University Brian Jackson, Northeastern State University Justin Halpern, Northeastern State University
ABSTRACT
This article estimates the economic impact in Oklahoma of spending on medical research financed by the Oklahoma Tobacco Settlement Endowment Trust (TSET). This is first done by the popular approach of input-output modeling, to estimate the impact of this spending on economic activity. It is then estimated via consideration of potential benefits to Oklahoma consumer welfare. Summary results are that the modest multiplier impact of this spending is amplified by the large extent to which TSET investments have helped state researchers obtain federal grant financing. Welfare benefits seem to be large and reflect innovative and translational research contributions. JEL Classification: I18, L38, L88
INTRODUCTION
Oklahoma’s Tobacco Settlement Endowment Trust (TSET) was created by a constitutional amendment, passed by the voters as State Question 692, on November 7, 2000. It is financed by a portion (currently 75%) of the payments Oklahoma receives every year from tobacco manufacturers, through Oklahoma’s participation in the Tobacco Master Settlement Agreement, negotiated between 46 state attorneys general and the four largest US tobacco manufacturers.1 In its creation the Oklahoma TSET was mandated to fund:
1. Clinical and basic research and treatment efforts to prevent and combat cancer, and other tobacco-related diseases. 2. Cost-effective tobacco prevention and cessation programs. 3. Programs designed to maintain or improve the health of Oklahomans or to enhance the provision of health care services to Oklahomans, with particular emphasis on such programs for children. 4. Programs and services for the benefit of the children of Oklahoma, with particular emphasis on common and higher education, before- and after-school programs, substance abuse prevention and treatment programs, and services designed to improve the health and quality of life of children. 5. Programs designed to enhance the health and well-being of senior adults.
99 Within this mandate the current TSET strategy is to focus on the leading causes of preventable death in Oklahoma: cancer, attributable to tobacco use; and heart disease, attributable to obesity and tobacco use. Total TSET expenditures for FY2014 will be approximately $42.96 million of which $10.02 million, or 23.3%, is programmed to be spent on research. The TSET’s budget allocation for research in FY13 was also 23.3%. This paper estimates and assesses the economic impact of this TSET research spending in Oklahoma. It first does this through the use of standard techniques of input- output analysis, to estimate the contribution of TSET research spending on the overall economic activity in the state. It then explores some top-level perspectives on the value of health research in Oklahoma, which provide insight into the payoff that consumers may be getting from this research investment.
Medical Research Spending in Oklahoma as Compared to Other States
Any assessment of the role of research in the state’s economy should start by noting that Oklahoma has consistently ranked toward the bottom of the 50 states in terms of research and development spending. In 2010, the most recent year for which these data were available, Oklahoma ranked 47th of the 50 states in research spending per dollar of GDP (70 cents spent on research for every $100 of state domestic product). Oklahoma was above only Arkansas, Louisiana, and Wyoming. In 2009 Oklahoma was ranked 45th and from 2004 through 2008 it was ranked between 45th and 46th. Oklahoma’s academic, i.e., university, spending on research and engineering made up just under one third of total research spending and its rank in academic research spending per dollar of state domestic product has also been stably at approximately 45th.2 The question of whether Oklahoma should be committed to achieving a higher ranking on these indices is essentially a political one. The answer depends on voters’ and policymakers’ perceptions of the value of research spending. However, the experience of other states suggests that a strong research environment is generally associated with stronger growth in high-tech industries, which is associated with clean industry, desirable jobs, and higher standards of living. At the national level, a substantial portion of research and “high-tech” spending is in medicine, and this would also be expected to support a more attractive environment for medical practice and innovation in the state. At the University of Oklahoma, which conducts over half of the academic research and development in the state, approximately 40% of R&D spending is in the life sciences, which matches the life-science proportion of total academic research spending for the state as a whole.3 This is well below the 57% national average for life sciences research, a category that includes medicine, biology, botany/horticulture/agriculture, and related sciences, but does not include environmental research. Further darkening the picture for medical research in Oklahoma, only 26% of academic research done in the life sciences (state-wide) is medical research, as compared to a national average of 55%.4 Hence, the proportion of academic research and development funding allocated to medical research in the state comes out to around 10%, as opposed to a national average of around 31%.5 Thus, relative to the size of the Oklahoma economy, Oklahoma’s overall level of spending on research and development is small. Its level of spending on academic research and development is also small and, within the category of academic research spending, its spending on medical research is also small. To the extent that research spending helps improve the quality of medical training, helps attract better physicians, helps respond to the health research needs of the community, and helps translate research from outside the state
100 for within-state application, this low level of spending is a concern.
ESTIMATING THE IMPACT OF TSET RESEARCH FUNDING ON ECONOMIC ACTIVITY IN OKLAHOMA WITH INPUT-OUTPUT ANALYSIS
TSET funding for research is allocated and managed through the research operations of the Stephenson Cancer Center (Stephenson), which includes the Oklahoma Tobacco Research Center (OTRC), and the Oklahoma Center for Adult Stem Cell Research (OCASCR). Table 1 shows TSET research spending (historical and budgeted) from FY 2012 through FY 2014. Researchers have been able to leverage TSET funding to bid for National Institutes of Health (NIH), and other federal and foundation grants. The budgeted spending by year from those sources is also shown on Table 1. Note that grant funding, particularly from the NIH, often comes in the form of large, multi-year awards. Table 1 reflects only budgeted spending by fiscal year. Since Stephenson and OCASCR activities are currently in expansion, total amounts awarded by year (including multi-year awards) are higher than is shown on this table. The standard approach for estimating the impact on economic activity within a state that is attributable to this kind of spending is via input-output (I-O) analysis. This involves detailed economic modeling to estimate the effects of stimulus programs, regulatory changes, and business relocations. These models, most commonly used by businesses and regional economic development authorities, are built using census and other survey data on revenue levels, sales channels, and purchasing channels, by industry. 6 They incorporate a tracking of how the production in one industry tends, on average, to draw from and consume the output and intermediate goods produced by other industries. They also track the geography of the flow of goods and the extent to which final goods produced in one county consume intermediate goods produced in that same county, in that same state, or in the same country. The art of I-O model construction is in the estimation of results where data are missing and in the reconciliation of contradictory input data. Subaggregated census data and sales tax data are used for these reconciliations.7 The two predominant I-O models used in the United States are the Regional Input- Output Modeling System (RIMS II) and IMPLAN (IMpact analysis for PLANning). RIMS II is maintained by the Bureau of Economic Analysis in the U.S. Department of Commerce. IMPLAN was originally developed by the US Forest Service (largely for USDA applications). IMPLAN is now very widely used, even by the Bureau of Economic Analysis that maintains RIMS, and is maintained by MIG, Inc. (previously known as the Minnesota IMPLAN Group, Inc.) A key advantage of IMPLAN over RIMS II is that it provides explicit modeling of economic flows via transfer payments and subsidies, in addition to economic flows attributable to the purchase and sale of labor, goods, and services. Regional development economists at Northeastern State University used IMPLAN Version 3 for this analysis. Since IMPLAN matrices are estimated and released to users with some time lag, and the most recently available for this study were based on 2011 data. Thus, these same matrices were used to estimate impacts in FY 2012, 2013, and 2014. Through the use of input-output analysis, the impact of TSET spending is captured by treating TSET research funding, plus NIH and foundation grant money leveraged off of TSET funding, as a purchase of research and development services (NAICS 5417, IMPLAN Sector 376). These purchases have a direct effect in that they increase the
101 revenues of the organizations financed, and thereby directly affect gross state product. They have indirect effects, in that those industries receiving TSET funding will normally purchase items from other industries in the state, and thereby stimulate those industries. Finally, increased payroll in the affected industries will lead to increased demand in other industries through the spending behavior of employees, an induced effect. A multiplier matrix simply tracks, by industry, the effect that a $1 purchase in one industry will tend to have on each industry sector. These matrices have one row per sector with one column within each row for each sector-to-sector interaction. The analyst needs data on the value added, payroll, and employment effects attributable to a $1 impact per industry. The analyst can then use a multiplier matrix for each county or region to calculate indirect and induced effects. For research and development services, a $1 increase in Oklahoma research spending is associated with a 39 cent indirect effect, attributable to increases in spending for Oklahoma-based research-related firms (summing across all other sectors in Oklahoma) and a 41 cent induced effect of additional Oklahoma spending attributable to the additional employment and salaries for these firms.7 The total effect of this $1 is thus $1.80 in increased economic activity. Since not all of this spending, by IMPLAN Sector 376 and by sectors affected by IMPLAN Sector 376, will go to Oklahoma firms, one can separate the $1.80 total effect into a value added component and a residual component. The value added component is that portion of the total economic activity stimulated that goes to Oklahoma firms. The residual component goes to business operations outside of Oklahoma. The multiplier measuring the Oklahoma impact of each additional dollar spent on Oklahoma research is $0.93 and the multiplier for the non-Oklahoma portion of this impact is $0.87. To see how this works, suppose TSET-funded research activity requires the purchase and installation of a new imaging machine. Some 20% of the components for this machine may be manufactured in Oklahoma, with 80% manufactured in other states or other countries. This 20%, less any imports to Oklahoma for the manufacture of these components, along with all Oklahoma-based transportation, installation, and set-up costs, would be counted as Oklahoma value added. The 80% would be included in total impact but would not be counted as Oklahoma value added. The numbers used for the multiplier analysis are not based on a specific tracking of TSET-funded purchases. Rather they are based on estimated industry averages for spending patterns by sector. A listing of these effects, together with estimates of the total employment, Oklahoma value added, and total dollar impact of TSET research spending is provided on Table 2. The convention in economic impact analyses like these, which use IMPLAN or RIMS II multipliers, is to emphasize the total effect of the subject increase or decrease in spending. This is because the total effect describes the total amount of money that the policy change will cause to be spent by Oklahoma firms or entities. The impact of TSET research spending on Oklahoma-based industrial activity is, however, better estimated by reference to the value-added totals on the table. The implication of these results, in terms of impact on state-level industrial activity, is that the TSET investment has been multiplied by roughly a factor of 4. This is because for each dollar TSET has invested in research, Oklahoma investigators have been able to obtain $3 in federal or foundation grant funding to extend or expand that research.
102 ESTIMATING THE VALUE OF MEDICAL RESEARCH IN OKLAHOMA
The input-output analysis of the previous section was used to estimate the impact of TSET research spending on state-level economic activity. The analysis of this section explores how one can estimate, or at least think about, the value to consumers of TSET research spending. Thus, while the analysis of the previous section was aimed at measuring an impact on gross state product, the analysis of this section tries to provide insight regarding an impact on state-level welfare and well-being. Policy oriented research, such as economic cost-benefit analysis, needs to estimate the value of an investment at a level of aggregation and abstraction appropriate to the decision being made. At the level of a procedure, if a doctor is deciding whether or not to run a test for a patient, her estimation of costs versus benefits should be defined in terms of the expected treatment outcomes for her patient. At the level of a staffing decision, if a project manager is deciding whether or not to hire an additional employee, she bases her decision on the expected contribution versus cost of that employee. At the level of an organizational expansion decision, if a pharmaceutical company is deciding whether or not to fund an additional project, it trades off the costs versus benefits of that level of the project. At the legislative level, if a state is deciding on an appropriate level of research funding, then its leaders need to evaluate the benefits versus costs of increased or reduced funding relative to the level of the aggregate budget. Since an estimation of the value of medical research in Oklahoma is, by its nature, an aggregate-level assessment, any such assessment will necessarily involve some abstraction. Numerous estimations of the value of medical research, at the aggregate level, have been made. An index of this literature, published in the Bulletin of the World Health Organization (2004), lists 31 “key studies” on this topic.9 A widely cited and influential analysis was provided by Kevin Murphy and Robert Topel (1999), economists from the traditionally conservative University of Chicago School of Business.10 Their summary finding is that, from 1970 through 1998 in the United States, improvements in healthcare and clinical practice, combined with gains (or losses) attributable to lifestyle and nutrition, added about as much to our overall well-being as all other improvements in U.S. material wealth combined.11 Given that real per-capita GDP increased by 79% during this period, this is a dramatic claim.12 Nevertheless, when one considers the enormous improvements in treatment for cancer and heart disease that were achieved during these years, it is clear that the magnitude of the benefits from health care innovation over those decades was very large. The Murphy and Topel estimation was made at the aggregate level, by comparing health gains for a population-average individual to health research spending, and then estimating the portion of those gains, about one-third, attributable to health treatment innovation. An alternative estimation, constructed by summing estimates of costs and benefits across individual health research programs, and looking at gains for the 1992-2005 timeframe, was conducted for Australia (Access Economics, 2008). This estimated $2.17 in benefit for every $1.00 spent in health research. A related discussion of the controversy over the competing priorities of traditional research versus diffusion in clinical practice and public health innovation is provided by Rust et al. (2010). Referring to the general tenor of these studies, the return on investment that they estimate is indeed remarkable. To put the scope of these advances in perspective, note that this period saw the advent of colorectal screening (late 1960s through early 70s, an advance that has given us a 40% decline in mortality since 1975), 13 the invention and introduction
103 of the CT and MRI scans (mid 70s), the introduction of chemotherapy after cancer surgery (mid 70s), a virtual cure for testicular cancer (various discoveries now yielding a 95% cure rate), lumpectomy for breast cancer (late 70s), mammography (late 70s), laparoscopic surgery (early 90s), the discovery that melanoma is linked to sun exposure (70s-90s), and dramatic improvements in chemo and radiation therapy (throughout this period). Between 1970 and 2005, the life expectancy of the average American increased by 6.6 years; with 4.7 of these years, over 70% of the increase, being due to reductions in deaths from cardiovascular disease.14 In recent years, the level of total U.S. research and development has hovered at around 2.5% of GDP.15 At the national level, approximately 25% of this spending is in academic R&D and approximately 40% of that 25% is in academic medical R&D. Thus approximately 10% of all U.S. R&D spending, an amount equal to 0.25% of U.S. GDP, is allocated to medical research and development at academic institutions.16 Attributing any substantial portion of this increase in life expectancy to such a small-percentage investment of GDP implies high returns on investment.
Judging Efficiency at the Individual Program Level
Given that one accepts the cost-effectiveness of some public funding for medical R&D, in general, it is still natural to be concerned about the particulars of the allocation and management of such funds. This is the program oversight responsibility that is normally handled at the executive or legislative level of public resource management. For observers who are far removed from the operations of individual research projects, managerial effectiveness is hard to judge. Naturally, individual programs normally define performance metrics and one can use those in an evaluation. An equally valid and perhaps more common approach is to compare the performance of a program to some external benchmark, such as the performance of a similar program in other states. Since special circumstances do exist and do arise, these types of evaluations are subject to error, but they can still be useful. An itemized analysis of the managerial effectiveness of the TSET would be too detailed for this current paper but it should be noted that this approach, of yardstick competition, has been integral to the TSET strategy for research funding. The TSET’s core research initiatives have been in the funding of the Stephenson Cancer Center (Stephenson) and the Oklahoma Center for Adult Stem Cell Research (OCASCR). In both cases, the plan has been that researchers will leverage TSET funding to engage in competitive bidding for external funds. As shown in Table 1, both the Stephenson and the OCASCR have been successful in doing this, obtaining more than $3 of external money for every $1 of state money provided by the TSET. This indicates that third-party evaluators have been judging TSET-funded projects as promising enough to provide external funding at a 3-to-1 ratio. This provides a substantial third-party vote of confidence for TSET funding priorities and TSET project management. Like yardstick competition, another commonly advocated approach is to use market forces for program evaluation. Variations on this approach assume that worthwhile R&D projects will, at least to some extent, be able to obtain private financing. The bulk of pharmaceutical and medical device development funding is provided by this means. The idea is that the years of monopoly pricing that may be made available by patent protection will provide sufficient incentive for private investment for the development of these products.
104 A well-documented problem with this market-based approach is that, although it can be useful for providing incentives for product development (the ‘D’ in R&D), it does not provide much incentive for the foundational research that makes product development and innovation possible.17 The problem is that, if a company is not able to protect the information it generates and sell it as a product, there will be insufficient incentive to invest. The research has to be marketable. Thus, for example, it is easier to find private funding for pharmaceutical research projects than for research on how to effectively organize quit-smoking campaigns. It’s also easier to fund applications of gene therapy than basic research in gene therapy.
Research Financing and the Transfer of Technology to the State
The standard explanation for the underfunding of medical research (and research in general) is that research projects that are worthwhile may not be research projects that will be profitable. This is because many more people, companies, states, or even countries may benefit from a research project than those who paid for it. As mentioned above, no matter how important a discovery may be, if a company cannot market that discovery, the investment in making that discovery will not be profitable.18 A classic example is the discovery of DNA. Although many will benefit, there is no marketable or patentable product coming directly from such a general discovery. Just as the inability to capture the benefits of research may discourage businesses from investing in activities deemed too far from marketability, state legislators and other leaders may be discouraged from investing in research that provides general, as opposed to state-specific, benefits. On the other hand, by investing at the state level in research and development a state may increase its capacity to take advantage of advances made elsewhere. Thus, the process can work in reverse. This is to say that, Oklahoma researchers, in addition to advancing their fields in ways that help cancer or public health research outside of Oklahoma, may also help the state of Oklahoma be better able to import and take advantage of discoveries made elsewhere. Hence, it is not clear that the potential for technology transfer will always cause an underfunding of research. In a state like Oklahoma, where the ratio of research and development expenditure to overall economic activity is low, it is very possible that an expansion of local research provides spillover benefits that outweigh spillover losses. This analysis is written to provide insight and perspective on the overall effectiveness of the TSET research mission. Any kind of critical examination of specific aspects of the management of TSET funds is beyond its scope. Those interested in such detailed evaluations would need to obtain information on individual TSET projects and could, for example, interpret or estimate the extent to which this research is serving the specific needs of Oklahoma. Some of the TSET projects, especially those involving adult stem cell research and general oncology practice, are more in the categories of pure research, and these are activities for which the TSET would expect substantial federal grant funding assistance. Other projects, such as OTRC projects on the epidemiology of tribal smoking in Oklahoma and cigarette taxation at tribal smoke shops in Oklahoma, are more specific to Oklahoma but may also attract federal dollars. Others, such as an investment made by the OTRC in data capture for some community clinics, are oriented toward specific Oklahoma applications that, at least in the near term, are less likely to attract federal dollars.
105 SUMMARY
As a simple metric this study finds that, for each dollar of TSET funding, the Oklahoma researchers hired and projects launched with these funds are obtaining approximately $3 of additional grant funding from external, mostly National Institutes of Health, sources. This commitment of external funding suggests that the projects launched by TSET funding have been viewed, by external evaluators, as worthwhile and likely to bear fruit. The ability of TSET researchers to leverage their funding to obtain federal grant money has also multiplied the impact of TSET research spending on Oklahoma economic activity, with $1 of TSET investment (after accounting for the $3 of federal grant funding it stimulates) tending to generate roughly $4 in total Oklahoma-based research spending. Input-output analysis of the impact of TSET spending on Oklahoma economic activity found that the total impact is nearly double this direct effect. Thus a dollar of TSET funding for research, which yields about $4 in total research funding, is estimated to yield roughly $7.30 in total economic impact. As discussed in the body of this paper, however, the conventional interpretations of such economic impact multipliers can be misleading. This is because the impact on Oklahoma state product, i.e., the impact on Oklahoma GDP, is less than the total economic impact of these programs. After discounting the extent to which increased Oklahoma spending is served by out-of-state suppliers, the net impact of TSET spending on Oklahoma state product, of $3.72 per dollar of TSET spending, is close to the above $4 of direct impact. One can thus think of the stimulus effect of TSET spending on research as being roughly equal in magnitude to the sum of direct TSET funding plus leveraged dollars in federal and foundation grants that are brought to Oklahoma because of TSET research funding. The input-output analysis was used to estimate the impact on economic activity that is attributable to TSET spending. Since this economic activity contributes to overall state product, the stimulative impact of this spending is commonly thought of as a benefit. Alternatively, one can interpret the benefit of a program in terms of the value of the goods and services it provides or, in this case, its estimated contribution to Oklahoma health and welfare. Estimating the market value of basic research is a notoriously difficult problem. Take for example, the discovery DNA. Since the results of this discovery are provided free-of- charge to researchers working on downstream applications, the discovery has no observable market price. Still, nearly all would agree that the discovery of DNA was valuable. To discuss the benefit of TSET research spending in Oklahoma, this paper thus explored two perspectives. The first compared Oklahoma spending on medical research as a share of gross state product to national levels of spending. This established that Oklahoma is consistently low in medical research funding. This suggests that increases in Oklahoma spending may stimulate the importation of advances made elsewhere for applications in Oklahoma, and that additional spending may also help improve the profile of the state as a destination for health care researchers and professionals. The second perspective looks at the value of medical advances to society. Here one sees that general spending on medical R&D has yielded a very high return on investment. Thus, it seems likely that, given the validation implicit in the ability of TSET-funded projects to obtain large amounts of external support, TSET research spending is likely to provide a high return on investment. These are perspectives, not measurements, of the likely effectiveness of TSET
106 spending on research. They are necessarily abstract, aggregate-level ways of looking at what kind of return Oklahoma can expect to get to get from the TSET medical research spending program as a whole. Micro-level assessments of the value of individual TSET projects would require review at the project level. This paper has not attempted such project-level evaluations.
107 REFERENCES
Access Economics. 2008. Exceptional returns: The value of investing in health R&D in Australia II. Canberra (Australia): Australian Society for Medical Research. Available: www.asmr.org.au/ExceptII08.pdf (accessed 2009 Jan. 21). Arrow, K. 1962. “Economic Welfare and the Allocation of Resources for Innovation.” In Richard Nelson (Ed.), The Rate and Direction of Inventive Activity. Princeton Uni- versity Press, Princeton, N.J. Buxton, M., S. Hanney, and T. Jones. 2004. “Estimating the Economic Value to Societies of the Impact of Health Research: A Critical Review.” Bulletin of the World Health Organization, 82: 733-739. Gilriches, Z. 1984. R&D, Patents, and Productivity. National Bureau of Economic Re- search, University of Chicago Press, Chicago, IL. Miller, R., and P. Blair. 2009. Input-Output Analysis: Foundations and Extensions, 2nd Edition. Cambridge University Press. Murphy, K. and R. Topel. 2003. “The Economic Value of Medical Research.” In Kevin M. Murphy and Robert Topel, (Eds.), Measuring Gains from Medical Research: An Economic Approach. University of Chicago, Chicago, IL. ______. 2006. “The Value of Health and Longevity.” Journal of Political Economy 114: 871-904. Muskin, S. 1979. Biomedical Research: Costs and Benefits. Ballinger Publishing, Cam- bridge, MA. National Science Board. 2012. Science and Engineering Indicators 2012. Arlington VA: National Science Foundation (NSB 12-01). Rust, G., D. Satcher, G. Fryer, R. Levine, and D. Blumenthal. 2010. “Triangulating on Success: Innovation, Public Health, Medical Care, and Cause-Specific US Mortality Rates Over a Half Century (1950-2000).” American Journal of Public Health 100: S95-S104. State Science & Technology Institute. 2012. “R&D as a Percentage of Gross Domestic Product by State, 2003–2008.” SSTI Digest 17(4). U.S. Department of Commerce, Bureau of Economic Analysis, 1997, Regional Multipli- ers: A User Handbook for the Regional Input-Output Modeling System (RIMS II), 3rd ed. U.S. Government Printing Office. U.S. National Science Foundation, National Center for Science and Engineering Statistics. 2011.Academic Research and Development Expenditures: Fiscal Year 2009. Detailed Statistical Tables NSF 11-313. Arlington, VA. Available at http://www.nsf.gov/statis- tics/nsf11313/. ______2013a. National Patterns of R&D Resources: 2010–11 Data Update. De- tailed Statistical Tables NSF 13-318, Arlington, VA. Available at http://www.nsf.gov/ statistics/nsf13318/. ______2013b. Higher Education Research and Development: Fiscal Year 2011. Detailed Statistical Tables NSF 13-325, Arlington, VA. Available at http://www.nsf. gov/statistics/nsf13325/.
108 ENDNOTES
* Laux is a current recipient of Oklahoma TSET funding and the research for this paper was originally commissioned as an input-output analysis to measure the impact of TSET research funding on Oklahoma economic activity. The decisions to expand this effort, to include commentary on the potential welfare impacts of TSET research spending, and to refine the analysis for publication, were made by the authors. 1 For information on the Master Settlement Agreement see the website of the National Association of Attorneys General, http://www.naag.org/tobacco.php (referenced on 3/12/2014). 2 The 2009-2010 numbers come from National Science Foundation, 2013a (NSF 2013a). Pre-2009 numbers are from State Science and Technology Institute (SSTI), 2012, which sources Census Bureau special tabulations (2011) of the 1989–2008 business information tracking series. Data for FY 2011, are available for academic R&D spending from the National Science Foundation and for state economic activity levels from the Bureau of Economic Analysis. Using these 2011 data, Oklahoma ranked 47th among the states in terms of dollars spent in academic R&D per dollar of gross state product. 3 From NSF, 2013b, Table 15. 4 Almost 90% of the academic R&D in the medical sciences reported to the National Science Foundation by universities in Oklahoma was attributed to the University of Oklahoma (OU). Within OU, 47% of total reported life sciences R&D was in the medical sciences. Thus, even within OU, the percentage allocation to medical sciences is below the national average. 5 Op. cit., NSF, 2013b. This calculation requires a compilation of data for all Oklahoma Universities from Table 15. 6 The break out of which components of business output and purchasing fall within which industry is tracked by North American Industry Classification System (NAICS) code. The census and other surveys then acquire and tabulate data by NAICS code. 7 For an introduction to input-output analysis, see Miller and Blair, 2009. A handbook covering the details of RIMS II is available from the U.S. Department of Commerce, Bureau of Economic Analysis, 1997. 8 The induced effect, as estimated, takes increased salaries, from both the direct effects on research spending and the indirect effects, into account. 9 Buxton et al., 2004. 10 Murphy and Topel’s analysis, “The Economic Value of Medical Research,” was first published by the Lasker Foundation, in 1999, http://www.laskerfoundation.org/reports/ pdf/economicvalue.pdf. It was later published as Murphy and Topel, 2003. Murphy and Topel, 2006, provides an update on these arguments. Mushkin, 1979, is another influential work on the value of medical research. While Murphy and Topel base their estimates on willingness to pay per life year gained, Mushkin’s work is based on productivity gains, and estimates a 47% average return on investment from medical research conducted between 1930 and 1975. 11 Murphy and Topel, 1999, p 3. 12 This, along with many other economic data series, is available from the Federal Reserve Bank of Saint Louis, in their “FRED” collection of economic data. The FRED label for this specific series is “USARGDPC”, accessed on 25 Nov 2013 at http://research.stlouisfed. org/fred2/series/USARGDPC.
109 13 http://www.cancerprogress.net/timeline/major-milestones-against-cancer, visited 13 March 2014. This website is maintained by the American Society of Clinical Oncology. 14 NIH Research Portfolio Online Reporting Tools, “Heart Disease,” http://report.nih.gov/ nihfactsheets/ViewFactSheet.aspx?csid=96, visited 13 March 2014. 15 SSTI, 2012. 16 National Science Board. 2012. Note that this is 40% of academic R&D in all fields and thus the bases for these calculations are different than for those referenced in footnotes 1, 2, and 4. 17 For a classic paper on this, see Arrow, 1962. Griliches, 1984, provides a general introduction to this line of research. 18 Ibid., Arrow, 1962.
110 TABLE 1: TSET AND RELATED EXTERNAL RESEARCH SPENDING ($'s, millions) FY12 FY13 FY14* Stephenson, Total 19.8 27.9 29.1 TSET 4.2 7.4 6.8 NIH/Foundations 15.6 20.5 22.3 OCASCR, Total 2.1 7.6 7.6 TSET 2 2 2 NIH/Foundations 0.1 5.6 5.6 Grand Total 21.9 35.5 36.7 *Based on mid-year receipts.
TABLE 2: TSET ECONOMIC IMPACT Economic impact of TSET dollars spent on research (employment, $'s millions) FY12 FY13 FY14 Employ- $ Value $ Total Employ- $ Value $ Total Employ- $ Value $ Total ment Added Efffect ment Added Efffect ment Added Efffect Direct 149 10.3 21.9 242 16.7 35.5 249.9 17.3 36.7 Indirect 74.2 4.8 8.5 120 7.8 13.8 124.3 8.1 14.2 Induced 71.9 5.2 9.0 117 8.4 14.5 120.6 8.7 15.0 Total 295.2 20.3 39.4 479 33.0 63.8 494.7 34.1 66.0
111 112 HOW NON-PROFIT INSPECTION SERVICES CAN CORRECT FOR CREDENCE GOOD TYPE MARKET FAILURES
John McCollough, Lamar University
ABSTRACT
The problems associated with asymmetric information and credence goods are a common worry to consumers who require the services of technicians with expert knowledge. A study was designed to see if the concerns of consumers were justified. Specifically, the study looks at two different samples of vehicle owners and the repair costs associated with a vehicle state safety inspection. In one sample the vehicles were inspected by a non-profit, state affiliated inspection station while in the second sample the vehicles were inspected by a for-profit vehicle inspection station. The results suggest that those vehicles inspected at a for-profit inspection station had higher repair costs than those vehicles inspected by a non-profit vehicle inspection station. JEL Classification: D82, L15, D8
INTRODUCTION
The issue of credence goods is a special case of asymmetric information and as such, it can lead to market failure. More specifically, credence goods deal with service goods provided by a technician with expert knowledge and the consumer’s knowledge is much less than the technician’s (Rasch and Waibel, 2012). There are many common examples of this. Because of this asymmetric knowledge between the technician and the consumer, there can be an incentive for the service provider to exploit the consumer, and as a result of this exploitation a market failure can arise. With respect to vehicle repairs, Schneider, (2012) estimates the welfare loss in this market at $8.2 billion. It could be that a majority of technicians within any one particular trade are honest, but if perceptions spread among consumers that this trade group has a proclivity toward exploitation then less service will be demanded by consumers than is socially optimal (Dulleck and Kerschbamer, 2006). As a result, this market failure may even result in environmental damage as consumers decide to forgo the chance of being exploited and dispose of a product that could have been repaired for further reuse (McCollough, 2010). There are many common examples of credence goods. As suggested above, one example would be services provided by an auto mechanic. Indeed, vehicle repairs rank first in customer complaints. However, there are many other examples which range from services provided by your local roofer or plumber or even services provided by the medical profession. Because of the consumer’s reliance on the mechanic’s expert knowledge, the consumers are at an information disadvantage and can, therefore, be easily exploited.
113 Exploitation can take the form of overcharging for service, providing more service than is needed, or perhaps even charging for services that never took place (Darby and Karni, 1973,Webbink,1978). The objective of this paper is to find evidence of this type of exploitation and to find out if consumers’ fears and suspicions are justified. This paper also attempts to quantify the exploitation. In addition, this paper will also show how state-affiliated agencies, acting in the role of safety inspectors without profit motives, can play an important role in correcting the market failure associated with credence goods. The hypothesis set out in this paper is that repair costs associated with vehicle safety inspections provided by quasi-governmental agencies will be statistically less than repair costs associated with inspections provided by privately owned, for-profit service stations. The reason for this is that for-profit service stations have an incentive to provide more service than is required. For example, a for-profit service station might require brake work or perhaps a tire replacement when, in fact, these services are not really needed in order for the vehicle to pass inspection. Worse yet, the for-profit service station might require certain repair work before the vehicle can pass the safety inspection, but then never provide the service, charging the customer for work that never took place. On average, any difference in repair costs should represent the cost of the market failure. A test was designed which compares the repair costs associated with vehicle safety inspection for residents from the state of Pennsylvania and for residents from New Jersey. In Pennsylvania, vehicle owners must have their vehicle inspected by a for profit service station, while in New Jersey the residents can choose to have their vehicle inspected by either a for-profit service station or a not for-profit vehicle inspection station. When New Jersey residents have their vehicle inspected by a non-profit, state-affiliated, vehicle inspection station, the vehicle is actually inspected by a for-profit, private firm that has been contracted to perform all state safety vehicle inspections. Safety inspectors do not work as state employees, rather they work for the firm which provides the inspections. Neither the firm nor the inspectors have a profit motive. These inspectors can only tell the vehicle owner what needs to be fixed before the vehicle can pass inspection. The inspection stations are prohibited from performing any repairs. The vehicle owner will then fix the problem at a service station of his or her choosing and then come back to the state-affiliated inspection station for an inspection sticker as proof that the vehicle passed its safety inspection. Following this introduction, the paper gives an overview of the current literature, as well as a presentation of the significant studies conducted in this area. After the literature review, a description of the data and empirical model used in this study will be forthcoming. A discussion of the empirical results will then follow. Finally, the paper ends with a conclusion and policy discussion. Table 1, which is at the end of the paper presents the empirical results.
LITERATURE REVIEW
Due to the nature of credence goods, and the fact that the expert knowledge required to perform the service is asymmetric, the services provided can often be price insensitive with low price elasticity’s (Peppers and Rogers, 2006). The lower the price elasticity for the service, the easier it is for disreputable service providers to take advantage of the consumer. The literature typically cites lack of competition as the cause for price insensitivity.
114 Geographic locations are a prime determinant in how competitive vehicle repairs and vehicle inspection services are. Typically, the denser the geographic location, the more competition there is and, hence, the ‘switching costs’ are low. In other words if it is easy for consumers to find other service providers then this makes it more difficult for service providers to overcharge. Rasch and Waibel (2012) state that overcharging for vehicle repairs occurs more frequently in less densely populated, non-competitive locations. They find that non- competitive, low density locations just off the interstate overcharge since there is less chance of repeat business. Customers at these locations are mainly one time customers just passing through. They conclude that in more dense geographic locations where competition is higher, service providers are dependent on repeat business. As service providers seek repeat customers, protection of their reputation can ‘discipline’ service providers, especially when there is a possibility of repeat business by customers (Schneider, 2012). However, Hubbard (2002, pg 466) warned that “Incentives are weaker when consumers are naive about sellers’ private objectives, believe that sellers are homogeneous, or when switching costs are high.” As an example of this, Hubbard (1998) finds that independently owned service stations are more likely to pass vehicles for inspection than chain store service stations, new car dealerships, and tune-up shops, because the latter work on commission whereas the independent shop is motivated by repeat business. He also finds that the more inspectors there are at a service station, the more likely a vehicle is to fail. In addition, Schneider (2012) states that higher quality service can be provided by those technicians looking for repeat business, but the prospect for repeat business must be likely. In a follow-up study, Hubbard (2002) finds that the reputation effect does pay off. More specifically, he finds that consumers are 30 percent more likely to utilize a service station in the future if that service station had passed the vehicle for inspection in the recent past. Biehal (1983) also finds that consumers make choices with respect to auto repair services based on previous experiences with repair facilities. However, with respect to annual state vehicle safety inspections, the desire for repeat business can actually create a moral hazard problem. For example, Hubbard (1998) found that in California private inspection facilities pass vehicles at twice the rate of state inspection facilities, except in cases when the emission repairs are covered under a warranty for late model, low mileage vehicles that are being inspected at new car dealerships. Interestingly, Hubbard also found that the inspection failure rate was even slightly lower when the service provider was located in a more competitive location, and this he attributes to ‘low switching costs’ and the ease in obtaining a second opinion. Schneider (2012) finds that initial diagnostic fees are lower for possible repeat customers. This suggests that when reputation was important, the service provider charged a lower up front diagnostic fee, but Schneider (2012) found no difference in repair recommendations, repair prices or the number of legitimate repairs when the service mechanic was trying to protect his reputation. Schneider concludes that the ability of consumers to discipline service providers with the possibility of repeat business is ‘fruitless’. So, how can consumers protect themselves from unscrupulous service providers? One way is to obtain a second opinion. However, second opinions are usually expensive with respect to either money or time (perhaps both) for the consumer and the service provider, particularly when it is cheaper to provide the diagnosis and the repair service together as opposed to the repair service and diagnosis taking place separately (Emmons,
115 1997). In addition, it is unclear to the consumer if a proper diagnosis was even performed. Pesendorfer and Wolinsky (2003) suggest that in competitive markets with competitive prices, efforts by the service provider to provide proper diagnosis might be sub-optimal. Therefore, barring a second opinion it is difficult for consumers to determine if they actually were taken advantage of because who, other than the service provider, can really judge if service was required or not. With respect to vehicle repairs mandated by an annual vehicle inspection, second opinions can be costly to the consumer, particularly with respect to time. Typically consumers must leave their vehicle for half a day or more with the service provider who is giving the second opinion. Most likely, alternative transportation must be arranged. Customers then find themselves in a dilemma. If the vehicle inspection station does the repair work itself, which is usually the case in Pennsylvania, then the customer must decide to either go ahead and trust the inspector to do the repairs while the vehicle is still queued up. Or, does the customer take the vehicle in for a second opinion, requiring additional time and expense. Another common strategy that a consumer can use is to ask for the old part back after the part was replaced. This helps to prevent fraudulent billing and overcharging for work that was not performed. But it still does not prevent ‘over-treatment’, which is providing more repairs than necessary (Dulleck and Kerschbamer, 2006) States can help to protect consumers from fraudulent repairs by requiring licensing or certification of service providers. Unfortunately, this can increase barriers to competition which reduces competition when more competition should be encouraged. Often time consumer publications and rating agencies such as AAA membership clubs or Angie’s list allow consumers to undertake information searches to help sort out reliable service providers from the unreliable ones. However, Biehal (1983) suggests that consumers are not as proactive in their information search as they need to be. For example, from a survey of customers who recently had their vehicles repaired Biehal found that 31.7 percent of the respondents felt their bills were unreasonable and one-fourth were dissatisfied with their service. But, the level of customer dissatisfaction decreased as the amount of customer external information for auto repair services increased. Whether or not a repair will fall into the hands of a reputable or disreputable mechanic, the consumer needs to weigh the expected benefits against the cost of a repair. The expected benefits of a repair includes both an expectation that a repair will be completed correctly as well as an expectation that the product will have its useful life extended. The more trust a consumer has in a mechanic (either because the mechanic is certified, or has an excellent “word of mouth” reputation, or has been endorsed by a rating agency) the higher the expectation that a repair will be made correctly. If this expectation is low then it is more likely that the consumer will forgo the repair in favor of choosing to replace the product (McCollough, 2010, pg 189)
DATA DESCRIPTION AND EMPIRICAL MODELING
An empirical test was designed to see if the repair bills associated with vehicle state inspections are statistically different for vehicle owners who go to a non-profit, state- affiliated inspection station as opposed to those who go to a for-profit, private inspection station. Obviously, if the repair bills at the private inspection stations are statistically higher then this would suggest evidence of market failures due to asymmetric information. In other words, according to the literature, technicians with asymmetric and expert information
116 regarding the repair and maintenance of a product are thought to have an incentive to cheat the customer and are in fact doing so. This then would suggest that there is role for government with respect to vehicle inspection services because the government would be able to cut back on unnecessary repair bills for the consumer by providing an initial diagnostic and inspection service as in the case of New Jersey’s vehicle safety inspection program. On the other hand, if repair bills associated with vehicle inspections from a private vehicle inspection station are not statistically different from the state-affiliated vehicle inspection station, then this might suggest that there is no need for governments to be involved in the vehicle inspection business. Finally, if the repair charges associated with vehicle inspections are statistically less at private vehicle inspection stations than for the state-affiliated vehicle inspection stations, then one might conclude that the reputation effect is at work and that private inspection stations are working hard to keep customers satisfied. However, it is very troubling to think that private inspection stations could possibly be overlooking necessary and important repairs at inspection time because they are afraid of losing potential long term clients. On the other hand, it is just as unsettling to think that the state-affiliated inspectors could be overlooking necessary and important repairs which are being caught by private sector inspectors. The data for the empirical test was taken from the 2005 BLS annual consumer expenditure survey. In this data set households are chosen at random from around the country and the head of the household keeps a bi-weekly diary on day to day expenditures. In addition, the head of household responds to a detailed monthly survey with respect to purchases that are not routine and do not occur on a daily or weekly basis. During the interview the respondents are asked to list their vehicle repair expenditures as well as annual vehicle registration fees and vehicle inspection fees. Additional information is also collected on the vehicle’s make and model, age and mileage, as well as if the vehicle was purchased new or used. Respondents to the survey from both New Jersey and Pennsylvania were chosen for the empirical test. Both New Jersey and Pennsylvania have an annual vehicle inspection program. However, the major difference between the two states is that Pennsylvania requires its residents to have their vehicle inspected once a year by a private inspection station. These privately owned service stations could be a large chain of service stations, or proprietarily owned service station, or maybe even a car dealership. In this case, the vehicle owner pays the for-profit vehicle inspection station a fee for the state inspection. If a repair is required to pass the inspection, the vehicle owner can then opt to have the inspection station do the repairs or have a different service station do the work. The car is then re-inspected, and if it passes, the vehicle gets its annual inspection sticker. Most vehicle owners simply choose to have the original inspection station perform the repairs since it more convenient and will save time. New Jersey residents, on the other hand, can opt to have their vehicle inspected by one of many state-affiliated vehicle inspection stations located around the state or by a privately run inspection station. In the past the state of New Jersey would actually provide inspection services as an alternative to the privately run inspection stations. However, at the time the survey was conducted, New Jersey no longer provided the inspection service itself. Instead, they sub-contracted this service out to a private firm. This firm is prohibited from performing any repairs; they only provide the inspection service. Therefore, in the case of New Jersey, there is no incentive to cheat the vehicle owner at a state-affiliated inspection station by requiring unnecessary repairs. If the inspectors at the state-affiliated inspection stations find a problem with the vehicle, then the owner must go to a service
117 station of his or her choice, have the problem fixed and return to the inspection station for a final inspection and inspection sticker. There is no charge to New Jersey residents who use this inspection service, but there is an inspection fee for those New Jersey residents who opt to use the inspection service of a private inspection station. Therefore, the overwhelming majority of residents in New Jersey simply have their cars inspected by the state affiliated inspection stations. Other than New Jersey residents having the option to go to a state-affiliated inspection station, there are two other important differences between New Jersey and Pennsylvania. First, in New Jersey residents are required to have their vehicle inspected only once every 2 years as opposed to Pennsylvania where vehicles are inspected annually. Secondly, New Jersey does not mandate vehicle inspections for vehicles that are less than five years old. Only New Jersey and Pennsylvania respondents who had reported owning only one vehicle were selected for the empirical test. The reason for this is that from the data set it is impossible to determine which vehicle was being inspected for households with two or more vehicles. New Jersey respondents were also deselected from the data set if their vehicle was newer than 5 years old since, as previously stated, those vehicles are not required to have an inspection. New Jersey residents are only required to have their vehicles inspected every other year, while in Pennsylvania the vehicle must get inspected once a year. Therefore, the only difference, for example, between a 2005 Honda CRV owned by a New Jersey resident and a 2005 Honda CRV owned by a Pennsylvania resident is that the New Jersey vehicle had not been inspected for two years while the Pennsylvania vehicle was just inspected the year before. Because of this fact one would expect that repair bills for Pennsylvania vehicles to be half the amount of repair bills for New Jersey vehicles, since Pennsylvania vehicles were just inspected the year before. To correct for this discrepancy the repair costs reported by Pennsylvania residents were doubled. A total of a 128 vehicles were selected for the empirical test. There was a difference in the number of vehicles selected by state (ie, 120 for Pennsylvania and 28 for New Jersey) The difference in vehicles by state results from the fact that vehicles 5 years old or younger do not need to be inspected in New Jersey and vehicles in Pennsylvania are inspected twice as often as those vehicles in New Jersey. Various vehicle repair bills were cumulated and totaled for each survey respondent. The total of the repair bills per vehicle constitutes the explanatory variable. The following types of repairs were included as the explanatory variable; brake work, tire repair, tire purchases and mounting, front end alignment, wheel balancing and wheel rotation, steering or front end work, electrical system work, engine repair or replacement, exhaust system work, engine cooling system work, clutch or transmission work, motor tune-up, battery purchase and installation, and finally, other vehicle services, parts, and equipment. Survey respondents reported other types of vehicle repairs. However, these repairs were most likely not associated with passing a vehicle inspection, such as air conditioning repair, tune-up, body work, or radio repair.
The empirical test is modeled as follows. C = b1(S) + b2(F/D) + b3(N) + b4(R) + b5(MSRP) + b6(Y) + b7(M*A) The variables are defined as follows: C = This is the total of repair bills during the month of the vehicle owner’s annual state inspection as in the case of Pennsylvania residents or during the month of the annual car registration for New Jersey residents.
118 S = A value of ‘0’ was assigned to vehicles owned by New Jersey residents and a value of ‘1’ was assigned to vehicles owned by Pennsylvania residents. F/D = A value of ‘0’ was assigned to a vehicle if it was manufactured by a foreign car manufacturer such as Toyota or Volvo. A value of ‘1’ was assigned to the vehicle if it was manufactured by a domestic car manufacturer such as Ford. These values were assigned regardless of where the vehicle was manufactured. N = A value of ‘0’ was assigned if the vehicle was purchased by the owner as new and a value of ‘1’ was assigned to the vehicle if the owner purchased it as used. R = This is a general vehicle reliability index found on the website ‘autos.msn.com’ for any specific make, model, and year. This rates the reliability of the vehicle’s serviceable items such as the engine, transmission, brakes, steering and suspension, etc. MSRP = This is the manufacturer’s suggested retail price found on the website ‘autos.msn. com’ for any specific make, model, and year Y = This represents the average number of miles driven per year. M = This represents the total number of miles on the vehicle.
EMPIRICAL RESULTS AND ANALYSIS
The purpose of this paper is to find empirical evidence of problems associated with asymmetric information and credence goods. Therefore, determining the factors that explain vehicle repair costs associated with state inspections is not the reason for running the empirical tests. Rather the empirical test is more narrowly focused than that, and that is to find out if the cost of vehicle repairs associated with a state inspection is statistically different depending on whether the vehicle is inspected by a private inspection station or a state-affiliated inspection station. Therefore, in this regression the explanatory variable of interest is the state variable. Table 1 reports the regression results for the empirical model. The regression results show that the state variable is positive and significant at the 6.8 percent level. The positive coefficient of $166.17 suggests that those vehicles inspected by privately owned service station can expect to have, on average, an additional $166.17 in repair bills during the month of their state inspection. This amount is relevant on a bi-annual basis since, as stated above, repair bills for Pennsylvania residents were doubled to account for the fact that they are inspected twice as often as New Jersey vehicles Since New Jersey residents are having their vehicles inspected every other year, then they can expect to save $166.17 every time they go in for a state inspection. The finding from the regression analysis supports the literature on credence goods and asymmetric information, meaning that service technicians with superior and expert knowledge over the customer have an incentive to cheat the customer, and they are, in fact, doing so. This ‘cheating’ or even the belief that vehicle owners will be cheated is what creates the market failure. It should be pointed out that Poitras and Sutter (2002) have reported the results of a similar study which looks to see if vehicle inspections can increase vehicle repair cost. They use a dataset of 733 vehicle inspections for vehicles that were 12 years or older in 50 different states between the years of 1953 – 1967. They find that state inspections do not increase repair costs (ie, repair revenue for the inspecting facility). The difference in these results and the results reported here are most likely attributed to the time period used in both studies, the type of vehicles used in the study, and the fact that this study categorizes
119 vehicles inspected into those inspected by a for-profit or a non-profit inspection facility. The regression results also show vehicles produced by foreign manufacturers have statistically higher repair costs during the month of their state inspection at the 9.3 percent level of significance. This result suggests that repair costs associated with vehicles from foreign manufacturers is $129.63 higher than for vehicles from domestic manufacturers. There could be a number of reasons for this result. First, it could be the case that parts cost more for vehicles from foreign manufacturers as opposed to domestic manufacturers. Second, for whatever reason, it could be that vehicles from foreign manufacturers are slightly more complicated for American mechanics to work on. Perhaps American mechanics have a good deal more experience and training with vehicles made by domestic manufacturers. The only other variable that was statistically significant was total mileage. It was positive and significant at the 5.2 percent level. The value of the coefficient suggests that for each addition mile on the vehicle, owners can expect to pay an additional $.002. This result should be the most intuitive of all the explanatory variables. The higher the mileage on the vehicle, the more it cost to maintain The remaining variables all turned out to be insignificant, including the number of miles driven over the most recent year. The coefficient of determination was .229. The consumer expenditure survey data lacked one or two other relevant pieces of information which could have increased the coefficient of determination. This would be the labor rates charged per vehicle repair shop and information regarding each shop’s productivity. However, as stated above, the primary focus of this paper is to find empirical evidence on the problems associated with asymmetric information and credence goods. The data set yields socio-economic characteristics on the survey respondents. However, it is interesting to point out that each of these socio-economic characteristics was highly insignificant. Meaning that, on average, a survey respondent’s race, gender, age, education level or income level was insignificant in determining how much he or she paid for vehicle repairs in the month of the state inspection. Suspicions that a vehicle owner was being taken advantage of based on his or her gender, race, age, etc. were unfounded in this empirical test.
CONCLUSION
Credence good types of services provided by technicians that are characterized as yielding asymmetric information leave the consumers at an information disadvantage. This creates an opportunity for unscrupulous service providers to take advantage of the consumer. Indeed, from time to time one hears stories in the news media of consumers being taken advantage of. As a result, a market failure arises. An empirical test was designed and reported on in this paper to see if the consumer’s fears are warranted. The empirical test in this paper looks at vehicle owners who have had their vehicles inspected by either a for-profit inspection service station or a state affiliated, non-profit inspection station. The results from this test indicate that if your vehicle is inspected by a state-affiliated, non-profit, inspection station rather than a for-profit, inspection station, the vehicle repair bills will be less. The amount, as reported from the regression results, is $166.17 on a bi-annual basis. This is a meaningful savings for owners who utilize the non-profit, state affiliated inspection stations. However, there are a number of factors to consider when interpreting these results. First, there is the cost of providing vehicle
120 inspection services. Those vehicle owners in New Jersey do not directly pay an inspection fee whereas a direct fee is paid to the for-profit inspection station in Pennsylvania (when accumulating repair costs, the cost of the vehicle inspection fee was not included). The bi- annual savings of $166.17 in repair bills from New Jersey residents have to be compared to the cost of running the vehicle inspection stations in New Jersey. The costs of running the state-affiliated’ inspection stations are funded by the New Jersey taxpayers. Secondly, we do not know for sure if vehicle owners in Pennsylvania are getting more thorough and higher quality inspections. Perhaps the state-affiliated inspectors from New Jersey are simply shirking their duties and passing vehicles that, in reality, do require some repair and maintenance. There is no way to tell for certain. Although it is beyond the scope of this paper, it might be possible to look at highway traffic accident data and see if there is a correlation between increased accidents and vehicles inspected by state-affiliated inspection stations. However, the literature with respect to this topic is inconclusive and shows conflicting results between vehicle safety inspections and their effectiveness at preventing accidents (For example, see Fosser 1992, White 1985, or Merrell, D., Poitras, M. & Sutter, D. 1999) Finally, it need not be that inspections stations that simple provide diagnostic inspection services only, do not have to be state-affiliated or state operated. This type of service could just as easily be provided by the private sector, and perhaps the private sector could perform the services more efficiently than the state affiliated or state inspection station. If so, then perhaps it might just be possible that these private sector companies could run their services more efficiently than the state affiliated inspection facilities.
121 REFERENCES
Biehal, G. J. (1983). Consumers’ prior experiences and perceptions in auto repair choice. The Journal of Marketing, 82-91. Darby, M. & Karni, E, (1973). Free Competition and the Optimal Amount of Fraud. Journal of Law and Economics. 16, 67-88. Dulleck, U., & Kerschbamer, R. (2006). On doctors, mechanics, and computer specialists: The economics of credence goods. Journal of Economic Literature, 5-42. Emons, W. (1997). Credence goods and fraudulent experts. The Rand Journal of Economics, 107-119. Fosser, S. (1992). An experimental evaluation of the effects of periodic motor vehicle inspection on accident rates. Accident Analysis & Prevention, 24(6), 599-612. Hubbard, Thomas (1998), An Empirical Examination of Moral Hazard in the Vehicle In- spection Market. Journal of Economics. Vol. 29, no. 2 pg 406 – 426. Hubbard, Thomas (2002). Ho Do Consumers Motivate Experts? Reputational Incentives in an Auto Repair Market. Journal of Law and Economics. Vol. 45. No. 2, Part 1. (Oct. 2002), pp.437-468 McCollough, J. (2010). Consumer Discount Rates and the Decision to Repair or Replace a Durable Product: A Sustainable Consumption Issue. Journal of Economic Issues, 44(1), 183-204. Merrell, D., Poitras, M., & Sutter, D. (1999). The effectiveness of vehicle safety inspections: An analysis using panel data. Southern Economic Journal, 571-583. Peppers, D., & Rogers, M. (2006). Consumer Segmentation Strategies. Pricing on Purpose: Creating and Capturing Value, 197. Pesendorfer, W., & Wolinsky, A. (2003). Second opinions and price competition: Inefficiency in the market for expert advice. The Review of Economic Studies, 70(2), 417-437. Poitras, M., & Sutter, D. (2002). Policy ineffectiveness or offsetting behavior? An analysis of vehicle safety inspections. Southern Economic Journal, 922-934. Rasch, A., & Waibel, C. (2012). What drives fraud in a credence goods market?-Evidence from a field experiment (No. 03-07). Cologne Graduate School in Management, Economics and Social Sciences. Schneider, H. S. (2012). Agency problems and reputation in expert services: Evidence from auto repair. Journal of Industrial Economics. Webbink, D. W. (1978). Automobile Repair: Does Regulation or Consumer Information Matter?. The Journal of Consumer Research, 5(3), 206-209. White, W. T. (1986). Does periodic vehicle inspection prevent accidents?. Accident Analysis and Prevention.
122 TABLE 1 – REGRESSION RESULTS FOR THE EMPIRICAL MODEL
Variable Coefficient t – value State 166.17 1.840 Foreign or domestic 129.63 1.692 Purchase new or used 29.19 .394 Reliability index - 21.50 -1.154 MSRP .005 1.009 Annual mileage -.005 1.225 Total mileage .002 1.956 R-sq .229
123 124 THE IMPACT OF CHANGES IN THE DOW JONES INDUSTRIAL AVERAGE LIST ON PRICES AND TRADING VOLUMES Geungu Yu, Jackson State University Phillip Fuller, Jackson State University Patricia A. Freeman, Jackson State University
ABSTRACT
The price-pressure hypothesis (PPH) assumes that a temporary increase (or decrease) in returns and trading volumes occurs around the announcement day when firms are added to (or deleted from) a market index. On September 10, 2013, the Dow Jones Industrial Averages Index Committee announced that Goldman Sachs Group Inc. (GS), Visa Inc. (V) and Nike Inc. (NKE) would be added to the Dow Jones Industrial Average (DJIA) and Bank of America Corp. (BAC), Hewlett-Packard Co. (HPQ) and Alcoa Inc. (AA) would be deleted from the DJIA after the close of trading on September 20, 2013. According to the Index Committee, GS replaced BAC, V replaced HPQ and NKE replaced AA. This event study analyzes the effects that these changes have on the prices and volumes of these stocks. Changes of prices and trading volumes of the firms added to the DJIA are statistically significant enough to support the PPH.JEL classifications:G14
INTRODUCTION
According to S&P Dow Jones Indices LLC (2014), key facts of Dow Jones Industrial Average Index are as follows: 1) The index is maintained by the Averages Committee. Components are added and deleted on an as-needed basis. 2) For the sake of continuity, such changes are rare, and typically occur following corporate acquisitions or other significant changes in a component company’s core business. 3) While stock selection is not governed by quantitative rules, a stock typically is added only if the company has an excellent reputation, demonstrates sustained growth and is of interest to a large number of investors. 4) Maintaining adequate sector representation within the index is also a consideration in the selection process. 5) The index is price weighted. This study examines the composition changes to the DJIA announced on September 10, 2013. The purpose of this study is to determine if recent changes in the DJIA caused any significant impact on the prices and volumes of stocks that were either added to or deleted from the DJIA. The efficient market theory (EMT) suggests that including a stock in or removing a stock from the DJIA should not affect either a stock’s price or volume if the change does not convey any new information. However, the price-pressure hypothesis (PPH) assumes that a temporary increase (or decrease) in returns and volume results as firms are added to (or deleted from) an index around the announcement day. Several studies
125 have been conducted to examine these important issues. Prior studies focused on changes in the composition of the S&P 500, FTSE 100, Australian All Ordinaries and DJIA. This paper is organized as follows: the first section is a literature review; the second section describes the methodology; the third section explains the findings; the final section sets forth a summary and conclusion. There are three tables presenting the key descriptive and analytical statistics of this study.
LITERATURE REVIEW
Harris and Gurel (1986) confirmed the PPH in examining prices and volume surrounding changes in the composition of the S&P 500. The PPH assumes that investors who accommodate demand shifts must be compensated for the transaction costs and portfolio risks that they bear when they agree to immediately buy or sell securities, which they otherwise would not trade. The PPH and EMH are similar in that both suggest that long-run demand is elastic at the full-information price, but they differ in that the PPH hypothesizes that short-term demand curves may be less than perfectly elastic. They found that immediately after an addition is announced, prices increased by more than 3 percent, but the increase was nearly fully reversed after two weeks. Lamoureux and Wansley (1987) supported the PPH. By examining market responses to changes in the S&P 500, they found that stocks added to (or deleted from) the index experienced a significant positive (or negative) announcement day excess return. The average announcement day trading volume for firms added to the S&P 500 was substantially larger than the average pre-period trading volume of traded stocks. Pruitt and Wei (1989) also supported the PPH by showing that institutional holdings increased when listing occurred. Sahin (2005) analyzed the valuation and volume effects of 219 additions of Real Estate Investment Trusts (REITs) to various S&P indices since 2001. Salin’s analysis supported the PPH. The study found that the inclusions of REITS in various S&P indices experienced approximately a 5 percent market-adjusted abnormal return on average at the time of the announcement. Chan and Howard (2002) examined additions to and deletions from the Australian All Ordinaries Share Price Index (AOI). They found significant changes in daily returns and volume around the change date, which supported the PPH. They believed their findings, which were contrary to some findings based on the S&P500, were due to institutional differences in how changes in the composition of the AOI and S&P 500 are determined. Gregoriou and Ioannidis (2006) examined changes in the FTSE 100. They found no evidence that suggested that changes of the FTSE 100 supported the PPH. However, their findings were consistent with the information cost and liquidity explanation in that inclusion in (or deletion from) the FTSE 100 list increased (or decreased) the likelihood that they would be widely followed. Their study supported Merton’s attention hypothesis in that the changes in the FTSE 100 affected the likelihood of the market’s attention. Beneish and Gardner (1995), examining changes in the composition of the DJIA, found that the price and the trading volume of newly added DJIA firms were unaffected. However, firms removed from the index experienced significant price declines, which was consistent with the PPH. They believed that the market demanded an extra-return premium for higher trading costs due to relatively less information available to those stocks removed from the index. This suggested that the short-term demand curves of firms removed from
126 the index would not be perfectly elastic, supporting the downward-sloping demand curve hypothesis. Poloncheck and Krehbiel (1994) compared the price and volume responses associated with changes in the DJIA and Dow Jones Transportation Averages. They found that firms added to the roster of the DJIA experienced significantly positive abnormal returns and significantly greater trading volume on the event date; however, firms added to the Transportation Average experienced neither event period abnormal returns nor increased trading volume. They attributed the lack of significant effects on the Transportation Average to much less media attention, supporting Merton’s (1987) attention hypothesis.
METHODOLOGY
Table 1 shows profiles of the additions and deletions of DJIA constituents effective with the close of trading on Friday, September 20, 2013 with the key descriptive statistics as of close on Feb. 14, 2014. The actual trading with the new constituents began on Monday, September 23, 2013. The average retail price per share of the added stocks is eight times higher than that of the deleted stocks; the market caps and dividend yields are about in the same ranges, but the average P/E ratio of the added stocks is 1.4 times higher than that of the deleted. The significantly higher average price of the new additions means that these stocks will influence the index value in much greater proportion due to the fact that DJIA index is price-weighted. Another conspicuous difference is on PEG, Price/Earnings to Growth ratio. That is, the average PEG of the added stocks is 1.77, which is absolutely superior to that of the deleted stocks, -1.63. An implication of this comparison is that investors and the index observers should pay attention to the PEG ratio in particular for identifying likely candidates for additions or deletions. As shown in TABLE 1, three firms, GS, V, and NKE, were added and three firms, BAC, HPQ, and AA, were deleted. For all six firms, daily stock price and trading volume data were collected from historical data provided by Commodity Systems, Inc. for the period from June 27, 2013 to December 16, 2013, spanning 121 days, 60 days before and after September 23, 2013, the first day of trading reflecting the changes. Actual rates of return data will be calculated for 59 days before and after Monday, September 23, 2013. DIA, DIAMONDS Trust Series I (ETF) was used for market proxy with the data collected for the same period.
An event study was conducted to evaluate the impact on returns and volume on the two portfolios. The market model was used to calculate excess returns or the prediction error as follows:
PEt = Rt – [a + (b*RMt)] (1)
where PEt = the prediction error for market period or day t,
Rt = the logarithmic return of the stock for day t, defined by ln (Pt/Pt-1) or ln (Pt) – ln (Pt-1),
RMt = the logarithmic market return for period t or day t, and a and b are ordinary least squares estimates of the coefficients of the market model.
127 A positive (or negative) prediction error means that the underlying stock price increased (or decreased) more than was predicted. As in Beneish and Gardner (1995) and Gregoriou and Ioannidis (2003), prediction errors are examined over the 120-day period that extends from 59 days before to 59 days after the changes were announced. Average prediction errors, APE, are computed by dividing the prediction errors by the number of firms in the sample on each day t. To assess the presence of abnormal returns, the average prediction errors are cumulated over intervals of k days from t through t+k to obtain cumulative average prediction errors, CAE. That is,
CAEt,t+k = ∑APEi i = t, t + 1, t + 2, t + 3, . . . , t + k. (2)
Following the procedure used by Beneish and Gardner (1995) and Gregoriou and Ioannidis (2003) to test the null hypothesis that CAE equals zero, the following t-statistic with 79 degrees of freedom was computed:
2 1/2 t = CAEt,t+k /[ks APE] (3)
80 2 1 2 s APE = ∑(APEt − APE) (4) 79 t=1 2 where s APE is an equally weighted portfolio variance estimate and APE is the mean average prediction error for the 80-trading-day estimation period. The behavior of trading volume is analyzed based on the procedures used by Beneish and Gardner (1995) and Polonchek and Krehbiel (1994). Three announcement periods are examined: 1) the day of the announcement; 2) the day of the announcement and the day before the announcement; and 3) the day before the announcement, the day of the announcement, and the day after the announcement. First, trading volume is evaluated around the announcement with the mean volume in the prior eight weeks adjusted for changes in the market volume. Then, the mean trading volume for the eight weeks prior to and after the announcement period of DJIA changes (excluding days -1 to +1) are compared. Following the procedure of Beneish and Gardner (1995), Polonchek and Krehbiel (1994) and Gregoriou and Ioannidis (2003), trading volume is examined using the market- volume adjustment approach. The null hypothesis is that this ratio is 1. The relative trading volume, VR, is measured for firm i by the following equation:
VRit = (VOLit/VOLmt) (5)
where VOLit is the natural logarithm of trading volume of security i traded in period t of added (deleted) firms and VOLmt is the natural logarithm of trading volume for DIA in period t. The natural logarithm is used to compensate for the fact that daily volume distributions have been found to be skewed to the right and leptokurtotic (Polonchek and Krehbiel, 1994). Ajinkya and Jain (1989) found that natural log transformations of the volume measures are approximately normally distributed. Following the procedure of Beneish and Gardner (1995) and Gregoriou and Ioannidis(2003), the t-test is used to test the hypothesis of no significant statistical difference.
128 FINDINGS
Table 2 summarizes the results of the tests to determine if stock prices are affected when stocks are either added to or deleted from the DJIA list. As can be seen from Table 2, the stock returns of firms added to the DJIA are significantly affected by their inclusion. For example, the CAE is significant at the 1 % level for the added firms: on the day after announcement day (Day -9) is +2.183% (t = 2.8910); on the day before first trading day (Day -1), +2.626% (t = 3.4767); and on Day +4, +2.111% (t = 2.7962). The stock returns of firms deleted from the DJIA are mostly not affected by their deletion. However, the CAE on the Day +4 for firms deleted from the DJIA is statistically significant at the 5 % level; on the Day -3, +2.1% (t=1.8952). Our findings suggest that changing the composition of the DJIA does provide significant new information or pricing pressure as proposed by the PPH in that the stock returns of the added firms are significantly affected by the decision. Table 3 presents the results of the tests on market-adjusted trading volume effects. For the five sets of trading periods, there are significant increases in trading volume on the days surrounding the first trading day compared to the prior eight weeks for firms that were either added to or deleted from the DJIA. For the case of stocks added, the magnitude of trading volume increases is much more significant around the first trading day after the DJIA list changes become effective, compared to the days around the announcement day (Day -10). For example, the mean VR of days -1 ~ +1 is 7.648 times the benchmark volume, statistically significant at the 1 percent level; the mean VR of days -9 ~ -8 is 3.343 times the benchmark volume, significant only at the 10 percent level. Overall, the volume analyses support the PPH.
SUMMARY AND CONCLUSION
This study examines the 2013 changes to the DJIA that had not previously been examined. In recent years, the stock markets have experienced several operational changes. For example, high-frequency trading has become a dominant force; individual investors are able to quickly alter their portfolios for tactical and/or strategic reasons by engaging in online trading; individuals and institutions can easily invest in DJIA by purchasing Diamonds ETF fund (DIA). Probably because of these changes, significant additional information was provided when stocks were announced to be included in the DJIA list on September 10, 2013. Trading volumes changed significantly both for added and deleted stocks around the first trading day. Therefore, the findings of this study are consistent with prior research findings supporting the PPH. In essence, changing the composition of the DJIA in September 2013 provided the market with significant changes in returns and trading volumes particularly for the added firms.
129 REFERENCES
Ajinkya, B., & Jain, P. (1989). The Behavior of Daily Stock Market Trading Volume. Journal of Accounting and Economics, 11(4), 331–359. doi:10.1016/0165- 4101(89)90018-9 Beneish, M., & Gardner, J. (1995). Information Costs and Liquidity Effects from Changes in the Dow Jones Industrial Average List. The Journal of Financial and Quantitative Analysis, 30(1), 135–157. doi:10.2307/2331257 Chan, W. H., & Howard, P. (2002). Additions to and Deletions from an Open-Ended Market Index: Evidence from the Australian All Ordinaries. Australian Journal of Management, 27(1), 45–74. doi:10.1177/031289620202700103 Gregoriou, A., & Ioannidis, C. (2006). Information Costs and Liquidity Effects from Changes in the FTSE 100 List. The European Journal of Finance, 12(4), 347–360 doi:10.1080/13518470500249340 Harris, L., & Gurel, E. (1986). Price and Volume Effects Associated with Changes in the S&P 500 List: New Evidence for the Existence of Price Pressures. The Journal of Finance, 41(4), 815–829. doi:10.2307/2328230 Lamoureux, C., & Wansley, J. (1987). Market Effects of Changes in Standard & Poor’s 500 Index. The Financial Review, 22(1), 53–69 doi:10.1111/j.1540-6288.1987 tb00318. Merton, R. (1987). A Simple Model of Capital Market Equilibrium with Incomplete Information. The Journal of Finance, 42(3), 483–510 doi:10.1111/j.1540-6261.1987. tb04565.x Polonchek, J., & Krehbiel, T. (1994). Price and Volume Effects Associated with Changes in the Dow Jones Averages. The Quarterly Review of Economics and Finance, 34(4), 305–316. doi:10.1016/1062-9769(94)90016-7 Poterba, J., & Shoven, J. (2002). Exchange-Traded Funds: A New Investment Option for Taxable Investors. American Economic Review, 92(2), 422–427. doi:10.1257/000282802320191732 Pruitt, S., & Wei, K. J. (1989). Institutional Ownership and Changes in the S&P 500. The Journal of Finance, 44(2), 509–513. doi:10.2307/2328603 Sahin, O. F. (2005). The Impact of Standard & Poor’s Index Inclusions on Real Estate Investment Trusts. The Southern Business & Economic Journal, 28. S&P Dow Jones Indices LLC. (2014). Retrieved 14 February 2014, from http://djindexes. com
130 131 132 133 134 US-AUSTRALIA TRADE BALANCE AND EXCHANGE RATE DYNAMICS Matiur Rahman, McNeese State University Muhammad Mustafa, South Carolina State University
ABSTRACT
This paper seeks to explore the dynamics between changes in nominal bilateral exchange rate and nominal trade balance in the US-Australia case. Monthly data are utilized from January, 1995 through June, 2014. The unit root tests find nonstationarity of each variable in level with I(1) behavior. Both variables are cointegrated as unveiled by and tests. The vector error-correction model (VECM) shows unidirectional long-run causal flow from lagged exchange rate changes to the current change in trade balance with interactive short-run feedback effects. The Impulse Response analysis does not reveal any clear patterns. So, the potency of exchange rate policy to influence trade balance remains in doubts for these two countries. JEL classifications: F10, F14, F15
INTRODUCTION
The interaction between trade balance and exchange rate is an important topic of international economics. The traditional method of assessing the impact of currency devaluation was to estimate the well-known Marshall-Lerner (ML) Condition. According to this condition, if the sum of import and export demand elasticities add up to more than unity, devaluation or depreciation could improve the trade balance in the long run. The concept of J-curve was introduced by Magee in 1973. This described the phenomenon of initial deterioration in trade balance in the short run and subsequent improvement in the long run resulting in a pattern of movement that resembles the letter J. since 1973, researchers published numerous academic articles for many countries testing the validity of the J-curve. They produced mixed empirical results. A detailed review of literature on this topic is available in Bahmani-Oskooee and Ratha (2004). Given the implications of the J-curve for the conduct of macroeconomic stabilization polices, its empirical estimation has been a subject of interest. A number of studies have estimated the effect of a change in the real exchange rate on the balance of trade and have confirmed the existence of the J-curve (Artus 1975, Miles 1979, Spitaller 1980, Helkie and Hooper 1987, Krugman and Baldwin 1987, and Marwah and Klein 1996). However, Rose and Yellen (1989), using the data on the U.S. bilateral trade with the G-7 countries as well as the aggregate U.S. trade, did not find any statistically significant evidence for the J-curve. Rose and Yellen’s findings are important because theirs is the first time series econometric study that refutes the empirical validity of the J-curve. The United States is Australia’s most important economic partner country. The trade and investment links have been deepening under the Australia-United States Free Trade Agreement (AUSFTA) since January 1, 2005. Historically, Australia has persistently huge trade deficit with the USA as the record shows for 1985-2012. On average, annual total
135 imports from the USA have been more than double of its annual total exports to USA. The focus of this study is to reassess the dynamic relationship between changes in bilateral nominal exchange rate and trade balance involving the USA and Australia. The rest of the paper proceeds in the following sequence. Brief review of the related literature, empirical methodology, empirical results and conclusions with some policy implications.
BRIEF REVIEW OF THE RELATED LITERATURE
Numerous studies investigated the relationship between exchange rate and trade balance revolving around the J-curve phenomenon that has been analyzed extensively for a wide variety of countries employing different data sets and econometric techniques. This area of research has met with mixed results. Examples of papers finding support for the J-curve include Marwah and Klein (1996), Bahmani-Oskooee and Alse (1994) and Hacker and Hatemi (2003). Evidence of a weak or ‘delayed’ J-curve has also been found by several authors such as Rosensweig and Koch (1988), Yusoff (2007) and Bahmani-Oskooee and Bolhasani (2011). Other authors such as Rose and Yellen (1989), Rose (1991), Hsing (2009), Hsing and Savvides (1996), and Mehmet and Mushtag (2010) have not found any evidence of a J-curve in the data. Using data from 14 countries, Miles (1979) found no evidence for the J-curve effect suggesting that devaluation caused only a readjustment between various accounts of the balance of payments and that it did not improve the trade balance. On the other hand, using a four-country sample, Bahmani-Oskooee (1985) found evidence of a J-curve for Greece, India and Korea, while rejecting the J-curve effect for Thailand. Using a dynamic general equilibrium model, Brissimis and Leventankis (1989) confirmed evidence of the J-curve for Greece. Utilizing new time series econometric methods and a sample of 19 developed and 22 developing countries, Bahmani-Oskooee and Alse (1994) only found evidence of a J-curve effect for four countries (Costa Rica, Ireland, the Netherlands and Turkey). Backus, et al. (1998) found statistically significant evidence for the presence of a J-curve for Japan. Using a similar technique, Demirden and Pastine (1995) found strong evidence of J-curve effects for the USA. In order to mitigate aggregation bias that could result from aggregate data, some studies have moved to bilateral trade data to investigate the J-curve effect. For instance, Marwah and Klein (1996) used quarterly data for the US and Canada with their major trading partners and found some evidence of the existence of the J-curve. Bahmani- Oskooee and Ratha (2004) examined the US trade balance with industrialized countries and found no specific pattern of a J-curve. Wilson (2001) used VAR methods to examine J-curve effects for three Asian countries (Singapore, Malaysia and South Korea), but found evidence of a J-curve only for South Korea. Rose (1990) examined the relationship for a sample of developing countries and found no evidence of the J-curve. Bahmani-Oskooee and Ratha (2004) considered 18 major trading partners of the United States (Australia, Austria, Belgium, Canada, Demark, Finland, France, Germany, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Spain, Sweden, Switzerland and U.K.) and were unable to discover any J-curve pattern in the short run, although real depreciation of dollar revealed favorable effects on the U.S. trade balance in most cases. In contrast, Mahdavi and Sohrabian (1993) found evidence of a delayed J-curve for the USA. Demirden and Pastine (1995) also found evidence of the J-curve for the USA.
136 Kale (2001) found evidence of the J-curve for Turkey. Narayan (2004) concluded that New Zealand’s trade balance exhibited a J-curve pattern following a depreciation of the New Zealand dollar. Kulkarni (1994) found the evidence of the J-curve phenomenon for Egypt and Ghana. In this study, Kulkarni also suggested the possibility of a shifting J-curve phenomenon for these countries over time. In another study, Kulkarni and Bhatia (2002) found the evidence of J-curve in six out of seven different countries (the Philippines, Kenya, Japan, Indonesia, Mexico, China, and Spain) with the exception of China. The dynamics of consumption smoothing and capital formation of small-open- economies of LDCs give rise to the S-curve in the presence of productivity shocks only (Senhadji, 1998). For these countries, the trade balance determines the net foreign exchange receipts while the terms of trade determine their purchasing power. Additionally, Bahmani- Oskooee (1986b) found evidence of a W-curve for the U.S. current account using quarterly data for 1973-1985. This describes that subsequent to depreciation of the dollar, the current account deteriorated for two quarters and then started improving for five quarters, again deteriorated and finally improved. Rahman and Islam (2006) examined the dynamics of Taka-Dollar exchange rate and Bangladesh trade balance using quarterly data for 1972- 2003. They found evidence of J-curve with significant deterioration in the short run and slow improvement in the long run. Several studies have employed Australian data to examine the J-curve phenomenon. Arndt and Dorrance (1987) adopted a descriptive approach to infer that the Australian trade balance has exhibited J-curve behaviour. However, using a more advanced statistical approach, both Flemingham (1988) and Karunaratne (1988) found no evidence of the J-curve for Australia. Utilizing bilateral trade data and cointegration methodology, Bahmani-Oskooee et al. (2005) find a J-curve effect for only 3 of the 23 Australian trading partners which they examined. Another paper by Bahmani-Oskooee and Wang (2006) examined the J-curve effect using bilateral trade data between Australia and the US for 108 industries. They found evidence of the J-curve only for 35 industries.
EMPIRICAL METHODOLOGY
The standard estimating base model is specified as follows: TB = f(ER) Where, TB = US-Australia nominal trade balance, and ER = US dollar per Australian dollar (nominal bilateral exchange rate). Monthly data are used from January, 1995 through June, 2014. In terms of years, the sample period may deem relatively inadequate for meaningful cointegration analyses. However, the use of high frequency monthly data may help partially compensate for this deficiency (Zhou, 2001). Due to the application of monthly data, home country and foreign country GDP data are excluded in this paper for consideration since they are available either annually or quarterly. The data are obtained from the direction of Trade and the International Financial Statistics, published by the IMF. Another data source includes the Australian Bureau of Statistics. Finally, the estimating model in linear form in level is expressed as follows: TB = a + bER + e (1) Prior to testing for cointegration, the time series properties of the variables involved are examined. To test for unit root (nonstationarity) in the variables, the modified Dickey- Fuller test, the modified Phillips-Perron test (Elliot et al., 1996; Ng and Perron, 2001) and their counterpart KPSS (Kwiatkowski, Phillips, Schmidt and Shin, 1992) test for no unit root (stationarity) are implemented instead of the standard ADF and PP tests for their
137 high sensitivity to the selection of lag-lengths. It is important to examine the time series properties of variables since an application of Ordinary Least Squares (OLS) to estimate a model with nonstationary time series data results in the phenomenon of spurious regression (Granger and Newbold, 1974) invaliditating the inferences through the standard t-test and joint F-test (Phillips, 1986). To be cointegrated, nonstationary time series variables must possess the same order of integration, i.e., each variable must become stationary on first- order differencing depicting I (1) behavior. Second, the cointegration procedure, as developed in Johansen (1988, 1992, and 1995) and Johansen and Juselius (1990), is implemented that allows interactions in the determination of the relevant macroeconomic variables and being independent of the choice of the endogenous variable. It also allows explicit hypothesis testing of parameter estimates and rank restrictions using likelihood ratio tests. The empirical exposition of the Johansen-Juselius methodology is as follows: (k-1) ∆Vt= τ+ ΩV(t-1)+ ∑(j=1) Ωj ∆V(t-j)+ mt (2) where, denotes a vector of ER and TB, and. Here, is the speed of adjustment matrix and is the cointegration matrix. Equation (2) is subject to the condition that is less-than- full rank matrix, i.e., r < n. This procedure applies the maximum eigenvalue test and trace test for null hypotheses on r. Both tests have their trade-offs. test is expected to offer a more reliable inference as compared to test (Johansen and Juselius (1990), while test is preferable to test for higher testing power (Ltkepohl, et al., 2001)). However, the Johansen-Juselius test procedure is also not immune to super sensitivity to the selection of lag-lengths. The optimum lag-lengths are determined by the AIC (Akaike Information Criterion), as developed in Akaike (1969). Third, on the evidence of cointegrating relationship between the variables, there will exist an error-correction representation (Engle and Granger, 1987). The vector error- correction model takes the following form: k k ∆TBt= β1 e(t-1)+ ∑(i=1) ϕi ∆TB(t-i) + ∑(j=1) δj∆ER(t-j) + ut (3) Equation (3) corresponds to original equation (1). Here, is the error-correction term of equation (3). If is negative and statistically significant in term of the associated t-value, there is evidence of a long-run causal flow to the dependent variable from the relevant explanatory variables. If δ’s and ϕ’s do not add up to zero, there are short-run interactive feedback relationships in equation (3).
EMPIRICAL RESULTS
To describe the data distribution of each variable, the following standard statistical descriptors are reported:
As observed in Table 1, the distribution of US-Australia trade balance (TB) is slightly skewed to the left and that of exchange rate (US $/AU $) is slightly skewed to the right. The numerics of respective kurtosis and Jarque-Bera statistic suggest near-normal distribution of each variable. Moreover, the simple correlation between these variables is -0.836.
138 Furthermore, some degree of comovement is observed between bilateral trade balance and exchange rate (Appendix-A). To examine the non-stationary property of each time series variable, DF-GLS, Ng- Perron, and KPSS tests are implemented. They are presented as follows:
*The modified Dickley-Fuller (DF-GLS) critical values are -2.653 and -1.954 at 1% and 5% levels of significance, respectively. The modified Phillips-Perron (Ng-Perron) critical values are -13.00 and -5.70 at 1% and 5% levels of significance, respectively. The KPSS critical values are 0.70 and 0.347 at 1% and 5% levels of significance, respectively.
In Table 2, DF-GLS and Ng-Perron tests fail to reject the null hypothesis of unit root (non-stationarity) at both 1% and 5% levels of significance. Their counterpart, the KPSS also rejects the null hypothesis of no-unit root (stationarity) at the same levels of significance leading to an identical conclusion. On first-differencing, each variable becomes stationary depicting I(1) behavior, as observed above. Since both variables are nonstationary in levels with I(1) behavior, the Johansen- Juselius procedure is applied for cointegration between the variables. The and test results are reported as follows:
*denotes rejection of the null hypothesis at the 0.05 level of significance. ** MacKinnon-Haug-Michelis (1999) p-values
Table 3 reveals that both and test results clearly reject the null hypothesis of no cointegration between US-Australia trade balance and exchange rate at 5% level of significance confirming a long-run converging equilibrium relationship between the variables. In light of the above, a bivariate error-correction model (ECM) is estimated. The estimates of ECM (3) are reported as follows: ∆TBt= -0.6045 e(t-1) - 0.1721 ∆TB(t-1) - 0.1730 ∆TB(t-2)+ -6.3143∆ER(t-1) (-5.6731) (-3.1943) (-2.3675) (-1.9600) + 11.5052 ∆ER(t-2) (2.3841) R2 = 0.40, F = 27.39, AIC = 12.76
139 As observed , the coefficient of the error-correction term has expected negative sign, and it is statistically highly significant in terms of the associated t-value within parenthesis. This confirms long-run causal flow from change in exchange rate towards the current change in trade balance. The numerical coefficients of lagged changes in trade balance and exchange rate with their statistical significance in terms of the associated t-values unveil short-term interactive dynamics between the above variables. To add further, the Impulse Response analysis that shows how trade balance responds to a given exchange rate shock (Appendix-B) reveals no clear patterns. In fact, figure 2 in this Appendix unveils no improvement in Australia’s chronic trade deficit with the USA even if its currency is allowed to depreciate against US dollar. Counter-intuitively, the bilateral trade deficit may even worsen further.
CONCLUSIONS WITH SOME POLICY IMPLICATIONS Time series monthly data on bilateral nominal exchange rate and nominal trade balance between the USA and Australia are nonstationary in levels depicting I(1) behavior. Both variables are found co-integrated, based on both tests. There are evidences of long- run unidirectional causal flow from exchange rate changes to changes in trade balance with short-run interactive dynamic feedback effects. The impulse response analysis in figure 3 (Appendix A) shows that a deliberate policy of currency depreciation against US dollar is very unlikely to cure its chronic trade deficit with the USA. So, Australia should pursue other macroeconomic policy measures to improve its persistent trade deficit with the USA in lieu of exchange rate policy as the findings cast doubts on the potency of such policy. Long-run macroeconomic stabilization policies and export promotion through targeted marketing strategies in conjunction with accelerating export facilitation services are likely to be fruitful to reduce Australia’s chronic trade deficit with the USA.
140 REFERENCES
Akaike, H., (1969), “Fitting Autoregression for Prediction,” Annals of the Institute of Statistical Mathematics, 21, 243-247. Arndt, H. W. and Dorrance, G., (1987), “The J-curve,” The Australian Economic Review, 1st quarter, 9-19. Artus, J., (1975), “The 1967 Devaluation of the Pound Sterling,” IMF Staff Papers, 22, 595-640. Backus, D.K., Kehoem P.J. and Kydland, F.E., (1998), “Dynamics of the trade balance and the term of trade: the J-curve?” American Economic Review, 84, 84-103. Bahmani-Oskooee M. and A. Ratha., (2004), “The J-curve: a literature survey,” Applied Economics, 36, 1377-1398. Bahmani-Oskooee, M and Wang, Y., (2006), “The J-curve: China verses her trading partners,” Bulletin of Economic Research, 58(4), 323-43. Bahmani-Oskooee, M., (1985), “Devaluation and the J-curve: some evidence form LDCs,” The Review of Economics and Statistics, 67, 500-504. Bahmani-Oskooee, M., Goswami, G., and Talukdar, B., (2005), “The bilateral J-curve: Australia versus her 23 trading partners,” Australian Economic Papers, 44, 110-120. Bahmani-Oskooee, Mohsen, and Janardhanan Alse., (1994), “Short-run versus long- run effects of devaluations: error correction modeling and cointegration,” Eastern Economic Journal, 20(4), 453-464. Bahmani-Oskooee, Mohsen, and Marzieh, Bolhasani., (2011), “How sensitive is U.S.- Canadian trade to the exchange rate: evidence from industry data,” Open Economies Review, 22(1), 53-91. Brisssimis, S.N. and Leventakis, J.A., (1989), “The effectiveness of devaluation: a general equilibrium assessment with reference to Greece,” Journal of Policy Modeling, 11(2), 247-271. Demirden, T. and Pastine, I., (1995), “Flexible exchange rates and the J-curve: an alternative approach,” Economics Letters, 48, 373-377. Elliot, G., Rothenberg, T.J. and Stock, J.H., (1996), “Efficient test for an autoregressive unit root,” Econometrica, 64, 813-836. Engle, R.F. and Granger, C.W.J., (1987), “Cointegration and error correction: representation, estimation and testing,” Econometrica, 55, 251-276. Flemingham, B.S., (1988), “Where is the Australian J-curve?” Bulletin of Economic Research, 40(1), 43-56. Granger, C.W.J. and Newbold, P., (1974), “Spurious regressions in econometrics,” Journal of Econometrics, 2, 111-120. Hacker, Scott, and Abdulnasser Hatemi., (2003), “Is the J-Curve effect observable for Small North European Economies?” Open Economies Review, 14(2), 119-134. Helkie, W.L. and Hooper, P. (1987) The U.S. external deficit in the 1980s. International Finance Discussion Paper No. 304. Board of Governors. Federal Reserve System. Washington, D.C. Hsing, Yu., (2009), “Test of the J-curve for six selected new EU countries,” International Journal of Economic Policy in Emerging Economies, 2(1), 76-85. Hsing, Han-Min and Savvides, Andreas., (1996), “Does a J-curve exist for Korea and Taiwan?” Open Economies Review, 7(2), 127-145. Johansen, S., (1988), “Statistical analysis of cointegration vectors,” Journal of Economic
141 Dynamics and Control, 12, 231-254. Johansen, S., (1992), “Testing structural hypothesis in multivariate cointegration analysis of the PPP and the VIP for U.K.,” Journal of Econometrics, 53, 211-244. Johansen, S., (1995), “Likelihood-based inference in cointegrated vector autoregressive models,” Oxford: Oxford University Press. Johansen S., and Juselius, K., (1990), “Maximum likelihood estimation and inference on cointegration with applications to the demand for money,” Oxford Bulletin of Economics and Statistics, 51, 169-210. Kale, P., (2001), “Turkey’s Trade balance in the short and the long run: error correction modeling and cointegration,” International Trade Journal, XV, 27-56. Kulkarni, Kishore., (1994), “The J-Curve hypothesis and currency devaluation: A test of Kulkarni hypothesis with Egypt and Ghana,” Journal of Applied Business Research, reprinted in Kulkarni, Kishore (ed), Reading in International Economics, Serials Publications, New Delhi, India, 2004, 25-38. Kulkarni, Kishore and Bhatia, Alpana., (2002), “Empirical Evidence of the J-Curve Hypothesis,” Indian Economic Journal, reprinted in Kulkarni, Kishore (ed), Reading in International Economics, Serials Publications, New Delhi, India, 2004, 39-54. Karunaratne, N.D., (1988), “Macro-economic determinants of Australia’s current account 1977-86,” Weltwirtschaftliches Archive, 124(4), 712-28. Krugman, Paul and R.E. Balwin., (1987), “The persistence of the U.S. trade deficit,” Brookings Papers on Economic Activity, 1, 1-43. Kwiatkowski, D., Phillips, P.C.B., Schmidt, P. and Shin, Y., (1992), “Testing the null hypothesis of stationarity against the alternative of a unit root,” Journal of Econometrics, 54, 159-178. Ltkepohl, H., Saikkonen, P. and Trenkler, C., (2001), “Maximum eigenvalue versus trace tests for the cointegrating rank of a VAR Process,” Econometrics Journal, 4, 287-310. Magee, S., (1973), “Currency contracts, pass-through, and devaluation,” Brookings Papers. Mahdavi, S. and Sohrabian, A., (1993), “The exchange value of the dollar and the U.S. trade balance: An empirical investigation based on cointegration and Granger causality tests,” Quarterly Review of Economics and Finance, 33, 343-358. Marwah, K. and Klein, L.R., (1996), “Estimation of J-curve: United Stated and Canada,” Canadian Journal of Economics, 29, 523-539. Miles, Marc A., (1979), “The effects of devaluation on the trade balance and the balance of payments: some new results,” Journal of Political Economy, 87, 600-620. Mehmet, Yazici and Mushtag, Ahmad Klasra., (2010), “Import content of exports and the J-curve effect,” Applied Economics, 42(6), 769-776. Narayan, P.K., (2004), “New Zealand’s trade balance: Evidence of the J-curve and Granger causality,” Applied Economics Letter, 11, 351-354. Ng, S. and Perron, P., (2001), “Lag length selection and the construction of unit root tests with good size and power,” Econometrica, 64, 813-836. Phillips, P.C.B., (1986), “Understanding spurious regressions in econometrics,” Journal of Econometrics, 33, 311-340. Rahman, Matiur and Islam, Anisul., (2006), “Taka-Dollar Exchange Rate and Bangladesh Trade Balance: Evidence on J-curve or S-curve?” Indian Journal of Economics and Business, 5, 279-288. Rose, Andrew K. and Yellen, Janet L., (1989), “Is there a J-curve?” Journal of Monetary Economics, 24, 53-68. Rose, A.K., (1990), “Exchange Rates and the Trade Balance: Some Evidence from
142 Developing Countries,” Economics Letters, 3, 271-275. Rose, Andrew., (1991), “The role of exchange rate in a popular model of international trade: does the Marshall-Lerner condition hold?,” Journal of International Economics, 30, 301-316. Rosensweig, Jeffery A. and Koch, Paul D., (1988), “The U.S. dollar and the delayed J-Curve,” Federal Reserve Bank of Atlanta, Economic Review, 2-15. Senhadji, A.S., (1998), “Dynamics of trade balance and the terms of trade in LDCs: the S-curve,” Journal of International Economics, 46, 105-131. Spitaller, E., (1980), “Short-run effects of exchange rate changes on terms of trade and trade balance,” IMF Staff Papers, 27, 320-348. Wilson, P., (2001), “Exchange rates and the trade balance for dynamic Asian economies: does the J-curve exist for Singapore, Malaysia and Korea?” Open Economies Review, 12, 389-413. Yusoff, Mohammed., (2007), “The Malaysian real trade balance and the real exchange rate,” International Review of Applied Economics, 21(5), 655-667. Zhou, Su., (2001), “The power of cointegration tests versus data frequency and time spans,” Southern Economic Journal, 67, 906-921.
143 Appendix-A Comovement between Bilateral Trade Balance and Exchange Rate US-AUSTRALIA TRADE BALANCE TB =Trade balance between US and Australa and ER = US$/AU$
0
-400
-800 1.8 -1,200 1.6
-1,600 1.4
-2,000 1.2
1.0
95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11
ER TB
TB
144 145 146 SAVING FOR SUSTAINABILITY: WHY A 10% PERSONAL SAVING RATE IS TOO LOW Laura L. Coogan, Nicholls State University John Lajaunie, Nicholls State University Shari Lawrence, Nicholls State University
ABSTRACT
This paper analyzes and discusses expected outcomes for an individual consistently saving 10% throughout his/her work life. Including consideration for Social Security retirement benefits, we find the 10% saving rate would require continuous employment and an aggressive portfolio allocation to provide over 20 years of consumption security in retirement. These results suggest that a personal saving rate of 10% should be considered as a minimum, and a higher saving rate is required to provide consumption security throughout a worker’s career and retirement. JEL Classification: D14
INTRODUCTION
While almost everyone is aware that one should eat at least five servings of fruits or vegetables a day and that weight, cholesterol, and blood pressure readings should be maintained below certain thresholds, a simple rule for how much an individual should save does not seem to be as well established. Just as these health measures lead to a better life, might a saving “rule of thumb” help the financial security and sustainability of U.S. households? The financial press (and the many gurus found there) frequently state that everyone should have six months of living expenses set aside as an emergency, but should saving goals end with the short-term? Additionally, it is often stated that if one’s employer offers a matching contribution in the firm’s retirement plan, the employee should save at least enough to capture all of the matching funds, but is that enough to provide for future needs? What level of saving should a worker without an employer-sponsored plan maintain? While there are various opinions and suggestions on how an individual’s net worth should be divided between short-term assets (such as bank accounts and C.D.s), long-term assets (such as bonds and stocks), and real assets (such as real estate and gold), a minimum and/or a preferred level of saving over the work life is not as widely reported. However, some popular personal finance books support a 10% saving rate, but after closer examination, we believe this level is too low to provide consumption security throughout a worker’s lifetime, including retirement.1 This paper aims to clarify this discussion by presenting a life-cycle model where saving is stated as a percentage of earnings and net worth is measured by years of consumption available. By using years of consumption as the measure of wealth, this approach differs from the models where the focus is on specific dollar amounts saved and/or portfolio allocation.
147 These ideas are important because many people are not exposed to the basic tenets of finance. For example, consider how saving is discussed in the educational setting. In most economics courses, saving is used as a source of funds for macroeconomic capital and investment (which is vital for long-term economic growth) or as an alternative to consumption (and a restriction to short-term growth). Meanwhile, in most finance courses, saving is used as the resource of investment decisions (and portfolio allocation is the focus). In this paper, the positive and negative macroeconomic externalities provided by personal saving are set aside, and portfolio allocation is simplified. Here, the goal of saving is to provide the individual with retirement consumption as well as provide a safety net from income interruption and uninsured losses throughout the work life. In other words, the goal of this research is to stress the private benefits of savings, including personal economic sustainability. The paper proceeds by reviewing data on the behaviors and beliefs about saving and life expectancy, and then it presents a base case that analyzes how a 10% saving rate would provide for an individual in retirement. After this base case is presented, alterative assumptions are discussed with some obvious and not so obvious results. These alterative assumptions include examining workers with periods of unemployment, worker with higher incomes, and the issue of how home ownership impacts retirement saving and asset accumulation. This paper concludes by highlighting the key points of these various projections.
AN OVERVIEW OF SAVING AND LIFE EXPECTANCY
Over the past two decades, the monthly U.S. saving rate has varied between 2.0% and 8.3% of disposable income (after-tax income) while averaging 5.1% (as calculated by the Bureau of Economic Analysis using the NIPA accounts), and this is below the minimum of 10% that will be analyzed below.2 The tendency to save appears to differ significantly across cultures. OECD countries with high household saving rates over the past decade include France, Germany, and Switzerland (with saving rates varying between 8% and 12% of disposable income) while much lower levels of saving were observed in Greece, Denmark, and Estonia (all of which reported negative saving rates during some years), and the U.S. saving rate tends to be in the lower-half of developed countries (OCED, 2013). According to the 2010 Survey of Consumer Finances (“SCF”), just over 50% of U.S. households save, and the probability of a household saving increases with income. While the percent of families who save tends to be pro-cyclical, higher income households tend to have a greater probability of saving. In the four SCFs between 2001 and 2010, the percentage of families in the lowest income quintile who save varied from 30.0% to 34.0%, and between 80.6% and 84.8% of the families in the top decile saved (Bricker, et al, 2012, Table 1). SCF respondents also reported that the dollar amount of savings a household would need in case of an emergency increases with income. However, the average estimate from each income quintile with respect to the percentage of income needed for emergencies was fairly consistent (varying from 8.9% to 14.1% of usual annual income), and the percentage of income needed for emergencies did not monotonically increase or decrease with the income quintile. The median ratio of estimated emergency saving to usual annual income was 10.8%, or between one and two months of average income, and liquidity and retirement were given as the two most important reasons for saving (ibid, Table 3.1 and Table 3).
148 Saving behavior may be a habit passed from one generation to the next, or individuals from different cultures may materially differ in their perception about current and future conditions and the need for savings. Setting aside these customs and perceptions, the focus of this paper is to provide a simple and understandable standard of saving of at least 10% of earnings by using a life-cycle model to predict the outcome if this level of saving is maintained. Based on the statistics noted above, it is likely that many people would find this suggested level of saving to be high, but it should also be noted that the worker following this rule does not retire affluent. For example, assume that someone is consistently employed beginning at age 25 and retires at 67 and saves 10% of earnings throughout his/her career. This worker should have enough resources in retirement (combining income from savings, drawing down on the balance of savings, and Social Security retirement payments) to sustain their pre-retirement consumption through age 90, but it is up the individual to decide if that enough. For example, a worker may have the expectation of exceptional longevity or, alternatively, this worker may believe that there will be increasing expenses in retirement, and both expectations would require a higher saving rate throughout the worker’s career. While individuals need to make their own judgment on how important is it to have resources into their nineties, the U.S. Life Tables show that approximately 1 in 5 adults who reach 25 (the start of the work life in this model) will live into their nineties, and the likelihood of a long life increases with age. The U.S. Life Tables also show that women are more likely to reach these advanced years (and approximately 1 in 4 females who reach their 25th birthday will live into their nineties).3 Under the base-case saving model with a 10% saving rate, individuals who live into their nineties will have only Social Security providing income and those benefits, on average, provide less than one-half of the worker’s average consumption. In a recent analysis of poverty in the United States, the tendency of outliving one’s resources was noted. In a November 2013 Congressional Research Service report, 12.1% of the 65 to 74 year olds were considered poor or “near poor”, and that percentage increased to 17.8% for those who are 75 years of age and older (Gabe, 2013, pg. 9).4 While the Social Security and Medicare programs have significantly reduced the level of poverty among the elderly (which was over 25% prior to 1975 (Gabe, 2013, Figure 2)), individuals may want to consider the implication of a longer-term survival horizon on consumption sustainability as labor market opportunities tend to decrease with age. Due to economic and health issues, workers entering their 60s and 70s with limited resources have few options to generate income or wealth at that advanced age. While many workers may plan on worker longer, few actually do, and labor force participation rate (the percent of the civilian population working or looking for work) decreases with age. Currently, over 80% of the population between the ages of 25 and 54 are in the labor force, but the participation rate drops to about 64% for those in their late 50s and early 60s, and it continues to drop well before the age of full Social Security retirement benefits. The labor force participation rate is below 32% for those 65 to 69, and it is below 18% for those 70 or older (BLS, Household Data, 2014).
BASE-CASE MODEL ASSUMPTIONS, A WORKER EARNING $40,000 ANNUALLY
The financial returns and average earnings for the base-case assumptions use U.S. averages with some rounding for readability; however, we do not propose that this is a
149 portrait of the “average” U.S. worker. Our aim is to use assumptions to construct a model that provides easily understandable output (which is the number of years of consumption security in retirement) and then show how assumptions changes alter the output. While the appropriateness of these assumptions is always debatable, whether they are conservative or aggressive will be left for the reader to decide. As with all economic models, changes in these assumptions may significantly alter the results, and some of the more significant and interesting alternative assumptions will be discussed in the sections that follow. In order to provide a starting point to the discussion, the details of the base-case assumptions include:
1. All growth rates are real growth rates and all dollar amounts are adjusted to 2014 levels. 2. The worker is single and has completed his/her education, initial job sorting, and train- ing by the age of 25 and begins employment. 3. The worker maintains steady and uninterrupted employment starting on his/her 25th birthday through age 66 and retires on his/her 67th birthday. 4. Average earnings over the lifetime are $40,000 per year. 5. Earnings have a real growth rate of 0.5% annually. 6. Throughout his/her work life the worker saves 10% and consumes 90% of labor mar- ket earnings as consumption. Labor market earnings are adjusted for Social Security tax of 6.2% (based on program specifics 2014).5 7. Saving is divided into two asset classes: short-term assets (e.g., bank accounts and C.D.s), long-term assets (e.g., bonds and stocks). Short-term funds earn a real re- turn 0.25% annually. Long-term funds (a portfolio of 60% stocks and 40% corporate bonds) earn a real return 5.25% annually.6 8. At the beginning of the work life, all saving is directed to the short-term account (which is an emergency and income replacement account) until 0.5 years of consump- tion is saved. This takes approximately four and one-half years. 9. After the emergency and income replacement fund is established, the worker contin- ues at the same saving rate of 10% of labor market earnings, and this saving is split between the short-term accounts (10% of dollars saved) and long-term accounts (90% of dollars saved).7 10. Once retired, the short-term and long-term funds are combined into an account that has a real return of 1.75% annually. This is based on approximately 30% held in the long-term funds (stock and bond portfolio) and 70% held in short-term (and less vola- tile) assets. 11. Accumulated savings is economically the same as net worth (assets minus outstand- ing debt). In other words, contribution to savings is an increase in net worth, and debt cannot be used to fund savings. (The terms “savings”, “net assets” and “net worth” are used interchangeably.) 12. Consumption in the first year of retirement is the same as the consumption in the last year of the work life. After that, real consumption increases by 0.5% throughout re- tirement. 13. Retirement consumption is provided by Social Security payments (based on 2014 rules), earnings on assets (the combined short-term and long-term assets accumula- tion), and withdrawal of the assets from the saving accounts.8
For a base-case model, are these assumptions reasonable? The real returns for short-term and long-term assets are based on past performance, and the portfolio
150 allocations are not overly aggressive (although individuals may not have the tolerance for risk that investment in the stock market requires). The suggested level of saving is about double the historical U.S. average, but this model assumes that the saving and consumption decisions are under the control of the worker. While saving for the future reduces current consumption, it is assumed that a worker without dependents who earns at approximately the U.S. average of $40,000 per year is able to save at this level and still have enough after-savings earnings for consumption (BLS, Earnings Data, 2014).
DISCUSSION OF THE BASE-CASE MODEL RESULTS AND THE CHALLENGES TO THE BASE ASSUMPTIONS
With these assumptions, the worker saving 10% of labor market earnings retires with approximately 11 years of consumption available in accumulated financial assets. A combination of Social Security retirement benefits, income from the retirement assets, and drawing down of those retirement assets allows this worker to sustain the same level of consumption through his/her 90th birthday. This period of consumption is beyond the life expectancy of a 67-year worker (which is age 84), so the probability that a worker will outlive his or her assets is less than fifty percent. However, once those reserves are depleted at that advanced age, the 10% saver’s sole source of income and consumption is Social Security benefits, but those payments would provide less than one-half of the individual’s average level of consumption. These results are summarized on the first line of Table 1. As discussed above, this model could easily be challenged by the life expectancy or the growth rate of assets assumptions, but a worker faces more uncertainty than those two factors. Other assumptions used, including the condition of continuous employment and that consumption in retirement will have very modest growth as compared to pre- retirement levels, eliminate a great deal of uncertainty that a real worker faces. While these conditions are necessary to develop this model, it becomes apparent that the 10% saving rate does not provide the worker with security against these and other downside risks. Continuous employment, steady real increases in earnings, and predictable retirement consumption may have been much easier to obtain in the past than it will be for workers now and in the future. While the U.S. had an extensive period of economic growth during the 1990s and early 2000s, the recent “Great Recession” provided a reminder to workers of the many challenges the business cycle places on the individual. During this downturn, unemployment rates, underemployment rates, and the length of the unemployment period were all higher than the average of the previous two decades. While these conditions confronted workers at all ages, it appears that older workers face greater challenges if they lose their jobs. In addition to the income lost during a period of unexpected unemployment, it is likely that when the worker regains employment, earnings will be lower in the new job. A recent government report highlights some of these conditions. Since the 2008 recession, the number of workers unemployed for more than 26 weeks (considered as the long-term unemployed) has increased for workers of all ages. For unemployed workers over the age of 55, 55% have been looking for work for more than six months, and for unemployed workers under 55, 47% of them need this length of time to complete a job search (GAO, 2012, Figure 8). It appears that as a worker ages, the time to find a new job increases while the ability to recover to the previous level of earning decreases (GAO, 2012, Figure 10). In addition to the business cycle, as workers age, the probability of
151 illness increases and medical complications are often accompanied by unemployment and income interruption. The strong relationship between illness and financial hardship (including personal bankruptcy), has been repeatedly reported in the popular press over the past decade (see Cussen, 2010; Factcheck, 2008; and Mangan, 2013). A Health Affairs study found that between 30% and 50% of personal bankruptcy have health and medical issues as the root cause (and the percentage varies based on how closely linked the illness is to the bankruptcy filing) (Himmelstein, et al, 2005). Finally, in addition to the risks of unemployment from economic forces and job interruption due to medical conditions, some workers may simply prefer to not work at some point in their life, and common reasons for labor market interruptions include using time for care-giving and education. These employment trends are summarized by the labor force participation rate, which reaches a maximum of over 80% for workers in their 30s and 40s, but begins to drop after age 50. Between the ages of 55 and 64, average labor force participation drops to 64%. For workers between the ages of 65 and 69, less than 32% of the civilian population remains active in the work force, and for workers aged 70 and older, that ratio falls below 18% (BLS, Household Data, 2014). For all of these reasons, the assumption of continuous employment may be an aggressive and overly optimistic base-case condition. Another assumption in the base-case model that is challenging to support is that pre- and post-retirement consumption will be consistent. While this assumption is based on two, counter-balancing, economic forces, it may be difficult for an individual to plan the level of retirement consumption. On one side, retirement should decrease work-related expenses (such as commuting and child/elder care expenses), but these reductions are balanced by increases in everyday expenditures (since there is more time for activities and travel) and increases in medical services consumption. Unless a worker has significant work-related expenses, it seems likely that consumption would increase in retirement, as the possibility of catastrophic medical condition increases with age. While our model uses the assumption that pre- and post-retirement consumption remains constant, it does so without empirical support and acknowledges that it is in contrast with the conditions often provided in the financial press (where it is often assumed consumption in retirement will decrease). Summarizing the base-case model, a 10% saving rate throughout the work life would contribute significantly to consumption security and sustainability in retirement, but that level of saving does not provide much insurance against periods of unemployment, uninsured losses, increasing consumption in retirement, or other downside risks.
ALTERNATIVE ASSUMPTIONS, INCLUDING HIGHER AVERAGE EARNINGS
The results of most alternative assumptions change the outcomes as expected, but two alternatives provided results that were not so obvious. The easier to understand alternatives and results will be discussed first and these alternative assumptions are grouped by those related to the actions of the worker, those resulting from labor market variation, and those associated with changes from financial market returns and government policy. Two changes in labor market conditions led to results that are not commonly or frequently discussed are presented at the end of this section. Of the conditions that can be controlled by the worker, perhaps the most obvious is that higher saving rates would provide for longer periods of consumption security. Other conditions that the worker may be able to adjust include beginning work earlier
152 and retiring later, and either of those changes would increase the assets available for retirement consumption. Alternatively, beginning work later and retiring earlier will reduce the retirement assets. Post-retirement consumption greater than pre-retirement consumption would deplete the assets more quickly than projected while lower post- retirement consumption would increase the life of the assets. Similarly, any use of the savings before 67, or any interruption in earnings and savings, would reduce the resources available during retirement. While the exact impact of an interruption can vary based on the timing within the career, each year away from the labor market decreases the retirement savings by more than the lost year’s savings, since total savings is reduced in three ways: 1) accumulated savings must be used for current consumption; 2) there are no contributions to savings during the period outside the labor market; and 3) the savings generates less earnings (due to the lower level of accumulated assets).9 As would be expected, the longer the earnings interruption and the earlier in the work life the interruption occurs the greater the decrease in final retirement savings. While an earnings interruption for illness and injury may be beyond the worker’s control, the loss of contributions or the drawing down from savings for unplanned (or uninsured) medical expenses will impact the accumulation of retirement assets similar to any other earnings interruption. Details of this type of change in assumption, for example, retiring early at age 66, are shown in the second line of Table 1. Leaving the labor force one year earlier than in the base case would reduce the years of retirement consumption by almost three years (from 90 to 87). This reduction holds whether the Social Security payments are started at 66 (at a reduction from the full retirement payment amount) or if the worker consumes out of saved assets for the first year of retirement. Another condition that is under the control of the worker includes the mix of assets in the portfolio. Shifting assets from the short-term to the long-term asset class would likely increase retirement savings, while holding proportionately more assets in the short-term account would likely decrease total savings. These decisions would also alter the risk of the retirement portfolio. Variations from the labor market could include non-voluntary employment interruptions resulting from the business cycle, the creative destruction of jobs as the labor market evolves, or the geographic shift of the worker, firm, or industry. Again, any earnings interruption would require a higher saving rate throughout the work life to provide the same level of retirement consumption as the base case. Also as expected, financial market forces and government policy will alter the accumulation of assets available for consumption in retirement. Higher rates of return would increase the assets at retirement while lower rates return would decrease those assets. Any reduction in the current Social Security benefits schedule would require the worker to save more than the 10% of earnings assumed above if the worker wanted to maintain the length and level of consumption previous described. The most interesting results were found when changes to the pattern of expected life time earnings were altered. Both the level of average earnings and the growth rate of earnings had a noticeable impact on the required minimum saving rate, and higher than average earnings or faster earnings growth required an increase in the percent of earnings saved in order to maintain the same level of retirement consumption. As average earnings rise above the $40,000 per year used in the base case, the replacement ratio provided by the Social Security retirement benefits decreases, and in order to maintain the same level of retirement consumption, the worker’s savings must replace a greater percentage
153 of consumption. For example, comparing two workers using the same 10% saving rate, a worker with average annual earnings of $80,000 would deplete his or her savings approximately four years sooner than a worker earning $40,000 per year. The results for a worker with higher earnings are shown in the third line of Table 1, and if this higher earner maintains the base-case saving rate of 10%, their retirement assets are depleted by age 86. Alternatively, if the higher-earning individual wanted to have the same asset longevity in retirement, that worker’s saving rate would have to increase to between 11% and 12%. As the average earnings over the work life increases, the portion of retirement consumption replaced by Social Security decreases. This situation indicates that higher-earning individuals need higher personal saving rate to maintain consumption in retirement. Returning to Table 1, at the start of retirement, Social Security benefits replace approximately 49% of the consumption for worker with average earnings of $40,000, but Social Security only replaces 37% of consumption for the worker whose average earnings are $80,000. For a worker with average earnings of $100,000 (who is at the upper limit of the ninth decile of wage earners according to BLS Earnings Data (2014)), Social Security only replaces 33% of consumption at the start of retirement. Workers should understand that the replacement rate of Social Security retirement benefits decline as average life time earnings increase, and those workers need to plan appropriately. The second assumption alteration that provided an interesting result was that faster earnings growth (holding the lifetime average constant), requires a higher saving rate. This result is due to the condition that the savings from the first part of the worker’s career do not contribute as much to retirement when compared to the higher level of consumption later in life. For example, if a worker experiences 1% real annual growth in earnings and consumption throughout life (which is twice the base-case assumption), the retirement savings would be depleted approximately six years earlier than the worker with the lower growth rate. Details of these results are shown on the fifth line of Table 1, and the individual with the higher earnings growth rate would have to increase the saving rate to approximately 12% of labor market earnings in order to have the same consumption longevity as the worker with the lower growth rate of earnings. After this review of these alternative assumptions, it appears that in order to have a saving plan provide sustainability through life’s ups and downs, the worker should save more than the 10% established for the stable and predictable world. As a final addition to the summary results from the alternative assumptions, the last line of Table 1 shows the increase in retirement consumption security if the base-case worker saves 15%. If consumption security in retirement is the worker’s goal, clearly there are benefits to increasing savings in the financial markets throughout one’s work life. The following section considers a common decision outside of the labor and financial markets that is often associated with retirement consumption security, and the benefits and costs associated with home ownership are discussed next.
ANOTHER ALTERNATIVE: HOME OWNERSHIP
The above analysis assumes the worker is a renter, which leads to the question on how the purchase of a home would impact the base-case results. Home ownership and retirement saving are not necessarily related. The home equity (gained through principal payments or outright purchase) could be considered as either additional savings or it could be part of the 10% annual saving (which would put the real estate into the retirement
154 portfolio). Whether the home equity is treated as additional saving or as part of the primary portfolio (and it is funded from the 10% of annual savings) will have major implications on the availability of assets for consumption in retirement. If the home purchase is treated as a consumption good (with all housing expenses, including principal, paid from after-saving earnings), then the above results would hold, and the retiree would have home equity in addition to the previously discussed assets. While it might sound odd to consider all home expenses as consumption items, such expenses as maintenance, insurance, property taxes, and mortgage interest are clearly consumables. Only the equity and growth of the real estate value contributes to net worth accumulation. If the worker can pay for the home outside the 10% of earnings saved, the worker is increasing the assets available for consumption in retirement. On the other hand, if home equity payments reduce retirement savings, the worker is altering the retirement portfolio to include an asset that is fixed in place and historically has had a real growth rate of less than 1%.10 This would reduce the expected value of the retirement assets, reduce the liquidity of those assets, and reduce the years of consumption available to the retiree. If the home is part of the retirement assets, the cost of the home relative to the annual earnings also has impact on retirement sustainability. The more expensive the home (relative to the worker’s annual earnings), the greater the reduction of the more liquid assets the worker has for consumption in retirement. This issue is captured in the common expression of someone being “house poor”, a condition where home ownership crowds out other forms of consumption. While it is fully acknowledged that home ownership provides a dividend in the form of housing (which is roughly equivalent to the expenditure for rent), it must also be pointed out that this model assumes that the individual either consumes or saves all earnings. 11 If a worker includes a home as part of the retirement portfolio and reduces the amount of retirement savings in order to purchase and maintain the home, the worker will be decreasing the amount of financial assets available in retirement. The home cannot provide for the necessary consumption (other than housing) after this worker has left the labor force, but the worker needs more than housing to sustain consumption for several decades in retirement.
CONCLUDING REMARKS
If the future was predictable, retirement savings (and many other decisions) would be easy. However, the world continues to change and the average worker needs to plan within this framework of uncertainty. At the very minimum, workers should continuously save 10% of their earnings throughout their work life in order to provide some reasonable level of consumption security in retirement. Unfortunately, this level of saving does not provide a significant cushion against downside risks such as earnings interruption or other uninsured, unplanned expenditures. Perhaps a more practical “rule of thumb” is a saving rate of 15%, even for worker whose goals include maintaining continuous employment and working until age 67. While this higher level of saving provides more consumption security, workers who plan to be out of the labor market for some period, who have higher than average earnings, or who have experienced higher than average growth in earnings over the work life should consider greater saving ratios. Any combination of these events would require still higher levels of saving in order to provide the same base level of sustainability throughout retirement.
155 Additionally, if the worker believes that stock and bond market returns in the future will be lower than their historical average, that worker should save more. A higher saving rate is also needed if the worker believes that future Social Security benefits will be less than the current program pays. This analysis also concludes and acknowledges that significant saving may materially change many financial decisions throughout a worker’s life, and these suggested rates are two or three times the average U.S. saving rate. As the worker assesses the value of current consumption against future consumption sustainability, such decisions as whether one should live alone or have roommates, the timing and size of major purchase such as vehicles and homes, how much of current earnings should be consumed for leisure activities, and if or how large of a family to have may need to be considered. While no one looks forward to a reduction in current consumption, the reward for saving is more financial security and sustainability throughout one’s work life and retirement years.
156 REFERENCES
Arias E. United States Life Tables, 2008. National Vital Statistics Reports; vol 61 no 3. Hyattsville, MD: National Center for Health Statistics. 2012. Bricker, Jesse, Arthur B. Kennickell, Kevin B. Moore, and John Sabelhaus. “Changes in U.S. Family Finances from 2007 to 2010: Evidence from the Survey of Consumer Finances (SCF)” Federal Reserve Bulletin, vol. 98, no 2, (February 2012), pp. 1-80. Cussen, Mark P. “Top 5 Reasons Why People Go Bankrupt” http://finance.yahoo.com/ news/pf_article_109143.html March 22, 2010. Factcheck. “Health Care Bill Bankruptcies” http://www.factcheck.org/2008/12/health- care-bill-bankruptcies/ December 18, 2008. Federal Housing Finance Agency. All-Transactions Indexes (Estimated using Sales Prices and Appraisal Data), U.S. and Census Divisions through 2013Q1 (Not Seasonally Adjusted). http://www.fhfa.gov/DataTools/Downloads/Pages/House-Price-Index. aspx. Accessed: May 2013. Gabe, Thomas. Congressional Research Service, “Poverty in the United States: 2012,” RL33069. November 13, 2013, https://www.fas.org/sgp/crs/misc/RL33069.pdf, Accessed: December 2013. Guidolin, Massimo and Elizabeth A. La Jeunesse. “The Decline in the U.S. Personal Saving Rate: Is It Real and Is It a Puzzle?” Federal Reserve Bank of St. Louis Review, November/December 2007 Vol. 89, No. 6, pp. 491-514 Himmelstein, David U., Elizabeth Warren, Deborah Thorne and Steffie Woolhandler. “MarketWatch: Illness and Injury As Contributors To Bankruptcy” Health Affairs (2005). Accessed May 2014: http://content.healthaffairs.org/content/ early/2005/02/02/hlthaff.w5.63.long Mangan, Dan. “Medical Bills Are the Biggest Cause of US Bankruptcies: Study” http:// www.cnbc.com/id/100840148 25 Jun 2013 Morningstar, Inc., SSBI. Stocks, Bonds, Bills, and Inflation: Market Results Yearbook. OECD (2013), “Household Raving Rate” National Accounts at a Glance 2013, OECD Publishing. http://dx.doi.org/10.1787/na_glance-2013-10-en Tobias, Andrew (2010), “The Only Investment Guide You’ll Ever Need”. First Mariner Books, New York U.S. Government Accountability Office (GAO). Report to the Chairman, Special Committee on Aging, U.S. Senate; “Unemployed Older Workers; Many Experience Challenges Regaining Employment and Face Reduced Retirement Security” April 2012 U.S. Bureau of Labor Statistics (BLS). Weekly and hourly earnings data from the Current Population Survey. Series Id: LEU0252881500 Accessed: May 2014. http://www. bls.gov/news.release/wkyeng.t02.htm and http://www.bls.gov/news.release/wkyeng. t05.htm U.S. Bureau of Labor Statistics (BLS). Household data, not seasonally adjusted, A-13. Employment status of the civilian noninstitutional population by age, sex, and race for March 2014. Accessed: May 2014. http://www.bls.gov/web/empsit/cpseea13.htm U.S. Bureau of Labor Statistics (BLS). Consumer Price Index - All Urban Consumers, Series ID: CUUS0000SA0L2. U.S. city average: All items less shelter. Accessed: May 2013. U.S. Department of Commerce: Bureau of Economic Analysis, PSAVERT: Personal Saving Rate, montly data. Accessed via: http://research.stlouisfed.org/fred2/series/
157 PSAVERT/ downloaddata?cid=112, December 2013. Wasik, John. “The Simplest Wealth Plan Ever” Forbes (online), published May 7, 2014. Accessed via: http://www.forbes.com/sites/johnwasik/2014/05/07/the- simplest-wealth-plan-ever/ Link to William J. Bernstein’s pamphlet: https:// dl.dropboxusercontent.com/u/29031758/If%20You%20Can.pdf
158 ENDNOTES 1 For examples of the 10% saving rate suggestion, see Andrew Tobias’ “The Only Investment Guide You’ll Ever Need”. 2 Based on series “PSAVERT” as calculated by the U.S. Department of Commerce in Nov 2013. See Guidolin and La Jeunesse (2007) for a discussion on the two primary methods of calculating of the U.S. saving rate. 3 See Table 1 of United States Life Tables, 2008, second column of data. Of 100,000 live births, 98,293 survive to their 25th birthday 81,364 survive to their 67th birthday, and 22,347 survive to their 90th birthday. At all of these ages, the expected survival for females is higher. 4 “Near poor” is defined in this report as 125% of the poverty threshold. 5 Income is adjusted for Social Security tax and retirement benefits under the current program, but no adjustment is made for income tax on labor market earnings, capital gains, or dividends. While including income tax assumptions may seem practical for understanding the results, this paper treats income tax as part of the workers consumption bundle. Simulations using income tax added complication to the model without altering the general results. This tax assumption is conservative in that income tax as a percent of income tends to decrease after retirement (e.g. higher standard deduction, partial exemption for Social Security benefits, lower tax rate for capital gains). Therefore, this model may slightly understate the consumption available in retirement, assuming current and future U.S. tax policy is constant. 6 The returns on short-term assets are less volatile than the returns on long-term assets or real assets. Long-term returns are based stock and corporate bond real returns (total returns minus inflation, annually) for the 20 year periods (average holding period of these assets) since 1926 (Morningstar: SSBI data, S&P total return at http://www.spindices.com/indices/ equity/sp-500). 7 While not explicitly model (since the work maintains continuous employment and never uses the short- or long- term savings prior to retirement and income taxes are not included), withdrawal from short-term accounts carry no economic penalty (no transaction costs), but withdrawals from long-term accounts would likely carry a penalty (based on current law with respect to 401(k)-type accounts and IRAs). 8 As with other income taxes, the income tax burden on Social Security burden is not included in this model as such analysis is complex. However, it should be noted that the higher level of other taxable income in retirement, the more likely Social Security benefits will be subject to federal income tax, but at most, 85% of Social Security benefits are subjected to federal income tax. 9 For a detailed discussion on how labor market interruptions impact retirement savings, see the GAO report (2012). 10 A real growth rate of 0.72% for home values is based on comparison of home price index (Federal Housing Finance Agency) and CPI for all goods minus shelter (BLS CPI data) from 1984 through 2012. For housing price index, see http://www.fhfa.gov/ DataTools/Downloads/Pages/House-Price-Index.aspx. For the CPI for all goods minus shelter, see series CUUS0000SA0L2 (Consumer Price Index-All Urban Consumers, U.S. City Average). For methodology, see http://www.fhfa.gov/Media/PublicAffairs/Pages/ Housing-Price-Index-Frequently-Asked-Questions.aspx#quest17. 11 Remember, in this model, any post-saving earnings are consumed. While homeowners may argue that home ownership is cheaper than renting, if any difference (any gain) from home ownership is consumed, this difference does not benefit the worker in retirement.
159 160 AN EMPIRICAL RE-EXAMINATION OF THE FISHER HYPOTHESIS: PANEL COINTEGRATION TESTS Swarna (Bashu) Dutt, University of West Georgia Dipak Ghosh, Emporia State University
ABSTRACT There is no unanimity in the literature on the Fisher hypothesis. This study will re- visit this academic quandary with a powerful econometric test proposed by Pedroni (2004). The strength of this test is that the test statistic is able to accommodate short run dynamics, deterministic trends and different slope coefficients. The study will use monthly interest rate and inflation data for the G5 countries. The study starts with stationarity characteristics of the data and then applies the Pedroni panel cointegration tests. This will shed some light on the Fisher hypothesis and the mixed evi- dence that exists in the literature. JEL Classification: F410
INTRODUCTION
The Fisher hypothesis is one of the most controversial topics in financial economics, with serious ramifications for policy analysis. Fisher basically states that there should be a one-to-one correspondence between the nominal interest rate and the rate of inflation, resulting in stable (if not fixed) real interest rates. It implies that there is effectively no correlation between expected inflation and ex ante real interest rates. If this theory holds (as claimed), then in the long run the real interest rate would remain unchanged under monetary policy shocks. This study will extend Dutt and Ghosh (2007) studying the Fisher effect for G5 coun- tries in which they found mixed evidence in favor of the hypothesis. There is no unanimity among the researchers and by extension policy makers since the empirical literature is all over the place. Therefore this study will revisit this academic quandary with a power- ful econometric technique proposed by Pedroni (2004), where he introduces pooling of economic data which allows for one to vary the degree of heterogeneity among the panel members. It examines both the between dimension and within dimension residuals. The strength of this test is that the resultant “test statistic” is able to accommodate short run dynamics, deterministic trends and also different slope coefficients. This test statistic is “standard normal’ and free of nuisance parameters. This test has been used for studies of Purchasing Power Parity (hereafter PPP) and seems particularly suitable for an analysis of the Fisher effect. One property of cointegra- tion tests is that the span of the data (and not the number of observations) is important in increasing the power of the tests (Pedroni 2004). Increasing the span of the data may not be possible in some cases due to the lack of availability of data, and in others may introduce structural changes like regime changes which would call into questions the validity of the
161 results. Therefore, the study will use the Pedroni (2001, 2004) methodology to test for PPP. This procedure allows us to increase the power of the tests even when we don’t have access to a larger span of data. The next section presents a brief literature review. The third section describes the model that is estimated in this paper and the fourth section is a description of the data set. The fifth section is a description of the Pedroni panel cointegration procedure and the sixth section is a description of our empirical results. The final section contains some concluding remarks.
LITERATURE REVIEW
Fisher (1930) hypothesized that:
e it = rt + πt (1)
where it is the nominal interest rate and it is composed of two entities, namely the expected e rate of inflation (πt ) and the real interest rate (rt.) Based on this, it postulates a one-to-one correspondence between the nominal interest rate and the expected inflation rate, assuming the constancy of real interest rates over time. This theory has been extensively examined in the economics literature. The genesis was with Fisher (1930), where he tested the relationship between nominal interest rates and inflation for the UK and USA over decades and found “no apparent correlation.” But, when past inflation was substituted as a proxy measure for expected inflation, the “correlation coefficient” jumped from the 30’s into the 90’s. Thusprice changes do affect interest rates. This study starts with a brief survey of the different Fisher studies done over time. Fama (1975) examined US treasury bills for the period 1953-71 and found evidence that nominal interest rates did incorporate inflation rates, supporting the Fisher hypothesis. But following studies by Nelson and Schwert (1977), Carlson (1977), Joines (1977) and Tanzi (1980) did not find any evidence of Fama’s “joint hypothesis.” Then Mishkin (1992) found evidence supporting Fisher (high correlation between interest rates and inflation) but it changes over time. He reported that the hypothesis held over specific time intervals, but failed over others. Based on this observation he made the distinction between the short and long run fisher effect and leaned towards supporting the interest/ inflation nexus over the long run. This long run correlation was supported by Crowder and Hoffman (1996) who report a near one-to-one correspondence between nominal interest and inflation for the USA over the period 1952-92. It is also supported by Fahmy and Kandil (2003) for the USA over the decade of the 80’S and 90’S, using cointegration techniques. Tillman (2004) also supports the Fisher hypothesis for post-war data. USA data has been generally favorable to the Fisher hypothesis, but Canadian data has not. Dutt and Ghosh (1995) use cointegration techniques and separate the entire exchange rate period into fixed and floating rate regimes, but do not find evidence supporting Fisher for Canada in neither the fixed nor the floating exchange rate regimes. But contrary to this, Crowder (1997) finds evidence supporting Fisher for Canada. Mishkin and Simon (1995) find long run evidence supporting Fisher (but not so in the short run) for Australia. Again contrary to this, Hawtrey (1997) and Olekalns (1996)
162 find supporting evidence for Australia. Then there is Evans (1998) who finds no evidence supporting Fisher for the UK. But Muscatelli and Spinelli (2000) find that the long run Fisher relationship holds for Italy over the long run (1948-90.) Esteve, Bajo-Rubio and Diaz-Roldan (2004) find partial evidence supporting the Fisher hypothesis for Spain. The Atkins and Serletis (2003) study uses the autoregressive distributed lag (ARDL) model to examine Fisher for Norway, Sweden, Italy, Canada, UK and the USA, but finds little supporting evidence. Then again Atkins and Coe (2002) using the same methodology as Atkins and Serletis (2003), does not find any evidence of even a long run Fisher relationship for Canada and the USA. Interestingly enough, when they extend their study to examine for a “tax adjusted” Fisher correlation, they do not find any evidence of that either. Again Atkins and Sun (2003) find a long run (but not a short run) Fisher relationship for USA and Canada. Recent studies like Kaliva (2008) and Westerlund (2008) find significant evidence supporting the Fisher hypothesis.
The Model
According to the Fisher identity, we can write
Rkt = Et rkt + Et πkt (2)
where Rkt = k-period nominal interest rate at time t
rkt = k-period real interest rate at time t
πkt = inflation rate from time t to time t+k
The expected inflation cannot be observed. Assuming rational expectations, we will get
πkt = Et πkt + ekt
We can rewrite eq. 2 as
_ Rkt = Et rkt + πk ekt (3)
_ Rkt - πk = Et rkt ekt (4)
Expected value of ekt should be zero. Therefore if Rkt and πkt are both I(1), and Rkt-
πkt is stationary, then this would imply that the nominal interest rate and the inflation rate are cointegrated with a cointegrating vector of (1, -1). This would be an indication of a ‘full Fisher effect”. Even if the cointegrating vector is (1, -β) this would be evidence of a “partial Fisher effect.” Absence of cointegration would mean that nominal interest rate and the inflation rate do not move together over time, and therefore there is no long run relation between them according to Lee et. al. (1998).
DATA DESCRIPTION
This study estimates and tests the Fisher equation for the G-5 countries, United States, France, Germany, Japan, and the United Kingdom. All data were obtained from the OECD National Accounts database. All data is monthly. Data is available for the different
163 countries for different time periods, and therefore we have used different groups of the G5 countries to implement our analysis. The different groups are G2 - Germany, United States: June 1964 – June 2013 G3 - France, Germany, United States: January 1970 – June 2013 G4 - France, Germany, United States, United Kingdom: January 1978 – June 2013 G5 - France, Germany, Japan, United States, United Kingdom: April 2002 –June 2013
PEDRONI’S PANEL COINTEGRATION TESTS
Cointegration techniques are commonplace in the economics literature, when studying long run relationships between non-stationary variables. One point of concern has been the power of traditional cointegration tests. The problem with these tests is that they inherently suffer from low power and confidence. Increasing the time span of the variable series increases its credibility, but in reality it is a difficult proposition. The time span availability of the variables is not dependent on the researcher’s discretion. On the other hand if one blindly increases the data time span, the test strength will possibly increase but one could very well have introduced major policy shifts and structural economic changes. An example of this would be using pre-war and post-war data together, just to increase the time span. Another possibility is to increase the data frequency keeping time span the same. An example would be to use daily instead of weekly data or weekly data in place of monthly data. This increases the number of observations, but that does not necessarily increase the strength of the results. It has also been pointed out that the power of these tests depends more on the span of the data rather than the number of observations (Perron 1989, 1991). For example, if we consider a time span of 1969 to 2011, moving from annual to quarterly to monthly data will not appreciably increase the power, but increasing the span to 1960 - 2011 will increase the power of the tests. If increasing the time span of the data is not a practical solution (additional data may not be available, or it may introduce structural changes in the model) one alternative is to consider additional cross-sectional data instead of a longer time period, thus resulting in panel data. When considering panel data, it is important not to sacrifice differences between cross sections. One remedy to solve this dilemma has been proposed by Pedroni (2001 and 2004) where he introduces similar cross-sectional data over the available time period. This pooling of similar data will help in the above stated situation. One example would be where he pooled data from economically similar countries to study PPP (Pedroni, 2004.) The problem here is that simple pooling of time series data would involve “in model” heterogeneity. Here he has constructed “panel cointegration” test statistic (Pedroni, 2004) which allows for one to vary the degree of heterogeneity among the panel members. Moreover Pedroni (2001) has done residual based tests for the null of “no cointegration” for heterogeneous data. In Pedroni (2004) he extends the same test to include heterogeneous dynamics and slope coefficients. It examines both the between dimension and within dimension residuals. The strength of this test is that the resultant “test statistic” is able to accommodate short run dynamics, deterministic trends and also different slope coefficients. This test statistic is “standard normal’ and free of nuisance parameters.
164 Pedroni (2004) proposes the following way of testing for cointegration in a panel data setup. He proposes the following regression
yit= αi+δit+βXit+eit (5)
where yit = relevant variable where i= 1, 2….N observations and t= 1, 2….T time periods.
Xit = m-dimensional column vector for each member i t= time period under consideration
and βi == m-dimensional row vector for each member i
The variables yit and Xit are assumed to be I(1) for each member I of the panel, and under the null hypothesis of “no cointegration” eit will also be I(1). The parameters
αi and δi allow for differences between cross sections. The slope coefficient may also be different between cross sections. Pedroni (2004) proposes a set of residual based test statistics for the null of “no cointegration” which do not assume that the slope coefficient is the same in all cross sections.
First we test for the order of integration(non-stationarity) of the raw data series yit and xit. They are integrated of order one i.e., I(1.) The null is of no cointegration with an I(1) error structure. Here αi, δi and βi are allowed to be heterogeneous. The null is:
Ho : Panel series are not cointegrated, versus the alternative
HA : Panel series are cointegrated.
Here when we are pooling different data series, the slope coefficient iβ will not be of a common slope across different data series. The strength of these pooled tests is that the slope coefficients are not constrained to be the same, but rather allowed to be heterogeneous (i.e., allowed to vary across individual data series.) The tests distributional properties are that the standard central limit theorem (CLT) is assumed to hold for each individual series, as the time span grows. The advantage is that the error structure includes all auto regressive moving average (ARMA) processes.
The matrix structure is (m+1) x (m+1) in size where the off diagonal entities Ω2li capture the feedback between the regressors and the dependent variable, based on the invariance principle. Also cross sectional independence or process i.i.d. (independent and identically distributed) is assumed. This allows for the application of the standard
CLT even in the presence of heterogeneous errors. Here Ωi >0 ensures that there is no cointegration between yit. The invariance and cross sectional independence help construct the asymptotic properties of the test statistic. It allows the test statistic to converge asymptotically to the actual values.
(6)
(7)
These convergence results hold under standard assumptions. The assumption of sectional independence allows for “averaging” over the cross sectional sums of the panel statistic. Moreover it also reduces the effect of “nuisance parameters” due to serial correlation in the data as T→∞. This makes the computation a lot simpler. It also has another distinct advantage. Applying the limit T→∞ results in higher order
165 terms being eliminated prior to “averaging,” leaving only the first order terms of the time series. Pedroni considers two class of statistics. The first pools the residuals of the regression “within panel dimensions” and the second pools the residuals “between panel dimensions.” Similarly in equation (8) and (9)
(8)
(9)
and stand for “panel variance ratio statistic” and “panel t statistic” respectively. Equations (10) and (11) below pool the data “between panel dimension” to compute the group mean of the time series.
(10)
(11)
Pedroni (2004) then demonstrates the asymptotic distribution of the residual based tests for the null of “no cointegration” in heterogeneous panels. His results are fairly general and assumes “only finite second moments.” These results apply to all cases and for any number of regressors, when we measure the slope coefficients separately for each panel data series. He also conducts Monte Carlo simulations to study the small sample properties of the ‘statistic’ for different panel dimensions. He demonstrates excellent convergence of the “t” statistic (as “T” increases beyond 150 observations) keeping N fixed. Then he keeps “T” fixed and varies “N.” As the index becomes larger and larger the convergence properties becomes more stable. He also studies the strength and stability of his test statistic against various ‘alternative hypotheses.” Now regarding the data generating process, it is
yit = xit+ eit where
eit = øeit-1 + ηit and ∆ xit ~N(0,1)
ηit ~N(0,1), ø = {0.9, 0.95, and so on…}
The alternative hypothesis here is that the residuals eit is stationary. Pedroni uses the autoregressive (AR) process, rather than a moving average (MA) error correction process. The tests are powerful enough to show that using monthly data with more than 20 years of observations, it is quite easily possible to distinguish the cases from the null of “no cointegration” when the data is pooled. Moreover the Monte Carlo simulations show that: Case 1: For small panels, the group-rho statistic rejects the null of “no cointegration.” Case 2: For large dimensional panels, the panel –v statistic has the best power. The other statistics lie in between the two extremes of case 1 and case 2.
166 EMPIRICAL RESULTS
The inflation rates and interest rates for each panel are tested for the presence of unit roots using panel unit root test. At the 5 percent level, the G2 and G3 mostly have unit roots (for the G3 one statistic is against the presence of a unit root). For the G4 group the inflation series have an unit root whereas the interest series does not. For the G5 group the inflation series does not have an unit root whereas the interest rate series does. We then proceed to apply the Pedroni (2004) tests, which are a test of the null hypothesis that all the individuals in the panel are not cointegrated against the alternate hypothesis that a significant portion of the individuals are cointegrated. We also go on to estimate the Pedroni (2001) Fully Modified OLS (FMOLS) and Dynamic OLS (DOLS) tests which test whether the coefficient of the cointegrating equation is equal to one. The results given in table 2 are for the Pedroni (2004) tests and there is some evidence in favor of cointegration between inflation rates and interest rates for the G-2 group of countries. The standard model results in an acceptance of the null hypothesis
(H0: all countries in the panel are not cointegrated) whereas the time demeaned model shows evidence in favor of the alternate hypothesis of cointegration (hypothesis (H1: a substantial portion of the countries in the panel are cointegrated). This is (at best) weak evidence in favor of the weak form of the PPP hypothesis in the full data set. The results in table 3 are for the Pedroni (2001) test which is supposed to be carried out on a data set which is cointegrated and the null hypothesis is that the coefficient in the cointegrating equation is equal to one, which would be evidence in favor of the strong form of the Fisher hypothesis. Since we have weak evidence in favor of cointegration the results from strong form test are suspect. The null hypothesis of the existence of the strong form of the Fisher hypothesis is rejected in all four cases for the panel tests, indicating that strong form of the Fisher hypothesis does not exist for the panel or for individual countries. Tables 4-9 give the results of the Pedroni (2001, 2004) tests for the groups G3 (France, Germany, United States), G4 (France, Germany, United States, United Kingdom) and G5 (France, Germany, Japan, United States, United Kingdom). For the G3 countries the results are given in tables 4 and 5. The Panel Statistics in table 4 indicate rejection of the null in favor of the alternate hypothesis that a substantial portion of the countries are cointegrated as 6 of the eight statistics are significant. The Pedroni (2001) test results given in table 5 for the G3 countries indicates the rejection of the null hypothesis that the coefficient in the Fisher equation is equal to 1. Therefore there is some evidence for the weak form of the Fisher hypothesis for the G3 countries but no evidence in favor of the strong form of the Fisher hypothesis. For the G4 group of countries the results are given in tables 6 and 7. The Pedroni (2004) test results given in table 6 indicate that for the G4 countries 5 out of 8 statistics provide evidence in favor of rejecting the null hypothesis of no cointegration in favor of the alternate hypothesis of cointegration. The Pedroni (2001) statistics results given in table 7 indicate the rejection of the null hypothesis of the coefficient in the Fisher equation is equal to 1. Therefore the results provide evidence against both weak and strong form of the Fisher hypothesis for the G4 countries. For the G5 group the Pedroni (2004) statistics results given in table 8 indicate that the standard model is cointegrated whereas the time-demeaned model is not cointegrated. The Pedroni (2001) results for
167 the G5 countries given in table 9 indicate a rejection of the null hypothesis that the coefficient of the Fisher equation is equal to 1. Therefore the evidence is mixed in favor of the weak form of the Fisher hypothesis but against the strong form. There is some evidence in favor of the weak form of the Fisher hypothesis for the different groups of countries as shown in tables 2, 4, 6, and 8. The results presented in tables 3, 5, 7, and 9 however show that there is no evidence in favor of the strong form of the Fisher hypothesis.
CONCLUSION
We have looked at the evidence in favor of the Fisher effect for different groups of countries among the G-5 countries using panel data tests. These tests provide us with the opportunity for improving the power of cointegration tests when we don’t have access to a greater span of data. This is an important issue since the data for some countries is limited and carrying out panel data tests allow us to obtain robust results even with limited data. The evidence in favor of cointegration is weak at best. This implies that the evidence in favor of the partial Fisher effect is weak at best. There is no evidence in favor of the full Fisher effect for any of the groups or countries. The lack of evidence in favor of the strong Fisher effect indicates that while inflation and interest rates may move together for some countries, there is no one-to-one correspondence. On the other hand, weak evidence in favor of the partial Fisher effect indicates that there is some evidence that some degree of policy coordination has taken place over time, which is not surprising as these are some of the largest economies in the world. However, the weak evidence in favor of cointegration of the inflation rates and interest rates itself indicates that the countries do not have inflation rate targets. The European central bank (ECB) does have an inflation target, but it makes decisions for only Germany and France, and that too since 1999. Prior to that time the German central bank probably did pay more attention to inflation than other European Central Banks. In the United States too the primary concern of the Federal Reserve has been growth and stable prices, and not primarily stable prices. This would explain the lack of a Fisher effect.
168 REFERENCES
Atkins, Frank J.; Coe, Patrick J.; “An ARDL Bounds Test of the Long-Run Fisher Effect in the United States and Canada,” Journal of Macroeconomics, June 2002, v. 24, 2, pp. 255-66. Atkins, Frank J.; Serletis, Apostolos; “Bounds Tests of the Gibson Paradox and the Fisher Effect: Evidence from Low-Frequency International Data,” Manchester School, December 2003, v. 71, 6, pp. 673-79. Atkins, F.J. and Sun, Z., 2003. Using wavelets to uncover the Fisher effect. Discussion Paper 2003-09. Calgary: Department of Economics, University of Calgary. Bajo-Rubio, Oscar; Diaz-Roldan, Carmen; Esteve, Vicente; “Searching for Threshold Effects in the Evolution of Budget Deficits: An Application to the Spanish Case,” Economics Letters, February 2004, v. 82, 2, pp. 239-43. Carlson, J.A., 1977. Short term interest rates as predictors of inflation: Comment. American Economic Review.67, 469-475. Crowder, William J.; “The Long-Run Fisher Relation in Canada,” Canadian Journal of Economics, November 1997, v. 30, 4, pp. 1124-42. Crowder, William J.; “Hoffman, Dennis L.; “The Long-Run Relationship between Nominal Interest Rates and Inflation: The Fisher Equation Revisited,” Journal of Money, Credit, and Banking, February 1996, v. 28, 1, pp. 102-18. Dutt, Swarna D.; Ghosh, Dipak; “The Fisher Hypothesis: Examining the Canadian Experience,” Applied Economics, November 1995, v. 27, 11, pp. 1025-30. Dutt, Swarna and Ghosh, Dipak “A threshold cointegration test of the Fisher hypothesis: Case Study of 5 European Nations.” Southwestern Economic Review, Spring 2007, v. 34, 1, pp. 41-4. Evans, M.D.D., 1998. Real rates, expected inflation, and inflation risk premia. Journal of Finance. 53, 1:187-218. Fahmy, Yasser A. F.; Kandil, Magda; “The Fisher Effect: New Evidence and Implications,” International Review of Economics and Finance, 2003, v. 12, 4, pp. 451-65. Fama, Eugene F.; “Short-Term Interest Rates as Predictors of Inflation,” American Economic Review, June 1975, v. 65, 3, pp. 269-82. Fisher, I., “The Theory of Interest,” Macmillan, New York, NY, 1930. Hawtrey, K. M.; “The Fisher Effect and Australian Interest Rates,” Applied Financial Economics, August 1997, v. 7, 4, pp. 337-46. Joines, D., 1977. Short-term interest rates as predictors of Inflation: Comment. American Economic Review.67, 469-475 Kaliva, Kasimir; “The Fisher Effect, Survey Data and Time-Varying Volatility,” Empirical Economics, 2008, v. 35, 1, pp. 1-10. Lee, Jeung-Lak; Clark, Carolyn; Ahn, Sung K.; “Long- and Short-Run Fisher Effects: New Tests and New Results,” Applied Economics, January 1998, v. 30, 1, pp. 113-24. Mishkin, Frederic S.; “Is the Fisher Effect for Real? A Reexamination of the Relationship between Inflation and Interest Rates,” Journal of Monetary Economics, November 1992, v. 30, 2, pp. 195-215. Mishkin, Frederic S.; Simon, John; “An Empirical Examination of the Fisher Effect in Australia,” Economic Record, September 1995, v. 71, 214, pp. 217-29.
169 Muscatelli, Vito Antonio; Spinelli, Franco; “Fisher, Barro, and the Italian Interest Rate, 1845-93,” Journal of Policy Modeling, March 2000, v. 22, 2, pp. 149-69. Nelson, Charles R.; Schwert, G. William; “Short-Term Interest Rates as Predictors of Inflation: On Testing the Hypothesis That the Real Rate of Interest is Constant,” American Economic Review, June 1977, v. 67, 3, pp. 478-86. Olekalns, Nilss; “Further Evidence on the Fisher Effect,” Applied Economics, July 1996, v. 28, 7, pp. 851-56. Pedroni, Peter (2001), “Purchasing Power Parity in Cointegrated Panels,” The Review of Economics and Statistics, 83(4), 727-731. ------, (2004), “Panel Cointegration: Asymptotics and Finite Sample Properties of Pooled Time Series Tests with an Application to the PPP Hypothesis,” Econometric Theory, Vol. 20, No. 3, pp. 597-625. Perron, Pierre, (1989), The calculation of the limiting distribution of the least square estimator in a near integrated model, Econometric Theory, 5, 1989, 241-55. ------(1991) A continuous time approximation to the unstable first order autoregressive model: a case without an intercept, Econometrica, 59, 211-36 Tanzi, Vito; “Inflationary Expectations, Economic Activity, Taxes, and Interest Rates,” American Economic Review, March 1980, v. 70, 1, pp. 12-21. Tillmann, P., 2004. Testing for stationarity and prespecified cointegration under regime-switching: A note on the Fisher effect. Working paper. Bonn: The Institute for International Economics, University of Bonn. Westerlund, Joakim; “Panel Cointegration Tests of the Fisher Effect,” Journal of Applied Econometrics, March 2008, v. 23, 2, pp. 193-233.
170 TABLE 2: PEDRONI (2004) TESTS FOR PANEL COINTEGRATION: GERMANY, UNITED STATES – JUNE 1964 – JUNE 2013 ν-stat Rho-stat t-stat ADF-stat Panel Statistics Standard 0.8698 -1.9269 -0.6666 1.3921 Time demeaned 3.7894* -4.2430* -2.4984* -1.8270
NOTE: All reported values are distributed as N (0, 1) under the null hypothesis. An asterisk indicated rejection of the null hypothesis at the 10% level or higher.
171 TABLE 4: PEDRONI (2004) TESTS FOR PANEL COINTEGRATION: FRANCE, GERMANY, UNITED STATES – JANUARY 1970 – JUNE 2013 ν-stat Rho-stat t-stat ADF-stat Panel Statistics Standard 0.6873 -3.9694* -1.7949* 1.4159 Time demeaned 4.1062* -5.2421* -3.1087* -2.4759* NOTE: All reported values are distributed as N (0, 1) under the null hypothesis. An asterisk indicated rejection of the null hypothesis at the 10% level or higher.
TABLE 5: PEDRONI (2001) TESTS FOR PANEL COINTEGRATION: FRANCE, GERMANY, UNITED STATES – JANUARY 1970 – JUNE 2013 Country FMOLS t-stat DOLS t-stat France 0.11 -61.28** 0.13 -53.62** Germany 0.09 -52.45** 0.22 -27.34** United States 0.11 -43.96** 0.18 -28.03**
Panel results Without Time Dummies Between 0.11 -83.59** 0.18 -62.93**
With Time Dummies Between 0.03 -190.02** 0.07 -105.89**
TABLE 6: PEDRONI (2004) TESTS FOR PANEL COINTEGRATION: FRANCE, GERMANY, UNITED STATES, UNITED KINGDOM – JANUARY 1978 – JUNE 2013 ν-stat Rho-stat t-stat ADF-stat Panel Statistics Standard -0.1648 -0.7851 0.2043 2.5581* Time demeaned 3.7994* 3.8049* 2.5610* -1.7871* NOTE: All reported values are distributed as N (0, 1) under the null hypothesis. An asterisk indicated rejection of the null hypothesis at the 10% level or higher.
172 TABLE 8: PEDRONI (2004) TESTS FOR PANEL COINTEGRATION: FRANCE, GERMANY, JAPAN, UNITED STATES, UNITED KINGDOM – APRIL 2002 – JUNE 2013 ν-stat Rho-stat t-stat ADF-stat Panel Statistics Standard -1.3460 2.0620* 3.0140* 2.3197* Time demeaned -1.2303 1.4391 1.5109 1.4027 NOTE: All reported values are distributed as N (0, 1) under the null hypothesis. An asterisk indicated rejection of the null hypothesis at the 10% level or higher.
173 174 DIGITIZATION AND RECORDED MUSIC SALES WHITHER THE 'LONG TAIL'? Ian Strachan, West Texas A&M University
ABSTRACT
Recently, business models have reformed in the recorded music industry. Many have seen promise in the leveling of the playing field referred to as the ‘long tail effect’. Others predict a scaled-up presence of pop stars with an enhanced ‘superstar effect’. So which effect is evident in the data? In this research, point of sale data is tested during a transitional period from 2004 to 2008 for a ‘superstar’, ‘long-tail’, and combined effect. It is evident there is a superstar effect in digital formats, a long tail effect in non-digital formats, and a shrinking middle class of artists. JEL Classification: L26, L82, Z11
INTRODUCTION
Technological changes and digitization have been devastating to old business models in the recorded music industry. The key changes have included MP3s, peer-to- peer file sharing, subscription services, and, essentially, free music. This destruction opens up potential opportunities. But who benefits? Has this new economic reality changed the sales distribution profile in favor of pop stars or independents? There is potential for the industry to restructure in a decentralized way which fosters the relationship between the fan and the band. If artists and bands are free from constraining relationships with large companies with distribution monopolies and marketing bottlenecks such as terrestrial radio, then artists and bands can create their own rules. According to the French indie band Phoenix, there is ‘no one between you and your fans’ and there is more possibility for ‘indies to make the pop charts’ in the new paradigm (Phoenix 2010). But the broader context of the conditions of production in the music industry presents a new issue: the absence of scarcity goes hand-in-hand with the absence of property rights. This diminishes the chances for remuneration and may adversely affect the incentives for artists. In 1995, Robert Frank and Philip Cook published the book The Winner Take All Society. In it, they sought to explain why in entertainment and sports it is generally the case that there are a few very highly paid superstars at the top, and a very large group of lowly paid players and entertainers at the bottom. In a market economy, people are supposed to be paid their marginal product, but can it be the case that talent is so unequally distributed in society, or are there other possible explanations? According to Frank and Cook, the winner-take-all markets can arise whenever payoffs are determined by relative rather than (or in addition to) absolute performance (Frank
175 and Cook 1995: 24). The outcome is not necessarily desirable, since in a society where winner-take-all markets are prevalent, the unequal distribution of income has a depressing effect on economic growth, according to Frank and Cook. The reason is that the competition in these markets leads to a misallocation of resources in society. Those at the top earn more than they would in competitive markets based on absolute performance. In effect, the superstars in these markets have a degree of monopoly in the selling of their labor services. It is exploitation of the worker in reverse—it is now the capitalist who is exploited. Of course, in entertainment and sports where large amounts of public funds are often used to build venues for these performances, there is an additional cost to the consumers of entertainment and sports. Winner-take-all markets can be toppled by the development of technology. The creation of the Internet has led to a potential for reducing the costs of disseminating information about musical groups. Though it does not make the playing field totally level, fundamentally both the Rolling Stones and the local rock band have access to the Internet. In this case, technology increases competition and helps to eliminate the institutional structures (the record companies, radio, and so on) that helped create the winner-take-all market. When a few superstores negotiate contracts with a few record companies, it is not surprising that compensation is higher. The record companies believe that they can still profit on their investment, even at high rates of compensation for the superstore. However, if bands that are just starting out are able to get their music directly to the consumers without going through the record companies, both the musicians and the consumers will be better off. In his book, The Long Tail, Chris Anderson argued that the Internet and digital recordings would result in decreased distribution costs and inventory carrying costs which would increase product supply and variety. Following this, consumption patterns will diversify because preferences are better matched, and sales will move away from a small number of high-volume hits towards a larger number of low-volume niche products. These will make up a relatively larger proportion of revenues for retailers. Consumer search costs are lowered by online search, recommendation, and filtering tools (e.g. Brynjolfsson, Hu, and Simester 2011). This could help less-known and new artists gain exposure. According to the Long Tail thesis, the democratization of music production, distribution, and consumption should shift the music industry’s balance of power away from incumbents and in favor of these insurgent entrepreneurs. The winner-take-all market and long tail are paradoxical distributional outcomes which could both theoretically emerge from the same shift to digital technology in content-generation industries like the recorded music industry. As alluded to above, these may be countervailing trends. This notion is explained in more detail in the section below. Therefore, the purpose of this research is to discover the underlying distributional trends over time in the recorded music industry in light of the sweeping digital technology changes that have occurred.
WHO WILL LEAD THE MUSIC INDUSTRY INTO THE FUTURE?
Today, the Internet and digital technology have revolutionized the music industry. Signing with a major record company is no longer needed when the costs of recording and distributing technology have been dramatically reduced. New contractual arrangements and business models are being used by both established and
176 new artists. The fundamental research question addressed in this paper is whether these recent changes in technology coupled with entrepreneurial initiative in the form of new business models can change the ‘winner-take-all’ structure and create a ‘new artistic middle class’ that will better distribute the revenues generated by the industry. Hypothesis 1. There is a ‘superstar’ effect in the market for recorded music today. The superstar hypothesis is that sales have become relatively more skewed toward hits over time. If the distribution of recorded music sales has shifted toward top quantiles over time at the expense of all the other quantiles, this hypothesis will be supported (illustrated in Figure 1). Hypothesis 2. There is a ‘long-tail’ effect in the market for recorded music today that helps new or insurgent artists increase sales at the expense of established or incumbent artists. The long tail hypothesis is that sales have become relatively less skewed toward hits and superstars over time. If the distribution of recorded music sales has shifted toward bottom quantiles over time at the expense of all other quantiles, this hypothesis will be supported (Figure 1). The superstars hypothesis predicts that demand will become more homogeneous. The long tail hypothesis predicts that demand will become more heterogeneous. Paradoxically, the same technology change which leads to the leveling of the playing field in the long tail thesis could lead to the superstar thesis. Many researchers with relevant data believe the popular long tail theory has recently become dogmatic: nothing more than ‘web utopian fantasies’ (e.g. Gomes 2006). Hypothesis 3. There is some combination of the ‘superstar’ and ‘long-tail’ effects. The hybrid hypothesis is a combination of the effects of the above two hypotheses. If the distribution of recorded music sales has shifted such that both top and bottom quantiles increase at the expense of the middle quantiles, this hypothesis will be supported. This hypothesis could also be referred to as the ‘shrinking middle class’ hypothesis. Some recent studies have found a distribution becoming skewed toward the tail and the head simultaneously while the distribution is tucked towards the origin as absolute sales fall across the entire distribution, but relatively more for the middle quantile (Elberse and Oberholzer-Gee 2008, and Page and Garland 2009). This ‘hit-heavy, skinny tail’ implies greater inequality. However, unlike the superstars hypothesis, superstars in this distribution see a shrinking absolute share, even though their relative share is increasing. But this is also true of the observations in the tail – the ‘underdogs’ – because the tail becomes longer but thinner. Evaluation of these hypotheses is of great importance for the future of the music industry. The long tail theory itself became a hit among many industry operators and observers because it fit many peoples’ world view of increased competition and diversity. Some skepticism followed as creative industries continued to see shrinking sales. If there is a long tail effect, a leveling of the playing field is occurring, likely benefiting new entrepreneurs. If there is a superstar effect, consolidation by the few at the top of the hierarchy is occurring, likely benefiting incumbents. As Page and Garland (2009) ask, if indeed the tail of available niche products lengthens – a supply side effect – will it then ‘fatten’ with sales – a demand-side effect? This will depend upon how the demand for recorded music changes in response to the digital revolution (Tschmuck 2006). A change in the distribution of revenues from incumbent to insurgent entrepreneurs would imply increased creative freedom and innovation (e.g. Verboord 2011). On the other hand, investment in new talent could be dampened due to the
177 incumbent firms’ heavy losses. This latter prediction is suggested by Schumpeter’s later work (Schumpeter 1942, cited in Audretsch 2008). If innovation needs to be funded and large, incumbent firms can make the necessary investments. If startup costs are prohibitive to get a new creative act up and running, well-capitalized record labels may be necessary. But, both effects could cause innovation in business models as artists and firms cooperate or defect in enforcing copyrights, or shift to different approaches to making money. To examine these questions, this paper quantifies year-end sales in the recorded music industry over a five-year period from 2004 to 2008. Specifically, a random sample of 2,051 artists, 7,010 titles – which are albums, singles and videos – included on 1,836 labels are observed to see if and to what degree sales distribution profiles are changing from year to year. The changes are measured by a variety of techniques also used in other studies of this nature. But, no other research has addressed all of the above hypotheses in the recorded music industry with such a large sample or in such a recent, relevant time frame. This research will be proceed as follows: the next two sections will cover the data collection and a description of the data, respectively; following that are four sections on empirical results including Gini coefficients, absolute sales changes including summary statistics, and regression analysis results; the paper then finishes with concluding remarks; finally, all tables and figures referred to in the text follow after the conclusion section.
DATA COLLECTION
The data for this study comes from Nielsen Soundscan. Nielsen Soundscan collects point-of-sale transaction data on unit sales from over 14,000 retailers across the U.S. It comes from retail, mass merchant and non-traditional distribution channels (on-line stores, venues, mail order, and digital services). Big retail chains, such as Wal-Mart, Tower Records, Virgin Megastores, and Sam Goody – now For Your Entertainment (F.Y.E.) – report to Soundcan. The sales information collected by Nielsen is used by most music industry operators and has been used to create the Billboard music charts since 1991. They provide academic access packages as well. The necessary information for our research from the Nielsen system, pulled from ‘Chart and Sales History’ reports, are only available starting in 2004, due to a Nielsen system upgrade that July. But, the available time frame begins in week 1 of 2004 and goes through week 52 of 2009. Our focus is on a random sample of artists available during this time frame. Because of limitations in the system, the random sample covers 2004 through 2008. Artists were chosen randomly by a random letter generator and random page generator. Individual ‘artists’ are defined by Nielsen Soundscan. An ‘artist’ could be an artist’s stage name, an artist’s real name, a band name, an orchestra conductor’s name, a compilation of different conductors together, a movie soundtrack, a record label sampler, or other collaborations such as a movie or documentary for which copyright holders have an interest in tracking. Some artists release titles with a band as well as solo. If the solo artist’s name or stage name was listed on the page randomly selected, this became part of the sample. But any other artist-band collaboration the artists might be involved with was omitted if it was not listed on that page. The reason for this is that
178 an artist and band collaboration includes other creative inputs from other artists. It is the combination of inputs, i.e. defined by Nielsen’s listed ‘artist’, which were randomly chosen and so belong in the random sample. So in general, an individual artist could be an individual artist’s name or stage name with or without any collaboration by another artist, conductor, band, or bands. For example, if an artist showed up on the randomly chosen page, but a release recorded by this artist and her band showed up on another page, only the solo artist is included. We used the Chart and Sales History title reports because they displayed year-to- date (YTD) and release-to-date (RTD) unit sales broken into each sales format for each week. The broad format characterizations are albums, singles, or videos. These are then divided into YTD total units and YTD digital units. Total unit sales information for album releases also includes more specific formats: LPs, cassette tapes, CDs, DVDs, digital albums (bundled). Digital tracks have their own serial number and are categorized differently than digital albums and unbundled tracks, but are not included in this sample. Singles are shorter than albums with one or more tracks. Songs on a single can also come out on an album. Singles often contain a song or songs which are the most popular song or songs from an album and therefore serve a promotional purpose. Total unit sales for single releases specifically includes: CDs, 12-inch singles (12”), cassette tapes, digital singles (bundled), Maxi CDs. Total unit sales for videos specifically includes: UMDs (obscure), DVDs, video tapes. The detailed level of the formats included here is important because of the technology changes that have been occurring over the last few years and shifting consumer preferences. When a yearly sales figure of zero is observed for a given artist’s title, it is not known whether this title is available for sale or not. All that is known is that at some point in the past, the title was registered with Nielsen Soundscan. A band, for example, could have broken up years ago. This is important because the number of titles and artists with yearly sales of zero doubles from 2004 to 2008. So certain analyses below are done for two groups. The first group is non-conditional sales, which includes the zeros. The second group is conditional sales, which drops the zeros. So ‘conditional sales’ for the second group is conditional on sales being greater than zero, or nonzero.
DATA DESCRIPTION
Figure 2 displays all unit sales, bundled digital sales, and unbundled track sales from 2004 to 2008 for the sample used for this study. Non-digital sales are also shown, which is all sales minus digital sales. All unit sales rise from about 5 million in 2004 to peak at 7 million in 2006. From there, all unit sales decline to about 3 million – its lowest value – by 2008. The rise in digital unit sales increases consistently across this time frame from 43,000 in 2004 to 380,000 in 2008. Comparatively, digital sales are still dwarfed by CD sales. Digital sales are only 3% of CD sales in 2004 and 14% of CD sales in 2008. So, digital albums do not compensate for the sharp decline in physical CD albums. But at the same time, unbundled track sales rise drastically from 3 million to 43 million. This exemplifies an increased consumer preference for individual songs over the album as a unit. This is facilitated by the record label and retailer strategy of mixed bundling (Elberse 2009). Figure 3 confirms the above trends with sales per artist while average unbundled track sales do decline for artists in 2008. This decline is a symptom of the crowding of the market due to lower barriers to entry
179 catching up to the sales numbers. The supply of artists and bands has increased and so has competition. Figure 4 compares unit sales for the more popular formats: digital, CDs, and DVDs from 2005 to 2008. Figure 5 compares unit sales for the less popular formats: LPs, cassettes, Maxi CDs, 12 inch singles, video tapes, and UMDs. Although bundled digital unit sales numbers are available in 2004, the remainder of the formats are only available starting in 2005. For these formats, in order to find year-to-date (YTD) sales, release-to-date (RTD) sales for year t are subtracted from year t+1 RTD sales. Because the first year available is 2004, 2003 RTD figures are not available, and 2004 YTD numbers cannot be computed for the titles released before 2003. This is why Figures 4 and 5 only show a four year period. Looking at Figure 4, DVD sales peak in 2006 but then bottom out by 2008. This follows the pattern of overall unit sales in Figure 2. CD sales do the same, but the yearly changes are much more pronounced. Starting from 5.4 million in 2005 and peaking at nearly 7 million in 2006, CD sales settle to a new low by 2008 at 2.6 million. Bundled digital album sales rise consistently over the years, but are relatively small in volume. Figure 5 tracks the less popular formats over time. Perhaps not surprisingly given consumer technology trends, audio cassette sales plummet from 2005 to 2008. Another implication from Figure 5 is that LP sales are a niche market on the rise: increasing 107% from 2005 to 2008.
EMPIRICAL RESULTS
In this section, the distribution of sales is investigated for artists for each year of the sample by comparing changes in the proportion of the total sales captured by different percentiles of artists. To start, relative changes in the distribution of sales are analyzed by quantile. Subsequently, Gini coefficients are compared in the same man- ner. Following the analysis of Elberse and Oberholzer-Gee (2008), Table 1 shows sales distribution percentages corresponding to multiple quantiles of artists for each year in the sample. This shows, for all formats taken together, the percentage of total sales received by each decile of artists, as well as by each percentile above the 90th. These numbers imply that the market becomes more concentrated from 2004 to 2006. This is because each quantile of artists shown from the 60th on up to the 99th account for a smaller percentage of total sales from 2004 to 2006. The top 1% of artists increas- es until 2006 and is the driver this change. More concentration, in a relative sense, implies the tail of the distribution gets less. This suggests a superstar effect. However, later on the market becomes less concentrated. From 2006 to 2008 for artists there is less concentration because these same quantiles account for an increasing percentage of total sales. Less concentration implies the tail of the distribution receives relatively more sales. This suggests a long tail effect. Specifically, in 2004, 95% of artists ac- count for 2.65% of total sales. This drops to 1.08% by 2006, but rises again to nearly 2% in 2008. In 2004, 99% of artists account for 17.65% of total sales. This dips to a low of 7.35% in 2006, but rises again to 12.62% by 2008. Figures 6, 7, and 8 illustrate this distributional pattern for artists by graphing the percentage of total sales captured by each quantile listed for each year. This depicts the increasing and then decreasing trend in relative concentration.
180 GINI COEFFICIENTS
Another way to compare relative shifts in industry concentration over time is with Gini coefficients. A Gini coefficient has values between zero and oneand measures the degree of consolidation of a variable across individuals contributing to it. A higher Gini coefficient implies a higher concentration of the distribution. For example, a Gini coefficient of unity implies one individual sells 100% of the total sales while the remainder of individuals sell 0%. A lower Gini coefficient implies a lower concentration in the distribution of sales. The Gini coefficient results are displayed in Figure 12. All Gini coefficients are in the upper nineties which implies an extremely high concentration of sales. As with the analysis above, there is a trend of rising inequality and then declining inequality across the five-year time span. But for Gini coefficients, the peak concentration years range from 2005 to 2007. Many of these year-to-year changes for this measure are quite small which throws doubt upon their economic significance. Figure 13 shows the same Gini coefficient calculations but with zero sales dropped. A similar pattern is observed, however the decreasing concentration observed above in the later years is much less pronounced. In all three figures, there is rising inequality and then a leveling out in which consolidation of sales remains about the same or declines only slightly in the final year or two. So when only non-zero sales are taken into account, there is a superstar effect for the first three years – 2004, 2005, and 2006 – for albums, titles, and labels. Then, in 2007 and 2008, there is a slight decline in concentration. All of the Gini coefficients for artists end up higher at the end of the five-year period. This is a clear superstar effect, regardless of the slight decline in concentration observed in the final two years. Importantly, after removing the top selling artist, Johnny Cash, it is evident that he is the driver of these changes in the Gini coefficients. But that is the nature of a superstar, ‘winner-take-all’ business: if you remove the superstars, there is no business.
ABSOLUTE SALES CHANGES
This section looks at absolute changes by analyzing the location, scale, skewness, kurtosis, and inter-quartile measures from year to year. This illuminates the distributional shapes and, for example, how tails shift relative to the median. In th rd this section, quantile is abbreviated with a Q, such that Q.75 is the 75 percentile or 3 nd quartile. Q.50 stands for the median or 2 quartile. Table 2 shows absolute distributional changes from year to year across artists. For albums, singles, and videos together, the number of artists with zero yearly sales increases every year. For all formats, this number trends up from 477 artists in 2004 to 806 artists by 2008. The percentage of artists who sell one or more units, i.e. the percentage of artists that have non-zero sales, remains fairly constant at around 60%. For all formats, the number of artists reaching sales levels above the 50th percentile of unit sales increases steadily from 573 artists in 2004 to 947 in 2008. The number of artists with sales levels above the 90th percentile rises consistently from 117 in 2004 to 204 in 2008. The number of artists with sales above the 99th percentile rises from 12 to 21 over this five-year period. Summary statistics for the sales distributions for artists are revealed in Table 3.
The location of the median (Q.50) and scale, (Q.75-Q.25)/(Q.75+Q.25), remain the same
181 for each year. Scale remains the same because the 25th percentile is 0. The measure of skewness, (Q.75+Q.25-2Q.5)/(Q.75-Q.25), is positive and declines, i.e. becomes less skewed, steadily over time from 0.91 in 2004 to 0.75 in 2008. This means that the sales distribution becomes less concentrated which implies the long tail effect. Similarly, kurtosis, (Q.9-Q.1)/(Q.75-Q.25), generally declines over time from 15.75 in 2004 to 11.06 in 2007 with a slight rise again in 2008 to 11.75. A decline in kurtosis implies a more rounded peak and shorter, thinner tail for the distribution. These numbers provide evidence that the distribution of artist sales becomes less asymmetric and less concentrated. However, the decline in kurtosis means the sales distribution becomes less spread out. The inter-quartile measure (Q.75-Q.25), i.e. the difference between the 3rd and 1st quartiles, decreases consistently each year from 45.5 in 2004 to 16 in 2008. This implies that the middle class of artists is shrinking, which supports the hybrid hypothesis. The left tail inter-quartile measure (Q.50-Q.25)/Q.5) remains constant at 1. This is consistent with a long, flat tail. The right tail inter-quartile measure (Q.75- Q.50)/Q.50) declines each year starting from 21.75 in 2004 and ending at 7 in 2008. This also implies that the middle class of artists is shrinking, which supports the hybrid hypothesis.
REGRESSION ANALYSIS
There is much evidence that the sales distributions are changing from year to year. The next step is to test the statistical significance of these differences. In this section, unit sales are regressed on sales rank in the following log-linear form for annual sales:
ln(Sales+1)= β0+ β1∙ln(Sales Rank)+ε (1)
‘Sales’ are the total unit sales for each year from 2004 to 2008. ‘Sales Rank’ represents artists or titles ordinally ranked by the sum of their yearly unit sales. Sales is regressed on sales rank for artists as well as titles for total unit sales, digital unit sales, non-digital, and unbundled track sales. The coefficients on Sales Rank are then compared between years. Subsequently, the same regressions are performed with zero sales dropped. Firstly, β0 can be considered a measure of overall sales in a given year. Secondly, β1 can be considered a magnitude of how rapidly the share of yearly sales drops off as sales rank increases. The long tail hypothesis implies that β1 will decrease (e.g. Elberse and Oberholzer-Gee (2008)) in absolute value over time due to the institutional and technological changes as consumers diversify consumption as more variety is made available to them. So, the slope with which sales drops off will be less steep or become less negative over time. The intuition is that less popular artists will gain a larger share of sales over time. It is the same analogy for regressions with titles instead of artists. The superstar hypothesis would be just the opposite effect, with an absolute value of β1 increasing over time. The slope will become steeper or more negative from year to year. The same institutional and technological changes on the supply side allows consumers not only to ‘ride down the long tail’ to seek out unique, niche, and less popular content but also buy more of what the superstars have available because superstars can reach larger audiences than they could before the institutional constraints and bottlenecks were lifted.
182 Because both variables are in log form, β1 represents an elasticity. These elasticities will be negative because as sales rank increases for an artist, they are moving down the list with lower sales. As rank increases, sales decline. The β1 then can be referred to as a rank elasticity of sales or the elasticity of sales with respect to rank.
The β1s for each year are graphed for ease of comparison in Figure 11. To interpret these coefficients economically, recall that they represent elasticities. If 1β increases over time – decreases in absolute value – then sales become less elastic with respect to rank over time. For example, a 1% increase (decrease) in sales rank causes a decline (rise) in the average artist’s sales of 2.83% in 2004, and a decline (rise) in sales of 2.34% in 2008. All coefficients are statistically different from zero, p<0.01, 2 and all R s are high. β1 increases, or decreases in absolute value, for artists from 2004 through 2008 for all unit sales and non-digital sales. This implies a long tail effect: less concentration of sales over time. However, the β1 for digital units and for tracks declines, or increases in absolute value, each year from 2004 to 2008. This implies a superstar effect: less concentration of sales over time for these particular formats.
The tests for statistically significant differences between the 1β s between each pair of consecutive years show that these are all highly significant, p<0.01. So, the differences in concentration – the superstar and long tail effects – implied by the coefficients on sales rank each year are significant. For artists, the story for the regressions performed with only non-zero sales is similar. The coefficients on sales rank are graphed in Figure 12. All coefficients are statistically different from zero, p<0.01 and all R2s are high. For all unit sales and non- digital sales, β1 increases, or decreases in absolute value, each year for artists. This implies a long tail effect, but the year-to-year differences here are far less pronounced than above. This can be seen in Figure 12. Likewise, for digital units and tracks, the superstar trend is much less obvious. For artists the β1 on ln(Sales Rank) does not change much for digital sales. Unbundled track sales for artists show no obvious trend. For artists with non-zero sales only, the tests for statistically significant differences between the β1s between each pair of consecutive years are more nuanced. For all unit sales and non-digital sales, these are all highly significant, p<0.01. For bundled digital sales, the year-to-year differences are insignificant for 2004-2005, are significant at 5% for 2005-2006 and 2006-2007, and significant at 1% for 2007-2008. For unbundled track sales, the difference is significant at the 1% level from 2004- 2005, insignificant in between and then significant at the 1% level again from 2007- 2008. The same implications from the above regressions with sales of zero included still stand for all unit sales and non-digital sales when the zeros are removed from the regression: a significant long tail effect. For the two digital formats considered, the superstar effect is non-existent when only non-zero sales are considered in the regression. So, a superstar effect is evident in the digital formats: bundled digital albums and singles and unbundled tracks for both artists and titles. Consider this the ‘digital superstar effect’. A long tail effect is evident in overall unit sales and non-digital sales for artists. Consider this the ‘non-digital long tail effect’. However, when only non- zero sales are considered these trends are not as conspicuous, though they are still present. For non-zero artist sales, the digital superstar effect is diluted while the non- digital long tail effect is still there. In addition to the least squares approach above, quantile regressions are also used to estimate coefficients and the results above are corroborated. Therefore, for the sake of brevity the results are not reported here.
183 CONCLUSIONS
The recorded music industry is without question a winner-take-all industry where superstars like Johnny Cash and the Allman Brothers rule. Superstar artists still provide the most lucrative revenue streams for an ailing business. Observing the trends in the unit sales distribution over this five-year period, our study provides support for the superstar effect and the “winner-take-all” view of the music industry. Though this research provides some support for the long tail effect, it is for non-digital sales which would appear to be somewhat at odds with the idea that the Internet and digital technology has enabled niche artists to increase their market share. Why then does the superstar effect continue in the industry today? Though there are undoubtedly varied answers and the subject matter of another paper, we can perhaps make some conjectures. The fears that our music culture is being watered-down and homogenized by Internet technology and globalization are partially true. But this new technology and connectedness brings forth important benefits such as lower barriers to entry. Perhaps strangely, the notion of a ‘utopian’ long tail effect of a leveling playing field and a cornucopia of niches is also partially true. The evidence we find for a long tail effect is not a dramatic erosion of the empires of the superstars, but more akin to a persistent ‘trickling down’ of demand towards indie artists. The indies most likely are using digital and Internet technology to connect with their indie fans, but our evidence of a long tail effect in recorded music sales is found in the physical and analog world. Indie fan loyalty and reciprocity may be expressed in buying CDs and LPs at shows and record stores. For many indie artists and fans, cultural relevancy is more important than commercial gains – at least on the surface (e.g. Bourdieu 1993, 1984). The evidence we see for our hybrid hypothesis can be interpreted as a hollowing out of the middle class of artists. So although we observe a long tail effect, artists in the middle are not necessarily flourishing when it comes to record sales – perhaps they are losing the most in this situation. Our results are preliminary and subject to the limitations of the data collected through SoundScan. One problem is that SoundScan was created in the pre-digital, pre-Internet era and the organization of the data and the means of access reflects these limitations.
184 REFERENCES
Adler, Moshe. (1985). Stardom and Talent. The American Economic Review, 75 (1): 208-212. Anderson, Chris. (2004). The Long Tail. Wired Magazine, Issue 12.10 (October 2004), http://www.wired.com/wired/archive/12.10/tail.html (accessed 9/15/2008). Anderson, Chris. (2006). The long tail: why the future of business is selling less of more. New York: Hyperion. Audretsch, David B. Joseph Schumpeter and John Kenneth Galbraith: 2 sides of the same coin? Working Paper. March, 2008. Blackburn, David. On-line Piracy and Recorded Music Sales. Working Paper. December, 2004. Boldrin, Michele and David K. Levine. (2004). The Case Against Intellectual Monopoly. International Economic Review. Vol. 45. pp. 327-350. Bourdieu, Pierre. (1993). The Field of Cultural Production. Columbia University Press. Bourdieu, Pierre. (1984). Distinction: A Social Critique of the Judgment of Taste. Trans Richard Nice. Cambridge, MA: Harvard Univ. Press. Brynjolfsson, E., Y. Hu, & D. Simester. (2011). Goodbye Pareto Principle, Hello Long Tail: The Effect of Search Costs on the Concentration of Product Sales. Management Science, 57(8) 1373-1386. http://dx.doi.org/10.1287/ mnsc.1110.1371 Brynjolfsson, Erik, Yu Hu, and Michael Smith. (2003). Consumer Surplus in the Digital Economy: Estimating the Value of Increased Product Variety at Online Booksellers. Management Science. 49(11) 1580-1596. Brynjolfsson, Erik, Yu Hu, and Michael Smith. (2006). From Niches to Riches: Anatomy of the Long Tail. Sloan Management Review. 47(4) 67-71. Chevalier, Judith and Austan Goolsbee. (2003). Measuring Prices and Price Competition Online: Amazon.com and Barnesandnoble.com. Quantitative Marketing and Economics. 1: 203-222. Chung, Kee and Raymond Cox. (1994). A Stochastic Model of Superstardom: An Application of the Yule Distribution. The Review of Economics and Statistics, 76 (4): 771-775. Connolly, Marie and Alan Krueger. (2006). Rockonomics: The Economics of Popular Music. In Handbook on the Economics of Arts and Culture. Ed. Victor Ginsburgh and David Throsby. Elsevier. Crain, Mark and Robert Tollison. (2002). Consumer Choice and the Popular Music Industry: A Test of the Superstar Theory. Empirica, 29: 1-9. Demers, Joanna. (2006). Steal this music: how IP law affects musical creativity. GA: University of Georgia Press. Diplo [Video]. (2010). Retrieved 5/25/2010, from http://www.thecreatorsproject.com/ creators/diplo Elberse, Anita. (2010). Bye-Bye Bundles: The Unbundling of Music in Digital Channels. Journal of Marketing: May 2010, Vol. 74, No. 3, pp. 107-123. http:// dx.doi.org/10.1509/jmkg.74.3.107 Elberse, Anita and Felix Oberholzer-Gee. (2008). Superstars and Underdogs: An Examination of the Long-Tail Phenomenon in Video Sales. Working Paper. July
185 1, 2008. Fisher, Irving. (1925). Mathematical Investigations in the Theory of Value and Prices. New Haven: Yale University Press. Frank, Robert H. and Philip J. Cook. (1995). The Winner-Take-All Society. New York: The Free Press. Gomes, Lee. (2006). Many companies still cling to big hits to drive earnings. Wall Street Journal, August 2, Eastern Edition. Hamlen, William. (1991). Superstardom in Popular Music: Empirical Evidence. The Review of Economics and Statistics, 73 (4): 729-733. Hamlen, William. (1994). Variety and Superstardom in Popular Music. Economic Inquiry, 32 (3): 395-406. Hardin, Garrett. (1968). Tragedy of the Commons. Science. 162, 1243-1248. Heller, Michael. (1998). The Tragedy of the Anticommons: Property in the Transition from Marx to Markets. Harvard Law Review. 111 (3): 621-688. Hibbett, Ryan. (2005). What is Indie Rock? Popular Music and Society. 28:1, pp. 55- 77. Krueger, Alan. (2005). The Economics of Real Superstars: The Market for Rock Concerts in the Material World. Journal of Labor Economics, 23 (1): 1-30. Leeds, Jeff. (2005). The net is a boon for indie labels. New York Times. Dec. 27, 2005. Arts Section. (Accessed Online, 4/6/2009). Leeds, Jeff. (2007). The New Deal: Band as Brand. New York Times. Nov. 11, 2007. Music Section. (Accessed Online, 5/31/2009). MacDonald, Glenn. (1988). The Economics of Rising Stars. American Economic Review, 78 (Mar.): 155-167. Marshall, Alfred. (1947). Principles of Economics, 8th Ed. New York: MacMillan. Orlowski, Andrew. (2008). Chopping the Long Tail Down to Size, Wired’s Advice Could Seriously Damage Your Business. The Register. Nov. 7, 2008. Music and Media Section. (Accessed Online, 5/25/2010). Page, Will and Eric Garland. (2009). The Long Tail of P2P. PRS for Music. Issue 14. May 14, 2009. 1-8. Page, Will. (2008). Can 1,000 Fans Replace the Music Business? The Register. March 28, 2008. Page, Will. (2007). Is Live the Future of Music? The Report. Issue 181. November 29, 2007. Page, Will. (2006). Is the Price of Recorded Music Heading Towards Zero? Paper presented at Transmission Conference, November 30 – December 2, in Edmonton, AB, Canada. Phoenix [Video]. (2010). Retrieved 5/28/2010, from http://www.thecreatorsproject. com/creators/phoenix Posner, Richard A. (2004). “The Importance of Property Rights in the Common Law Tradition”. Property Rights: The Essential Ingredient for Liberty & Progress. Economic Education Bulletin: Proceedings of a Conference Sponsored by the American Institute for Economic Research. pp. 67-76. A.I.E.R., Great Barrington, MA: Nov. 4-5. Rose, Carol. (1986). The Comedy of the Commons: Custom, Commerce, and Inherently Public Property. The University of Chicago Review. 53: 711-781. Rosen, Sherwin. (1981). The Economics of Superstars. The American Economic Review. 71(5) 845-858.
186 Schumpeter, Joseph A. (1911). Theorie der wirtschaftlichen Entwicklung. Berlin: Duncker und Humblot. Cited in Audretsch, 2008. Schumpeter, Joseph A. (1942). Capitalism, Socialism, and Democracy. New York: Harper and Row Publishers. Cited in Audretsch, 2008. Spierdijk, Laura, and Mark Voorneveld. (2007). Superstars without talent? The Yule distribution controversy. SSE/EFI Working Paper Series in Economics and Finance, No. 658. December, 5. Tschmuck, Peter. (2006). Creativity and Innovation in the Music Industry. The Netherlands: Springer. Verboord, M.N.M. (2011). Cultural Products Go Online: Comparing the Internet and Print Media on Distributions of Gender, Genre and Commercial Success. Communications: the European Journal of Communication Research. 36(4), 441–462. http://dx.doi.org/10.1515/comm.2011.022
187 FIGURE 1: ILLUSTRATION OF THE SUPERSTAR AND LONG TAIL EFFECTS This is from Elberse and Oberholzer-Gee (2008) who examine both hypotheses of sales data for movies, another industry affected by the digital revolution.
FIGURE 2: TOTAL UNIT SALES EACH YEAR: ALL UNITS, TRACKS, DIGITAL, AND NON-DIGITAL, 2004-2008. For each year of the sample period. Included are all unit sales, unbundled digital track sales, bundled digital units sales (i.e. digital albums or singles), and non-digital sales (all unit sales less bundled digital sales).
188 FIGURE 3: AVERAGE UNIT SALES FOR ARTISTS: ALL UNITS, TRACKS, AND DIGITAL, 2004-2008. For each year of the sample period. Included are unit sales, unbundled digital track sales, and bundled digital unit sales
FIGURE 4: UNIT SALES FOR THE MORE POPULAR FORMATS: DIGITAL, CDS, DVDS, 2005-2008. Bundled digital albums and singles, CDs, and DVDs from 2005 through 2008.
189 FIGURE 5: UNIT SALES FOR THE LESS POPULAR FORMATS: LPS, CAS- SETTES, MAXI CDS, 12 INCH, VIDEO TAPES, UMDS, 2005-2008. LPs, cassettes, Maxi CDs, 12 Inch singles, video tapes, and UMDs.
FIGURE 6: PERCENT CUMULATIVE FREQUENCIES – PERCENT OF TOTAL SALES ACCRUING TO 4TH THROUGH 9TH DECILES OF ARTIST UNIT SALES, 2004-2008.
190 FIGURE 7: PERCENT CUMULATIVE FREQUENCIES – PERCENT OF TOTAL SALES ACCRUING TO 91ST THROUGH 99TH PERCENTILES OF ARTIST UNIT SALES, 2004-2008.
FIGURE 8: PERCENT CUMULATIVE FREQUENCIES – PERCENT OF TOTAL SALES ACCRUING TO 97TH THROUGH 100TH PERCENTILES OF ARTIST UNIT SALES, 2004-2008.
191 FIGURE 9: GINI COEFFICIENTS FOR ARTISTS, 2004-2008.
FIGURE 10: GINI COEFFICIENTS FOR ARTISTS, NON-ZERO SALES ONLY, 2004-2008. From 2004 through 2008 for only non-zero total unit sales, bundled digital unit sales, non-digital unit sales, and unbundled track sales.
192 FIGURE 11: COEFFICIENTS ON SALES RANK FOR ARTISTS, 2004-2008. For all unit sales, bundled digital albums and singles, unbundled digital track sales, and non-digital unit sales for each year.
FIGURE 12: COEFFICIENTS ON SALES RANK FOR NON-ZERO ARTISTS, 2004-2008. Coefficients on sales rank for artists with only non-zero sales for all unit sales, bundled digital albums and singles, unbundled digital track sales, and non-digi- tal unit sales for each year.
193 194 195 196 LOUISIANA MOTION PICTURE INCENTIVE PROGRAM: HOW WELL IS IT WORKING? Anthony J. Greco, University of Louisiana-Lafayette
ABSTRACT
This article explores incentives offered by states to induce the production of films within their borders. Specifically, it examines the efforts of Louisiana, the first state to offer some form of these incentives. Pro and con arguments relative to such incentives are presented. After reviewing legislation establishing incentives in Louisiana, the author examines relevant data to determine the economic impact of same on the state and concludes that the benefits have outweighed the costs. JEL Classification: H71
INTRODUCTION
As of the beginning of 2004, six states had Motion Picture Incentives (MPIs) designed to attract producers of motion pictures within their borders. Louisiana had been the first state to enact MPIs approximately a decade earlier in 1992. Hawaii and Minnesota followed in 1997; Missouri, in 1999; Virginia in 2001; and New Mexico in 2002. However, over the 2004-2009 period, 38 additional states enacted legislation creating MPIs. By the end of 2009, 44 states had some type(s) of MPIs. The six states lacking MPIs are: Delaware, Nebraska, Nevada, New Hampshire, North Dakota, and Vermont. The most common MPIs are: (1) Tax Credits, (2) Cash Rebates, (3) Grants, (4) Sales Tax Exemptions, (5) Lodging Exemptions, and (6) Fee-Free Locations. Of these, tax credits are the most significant. In fact, tax credits are the only type of MPI offered by Louisiana, one of the leading states involved in the production of motion pictures. Louisiana’s initial MPI legislation of 1992 enacted a tax credit for “investment losses in films with substantial Louisiana content.”1 In general, the tax credits offered by the various states remove a portion of the companies’ income tax, given that these companies satisfy certain stipulations. Cash rebates involve the reimbursement of a portion of the production companies’ qualified expenses. Grants are often, though not exclusively, tied to certain percentages of qualified expenses. While sales tax exemptions and lodging exemptions are basically self-descriptive, and while fee-free locations may offer the provision of rather complex services, the most common benefits offered are such things as traffic control by police officers and emergency standby crews (Luther (2010)). Of the MPIs listed above, 28 states offer tax credits, 17 offer cash rebates, 3 offer Grants, 28 offer Sales Tax exemptions, 33 offer lodging exemptions, and 6 offer fee-free locations. Three of the six that lack MPIs do not have at least one of the
197 major taxes to which the credits could be applied. That is, Nevada has no corporate or individual income taxes, Delaware has no sales tax, and New Hampshire does not tax general sales or wages. The three remaining states lacking MPIs (Nebraska, North Dakota, and Vermont) have, at least, considered legislation to install MPIs (Luther (2010)). Despite this, it is fair to say that states have generally moved heavily in the direction of offering MPIs in order to boast economic development within their respective borders. The question is “have they done so because such incentives have proved to be promotive of such growth or have they done so in a ‘monkey-see, monkey-do’ defensive posture?” It is frankly too soon to know. While the earlier states that moved to provide such incentives may have profited to some extent, perhaps the late-comers cannot, or perhaps, they can only hope to prosper by offering more lavish incentives. This could conceivably result in incentive-warfare similar to the price warfare experienced in various segments of the private sector in the early 20th century. Then, too, perhaps the benefits to be earned from such incentives have already reached critical mass and are purely illusionary beyond that point. If so, the increased incentive-thumping may merely devolve into a zero-sum game. Proponents of tax credits and other MPIs contend that these promote economic development and create substantial employment opportunities in the private sector while generating significant tax revenue for governmental entities. Critics of these incentives claim that the alleged benefits that they are said to generate are often based on fanciful estimates which cannot come to fruition. Further, the costs of the MPIs are said to be often understated. Also, many of the jobs created by film projects, critics claim, may only be of a temporary nature. Rather than granting tax relief on an industry-specific basis, critics of MPIs contend that states should implement tax systems that welcome all industries if they want to generate true wealth creation within their borders (Luther (2010)). However, given the recent nature of the provision of the MPIs, it is, as suggested earlier, too soon to draw any definitive conclusions about the long-run effectiveness of such incentives. Various studies conducted to evaluate the economic impact of state film incentives programs will be reviewed herein. In addition, a discussion of recent expansions and contractions in such state programs will be provided. The legislation of 2002 establishing the Louisiana film incentive program, as well as, subsequent amendments to the programs are discussed prior to an examination of the economic impact that the implementation of the program has had on the Louisiana economy. The paper closes with summary comments and conclusions relative to Louisiana film incentive programs.
REVIEW OF EVIDENCE
A number of studies have been conducted to evaluate the economic impact of state film incentives program. For example, Ernst and Young conducted a study for the New Mexico State Film Office employing three elements: (1) a survey of film industry employees and businesses related to the industry, (2) budget information submitted by film production companies within their applications to the New Mexico State Investment Council, and (3) qualifying expenditures by all film productions participating in the state’s film tax credit program. The study sought to assess the direct, indirect, and induced economic impact of the New Mexico Tax Credit Program
198 through (1) increased film production activity, (2) increased investment in New Mexico film studios and equipment, and (3) spending by tourists. At the time of the study, the New Mexico program, which was initiated in 2002, offered a film production tax credit of 25 percent of production expenses incurred during the production and post- production phases of each film produced in the state (Ernst and Young (2009)). The study concluded that the state’s film tax credit program had led to positive direct and indirect economic impacts attributable to film production spending activities, as well as, additional benefits from capital investments made to support the film industry’s growth and from film-related tourism. Specifically, the study found that film production activities generated 2,200 direct jobs in 2007. These 2,200 direct jobs led to 1,609 indirect jobs in other industries, resulting in a total employment expansion impact of 3,829 jobs associated with film production activities per se. Further, the study found that 5,381 (3,769 direct and 1,612 indirect) jobs were generated by film- related capital expenditures and film tourism spending. Hence, the total number of direct jobs generated by the program was found to be 5,989 and the total number of indirect jobs was 3,221. Overall, then, 9,210 jobs were found to be generated in 2007 by the state’s film tax credits. For the year (2007), 30 films were produced in New Mexico generating $253 million of spending benefiting the state’s economy. Further, these film expenditures generated increased state and local tax collections. For 2007, state tax collections emanating from film production activities reached $22.6 million. In addition, the study estimated that state tax impacts generated by capital expenditures undertaken in 2007 and from film tourism during 2008-2011 would total $21.5 million in 2007 dollars. Finally, the study concluded that, based on the 2007 value of present and future year tax receipts and the 2007 value of film production tax credits granted by the State of New Mexico, the state’s program earned $0.94 in additional state tax receipts for each dollar paid out in incentives. Coupled with the $0.56 earned by local governments within New Mexico, total state and local tax collections amounted to $1.50 for each dollar of state credits granted (Ernst and Young (2009)). The Center for Economic Analysis at Michigan State University conducted a study in early 2009 of the state-wide economic effects of audited expenditures of Michigan film productions in 2008, the year in which the state inaugurated its Michigan Film Production Credit. During the nine months following the passage of this program, 32 film productions were completed in the state generating $65.4 million dollars. Over $25 million was spent on direct wages and salaries with an excess of $40 million being spent on Michigan goods and services. Direct employment of Michigan residents totaled 2,763. An additional $28.4 million in state-wide expenditures were generated via the multiplier effect, resulting in a 2008 total of $93.8 million of state output emanating from film production expenditures. Further, 1,102 year-round equivalent jobs with total wage and salary income of $53.8 million were said to be generated from film productions via a multiplier effect (Miller and Abdulkadri (2009)). Film production expenditures were expected to increase over a four-year period. Both employment and output multipliers were expected to increase over time due to deepening value chains linked to infrastructure development within the state. Further, the report asserted that film expenditures would lead to increased migration into the state. The study, admittedly, did not consider the full bloom of motion picture and digital media production, such as the establishment of soundstages, production facilities, and other media production vehicles such as video game production. Finally, the study also did not consider real economic impacts on Michigan’s tourism industry
199 (Miller and Abdulkadri (2009)). Another study undertaken for a number of Convention and Visitor Bureaus in Michigan concluded that the economic impact of the state’s film tax credit program in 2009 and 2010 were significantly positive. For 2009, $119.1 million offilm- production expenditures were direct payments to Michigan residents and businesses. These direct impact payments generated estimated indirect impact payments of $190.2 million leading to an estimated combined total of $309.3 million of total Michigan economic output for 2009. The combined direct and indirect impacts on resident total income was $108.9 million and the combined impact on FTE resident employment was 2,631. All these measures rose in 2010. That is, combined economic output rose to $503.0 million. Total resident income increased to 172.5 million, with combined FTE resident employment rising to 3860. Further, indirect employment rose from 2009 to 2010 for all twelve industry groups within the state (Ernst and Young (2011)). State and local taxes generated from film productions showed increases, as well. For the state, the combined direct and indirect tax impacts resulting from film production activities were $15.2 million 2009 and $24.2 million for 2010 (an increase of over 58 percent). The combined direct and indirect local tax impacts was $4.3 million in 2009 and $6.8 million for 2010 (an increase in excess of 58 percent). Further, based on the assumption that film production employees were unemployed prior to the film productions undertaken, the study noted the additional fiscal impact of the reduction in state unemployment benefits resulting from jobs attributable to the film production activities. These reduced state unemployment benefits were $4.3 million for 2009, and $6.7 million in 2010 (Ernst and Young (2011)). In early 2010 Ernst and Young also issued an estimated impact study of the New York State Film Credit. The state had initiated the program in 2004 by offering a tax credit equal to 10 percent of qualified New York production activities. This state credit rate was increased to 30 percent in April 2008. In addition, there is a 5 percent credit applicable for qualifying New York City production expenses. The study found that $2,791 million of direct spending in 2009 generated total production value of $6,395 million via a multiplier of 2.29. Total multiplier-induced income was $3,351 million for 2009. Further, 11,262 direct jobs led to an increase of total employment of 32,027 via a 2.84 multiplier. In addition, the report projected that the average annual employment impact over the next five years would be 36,035 (Ernst and Young (2010)). A study of April 2010 funded by the Rhode Island Film Collaborative concluded that the state’s Motion Picture Production Tax Credit had been critical to the introduction and growth of the film and television production industry in the state. The study was designed to assess the economic impact of the credit on the state’s economy for the years 2005-2009. Positive results were found for the five-year period, as well as, for each of the five years relative to employment, wages and salaries, and employee state and local taxes generated. Overall, the total economic impact was $465.51 million. Wages generated for direct employees of film and television production companies were $181.7 million, while total wages generated for jobs created in other industries were $152.6 million. Further, the state’s tax credit was said to have created 4,184 FTE jobs for the 2005-2009 period (Mazze (2010)). The Massachusetts Department of Revenue has issued a report on the state’s film industry tax incentives wherein it concluded that the state’s film tax credit program resulted in $32.6 million in new spending in the state’s economy for 2009. In addition,
200 the report noted that $161.2 million in new in-state spending was generated by the program over calendar years 2006-2009. These numbers represent the net economic impact of the tax incentive program in that they allow for payments to non-residents and non-Massachusetts businesses, as well as state spending reductions needed to accommodate the tax credits and still maintain a balanced budget. For 2009, the tax credit program led to approximately 586 net new full time equivalent employees, with approximately 1,897 new FTE’s generated by production spending and its multiplier effects. State personal income over the 2006-2009 period rose by $115.4 million and total state revenue (state tax and non-tax revenue) rose by $36.3 million over the 2006- 2009 period (Department of Revenue (2011)). Connecticut’s Department of Economic and Community Development has found that its state’s film production tax credit program has generated an average of $995,401 of net state revenue over the 2006-2009 period. Its film production infrastructure tax credit program was found to have generated average annual net state revenue of $21,719 over the 2007-2010 period. However, the state DECD has recommended that the program should be continued, arguing that it is relatively new and has cost the state an insignificant amount of net revenue. The DECD believes an accurate assessment of other benefits accruing to the state from the program, such as net new jobs and procurement, needs to be undertaken by the state. Further, the third leg of the state’s incentive program, its digital animation tax credit, has generated an annual average of $510,159 over 2008 through 2010. The DECD has recommended that the state’s three film tax credit programs be maintained. It will analyze the performance of the programs every three years to examine the growth of the industry over time (McMillen and Smith (2011)). Meanwhile, the Tax Foundation, as mentioned earlier, continues to be critical of film tax credits as failing to fulfill their promise of overall economic growth and enhancement of tax revenue for states. The Foundation argues that the jobs created by such incentives are primarily temporary jobs and that competition among the several states has tended to benefit the movie industry more than local businesses or state revenue coffers. It asserts that 2010 was probably the peak for the state film incentives and notes that a number of states are ending their incentive programs finding, perhaps, that they cannot compete successfully in an over-saturated film incentive environment (Henchman (2011)). Further, most states are currently experiencing overall budgetary problems. Large expenses to cover Medicaid costs and other items combined with declining revenues due to the aftermath of the economic downturn have had a negative impact across the various states. The average deficit as a percentage of 2011 spending for all states is 17.6 percent. In fact, only six states do not project a budgetary shortfall for fiscal 2012. Therefore, several states have been engaging in budget reductions impacting spending on even essential services, such as education, health care, roads, and social services. It is, therefore, probably not surprising that some states may have cut or are contemplating cutting or even eliminating their film tax credit programs. Arizona and Washington ended their programs after 2010. Arkansas, Idaho, and Maine did not officially end their programs, but had not appropriated funds for them. Iowa, Kansas, and New Jersey suspended their programs. However, there is a push to reinstate the program in New Jersey, and the Iowa program was suspended due to the discovery of widespread fraud and abuse. The State of Alaska is considering non-renewal of its program. Tax review commissions in both Georgia and Missouri have recommended the elimination of the programs in their respective states. Further,
201 Rhode Island’s governor is seeking to end the state’s program. Connecticut, Michigan, and Wisconsin have reduced the generosity of their programs while New Mexico has put a cap on its program. Although Hawaii has maintained its program, legislators there have rejected efforts to expand it. The reader will recall the aforementioned studies recommending retention of programs in Connecticut, Michigan, New Mexico, and Rhode Island. All are presently still in effect despite the fact, as noted above, that the generosity of the Connecticut and Michigan programs have been reduced and that the program has been capped in New Mexico (Henchman (2011)). Of the 17 above states which have eliminated or reduced their programs or are considering doing so, two had passed their enabling legislation in 1997 and one did so in 2002. The remaining 14 had passed their enabling legislation over the 2005-2009 period, seven from 2007-2009. Probably, then, many of these states pursuing program elimination or reductions are acting under the pressure of strained state finances and not because they have become philosophically opposed to film tax credits or because they believe such programs to be ineffectual in promoting state revenues and economic growth. One would have to suspect this given the hesitancy of several states to eliminate their programs outright. Moreover, a number of states have taken positive action relative to film tax credit programs. Utah has enhanced the generosity of its tax credit while Wyoming had signed a five-year extension of its program. New tax credits have been put into effect in California and Virginia. Further, the governors of Ohio and Pennsylvania have chosen not to disturb or eliminate their existing state programs. Minnesota has restored funding for its program and there is a movement in Nevada to create a film incentive program. Obviously, Louisiana, and several other states, continue to actively utilize their programs to generate increased jobs and economic growth. Hence, there is a lack of unanimity relative to the merits and demerits of state film tax incentive programs. There will, undoubtedly be winners and losers among the several states employing such programs. Over time, one can expect a shakeout of state programs as those coming late to the stage or those experiencing continued tough economic conditions decide to pursue economic survival and development in other ways.
THE MPI EXPERIENCE OF THE STATE OF LOUISIANA
For decades, most film production was carried out in California and New York, which states remain the industry leaders. However, as indicated above, Louisiana actually became the first state to put MPIs into effect as early as 1992 though the state did not experience any substantive production activity until after the passage of its Motion Picture Incentive Act of 2002 and subsequent amendments to same in 2003, 2005, 2007, and 2009. The initial legislation created a 10 percent tax credit for investments between $300,000 and $1,000,000 and a 15% tax credit for investments exceeding $1,000,000. These credits applied to all production dollars inclusive of those spent out of the state. In addition, a 10 percent employment credit was provided on resident payrolls of $300,000-$1,000,000 and a 20 percent credit for in-state spending exceeding $1,000,000 (Acts 1 and 2 of the 2002 Extraordinary Session). The 2003 amendments made the credits transferrable and altered the thresholds for the 10 percent credit to investments between $300,000 and $8,000,000 and to investments exceeding $8,000,000 for the 15 percent credit. While the employment
202 credits remained unchanged, sales tax and use tax exemptions were added (Acts 551 and 1240 (2003)). Then in 2005, the state’s legislation was changed to a 25 percent tax credit for in-state spending (investment) only and a 10 percent additional credit for Louisiana resident payroll was provided. A 15 percent infrastructure credit was added, (due to sunset at the end of 2007), as well as, an audited expenditure report requirement for tax credit certification. The sales and use tax exemption however, was abolished (Act 456 (2005)). In October of 2005, the state’s Attorney General ruled that language in the 2005 law indicated that the infrastructure tax credit was 40 percent. In effect, the legislation had inadvertently added the 15 percent infrastructure credit to the 25 percent production credit. The 2007 amendments added further clarification of legislature intent regarding infrastructure, extended the sunset on the infrastructure credit by one year to the end of 2008 and set minimum thresholds and a $25 million per project credit cap for new infrastructure applications (Act 482 of the 2007 Regular Session). The 2009 amendments, descriptive of current incentive provisions, increased the tax credit for investments in excess of $300,000 to 30 percent and decreased the tax credit for expenditures on payroll for Louisiana residents to 5 percent of such payroll. It maintains the ability to transfer or sell any previously unclaimed tax credits to another Louisiana taxpayer or to the Governor’s Office of Film and Television. This transferability is important since most out-of-state production companies have no Louisiana State income tax liability. They can, in effect, monetize their credits through appropriate intermediaries by exchanging their credits for cash. However, transfers to the Governor’s office could be made for 85 percent of the face value of the credits (Act 478 (2009)). Hence, essentially the State of Louisiana went from granting tax incentives on a graduated scale between 10 and 15 percent on all production dollars and a graduated scale between 10 and 20 percent for resident payroll (2002-2005) to a single incentive rate of 25 percent for in-state productions spending only and a single 10 percent rate for resident payroll while adding a 15 percent sunset-provided infrastructure credit (2005). Then in 2009, the single incentive rate for in-state production expenses was increased to 30 percent while the single rate resident payroll incentive was lowered to 5 percent.
ASSESSMENT OF THE LOUISIANA MPI EXPERIENCE
The Louisiana MPI program began in earnest with the passage of the state’s Motion Picture Incentive Act of 2002. Hence, as of this writing, it is only in its ninth full year of operation (2003-2011), with eight years fully in the books, so to speak. Since the program is, thereof, still in its infancy, it may be premature to try to assess its impact on the State of Louisiana. However, available data relative to the program can, at least, provide some insight into the effectiveness of the Louisiana MPI Incentive program to date. The annual number of productions that have been certified by the State of Louisiana had increased from 5 in 2002 to 118 in 2010. The number of certified productions increased annually from 2002-2009, then fell off a bit (by 7.8 percent) from 2009 to 2010. Over the program’s eight full years of operation (2003-2010), there was a compound annual growth rate of 29 percent in the number of certified productions. In addition, there were 133 applications for
203 production incentives pending from 2010 and 58 pending for 2011. Total production budget spending rose over five years since the program’s inception and fell in only one (from 2006 to 2007) for which data are available. The compound annual growth rate of these expenditures over the 2003-2008 period was 22 percent. Actual budget dollars spent in Louisiana rose in seven years and fell in only one (from 2008 to 2009). The compound annual growth rate for this Louisiana spending over the 2003-2010 period was 30 percent. For the years 2002-2005, the percentage of production dollars spent in Louisiana remained about a third of the total production budget. This percentage increased to 41.6 percent from 2005-2006 and had reached levels of 83.1 and 72.7 percent, respectively, for 2006-2007 and 2007- 2008. This substantial increase in the percentage of total production dollars being spent within the State of Louisiana is largely attributable to the 2005 legislation which, as noted heretofore, had made the production credits exclusively applicable to in-state spending (investment). The higher the percentage of total production budget dollars spent in Louisiana, the greater, of course, is the economic impact of said production budget dollars on output, income, and employment within the state. The actual annual Louisiana payroll totals emanating from film production expenditures made within the state, as given in Table I in the “Tables” section at the back of this article, are used to derive annual estimates of final demand generated by these payroll figures. Estimates of final demand generated by the Louisiana payroll figures were derived through the procedure suggested by the Bureau of Economic Analysis (BEA) which involved dividing the annual Louisiana earnings (payroll) figures by the quotient obtained by dividing the BEA final demand earnings multiplier by the BEA direct effect earnings multiplier for the Motion Pictures and Video Industries Classification (NAICS 512100). Hence, dividing the final demand earnings multiplier for this industry in the State of Louisiana (.4670) by the direct effect earnings multiplier for said industry in the State of Louisiana (1.9493), yielded a quotient of .23957. The latter was then divided into each of the annual earnings figures in Table I to yield the estimates of final demand also shown in said table. Excluding the partial years of 2002 and 2011, the estimated final demand generated by the Louisiana payroll has increased in five years and decreased in two. The compound annual growth rate from 2003 to 2011 is 26 percent. Given these annual estimates of final demand generated by the Louisiana payroll attributable to film production, one can then derive the total output, employment, and value added effects of film production expenditures in Louisiana by applying the appropriate BEA final demand multipliers for NAICS 512100. In addition, we can also derive the annual effect of film production activity on earnings in the state of Louisiana by applying the appropriate BEA direct effect-multiplier for NAICS 512100. These data are reported in Table II. As with the Estimated Final Demand data listed in Table I, output generated, employment generated, value added generated and earnings generated, increased over five years and fell in two years over the 2003-2010 period. Again, the partial years 2002 and 2011, though reported in Table II, were excluded in the author’s range of examination. The common compound annual rate of growth of output generated, employment generated, value added generated, and earnings generated was again 26 percent. Therefore, despite the two years of decline in these data series, they have all experienced a healthy rate of growth over the 2003-2010 over which the Louisiana
204 motion picture incentive program has been fully operative: Coupling these with the upward trends noted above in both the number of movies produced in Louisiana and in the percentage of the associated production budgets being spent within the state, leads one to conclude that the production of movies has had a large positive effect on the private sector of the Louisiana economy and is likely to continue to do so into the future. However, since the State of Louisiana grants tax credit incentives to encourage the production of film projects within its borders, it does, as a consequence, give up some potential tax revenue. Hence, one can apply a benefit-cost analysis of the overall fiscal impact of the Louisiana motion picture tax incentive program on the State of Louisiana. The final benefits accruing to the state are the additional income tax and sale tax revenues resulting from the increased motion-picture production activity; and, as indicated above, the costs to the state are the tax revenues foregone by the state in its provision of the tax credit incentives. Annual tax revenues (sales and income tax) accruing to the State of Louisiana based on the earnings generated within the state by its motion pictures tax incentive program can be estimated for 2002-2011. The Louisiana state income tax ranges from 2 percent to 6 percent. The author applied an average rate of 4 percent to the state earnings. To derive the annual sales tax revenues resulting from the tax incentive program, the author focused on the broad categories of consumption spending relative to the market basket of goods and services employed by the Consumer Price Index and the percentages allotted to each within said index. For example, 4 percent of this market basket involves apparel while 5 percent of the same applies to entertainment and 7 percent applies to the “other” category. Expenditures made on these items would be subject to the state’s general sales tax of 4 percent. While food items take up 13 percent of the CPI market basket, the 4 percent state sales tax would apply only to those purchases made for prepared meals, as in restaurants. Purchases of food items made in grocery stores and the like are subject to a 2 percent state sales tax. Assuming that people, on average, eat at home five nights a week and eat prepared food outside the home two nights a week, the author applied the 4 percent state sales tax rate to only 28.6 percent of total food purchases (3.7 percent of the CPI market basket) and the 2 percent state sales tax rate to the remaining 71.4 percent of total food purchases (9.29 percent of the CPI market basket). In addition, the portion of the CPI market-basket allotted to private transportation is 16.082 percent. Over 9.9 percent of the CPI market basket is associated with the purchase of fuel and gasoline, purchases which are subject to federal excise tax. Hence, this leaves about 6.1 percent of the CPI market basket to the purchase of transportation goods and services subject to the state’s 4 percent sales tax rate. Accordingly, the author applied the state 4 percent sales tax rate to the 25.8 percent of annual earnings (the sum of the 4 percent of the market basket involving apparel, the 5 percent applying to entertainment, the 7 percent applying to “the other” categories, the 3.7 percent applying to food consumed in restaurants, and the 6.1 percent applying to transportation) and the 2 percent state sales tax rate to 9.29 percent of annual earnings noted above and added the tax revenue from each to estimate the annual totals of sale tax revenues generated as a result of the state’s motion picture tax credit incentives program. These annual sales tax totals were added to the annual income tax generated to yield its annual total tax revenue figures attributable to the state’s incentive program. It was found that total tax revenue generated by the state’s motion picture tax incentives program increased from $1,971,030 in 2003 to
205 $9,837,634 in 2010, experiencing five annual increases and two annual decreases over this period. Table III lists the annual total tax revenues accruing to the State of Louisiana and the annual tax credits granted under the state’s motion picture tax incentive program from 2002-2011. Since the dollar amounts in the last column of Table III are large negatives indicating that the annual amounts of tax revenue taken in by the State of Louisiana have been far less than the tax credits given by the state under its motion picture tax incentive program, said program would seem to be a failure in a pure fiscal sense. However, a different picture emerges in regard to the overall economic impact of the incentives program from Table IV which compares the total earnings generated by the program to the total tax credits granted under the same. As the reader can see, the total earnings generated within the State of Louisiana exceeded the total tax credits granted under the state’s motion picture tax incentive program in every year over the 2002-2011 with the exception of the partial first and last years of that period (2002 and 2011). This must be considered in conjunction with the aforementioned increasing number of productions being undertaken in the State of Louisiana, the increasing portion of the production budget being spent within the state, as well as, the increasing amount of output generated, employment generated and value added generated within the state in assessing the overall impact of the state’s incentive program. Doing so seems to suggest that the state’s program has been a success. In fact, Louisiana has been ranked among the top 10 American film locations (p3 update (2010)).
CONCLUSIONS
Louisiana has been in the forefront of the enactment of Motion Picture Incentives (MPIs) legislation designed to attract producers of motion pictures within their borders, having first passed such legislation in 1992. At present, all but six states have legislation providing various forms of MPIs. There are, of course, opposing views relative to such incentives. Proponents contend that MPIs promote economic development and create substantial employment opportunities in the private sector while generating substantial tax revenue in the public sector. Critics, however, contend, among other things, that the benefits claimed for such incentives are based on unrealistic estimates and that the costs associated with such incentives are often understated. Nevertheless, with Louisiana leading the way, states have been seeking to wrest film production enterprises from the traditional leading states of California and New York through the passage of legislation granting MPIs to film production companies. This paper has focused on the operation of the MPIs provided by the Louisiana legislation. It has detailed the provision of Louisiana’s MPI program as it has evolved through the passage of a number of state laws. Basically, the state offers tax credits for production activity occurring within its borders, as well as tax credits for wages paid to Louisiana residents. It provides for the transferability (sale) of these credits by companies not having any Louisiana tax liability. For a brief period, the state had also provided tax credits for infrastructure development within the state. The program actually began in earnest as a result of legislation passed in 2002; and, in truth, has only been in effect as of this writing for eight full years, 2003-2010. Therefore, it
206 may be a bit premature to try to formulate an assessment of the state of Louisiana’s MPI program, or, for that matter, that of any other state. However, one can, at least, examine the existing data in an attempt to determine the effects of the program to date. This is what the author has attempted to do within this paper. After examining the available data, it would seem that the fiscal costs of the state’s MPI program (the revenue foregone by the state through the provision of tax credits) far outweigh the fiscal benefits of said system (the tax revenue accruing to the state as a result of the MPI program). However, in the overall sense, the state’s MPI program as reflected in the substantial rate of growth of final demand, output, employment, value added, and earnings generated within the State of Louisiana, has been successful. The net benefits of the program, represented by the total earnings generated by the MPI, have exceeded the program’s associated costs in each of the program’s eight full years of operation. It remains to be seen if other states having MPI incentives (recall that 44 states do) will surpass Louisiana over time in terms of productions gained along, with their aforementioned concomitant benefits. However, it does not necessarily have to be a zero sum game where the gains made by one state are only made by inflicting losses on others. Perhaps the growth of motion pictures over time will accommodate increased productions in a large number of states. Further, while this paper has focused on incentives for motion pictures, several states, including Louisiana have incentive packages that apply to sound recording and digital media industries, as well. Consequently, there may be ample room for growth and success for several states in this more global electronic environment. Given Louisiana’s early start and growing success with its MPI program, as well as, its unique cultural and topographical features, it would seem that the state will reap positive net benefits from this program for many years to come. This is not necessarily true of all programs of the several states. As reviewed in this paper, some states have eliminated or reduced their incentive programs and others are considering such actions. Partly this is due to the tough budgetary conditions being encountered by most states. Then too, some state programs have proved to be more attractive than others due to differences in incentives offered in the various programs, as well as due to unique cultural and topological features found in those states. Overall, then, it is expected that there will be a gradual shakeout of states withdrawing from film tax credit programs. Louisiana’s tax credit program will be apparently one, which will survive and prosper due to its early start and the unique features it offers to the film industry.
207 REFERENCES
Combs, David( 2011) “State Budget Gaps: How Does Your State Rank?” Stateline Department of Revenue, Commonwealth of Massachusetts( 2011) A Report on the Massachusetts Film Industry Tax Incentives. Economics Research Associates( 2009) Louisiana Motion Picture, Sound Recording and Digital Media Industries: ERA Project No. 18014, prepared for State of Louisiana, Louisiana Economic Development, Chicago, Illinois. Ernst and Young( 2011) Economic and Fiscal Impacts of the Michigan Film Tax Credit, prepared for the Detroit Metro Convention and Visitors Bureau et al. Ernst and Young( 2009) Economic and Fiscal Impacts of the New Mexico Film Production Tax Credit, prepared for the New Mexico Film Office and State Investment Council. Ernst and Young( 2010) “Estimated Impacts of the New York State Film Credit,” prepared for the New York State Governors Office of Motion Picture and Television. Henchman, Joseph, Tax Foundation( 2011) Fiscal Fact, “More States Abandon Film Tax Incentives as Programs’ Ineffectiveness Becomes More Apparent,” Washington, D. C. Luther, Williams, Tax Foundation( 2010) Special Report, “Movie Production Incentives: Blockbuster Support for Lackluster Policy,” Washington, D. C. Louisiana Legislature, Acts 1 and 2, Extraordinary Session (2002). Louisiana Legislature, Acts 551 and 1240, Regular Session (2003). Louisiana Legislature, Act 456, Regular Session (2005). Louisiana Legislature, Act 456, Regular Session (2007). Louisiana Legislature, Act 478, Regular Session (2009). Mazze, Edward M.( 2010) The Economic Impact of the Motion Picture Production Tax Credit on the Rhode Island Economy for the Years 2005-200. McMillen, Stan and Smith, Catherine( 2011) Connecticut’s Film Tax Credits: An Economic and Fiscal Assessment for Connecticut Department of Economic and Community Development. Miller, Steven R. and Abdulkari( 2009) The Economic Impact of Michigan’s Motion Picture Production Industry and the Michigan Motion Picture Production Credit, Center for Economic Analysis, Michigan State University. www.p3update, viewed May 26, 2010.
208 209 210 211 Southwestern Society of Economists Past and Present Officers 1982 to Present 2014-2015 2011-2012 Adnan Daghestani, President Mihai Nica, President Barry University University of Central Oklahoma Chu V. Nguyen, President Elect & Program Chair Nicholas J. Hill, President Elect University of Houston - Downtown and Program Chair Ata Yesliyaprak, Vice President Jackson State University Alabama A&M University Susanne L. Toney, Vice President Mihai Nica, Secretary-Treasurer Hampton University University of Central Oklahoma Marshall Horton, Secretary-Treasurer Anne Macy, Editor Ouachita Baptist University Southwestern Economic Review Anne Macy, Editor West Texas A&M University Southwestern Economic Review West Texas A&M University 2013-2014 Susanne L. Toney, President 2010-2011 Savannah State University Geungu Yu, President Adnan Daghestani, President Elect Jackson State University and Program Chair Mihai Nica, President Elect Barry University and Program Chair Chu V. Nguyen, Vice President University of Central Oklahoma Univeristy of Houston - Downtown Nicholas J. Hill, Vice President Mihai Nica, Secretary-Treasurer Jackson State University University of Central Oklahoma Marshall Horton, Secretary-Treasurer Anne Macy, Editor Ouachita Baptist University Southwestern Economic Review Edward M. McNertney, Editor West Texas A&M University Southwestern Economic Review Texas Christian University 2012-2013 Nicholas J. Hill, President 2009-2010 Jackson State University Shari B. Lawrence, President Susanne L. Toney, President Elect Nicholls State University and Program Chair Geungu Yu, President Elect Hampton University and Program Chair Adnan Daghestani, Vice President Jackson State University Barry University Mihai Nica, Vice President Mihai Nica, Secretary-Treasurer University of Central Oklahoma University of Central Oklahoma Marshall Horton, Secretary-Treasurer Anne Macy, Editor Ouachita Baptist University Southwestern Economic Review Edward M. McNertney, Editor West Texas A&M University Southwestern Economic Review Texas Christian University
212 2008-2009 Edward M. McNertney, Editor Elizabeth Rankin, President Southwestern Economic Review Centenary College Texas Christian University Shari B. Lawrence, President Elect and Program Chair 2004-2005 Nicholls State University Elizabeth Rankin, President Geungu Yu, Vice President Centenary College Jackson State University Bruce C. Payne, President Elect Anne Macy, Secretary-Treasurer and Program Chair West Texas A&M University Barry University Edward M. McNertney, Editor M. Kabir Hassan, Vice President Southwestern Economic Review University of New Orleans Texas Christian University Lonnie Vandeveer, Secretary-Treasurer Louisiana State University 2007-2008 Edward M. McNertney, Editor Andrew Bacdayan, President Southwestern Economic Review Lamar University Texas Christian University Elizabeth Rankin, President Elect and Program Chair 2003-2004 Centenary College Marshall J. Horton, President Shari B. Lawrence, Vice President Southern Arkansas University Nicholls State University Elizabeth Rankin, President Elect Anne Macy, Secretary-Treasurer and Program Chair West Texas A&M University Centenary College Edward M. McNertney, Editor Bruce C. Payne, Vice President Southwestern Economic Review Barry University Texas Christian University Lonnie Vandeveer, Secretary-Treasurer Louisiana State University 2006-2007 Edward M. McNertney, Editor M. Kabir Hassan, President Southwestern Economic Review University of New Orleans Texas Christian University Andrew Bacdayan, President Elect and Program Chair 2002-2003 Lamar University Anisul M. Islam, President Anne Macy, Secretary-Treasurer University of Houston-Downtown West Texas A&M University Marshall J. Horton, President Elect Edward M. McNertney, Editor and Program Chair Southwestern Economic Review Southern Arkansas University Texas Christian University Elizabeth Rankin, Vice President Centenary College 2005-2006 Lonnie Vandeveer, Secretary-Treasurer Bruce C. Payne, President Louisiana State University Barry University Edward M. McNertney, Editor M. Kabir Hassan, President Elect Southwestern Economic Review and Program Chair Texas Christian University University of New Orleans Anne Macy, Secretary-Treasurer West Texas A&M University
213 2001-2002 Houston Baptist University Edward M. McNertney, President Elizabeth Rankin, Secretary-Treasurer Texas Christian University Centenary College Anisul M. Islam, President Elect John S. Kaminarides, Editor-in-Chief and Program Chair Southwestern Economic Review University of Houston-Downtown Arkansas State University Marshall J. Horton, Vice President Southern Arkansas University 1997-1998 Elizabeth Rankin, Secretary-Treasurer Jerry L. Crawford, President Centenary College Arkansas State University Edward M. McNertney, Editor Carl L. Montano, President Elect Southwestern Economic Review and Program Chair Lamar University 2000-2001 Waldo L. Born, Vice President Warren Mathews, President Eastern Illinois University Kingwood College Warren Matthews, Secretary-Treasurer Edward M. McNertney, President Elect and Houston Baptist University Program Chair John S. Kaminarides, Editor-in-Chief Texas Christian University Southwestern Economic Review Anisul M. Islam, Vice President Arkansas State University University of Houston-Downtown Elizabeth Rankin, Secretary-Treasurer 1996-1997 Centenary College Bill D. Rickman, President John S. Kaminarides, Editor-in-Chief Fort Hays State University Southwestern Economic Review Jerry L. Crawford, President Elect Arkansas State University and Program Chair Arkansas State University 1999-2000 Carl L. Montano, Vice President Waldo L. Born, President Lamar University Eastern Illinois University Warren Matthews, Secretary-Treasurer Warren Matthews, President Elect Houston Baptist University and Program Chair John S. Kaminarides, Editor-in-Chief Houston Baptist University Southwestern Economic Review Edward McNertney, Vice President Arkansas State University Texas Christian University Elizabeth Rankin, Secretary-Treasurer 1995-1996 Centenary College Donald L. Price, President John S. Kaminarides, Editor-in-Chief Lamar University Southwestern Economic Review Bill D. Rickman, President Elect Arkansas State University and Program Chair Fort Hays State University 1998-1999 Jerry L. Crawford, Vice President Carl L. Montano, President Arkansas State University Lamar University Warren Matthews, Secretary-Treasurer Waldo L. Born, President Elect Houston Baptist University and Program Chair John S. Kaminarides, Editor-in-Chief Eastern Illinois University Southwestern Economic Review Warren Matthews, Vice President Arkansas State University
214 1994-1995 Joseph A. Ziegler, President Elect Allyn B. Needham, President and Program Chair Needham Economic Research University of Arkansas Donald L. Price, President Elect Charles Becker, Vice President and Program Chair, Lamar University Texas Christian University Bill D. Rickman, Vice President Margaret O’Donnell Secretary-Treasurer Fort Hays State University University of Southwestern Louisiana Warren Matthews, Secretary-Treasurer Houston Baptist University 1989-1990 John S. Kaminarides, Editor-in-Chief Jim McMinn, President Southwestern Economic Review Austin Peay State University Arkansas State University Lynda Y. de la Via, President Elect and Program Chair 1993-1994 University of Texas-San Antonio Paul Kochanowski, President Joseph A. Ziegler VP for Membership Indiana University at South Bend University of Arkansas Allyn B. Needham, President Elect Margaret O’Donnell,Secretary-Treasurer and Program Chair, Bryan Industries University of Southwestern Louisiana Jerry L. Crawford, Secretary-Treasurer Arkansas State University 1988-1989 John S. Kaminarides, Editor-in-Chief David E. R. Gay, President Southwestern Economic Review University of Arkansas Arkansas State University Jim McMinn, President Elect and Program Chair 1992-1993 Austin Peay State University Charles Becker, President Lynda Y. de la Via, VP for Membership Texas Christian University University of Texas - San Antonio Paul Kochanowski, President Elect Charles J. Ellard, Secretary-Treasurer and Program Chair Pan American University Indiana University at South Bend Margaret O’Donnell, Vice President 1987-1988 University of Southwestern Louisiana Ray Perryman, President Jerry L. Crawford, Secretary-Treasurer Baylor University Arkansas State University David E. R. Gay, President Elect and Program Chair 1991-1992 University of Arkansas Joseph A. Ziegler, President Jim McMinn , VP for Membership University of Arkansas Austin Peay University Charles Becker, President Elect Charles J. Ellard, Secretary-Treasurer and Program Chair, TCU Pan American University Paul Kochanowski, Vice President Indiana University at South Bend 1986-1987 Margaret O’Donnell Secretary-Treasurer Gilbert Cardenas, President University of Southwestern Louisiana Pan American University Ray Perryman, President Elect 1990-1991 and Program Chair Lynda Y. de la Via, President Baylor University University of Texas-San Antonio David E. R. Gay, VP for Membership
215 University of Arkansas Charles J. Ellard, Secretary-Treasurer Pan American University
1985-1986 Thomas McKinnon, President University of Arkansas Gilbert Cardenas, President Elect and Program Chair Pan American University Ray Perryman, VP for Membership Baylor University Michael Crews, Secretary-Treasurer Pan American University
1984-1985 K.A.N. Luther, President Wake Forest University Thomas McKinnon, President Elect and Program Chair University of Arkansas Betty Slade Yaser, VP for Membership University of Houston-Clear Lake Michael Crews, Secretary-Treasurer Pan American University
1983-1984 Ernest R. Moser, President Northeast Louisiana University K.A.N. Luther, President Elect and Program Chair Wake Forest University Thomas McKinnon, VP University of Arkansas
216