Essays on Jobless Recovery

Essays on Jobless Recovery A Thesis Submitted to the Faculty of Drexel University by James Patrick DeNicco in partial fulﬁllment of the requirements for the degree of Doctor of Philosophy June 2013 ii

Acknowledgments

I would like to give a special thanks to my committee members, Dr. Christopher Laincz, Dr. Maria Olivero, Dr. Mark Stehr, Dr. Bang Jeon and Dr. Scott Dressler. I would also like to extend a thank you to the rest of the Drexel University Economics Department. All of you have been instrumental in this process from giving me ad- vice and guidance to being willing to sit through so many of my presentations with thoughtful attentiveness. Throughout my time here, I never felt like I was without help. I would be remiss if I did not single out the eﬀorts of my advisor, Dr. Christopher Laincz. He was always willing to sit and talk with me about my work, as well as read and re-read my papers countless times with red pen at the ready. He made me and my work so much better than it would have otherwise been. I really do appreciate all you have done for me. It was a pleasure and honor going through this process with all of the before mentioned people. Drexel University’s Economics Department is second to none in my opinion. It is the people that make it so. I would also like to thank my family, especially my wife Mary, for the patience I was shown throughout this process. They were supportive from start to ﬁnish. With- out that support this would never have been possible. I thank you all! iv

Table of Contents

List of Tables ...... vi

List of Figures ...... vii

Abstract ...... viii

1 Jobless Recovery in the United States ...... 10 1.1 Introduction ...... 10 1.2 US Data: Preliminary Analysis ...... 14 1.3 Analysis ...... 16 1.4 Robustness ...... 23 1.5 Forecasting ...... 25 1.6 Inﬂows to Unemployment ...... 27 1.7 Discussion ...... 28 1.8 Conclusions ...... 36

2 The Cost of Firing and the Speed of Hiring ...... 38 2.1 Introduction ...... 38 2.2 Background ...... 40 2.3 Literature Review ...... 43 2.4 The Model ...... 47 2.4.1 Households ...... 47 2.4.2 Representative Family Problem ...... 50 2.4.3 Firms ...... 51 2.5 STEADY STATE ...... 55 2.5.1 Parameters ...... 55 2.5.2 Equilibrium ...... 56 2.6 Impulse Response Functions ...... 58 2.7 Conclusions ...... 61

3 Employment-At-Will Exceptions and Jobless Recovery ...... 62 3.1 Introduction ...... 62 3.2 Literature Review ...... 65 3.3 Employment-at-Will Background ...... 68 3.4 Data ...... 70 3.5 Results and Analysis ...... 71 3.5.1 Single Variable Autocorrelation Regressions ...... 72 3.5.2 Panel VARs in Split Samples ...... 80 3.6 Robustness Tests ...... 83 3.7 Conclusions ...... 85 v

List of References ...... 88

Appendix A Chapter Two Steady State ...... 92

Appendix B Chapter Three Case Law ...... 94

Appendix C Tables ...... 97

Appendix D Figures ...... 114 vi

List of Tables

1 Historical Summary of U.S. Recoveries from Recessionary Periods . . 97 2 Unit Root Tests: Null Hypothesis for a Unit Root ...... 98 3 OLS VARs for Outflows Subsample...... 99 4 OLS VARs for Expansion Subsample...... 100 5 Robustness Test 1: OLS VARs for Outflows Subsample...... 101 6 Robustness Test 2: OLS VARs for Outflows Subsample...... 102 7 Forecast Model (Outflows: UR as Dependent Variable) ...... 102 8 Percent Industry Composition: Bureau of Labor and Statistics . . . . 103 9 Average Index Scores and Unemployment Rates from 1990 to 2010 . . 104 10 Average Changes in Unemployment Rates and State-Weighted GDP Growth, Conditional on the Presence of EWEs: Outflows and Expan- sion Periods...... 105 11 Average Changes in Unemployment Rates and State-Weighted GDP Growth, Conditional on the Presence of EWEs: Inflows and Contrac- tion Periods...... 105 12 Panel unit root tests: Summary For Differences in Unemployment Rates and State Personal Income-Weighted GDP, includes individual effects and a linear trend...... 106 13 Single Variables Analysis with Interaction Terms...... 107 14 Single Variable Analysis with Outflows and Expansion Subsamples. . 108 15 Single Variable Analysis with Inflows and Contraction Subsamples. . 109 16 OLS Panel VAR with Fixed State and Time Effects for Outflows and Expansion Subsamples...... 110 17 OLS Panel VAR with Fixed State and Time Effects for Inflows and Contraction Subsamples...... 111 18 OLS Panel VAR with Fixed State and Time Effects for Outflows and Expansion Subsamples with Union Density...... 112 19 Outflows: OLS Panel VAR with Fixed State and Time Effects and a Lagged Implementation Effect ...... 113 vii

List of Figures

1 US Unemployment (1948-2012)- BLS ...... 114 2 Out of Sample Forecast Results for Recovery from the 2001 Recession 115 3 Forecast Results from 2012 ...... 116 4 Forecast Results from 2012 ...... 117 5 JOLTS: Job Openings and Labor Turnover Survey (2001-2012) - BLS 118 6 Montly Participation Rate from (1948-2012) - BLS ...... 119 7 Median Usual Weekly Earnings of Full-Time Wage and Salary workers -BLS ...... 120 8 Industry Unemployment Rates: BLS ...... 121 9 Impulse Response Functions for Eﬀort and Wages ...... 122 10 Impulse Response Functions for Consumption Types ...... 123 11 Impulse Response Functions for Labor and Capital ...... 124 12 Recovery of Unemployment Rate Following a Negative Productivity Shock ...... 125 13 Percent Deviations From Steady State ...... 126 viii

Abstract Essays on Jobless Recovery James P. DeNicco

In this dissertation I focus on the topic of jobless recovery, which ex- plores the speed of recovery in unemployment rates post-recession, controlling for GDP growth. Chapter one furthers the empirical studies on the time series properties of the United States unemployment rate. Using vector auto regression models and controlling for changes in GDP, the unemployment rate and changes in the unemployment rate, I find structural breaks in 1959 and 1984 indicating that following a recession, the rate of decrease in the unemployment rate significantly slowed over time. Chapter one substantiates the phenomenon of jobless recovery in the United States and uses the timing of the structural breaks to review the possible causes and related theory, including industry composition, participation rates, entitlements and labor laws. Chapter two uses a representative, forward looking firm in capital and labor decisions to introduce separation costs into a discrete, dynamic, efficiency wage model in order to determine the effect of separation costs on both steady state unemployment rates and the hiring process following a negative productivity shock. I find higher separation costs cause higher steady states rates of unemployment and sclerosis of labor dynamics both in separations and hires following a recession. These findings provide a better understanding of the dynamics of post recession unemployment rates and show how firing costs may contribute to jobless recovery. In Chapter three, I study the effects on jobless recovery of diminishing the power of an employer to fire an employee through Employment-At-Will Exceptions (EWEs). I do so by using a dynamic panel with quarterly data ranging from 1976 to 2010 for the 50 ix states in the United States. I test both changes in state unemployment rates and state-weighted GDP growth in single variable regressions and VAR regressions. My contribution to the literature is threefold. First, I show two of the three EWEs contribute significantly to jobless recovery in the U.S. Second, I lend support to the predictions of theory that increased firing costs decrease the rate of hiring during recoveries. Third, I resolve differences in the various sources documenting the three types of EWEs in different states.

Chapter 1: Jobless Recovery in the United States

This chapter furthers the empirical studies on the time series properties of the United States unemployment rate. Using vector auto regression models and controlling for log differences in GDP, the unemployment rate and changes in the unemployment rate, we show that following a recession, the rate of decrease in the unemployment rate significantly slowed over time. We split our data sample to isolate the recovery portion of the unemployment cycle and find two structural breaks. Specifically the coefficients on the two structural breaks are positive and successively larger, which means controlling for GDP growth rates we are seeing weaker recovery in the unemployment rate , i.e. recoveries that are increasingly “jobless.” The first break is in 1959, and the second is in 1984 coinciding with the usual timing of the “Great Moderation.” Using the 7.85 percent unemployment rate at the end of 2012, recovery back to historical long run averages of 5.5 percent unemployment rates after the first structural break will take at least four additional quarters of 2.0 percent GDP growth. After the second structural break it will take at least another four additional quarters of the same growth to get back to the same unemployment rate. This chapter substantiates the phenomenon of jobless recovery in the United States and uses the timing of the structural breaks to review the possible causes and related theory, including industry composition, participation rates, entitlements and labor laws.

1.1 Introduction

The phrase “jobless recovery” became popular in the United States back in the 2000 recession, when it took seven straight quarters of GDP growth to result in decreases of the unemployment rate. With the 2008 recession, the phrase found new life in the beginning stages of recovery. The term is actually first found in print in The New York Times during the depression era of the 1930’s when the United States was experiencing its worst labor environment in history.1 The unemployment rate peaked just below twenty-five percent and took a decade to return to pre-depression levels. One definition of the concept refers to “An economic recovery, following a recession, where the economy as a whole improves, but the unemployment rate remains high

1http://www.npr.org/templates/transcript/transcript.php?storyId=113847257 11 or continues to increase over a prolonged period of time.”2 In this chapter we study jobless recovery as a comparative term looking at the relationship of GDP growth and unemployment rates over time. We consider a post-recessionary period as a jobless recovery if the speed at which the rate of unemployment declines is statistically and significantly slower than prior recessions. The statistical evidence found in this chapter lends credence to the applicability of the phrase when describing changes in the U.S. labor dynamics over time. We find that as U.S. GDP growth recovers after a recession, the size of decreases in the unemployment rate have lessened over time. Others have written papers dealing with jobless recovery but without isolating recoveries or looking at the long time series VAR relationship between actual unemployment rates and GDP growth. Goshen and Potter (2003) look at differences in the recovery from the 2001 recession compared to the recovery from the 1990-91 recession and find the summary evidence points to possible structural changes. They find the data suggest an increase in permanent job losses over temporary layoffs and inter-industry relocation of jobs may have created the 2001 jobless recovery. Aaronson, Rissman and Sullivan (2004) again focus on comparisons of the recovery from the 2001 recession looking at the effects of self-employment on jobless recovery. They use panel data of U.S. state unemployment rates from the Current Population Survey (CPS) back to 1979 to estimate predicted self-employment rates for the periods following the the 2001 recession until the fourth quarter of 2004. They find the predicted estimates are well below the actual value for that time frame. They con- clude that the joblessness of the recovery after the 2001 recession may be attributed to the temporary nature of the self-employed jobs. If these jobs are indeed temporary in nature, then as the labor markets recover and real wages rise there will simply be a shift back from self-employment to employment, keeping unemployment rates

2http://www.investopedia.com/terms/j/jobless-recovery.asp 12 persistently high with fewer workers moving from unemployment to employment. Faberman (2008) addresses jobless recovery using relatively new Business Employ- ment Dynamics (BED) data from the Bureau of Labor Statistics (BLS) covering the period from 1990-2006. He uses flow data to study job creation and job destruction as defined by Davis, Haltiwanger and Schuh (1996), again focusing mainly on the recoveries from the 1990-91 and 2001 recessions. Faberman extends his data back to 1947 using the BED and the previous estimates to create GMM predicted estimates of job creation and job destruction. Much in line with the structural breaks found in here, Faberman observes the magnitude of job flows began to steadily decline in the 1960’s and the volatility of job flows dropped sharply in the mid 1980’s. He attributes the jobless recovery from the 2001 recession to a persistent decline in the job creation rate and the recovery from the 1990-91 recession to an increase in the job destruction rate. He links them both to the reduction in volatility and increased persistence of job flows in the presence of aggregate shocks as seen in the Great Moderation period. (See Kim and Nelson (1999) and McConnell and Perez-Quiros (2000).) The novelty of this chapter is that it addresses the phenomenon of jobless recovery directly by isolating the recovery portion of the unemployment cycle using a long time series of quarterly data available from the BLS and BEA for national unemployment rates and GDP growth back to 1948. This approach allows us to quantitatively sub- stantiate the statistical significance of the comparative joblessness of recoveries over time. In addition, we can look at other macro variables that are changing over time in coordination with jobless recovery to begin to discern its causes. In order to better understand how unemployment dynamics have changed, it is worthwhile to generally describe the different post recession recovery periods over time. We do this in Table 1. We do not always follow the NBER official recession classification, but we will describe the instances we deviate from the NBER in the 13 table.3 Over time it appears that a cessation and reversal of the upward cycle of unemployment requires more persistent and greater growth. Visual inspection of Figure 1, the unemployment rate for the US, certainly suggests the speed of decrease in unemployment slowed over the past four post-recession periods. Using vector autoregression models with two found structural breaks and controlling for changes in GDP, we can show that following a recession, decreases in the unemployment rate significantly slowed over time. Specifically, the coefficients on the two structural breaks are positive and successively larger, which means with a certain growth level we are seeing weaker recovery in the unemployment rate. Thus we have “jobless recoveries.” The first structural break supports the notion from Blanchard and Simon (2001) that the Great Moderation is part of a trend that goes as far back as the 1950’s with a possible deviation in the 1970’s. The second structural break coincides with the period most often cited as the beginning of the Great Moderation in the 1980’s. These breaks suggest that the most recent changes in unemployment dynamics are linked to the documented changes in output and productivity dynamics. A large body of work examines both statistically and theoretically the various macroeconomic shifts that occurred throughout the Great Moderation in the US. We utilize the evidence presented here to discuss a number of the hypotheses put forward for jobless recoveries in the Great Moderation era. The rest of the chapter proceeds as follows: Section 1.2 summarizes the main data; Section 1.3 presents the empirical model and results; Section 1.4 presents the empirical models and results for our robustness checks; Section 1.5 examines the impact of jobless recovery through forecasting; Section 1.6 examines Inflows to unemployment; Section 1.7 discusses existing theory in light of the empirical evidence; and Section 1.8 concludes. 3There is actually a recession ending in Q1-1970, but the time period following it is so volatile that there is not enough of a recovery period to really examine. 14

1.2 US Data: Preliminary Analysis

The unemployment level and rate data come from the BLS, and the real GDP data is in chained 2005 dollars, from the BEA. Both series are seasonally-adjusted, quarterly, and span 1948 to 2012. The mean unemployment rate over the time period is 5.81% and the mean quarterly real GDP growth rate is 0.78%. We split the data into three periods: 1) before the fourth quarter of 1959 (1959Q4); 2) after 1959Q4 but before the fourth quarter of 1984 (1984Q4); and 3) after 1984Q4. The divisions stem from unknown structural break points we find, which are detailed in the following section. Both breaks indicate a statistically significant slowdown in the rate at which the unemployment rate falls post-recession. Here, we present a preliminary look at both unemployment and GDP behavior across the time divisions. Taking a summary look at the unemployment data, the average unemployment rate before 1959Q4 was 4.56%, 6.05% after the break until 1984Q4, and 6.11% thereafter. Looking at the periods following a recession for unemployment rates, the average negative change was -0.41% before 1959Q4, -0.25% after 1959Q4 until 1984Q4, and -0.17% after 1984Q4. These numbers demonstrate that the average negative movement of unemployment is smaller in absolute terms after each structural break. The average change during expansions was -0.20% before 1959Q4, -0.08% after 1959Q4 until 1984Q4, and -0.06% after 1984Q4. The slower mean rates of decline are informative confirming the flatter linear trend observable in Figure 1. These facts lend support to the idea that recovery from high unemployment will take longer after the structural breaks, but does not address the relationship between GDP and unemployment. The simple averages, of course, could coincide with differences in the length and/or strength of growth periods. Before 1959Q4, the average period of uninterrupted growth was 4.3 quarters, 6.4 quarters after the break until 1984Q4, and 17.0 quarters 15

of growth after that. Thus, the flattening of the unemployment rate recovery does not coincide with an increase in the frequency of recessions. The overall average growth rates during the three time periods were 0.91%, 0.87% and 0.65%, and the average growth rates during expansions in the three time periods were 1.53%, 1.21% and 0.78%. These figures indicate there has also been a decrease in the size of quarterly growth rates over all and during recoveries that coincide with the structural breaks. There is no link, however, between GDP and and unemployment rates in these summary statistics. The idea of a jobless recovery is that unemployment rates are slow to recover despite GDP growth. If slower recoveries in unemployment rates are due merely to slow downs in GDP growth, then we do not have a story of jobless recovery. This notion is addressed in this chapter using a VAR structure to test the system of unemployment rates, changes in unemployment rates and GDP growth as a whole. Calculating changes in employment per billion dollars of GDP growth during each time period allows us to look at the relationship between growth and at least one determining factor in changes in the unemployment rate. The economy grew in real terms by about $962 billion between 1948 and 1959Q4 with an increase in employment of 7.7 million. That equates to approximately 8,000 jobs for every billion dollars of GDP growth. Between 1959Q4 and 1984Q4 the economy grew in real GDP terms by $3.9 trillion with almost 41 million jobs added on net, or 10,600 jobs for every billion dollars of growth. From 1984Q4 through 2012Q4 real GDP increased $7.0 trillion with 37 million added in employment, resulting in 5,300 jobs added for every billion dollars in growth. While no causal relationship can be discerned from this simple exercise, it is quite striking that GDP growth after the structural break in 1984Q4 was about half as effective in increasing the number of employed than before the structural break. The prima facie evidence certainly suggests a case for increasingly “jobless recoveries” in the United States. However, the basic statistical presentation here does 16 not account for the non-linearity and asymmetry in unemployment rate movements. The higher the unemployment rate above long-run equilibrium, the greater the potential for rapid declines. Moreover, the changes to the simple averages could also be explained if a single outlier recession, the first oil shock being the most obvious candidate, behaved dramatically different from the long-run. The issue is important because if “jobless recoveries” are one time events associated with characteristics specific to certain recessions, that suggests the causes are then linked to those underlying features of specific recessions. On the other hand, if the trend is towards jobless recoveries in all recessions over time, it points towards searching for broader, systemic changes in the labor market. These issues we address in our time series analysis. The next section employs times series analysis to look for structural changes in unemployment rate behavior during both phases of the business cycle. That allows us to distinguish between explanations for jobless recoveries that would appear in only one or both phases. To preview, we find that post-recessionary periods certainly do behave differently over time, and the finding is quite robust. However, we detect little change in the behavior when unemployment is rising instead of falling. Thus, the labor market has changed during recovery periods, but not much over the actual recessions at least for the behavior of unemployment.

1.3 Analysis

The core approach of the analytical framework has its roots in Evans (1989), who describes US unemployment dynamics through the use of VAR’s. Evans’ paper addresses the behavior of the labor market from 1950 to 1985, asserting a structural break in 1974, which was chosen because it had the smallest standard error in his regressions. Evans speciﬁcation incorporated the unemployment rate, changes in growth and a structural break of the following form: 17

3 3 X X ∆Yt = ay + byi∆Yt−i + cyiUNt−i + dyD74 + eyt (1.1) i=1 i=1

3 3 X X UNt = au + bui∆Yt−i + cuiUNt−i + duD74 + eut (1.2) i=1 i=1 where ∆Y is real output growth, UN is the unemployment rate, and D74 is the structural break. Evans found a positive and significant coefficient for his structural break variable, modeled above, as well as for a time trend variable. However, Evans model is not ad- equate to study changes in the speed of recovery from high periods of unemployment. Because the specification above does not discern between the phases of the business cycle, it cannot capture the asymmetric dynamics of unemployment. (See Moosa et. al, 2004.) We analyze our times series data using VAR’s. Unlike Evans two variable system, we use three variables in order to adequately capture labor dynamics. We use both levels and differences of unemployment rates in our VAR with real changes in GDP. Because our focus is “jobless recovery,” it requires using differences in the unemployment rate. However, the unemployment level is an important factor in the size of the difference in unemployment rates. There are more likely larger declines in unemployment rates at the peak of a recession as opposed to near long run averages. We include a number of robustness checks and find support for the results of the main specification. In Table 2, we present unit root tests for the different lag lengths and model specifications. We test the variables using the Augmented Dickey-Fuller (ADF) test and the Dickey-Fuller Generalized Least Squares (DF-GLS) test. In all of our regressions we include either a constant or a constant and a time trend. The regressions are all run with between one and eight lag lengths for our endogenous variables. At a lag length of eight, we do not reject the unit root for the unemployment rate. However, 18 we firmly reject the unit root at a lag length of one, which is what we use in this chapter to capture the convexity of the unemployment rate cycle described in the paragraph above. The only other variable where we fail to reject the unit root in some tests was log differences in total factor productivity (DTFP), which was used as a robustness measure. We failed to reject the unit root for DTFP in the DF-GLS test with a constant for between five and eight lag lengths. Again, for that variable we only use a one period lag length for which we soundly reject the unit root in all tests. For the differences in the unemployment rate (DUR) and log differences in GDP (DGDP) we are able to reject the unit root in all tests performed. In order to ensure stationarity in everything that follows, DUR and DGDP are the only two variables for which we include up to eight lag lengths. Merz (1999) examines the time series properties of unemployment rate dynamics in different pieces. Merz uses the terms “Inflows” and “Outflows” to describe the two different movements comprising labor dynamics. Inflows are defined as those who become unemployed, while Outflows are defined as those who leave the state of unemployment. The idea gives insight into the different moving parts of the unemployment cycle, but still does not address the parts in separate phases of a cycle. For instance, Merz finds that Outflows have become more strongly countercyclical, but does not show which part of the asymmetric unemployment cycle drives the result. Therefore while the results from Merz (1999) lend support to Hall’s (2005) finding that the large majority of labor dynamics are due to fluctuations in the hiring rate as opposed to layoffs, it does not address the phenomenon of a “jobless recovery.” This chapter addresses the phenomenon directly by isolating the recovery portion of the unemployment cycle. In order to account for the asymmetry in unemployment dynamics, we need to separate the different phases of the unemployment cycle. We isolate the downward portion of the cyclical movements in the unemployment rate by splitting the sample. I 19 split the data into periods of Outflows and Inflows. While I use the same terminology as Merz, my definitions involve the description of periods of net movements instead of both movements in the same period. Outflows are characterized by periods of negative changes in the unemployment rate. Inflows are defined as periods of positive changes in the unemployment rate. As a robustness check we also split the data into two samples using changes in real GDP. One subsample consists of positive or expansion periods and the other consists of negative or contraction periods. This follows from the idea that unemployment and growth are inextricably linked. The results were supportive of the model using Inflows and Outflows. However, the regression equation from our system of equations with DUR as the dependent variable consistently showed better fits through the standards of AIC, SIC and adjusted R2, when using cyclical phases characterized as Outflows instead of Expansions. In other words, using the positive and negative movements of unemployment obviously describes the movements of unemployment better than the positive and negative movements of real GDP. Running a battery of regressions, interchanging unemployment rates, changes in unemployment rates, changes in GDP, participation rates, structural breaks and time variables, the overwhelming evidence shows that over time, changes in the unemployment rate have slowed during the downward phase of unemployment cycles. Including interaction terms only adds support to the basic regressions. In order to avoid biasing the regression results, we allowed the data to dictate the proper points for our structural breaks. We rely on the widely used Quandt- Andrews (QA) test for an unknown breakpoints. With jobless recovery as the focus of this chapter, we apply the QA tests to the portion of our VAR model with DUR, or changes in unemployment rates, as the dependent variable. We are looking for the breaks with regard to changes in unemployment rates, controlling for the level of the unemployment rate and GDP growth. We begin with no structural breaks in the specification and apply the AQ test to the entire (whole date range) split sample 20

to find the best place for a single structural break. If the result of the QA test is significant, we use that date as a structural break and test it for significance in the single regression equation from our VAR with DUR as the dependent variable.4 If the structural break is significant in the regression, we then lop off the beginning portion of data at that point and repeat the test going forward. We repeat this until we can no longer find a statistically significant structural break. For the Outflows susbsample, we find two structural breaks, which are both significant at the 1% level. Below are the main specifications for our VAR regressions in this chapter for both structural breaks and a time trend:

Structural Breaks

OF DURt = α + β1URt−1 + β2DURt−2 + β3DGDPt−1 + β4SB594t + β5SB844t + t (1.3)

OF URt = θ + γ1URt−1 + γ2DURt−2 + γ3DGDPt−1 + γ4SB594t + γ5SB844t + t (1.4)

OF DGDPt = ψ + φ1URt−1 + φ2DURt−2 + φ3DGDPt−1 + φ4SB594t + φ5SB844t + t (1.5)

Time Trend

OF DURt = α + β1URt−1 + β2DURt−2 + β3DGDPt−1 + β4QUART ERt + t (1.6)

OF URt = θ + γ1URt−1 + γ2DURt−2 + γ3DGDPt−1 + γ4QUART ERt + t (1.7)

OF DGDPt = ψ + φ1URt−1 + φ2DURt−2 + φ3DGDPt−1 + φ4QUART ERt + t (1.8)

Where DUROF is the change in the unemployment rate, UROF is the unemployment rate and DGDP OF is logged diﬀerences in GDP for the outﬂows subsample.

4For signiﬁcance of the QA test , we use Andrews’ (1993) critical values in Table 1 on page 840. 21

SB594, SB844 and QUART ER represent the structural break in the fourth quarter of 1959, the structural break in the fourth quarter of 1984 and the quarterly time trend. Looking at Table 3 we can see the results for our main specifications. While coefficients on the lagged variables in this VAR structure are biased, we see the expected signs on our coefficients. URt−1 represents a one period lag of the unemployment rate, and has an expected negative sign. The higher the unemployment rate (at peak is the highest in the cycle of course), the more likely there will be a negative change

in the unemployment rate. DGDPt−1 represents a one period lag of log diﬀerences in real GDP, and has an expected negative sign on the coeﬃcient. Real GDP growth

increases the likelihood of declining unemployment rates. DURt−2 represents a two period lag of the difference in the unemployment rate, and has the expected positive sign. The larger a change in the unemployment rate from two quarters past the larger will be this period’s change in the unemployment rate. Most importantly to this chapter are the results for the structural break and time trend variables. In both cases we see a positive sign with high levels of statistical significance. These results tell us, controlling for changes in GDP and controlling for levels of the unemployment rate, that during periods of Outflows changes in the unemployment rate have become less negative over time. It follows directly that recovery from elevated levels of unemployment following a recession will be slower as time goes on, using either a structural break or time variable representing the process. All of the above provide statistically significant evidence supporting the trends observed in the previous section.5

5These results hold with two signiﬁcant structural breaks for DUR or UR as the dependent variable with up to at least eight lags of DUR and DGDP . These results hold with two signiﬁcant structural breaks for DGDP as the dependent variable up to three lags of DUR and DGDP and for the 1984 structural break up until at least eight lags. 22

We repeat the same exercise explained above using the QA tests to ﬁnd unknown breakpoints for our Expansion subsample. Below are the resulting VAR speciﬁcations:

Structural Breaks

E DURt = α + β1URt−1 + β2DURt−2 + β3DGDPt−1 + β4SB601t + β5SB813t + t (1.9)

E URt = θ + γ1URt−1 + γ2DURt−2 + γ3DGDPt−1 + γ4SB601t + γ5SB813t + t (1.10)

E DGDPt = ψ + φ1URt−1 + φ2DURt−2 + φ3DGDPt−1 + φ4SB601t + φ5SB813t + t (1.11)

Time Trend

E DURt = α + β1URt−1 + β2DURt−2 + β3DGDPt−1 + β4QUART ERt + t (1.12)

E URt = θ + γ1URt−1 + γ2DURt−2 + γ3DGDPt−1 + γ4QUART ERt + t (1.13)

E DGDt = ψ + φ1URt−1 + φ2DURt−2 + φ3DGDPt−1 + φ4QUART ERt + t (1.14)

We still find two structural breaks, however, in different places. Looking at Table 4 we see the results for the Expansion susbample VAR’s. The first break in 1960Q1 is significant at the 1% level, but the second break in 1981Q3 is only significant at the 10% level. Even though the second structural break was weaker, the fact that the unknown structural breaks are found in such close proximity to the Outflows regressions add support to the idea that something occurred in these time frames causing slower recovery of unemployment after a recession. If we use the 1959 and 1984 structural breaks in this Expansion subsample, both breaks are significant at the 1% threshold with the same signs as the Outflows subsample regardless of the dependent variable. 23

1.4 Robustness

Due to the unconventional technique of using both levels and differences in the unemployment rate in order to fully capture the convexity of recoveries, the first robustness test uses a more traditional approach. The following VAR is for the Outflows subsample and includes only changes in the unemployment rates and GDP growth rates. We present a specification with four lags due to the quarterly data, but the results are robust from one to eight lags. Again we repeat our exercise of finding unknown breakpoints using this specification. The resulting VAR models are below:

Structural Breaks

4 4 OF X X DURt = α + γiDURt−i + βiDGDPt−i + β5SB602t + β6SB951t + t (1.15) i=1 i=1

4 4 OF X X DGDPt = θ + φiDURt−i + ωiDGDPt−i + ω5SB602t + ω6SB951t + t (1.16) i=1 i=1

Time Trend

4 4 OF X X DURt = α + γiDURt−i + βiDGDPt−i + β5QUART ERt + t (1.17) i=1 i=1

4 4 OF X X DGDPt = θ + φiDURt−i + ωiDGDPt−i + ω5QUART ERt + t (1.18) i=1 i=1

We still find two breaks, but again the specific quarters change. The first break point only changed from the first to the second quarter of 1960. However, when disregarding the level of the unemployment rate and focusing only on changes in the unemployment rate, the second structural break shifts about a decade to the first quarter of 1995. We put more weight on the main specification however, due the im- 24 portance of capturing the convexity of recoveries in unemployment rates as we have discussed previously. The 1990’s were generally a period of lower unemployment rates and one would expect smaller changes in the unemployment rate, controlling for GDP growth.6 Table 5 shows the results are indeed robust to the more conventional time series specification. On both structural breaks and the time trend there is still a positive and significant coefficient indicating that negatives changes in the unemployment rate have become smaller over time. We also want to know if these results are unique to a VAR including GDP growth or if they are still significant with alternative measures. We replace GDP growth with estimated quarterly TFP growth rates from John Fernald at the San Francisco Federal Reserve Bank. Again using QA tests to find unknown break points, we have our resulting VAR specifications below:

Structural Breaks

OF DURt = α + β1URt−1 + β2DURt−2 + β3DTFPt−1 + β4SB594t + β5SB844t + t (1.19)

OF URt = θ + γ1URt−1 + γ2DURt−2 + γ3DTFPt−1 + γ4SB594t + γ5SB844t + t (1.20)

OF DTFPt = ψ +φ1URt−1 +φ2DURt−2 +φ3DTFPt−1 +φ4SB594t +φ5SB844t +t (1.21)

6The results are robust to using the 1959Q4 and 1984Q4 structural breaks. 25

Time Trend

OF DURt = α + β1URt−1 + β2DURt−2 + β3DTFPt−1 + β4QUART ERt + t (1.22)

OF URt = θ + γ1URt−1 + γ2DURt−2 + γ3DTFPt−1 + γ4QUART ERt + t (1.23)

OF DTFPt = ψ + φ1URt−1 + φ2DURt−2 + φ3DTFPt−1 + φ4QUART ERt + t (1.24)

Convincingly, the structural breaks are in the exact same place for the Outflows subsample when using this measure of TFP growth as they were for GDP growth. Table 6 shows the results are indeed robust to this VAR specification, and are in fact stronger. Increases in productivity result in successively smaller changes in the unemployment rate at the structural breaks. Fernald calculates a number of different productivity measures along with this TFP, which he defines as output growth less the contribution of labor and capital. We also test the VAR using Fernald’s measure of utilization of capital and labor and utilization-adjusted TFP. The results are robust to both specifications and support the conclusion of this chapter that the U.S. is indeed experiencing jobless recoveries when compared to historical data.

1.5 Forecasting

While the results for the VAR with the changes in the unemployment rate on the left side of the equation are strong and robust, forecasting with the results for unemployment rates is helpful in showing the slowing trend of recovery from high rates of unemployment. In order to show the longer term, predictive power of the model, we perform an out of sample forecast using an Outflows model calculating the changes in the unemployment rate and predicting the new unemployment rate level. We repeat the process of finding unknown breakpoints with QA tests, but with the data cutoff before 2000. We then use the actual growth rates in my model and predict the re- 26 covery period following the 2001 recession, which resulted in an unemployment rate of 6.3% in June 2003. The results can be seen in Table 7 and Figure 2. Visually, the predicted values of the forecast model are very close to the actual values, and in fact, the average forecast error for the recovery from the 2001 recession is .08%. In order to understand the possible scale of the impact of slowing recovery times, Figure 3 uses the whole data range for the Outflows subsample. This forecast starts at the 7.85% unemployment rate at the end of 2012. Using an uninterrupted annual growth rate of 2.0%, the difference in the time for the unemployment rate to return to the historical long run average of about 5.5% before and after the structural break in 1959Q4 is about 4 quarters.7 The difference in the time for the unemployment rate to return to levels of about 5.5% before and after the structural break in 1984Q4 is 8 quarters. The model predicts that before the structural break in 1959Q4 the unemployment rate would return to 5.5% in the first quarter of 2014. The model predicts that it will now take until the first quarter of 2016, a two year difference. The forecast using the Outflows subsample might be overly optimistic. Using the regression model with the Outflows subsample ignores the fact that there are periods of increasing unemployment rates during expansions. Repeating the forecast exercise using the Expansion subsample reveals a much more dire scenario. Again starting at the 7.85% unemployment rate at the end of 2012, Figure 4 shows the prediction from the model with the Expansion subsample is that the U.S. will never really be able to get back to a 5.5% unemployment rate with 2.0% annual GDP growth. In order to get back to the historical long run average, the U.S. would need to experience more robust growth. The implications for longer unemployment recovery range from political with voting implications to various cost implications in terms of lost tax revenues, extended unemployment benefits, decreases in consumption and countless other examples. It is imperative to find causes and policy prescriptions for this neg-

72.0% is the average annualized quarterly growth rate during the most recent recovery. 27 ative trend. This chapter is aimed at starting that process.

1.6 Inﬂows to Unemployment

While this chapter is primarily concerned with the recovery portion of the unemployment cycle, we also investigate inflows to unemployment. If there have been shifts in the economy that slow down the hiring process due to increased costs of screening, then these same shifts could slow down the separation process. If a firm is already employing a productive worker then changes that make finding another equally productive worker harder may cause the firm to hold onto the current worker for a longer period of time, or until further declines in output. Performing the same procedures for Inflows to unemployment as we did for Outflows, we do not find a significant structural break through the QA test. This result by itself is interesting in that we may be observing differences in the firms decision making process when it comes to letting go of an employee rather than hiring one. This result supports two possible theories. One possible theory is that the separation process has been less effected over time than the hiring process. This outcome could be due to a shortsightedness of firm’s in the firing process. The other possible theory is that effects on Inflows during contractions and expansions cancel each other out. This outcome could be due to a calculation that during recession firms can unload less productive workers with less chance of workers claiming wrongful discharge or being successful if they do. Firms may carry unproductive workers during good times to avoid these possible separaton costs, and then dump them along with necessary cutbacks during recessionary periods. Regardless of the reason, we see the results are not symmetric for Outflows and Inflows. 28

1.7 Discussion

Like Merz (1999), Shimer (2005) and Hall (2005) challenge the conventional view on the dynamics of the unemployment rate with relatively new empirical evidence. The Job Openings and Labor Turnover Survey (JOLTS) greatly advanced our knowledge of separations. Traditional descriptions of unemployment dynamics would begin with a negative shock causing mass layoffs (or separations) that increase the number of unemployed. In turn, job-finding rates decline and the duration of unemployment rises. Hall and Shimer, backed by this empirical evidence, argue that separations play much less of a role in the unemployment dynamics than previously believed. They find that unemployment is high during a recession more due to firms reducing their hiring rates rather than increased separation rates. While there are changes in separation rates that accompany recessions, they are insignificant compared to regular, aggregate worker flows out of jobs. Looking at Figure 5 we can see this point in this most recent recession. While there was an increase in layoffs and separations in 2007, they are much smaller than the drop in both job openings and hirings. Further- more, we see the job openings and hiring rates respond much sooner to the economic downturn than layoffs and separations. Again, the results from Fujita and Ramey (2009) dampen the findings of Hall and Shimer with their statistical analysis showing that between 40 and 50 percent of fluctuations in the unemployment rate are due to changes in separation rates. However that still leaves between 50 and 60 percent of the fluctuations in unemployment rates to the job hiring rate. There are a number of models in contemporary literature to explain the relationship between hiring rates and changes in GDP. Hall (2005) goes through a number of papers focusing on search and matching models to describe employment dynamics. These models, going back to Mortensen and Pissarides (1994), are highly intuitive in that there exists some frictional unemployment in an economy due to the difficulty 29 of firms and workers finding matches satisfactory to both. Empirical evidence for these types of models can be seen in Valletta (2005), which investigates the negative short-term relationship between unemployment and vacancies in the Beveridge Curve. Valletta finds that regional disparities between jobs and workers can heavily influence the efficiency of workers and firms making matches. There are a number of possible explanations for the slower hiring rates that could be modeled using Mortensen and Pissarides (2004). An explanation that fits in per- fectly with search and matching models is the effect of increased labor participation on the unemployment rate. As people become discouraged and stop looking for work or find encouragement and enter the labor force to look for work, the participation rate will vary. Events that affect participation margins will affect the unemployment function in the search and matching model. Recovery from high rates of unemployment will be slowed as the number of participants increase. Sources affecting participation rates could be technological advances (such as the Internet) making the application process easier and less costly for job-seekers, changing demographics of the workforce resulting from larger numbers of women participating, or changes in the distribution of the labor force. Also the severity or characteristics of particular recessions may affect participation rates. With longer periods of high unemployment, non-working spouses may join the labor force to replace lost wages. Working in the other direction, as longer periods of unemployment persist, workers can become discouraged and give up on looking, which takes them off the unemployment roles completely. Peretto (2006) models the effect of participation margins on the unemployment rate. The paper investigates the effects of product and labor market frictions in a dynamic general equilibrium model with a three-states representation of the labor market: 1) Working or Employed, 2) Looking for work and cannot find it or Unem- ployed, and 3) Not participating in the labor market (maybe discouraged or content). One of the conclusions from Peretto’s model is that participation margins amplify the 30 effects of labor market frictions generating unemployment. With increased participation margins, there will be higher numbers of people who cannot find work. Looking at Figure 6, there has been a steady increase in the participation rate over time. Most of this increase is due to a larger role of women in the workforce. Coupling the increased participation rate over time with the pro-cyclical nature of participation rates, we may be seeing slowed hiring rates due to increased screening costs. With larger numbers of applications to sift through, the marginal cost of adding a worker increases. An implication of the invention of the internet is the ease with which job applications can be filled out and dispensed to potential employees. Not only will this increase the number of applications, but it will also exacerbate adverse selection. There will be larger numbers of less qualified people applying for jobs due to decreased application costs, further increasing the marginal cost of adding a worker. Hiring in the search and matching setting will also be affected by changes in the criteria of workers’ requirements on the supply side. The dramatic increases in social benefits seen in the U.S. over time may have led to an negative income effect on labor, which could result in a decrease in job search intensity and an increase in reservations wages. Using data from the Bureau of Economic Analysis, total government spending on social benefits to persons as a percentage of GDP has risen from 4.6% in 1960 to 9.7% in 1980 to 15.43% in 2010. Government assistance has increased in a number of areas, including healthcare, housing, food purchasing and income levels below certain poverty thresholds. From the U.S. Census Bureau for example, the percentage of people on medicaid has increased from 8.4% in 1987 to 15.9% in 2010 and the percentage increase in people on food stamps has outpaced the population growth rate 34.4% to 21.1% from 1990 to 2008. From the Social Security Administration, the number of people receiving Supplemental Security Income from the federal government has increased by 26.6% from just 2003 to 2010, while the population has only increased by 6.2%. 31

One of the more striking increases in entitlements has come in the form social security disability. Using data from the Social Security Administration and the Bu- reau of Labor and Statistics, there were about 152 people in the labor force per every worker receiving disability benefits in 1960. By 1970 that number quickly went down to 55, then 41 by 1990 and 18 by 2010. Workers on disability can still work and receive benefits as long as they make under a certain threshold or are not working too many hours. In bad times, disabled workers may be more willing stay at home or work part time, and accept less income supplemented by their benefits. In boom times there could be larger numbers of disabled workers either drawn into the labor force or moving from part time to full time work if the increase in wages and benefits is large enough. Looking at Figure 7, we see that real earnings have been generally increasing since the early 1980’s. However, it has been a volatile climb, and the real earnings at the end of 2012 are below the real earnings level in 1979. If wages have not sufficiently kept up with the increases in social benefits, then a decrease in search intensity and an increase in reservation wages could be very real contributors to jobless recovery. A second candidate for modeling the relationship between hiring rates and changes in GDP is the efficiency wage model from Shapiro and Stiglitz (1984). In this model involuntary unemployment can be explained by employers not being able to observe worker efforts. More recent papers like Alexopoulos (2003) extend the literature with monetary punishments for shirking instead of dismissals in order to fix issues with high wage volatility in the original model. Ichino and Maggi (2000) document different effects on the incentive to shirk by investigating a large banking firm with many branches all over Italy. They find there are significant relationships between shirking, which is defined here as misconduct and absenteeism, and individual backgrounds, group-interaction effects, and locational sorting. The effects of a changing composition of the labor force is an ideal candidate to 32 model the slow down of post recession hiring through efficiency wages. MacLeod and Malcomson (1998) look at the labor market conditions in which efficiency wages versus merit pay will be endogenously determined as optimal. They find that in highly capital intensive labor markets such as goods-producing and especially manufacturing sectors, the cost of vacancy is high and therefore efficiency wages will be optimal compared to merit pay. Efficiency wages will keep workers in their posts. An example of merit pay is a per item payment for goods produced or quantities harvested. Using merit pay in the service sector, however, is difficult because of firms’ inability to verify and quantify productivity. Often times in the service sector, firms depend solely on measures of quality. Quality is much more difficult to measure than quantity. The difficulty in verifying and quantifying productivity in the service sector makes shirking easier and more likely than in goods-producing industries. Firms in the service sector will have to pay hire wages to prevent shirking. Looking at Table 8, since 1948 the industry makeup of the United States has changed dramatically. The makeup of the U.S. labor force has moved away from being a goods producing, manufacturing heavy economy towards a service providing economy. We see distinct rises in Professional and Business services, Education and Health services and Leisure and Hospitality services. This shift may be contributing to jobless recovery in two ways. First, as demand increases after a recession, capital intensive industries hire quickly to keep costly capital from sitting idle. But with the large majority of the U.S. labor economy moving away from a manufacturing and goods-producing workforce, we may be seeing firms stretch employee productivity. Looking at Figure 8 we see the goods-producing sector is much more responsive to both the initial recession and the beginning stages of recovery. Second, if efficiency wages are being paid in the service sector, the higher marginal cost of labor will drive down employment. Another candidate for the slowdown in unemployment recovery is labor law. The 33

first large structural break in the sample was 1959Q4, which coincides with the passing of major, federal labor protection laws through the 1960’s and into the early 1970’s. Of note is: 1) The passing of the Work Hours Act of 1962, which ensured time-and-a-half pay for workdays over eight hours or work weeks over 40 hours. 2) The Age Discrimination Act in Employment (ADEA) of 1967, which “protects certain applicants and employees 40 years of age and older from discrimination on the basis of age in hiring, promotion, discharge, compensation, or terms, conditions or privileges of employment.” 3) The Occupational Safety and Health Act of 1970 to “assure safe and healthful working conditions for working men and women; by authorizing enforcement of the standards developed under the Act; by assisting and encouraging the States in their efforts to assure safe and healthful working conditions; by providing for research, information, education, and training in the field of occupational safety and health; and for other purposes.” 4) The Employee Retirement Income Security Act (ERISA) of 1974, which “is a federal law that sets minimum standards for most voluntarily established pension and health plans in private industry to provide protection for individuals in these plans. ERISA requires plans to provide participants with plan information including important information about plan features and funding; sets minimum standards for participation, vesting, benefit accrual and funding; provides fiduciary responsibilities for those who manage and control plan assets; requires plans to establish a grievance and appeals process for participants to get benefits from their plans; gives participants the right to sue for benefits and breaches of fiduciary duty; and, if a defined benefit plan is terminated, guarantees payment of certain benefits through a federally chartered corporation, known as the Pension Benefit Guaranty Corporation (PBGC).”8 It is hard to ignore the role of these new, larger costs of employment and operation on firms’ hiring practices. With increased marginal labor costs, firms may want to higher less workers and screen applicants more carefully,

8United States Department of Labor: http://www.dol.gov/dol/topic/discrimination/agedisc.htm 34

which slows down the hiring process. Again this can add to the adverse selection problem with the more qualified employees more likely already taken through better screening. This leaves the remaining pile of applicants less qualified, requiring stricter screening and even higher marginal labor costs. The structural breaks also coincides with changes in the legal environment of employment relationships, but this time through the state court systems. These changes, called Employment-At-Will Exceptions, are documented in Muhl (2001). There are three major categories: 1) Public Policy Exception (43 states have adopted this), under which “an employee is wrongfully discharged when the termination is against an explicit, well established public policy of the State;” 2) Implied Contract Exception (43 states have adopted this), which is ”applied when and implied contract is formed between an employer and an employee, even though no express, written instrument regarding the employment relationship exists;” and 3) Covenant of Good Faith and Fair Dealing Exception (11 states have adopted this), which “read a covenant of good faith and fair dealing into every employee relationship. It has been interpreted to mean either that employer decisions are subject to a ‘just cause‘ standard or that terminations made in bad faith or motivated by malice are prohibited.” Court opinions explaining these laws are written loosely (more so as they go on), leaving substantial room for legal interpretation and recourse on employers when an employee is separated. The precedent of Employment-At-Will goes back to 1884 in Payne v. Western & Atlantic Railroad, which states that employees can be fired for, “good cause, bad cause, or no cause at all.”9 Beginning with California in 1959 and spreading to the majority of states throughout the 1980’s and early 1990’s, state court systems diminished the power of U.S. firms to fire workers at will as governed by the doctrine of Employment-At-Will.10 While all of these labor laws and court opinions at the state and federal level are written to protect workers, there could be unintended conse-

9Payne v. Western & Atlantic Railroad, Supreme Court of Tennessee, 1884. 10Petermann v. Intl Brotherhood of Teamsters, 344 P.2d 25 (Cal. Ct. App. 1959 September). 35 quences of more cautious and slower hiring by firms incurring larger marginal costs of employment through screening and policy. Both the search and matching models and the efficiency wage models can be used to model changes in labor laws. The search and matching models would incorporate changes in legislation through the matching process. The laws would create an environment with higher cost causing the conditions for hiring to be more stringent. Efficiency wage models can also model these increased costs. DeNicco (2011) models labor dynamics in the efficiency wage model with forward looking firms, which internalize the cost of separating employees when they are hired and finds higher separation costs cause higher steady states rates of unemployment and sclerosis of labor dynamics both in separations and hirings following a recession. The Insider-Outsider Model from Olivier and Summers (1987) could also be used to model the effect of labor legislation on the relationship between unemployment and changes in GDP. Olivier and Summers model workers having power in compensation bargaining with the firm. Workers and firms can jointly decide to adjust salaries dependent on expected economic conditions (such as GDP growth). The stronger the workers’ power, the higher the wages they can demand. This may lead to overall compensation being too large, which will result in higher rates of layoffs from un- expected, negative shocks. Going forward the large costs of employment may stifle hiring. Olivier and Summers had unions in mind when they put together their model. With the percentage of union workers in the United States decreasing, the Insider- Outsider model would seem to lose applicability. However, if labor legislation is doing the work of the unions in increasing worker power, then the Insider-Outsider model may still be applicable to a slowdown in hiring rates. 36

1.8 Conclusions

Using vector auto regression models and controlling for log differences in GDP, the unemployment rate and changes in the unemployment rate, we show that the United States is entering into an era of “jobless recoveries.” We split our data sample into positive and negative changes in the unemployment rate, which we refer to as Inflows and Outflows, to isolate the recovery portion of the unemployment cycle. We find structural breaks for changes in the unemployment rate in the Outflows subsample in the fourth quarter of 1959 and 1984 using the Quandt-Andrews test for unknown breaks. When we incorporate the structural breaks into our split sample VAR’s, the coefficients with changes in the unemployment rate as the dependent variable are statistically significant, positive and successively larger. This result indicates that controlling for the unemployment rate and GDP growth, the U.S. is experiencing weaker recovery in unemployment rates. Using the 7.85 percent unemployment rate at the end of 2012, recovery back to historical long run averages of 5.5 percent unemployment rates after the first structural break will take four additional quarters of 2.0 percent GDP growth. After the second structural break it will take eight additional quarters of the same growth to get back to the same unemployment rate. When we replace the structural breaks with a time trend, we find the coefficient for the time trend is also statistically significant and positive. We conduct a number of robustness tests and the results all support our main specification. First we change the way we split our sample. We split the data into periods of positive and negative GDP growth, which we refer to as Expansions and Contractions. Using the same process of testing for unknown breaks, we find structural breaks in the first quarter of 1960 and the third quarter of 1981. The proximity of the structural breaks and the sign of their statistically significant coefficients when using the Expansions subsample support our main specification. Using a more con- 37 ventional VAR with only changes in the unemployment rate and log differences in the unemployment rate and using the same process for finding unknown break points, we find positive and statistically significant structural breaks in the second quarter of 1960 and the first quarter of 1995. The shift of the second structural break to 1995 shows the importance of including the level of the unemployment rate in our VAR’s. The 1990’s were a decade of relatively low unemployment rates. Due to the convexity of recoveries in unemployment rates, we would expect smaller negative movements at lower rates. However, the similar results in similar time frames are supportive of increasingly jobless recoveries in the U.S. over our sample. Finally, when we replace GDP growth with estimated TFP growth, we find structural breaks in the exact same quarters as our main specification. The coefficients are once again statistically significant, positive and even larger than in our main specification with changes in the unemployment rate as our dependent variable. Af- ter substantiating jobless recovery in the United States, we using the timing of the structural breaks to review the possible causes and related theory, including industry composition, participation rates, entitlements and labor laws. 38

Chapter 2: The Cost of Firing and the Speed of Hiring

Using a representative, forward-looking firm in capital and labor decisions, this chapter introduces separation costs into a discrete, dynamic, efficiency wage model in order to determine the effect of separation costs on both steady state unemployment rates and the hiring process following a negative productivity shock. This chapter builds on the general equilibrium models of Burnside, Eichenbaum, and Fisher (2000), Alex- opoulos (2003), and Alexopoulos (2004), which correct for empirical shortcomings in the original Shapiro and Stiglitz (1984) model. I compute equilibrium dynamics in order to investigate the impact of separation costs during a recession. I find higher separation costs cause higher steady states rates of unemployment and sclerosis of labor dynamics both in separations and hirings following a recession. My findings are applicable to comparisons of labor dynamics both in Europe versus the U.S. and in the U.S. over time. These findings provide a better understanding of the dynamics of post recession unemployment rates. I am also able to capture the larger role of hiring rates over separation rates in labor dynamics from Hall (2005) and Shimer (2005) by distinguishing between workers separated due non-cyclical reasons and those who are part of mass layoffs in a recession.

2.1 Introduction

Going back to the hysteresis literature of Blanchard and Summers (1987), the United States has come to be characterized by highly fluid labor markets with relatively low unemployment rates, while most of the European labor markets have grown sclerotic with high unemployment rates. Recent efforts categorizing differences in countries’ labor protection policies show that European labor laws have been far more worker friendly than those in the U.S. The OECD publishes an employment protection index, which rates countries on the friendliness of a their labor laws towards workers.1 The index rates countries on ten different categories, including administrative procedures for separations, conditions for fair and unfair separations, statutes of limitations and compensation for unfair dismissals, and various other categories that would in-

1www.oecd.org/employment/protection 39 crease a firm’s separation cost. The United States historically ranks the least friendly, while most of Europe ranks very high. However, throughout the 1980’s and 1990’s individual states within the United States instituted common laws, referred to as Employment-At-Will Exceptions, to increase the power of employees with regard to wrongful discharge. Using a representative, forward-looking firm in capital and labor decisions, this chapter introduces separation costs into a discrete, dynamic, efficiency wage model in order to determine the effect of separation costs on both steady state unemployment rates and the hiring process following a negative productivity shock. This chapter builds on the general equilibrium models of Burnside, Eichenbaum, and Fisher (2000), Alexopoulos (2003), and Alexopoulos (2004), which correct for empirical shortcomings in the original Shapiro and Stiglitz (1984) model. I build on these models by computing equilibrium dynamics to investigate the impact of separation costs during a recession. This chapter uses separation costs to explain observed differences in both steady state unemployment rates and dynamic movements of unemployment rates out of steady state. I find higher separation costs cause higher steady state rates of unemployment and a hardening of labor dynamics both in separations and hirings following a recession. My findings are applicable to comparisons of labor dynamics both in Europe versus the U.S. and in the U.S. over time. I am also able to capture the larger role of hiring rates over separation rates in labor dynamics, as argued by Hall (2005) and Shimer (2005), using an exogenous separation rate, taken from the average separation rate in the BLS. The exogenous separation rate allows for a distinction between workers that are separated due non-cyclical reasons and those who are part of mass layoffs in a recession. These findings provide a better understanding of the dynamics of post-recession unemployment rates. 40

2.2 Background

The motivation for exploring separation costs comes from the differences in Euro- pean and U.S. labor markets discussed above, as well as empirical evidence found in DeNicco (2011) and ADS (2006) investigating changes in wrongful discharge laws within the U.S. over time. Looking closer at differences in wrongful discharge laws in Table 9, we see average index scores for different categories pertaining to wrongful discharge and annual unemployment rates from 1985 to 2010 for 24 OECD countries. Analyzing the G7 countries, we see mixed results for different categories. Using the median possible score of three, I split the G7 countries into two groups for each category. One group is below three, and the other is at or above three. The G7 countries below three for “REG5 - Definition of Unfair Dismissal” have an average unemployment rate of 6.87% compared to 9.1% for countries at or above three. The G7 countries at or above three for “REG8 - Possibility of Reinstatement Following Unfair Dismissal” actually have a lower average unemployment rate of 7.1% compared to the 7.8% for countries below three. While both of the previous categories may be tangentially related to separation costs, the most compelling summary data come from the category closest to directly capturing different separation costs across countries, “REG7 - Compensation Following Unfair Dismissal.” G7 countries below three for REG7 plus the U.S. and Canada have an average unemployment rate of 6.31%, while the countries at or above three have an average unemployment rate of 9.06%.2 Including all countries with information for REG7, we still see a sizable difference between unemployment rates for the two groups of 6.79% to 8.54%. These last two differences are large, suggesting that my findings of the effects of separation costs on steady state unemployment rates are consistent with evidence from empirical data for these OECD countries. 2The OECD database does not have information on the U.S. and Canada for this category. Removing leaves an average unemployment rate of 5.48% 41

In DeNicco (2011) I look empirically at the phenomena of jobless recovery. Us- ing vector autoregression models and controlling for changes in GDP, I can show that following a recession, decreases in the unemployment rate significantly slowed in the United States over time. I find two unknown structural breaks for the the slow down in the relationship: one in 1959Q4 and the other in 1984Q4. Dur- ing both time frames we see major changes in labor law to protect workers rights. The goals of these laws ranged from instituting standard safety practices with the Occupational Safety and Health Act of 1970, preventing discrimination due to age or race with the Age Discrimination Act in Employment (ADEA) of 1967 and the Civil Rights Act of 1964, ensuring fair wages for time worked with the Work Hours Act of 1962, and regulating benefit packages with the Employee Retirement Income Security Act of 1974. While these federal laws have potential effects on recovery of unemployment rates, they are not direct components of a firm’s separation costs. In contrast, during the 1980’s and early 1990’s individual state courts joined in the federal effort to increase workers rights with Employment-At-Will Exceptions (EWEs from hereafter). These EWEs, documented in Muhl (2001), were not designed to protect workers’ rights to fair hiring practices or fair on the job treatment, but rather were designed to provide legal recourse to wrongfully discharged employees. There are three major categories of these laws governing wrongful discharge: 1) Public Policy Exception (43 states have adopted this), under which “an employee is wrongfully discharged when the termination is against an explicit, well established public policy of the State;” 2) Implied Contract Exception (43 states have adopted this), which is “applied when an implied contract is formed between an employer and an employee, even though no express, written instrument regarding the employment relationship exists;” and 3) Covenant of Good Faith Exception (11 states have adopted this), which “reads a covenant of good faith and fair dealing into every employee relationship. It has been interpreted 42

to mean either that employer decisions are subject to a ‘just cause’ standard or that terminations made in bad faith or motivated by malice are prohibited.” The EWEs are written loosely (more so as they go on), leaving substantial room for legal interpretation when an employee is separated. In essence, these laws increase the legal and administrative costs of separating an employee.3 Empirical evidence of negative employment effects due to wrongful discharge from the EWEs shows mixed results. Autor, Donohue III, and Schwab (ADS) (2004) summarize the results in different papers over time, including their own paper, Autor, Donohue III, and Schwab (2006). Dertouzos and Karoly (DK) (1992) use a difference in difference approach and find a three percent decrease in aggregate employment with the adoption of the Covenant of Good Faith and Public Policy Exceptions. DK find this result to be equivalent to the effect of increasing employer taxes on wages by 10%. However, the DK paper suffers from potential bias due to their choice of instrumental variables. Miles (2000) reinvestigates the effects of EWEs on employment and finds the complete opposite of DK. He finds no evidence of correlations between the laws and employment, but does not investigate the discrepancies in the findings. ADS (2006) reevaluate the issue more comprehensively and find statistically significant correlations between the adoption of the EWEs and reductions in employment. The ADS (2006) results disagree with the DK results in the magnitude of the negative effects on employment, finding an average reduction between 0.8% and 1.6% in the employment to population ratio. ADS (2004) explain the discrepancy between ADS (2006) and Miles (2000) as due to the classification of the timing of case law. ADS (2006) consider EWEs to be internalized by firms much earlier than Miles (2000) in most instances. ADS (2006) attempt to “locate the first case in a state that might trigger a client letter from attorneys warning about a change in law.” Miles (2000) does not consider the law to have effects until it has worked its way through the

3California and Nevada had the highest unemployment rates in the country during the recovery from the 2007 recession and are among the seven states that have all three laws 43 states’ court system for validation. The ADS (2006) approach is more in line with firms having rational expectations of cost. DeNicco (2012) investigates the effects of EWEs on changes in unemployment rates, controlling for GDP growth.4 He finds that the laws have significantly slowed recovery of unemployment rates post-recession. The evidence is substantial enough that it warrants a closer look at how separation costs affect firms’ behavior in employment decisions.

2.3 Literature Review

Traditional descriptions of unemployment dynamics begin with a negative shock causing mass layoffs that clog the supply of the unemployed labor force. This causes job-finding rates to go down and the duration of unemployment to rise. Hall (2005) and Shimer (2005) challenge the conventional views of unemployment dynamics with relatively new empirical evidence. In December 2000 the emergence of the Job Open- ings and Labor Turnover Survey (JOLTS) greatly advanced our understanding of the behavior of separations. The newer view, backed by this empirical evidence, is that hiring rates play much more of a role in unemployment dynamics than previously believed. While there are changes in separation rates that accompany recessions, they are relatively small compared to regular, aggregate worker flows out of jobs. Unem- ployment is high during a recession due to firms reducing their hiring rates more than it is due to increased separation rates. Fujita and Ramey (2009) dampen the findings of Hall and Shimer with their own statistical analysis showing that between 40 and 50 percent of fluctuations in the unemployment rate are due to changes in separation rates. However that still leaves between 50 and 60 percent of the fluctuations in unemployment rates to the job hiring rate. In my model, an exogenous separation rate allows for a distinction between workers separated due to non-cyclical reasons

4The third chapter of this dissertation 44 and those who are part of mass layoffs in a recession. Moreover, the fall in the hiring rate due to a negative productivity shock explains a large component of the cyclical unemployment. While Hall and Shimer focus almost exclusively on search and matching models to describe employment dynamics, this chapter investigates the effects of increased separation costs on the hiring process in the context of an efficiency wage model. The efficiency wage model, going back to Shapiro-Stiglitz (SS 1984), explains involuntary unemployment by appealing to a principal-agent problem in the firm-worker relationships due to imperfect observation of effort. Firms can detect shirking with a given probability and the punishment for being detected shirking is dismissal. Due to the inability to observe effort, firms must pay a higher than the going wage in order to prevent shirking. If this were not the case and the wages were market clearing, fired workers could immediately find another job upon being fired and there would be no penalty for shirking. Since all firms find it advantageous to pay this higher wage, the market does not clear and unemployment results. The major contribution of SS is the endogenous nature of unemployment in equilibrium. I use an efficiency wage model because its micro-foundations are more in the spirit of this research, where increases in the marginal cost of separating an employee will slow down a firm’s labor force decisions through the hiring process. There is merit and understandable intuition behind both the efficiency wage model and the search and matching model, but the prior has been paid less attention in dynamic modeling. It is quite reasonable that there exists a principal-agent problem in the firm-worker relationships due to effort being unable to be properly observed or monitored. Ichino and Maggi (2000) document different effects on the incentive to shirk. They investigate a large bank with many branches all over Italy and find there are significant relationships between shirking, which is defined as misconduct and absenteeism, and individual backgrounds, group-interaction effects, and locational sorting. MacLeod 45 and Malcomson (1998) look at the labor market conditions in which efficiency wages versus performance pay will be endogenously determined as optimal. They find that in highly capital intensive labor markets, the cost of vacancy is high and therefore efficiency wages will be optimal compared to merit pay. The paper most related to this chapter is Alexopoulos (2003). Alexopoulos extends the Shapiro-Stiglitz model, investigating growth and unemployment in a wage efficiency model with shirking. The Shapiro-Stiglitz model fails on two fronts to explain variation observed in the data: 1) to achieve stable unemployment rates with population and technology growth; and 2) low real wage variation with high employment variation. Alexopoulos is able to alleviate both these shortcomings of the Shapiro-Stiglitz model. Alexopoulos achieves low wage variation and high employment variation by using monetary punishments instead of dismissal for detected shirking and by setting wages proportional to the worker’s consumption purchased by the worker’s family. With risk-averse households there is smooth consumption across time and states of the world, and therefore low variation in wages. With positive technology shocks, firms have an increase in the marginal product of labor (aided by low wage variation) and employment rises, resulting in relatively larger variations in employment. While Alexopoulos (2003) fixes issues with growth in the original SS model, and credibly explains equilibrium levels of unemployment, it does not address the dynamics of hiring rates following a recession. This is where my chapter aims to extend the literature. My extension includes an intertemporal decision for a forward looking firm with respect to labor and capital investment as opposed to Alexopolous where the investment decision resides in the household. Instead of firms renting capital they now own the capital and households own the firm. The household’s intertemporal decision entails using profits for saving through zero equilibrium bonds or family provided consumption. Using zero equilibrium bonds and having the firm make investment de- 46 cisions allows the model to include financial frictions and capital adjustment costs in future work, but does not alter the core results. More importantly, I extend the model with a linear separation cost to the firm for the workers who are exogenously separated each period. This separation cost affects both steady state unemployment rates and rates of hiring following a recession. There is the potential for a larger portion of workers to be separated in a recession than the exogenous percentage separated each period. These workers will be considered laid off due to an economic downturn, and the separation costs will not apply to them. This model specification captures the idea of increased difficulty for employees to claim wrongful discharge as part of an economically driven downturn. Other papers have looked into the effects of firing costs on equilibrium rates of unemployment, hiring, and firing. Bentolila and Bertola (1990) use dynamic optimiza- tion with firing costs in a partial equilibrium model of labor demand with uncertainty. They find the effect of increased firing costs depend on the environment, but will affect firings more than hirings. Bentolila and Bertola (1990) do not look into equilibrium dynamics, however. Gavrel and Lebon (2008) use the search and matching model and find that the overall effects of increased firing costs on unemployment rates depend on the type of firing cost. They find that if firing costs come in the form of layoff taxes that help finance unemployment benefits then unemployment will decrease due to the firm internalizing the tax on dismissals. This result is due to a decrease in payroll taxes used to finance unemployment benefits. Some papers look at the effects of separation costs on levels of employment and wages in the efficiency wage model. De Palma (2000) models a segmented labor market and finds increased firing costs lower employment and raise wage discrimination. Baumann (2005) finds that as severance payments increase, higher wages must be paid to induce effort. This chapter extends the literature by looking at equilibrium labor dynamics out of steady state in order to see if increases in the costs of separating 47 an employee will lower the hiring rate following a recession and therefore lengthen the time of unemployment recovery. More specifically, I extend the literature by using a discrete time model that fixes the major empirical shortcomings in the original SS model to investigate whether separation costs applied to exogenously separated workers can explain differences in labor dynamics between the United States and Europe over time.

2.4 The Model

2.4.1 Households

Representative households own the ﬁrms and invest the proﬁts in each period through buying private bonds. Each household uses savings to provide an amount of consump-

f tion, ct , to each of the family’s members, as well as for future investment. Family provided consumption is the same for all members regardless of employment status. Individual household members can increase consumption by working through a one period contract with a specified wage rate wt and effort level et. Individuals choose their level of effort. Shirkers are detected with a probability 0 < d < 1. Workers receive a share s of their wage up front. If a worker is not caught shirking, they receive the rest (1-s) at the end of the period. If they are detected shirking they are not paid the (1 − s) portion. The consumption constraints given below differ from Alexopoulos (2003), only in the household’s saving mechanism. The difference here is that I use bonds for savings instead of family owned capital. In this model firms own the capital. This alteration allows for future work incorporating financial frictions, but does not alter the core results. The constraints below are for: 1) workers providing effort; 2) shirkers who are not caught; 3) shirkers who are caught; and 4) unemployed workers. This chapter uses t to represent quarterly time periods. The interest rate, rt, is paid on bond hold- 48

ings, Bt, this quarter. In quarter t households will choose the number of bonds, Bt+1, to hold in the next quarter. The consumption level of a worker not caught shirking

s is ct. The consumption level of a worker caught shirking is ct , and the consumption

u level of someone unemployed is ct . The consumption constraints are given below as:

f ct ≤ rtBt + Πt − (Bt+1 − Bt), (2.1)

f ct = ct + wt, (2.2)

s f ct = ct + swt, (2.3) and

u f ct = ct . (2.4)

Individuals value both consumption and leisure. The time t utility levels below are taken from Alexopoulos (2003), where T is the time endowment,e ˜t is the variable cost of providing effort and is the fixed cost of providing effort. θ ≥ 0 is a preference weight on leisure. This structure allows wages to be proportional to the worker’s consumption purchased by the family resulting in low variation in wages with risk averse families. Workers who provide effort receive utility from consumption purchased with wages and leisure time. The variable and fixed costs of effort take away from the time endowment and reduce utility. Workers who shirk and are not caught receive utility from consumption purchased with wages and leisure time equal to the full time endowment. Workers caught shirking receive utility from consumption purchased with the share of wages left after punishment and the leisure time equal to the full time endowment. Unemployed workers receive utility from family provided consumption and leisure time equal to the full time endowment. The different utilities are provided 49 below:

U(Eﬀort) = ln(ct) + θln(T − e˜t − ), (2.5)

U(Not Detected) = ln(ct) + θln(T ), (2.6)

s U(Detected) = ln(ct ) + θln(T ), (2.7) and

u U(Unemployed) = ln(ct ) + θln(T ). (2.8)

Firms make contract offers to workers who can accept or reject. If the worker accepts the job, he must then determine if he is going to abide by the effort contracted in the offer, or shirk and risk a monetary punishment. In order to induce effort, firms must ensure that the incentive compatibility constraint holds:

s ln(ct) + θln(T − e˜t − ) ≥ (1 − d)(ln(ct) + θln(T )) + d(ln(ct ) + θln(T )). (2.9)

In words, utility from providing effort must be greater than the expected utility from shirking, which is the average utility weighted by the probability of being detected. If this incentive compatibility constraint is satisfied, then the cost of providing effort, e˜t, will equal the effort put forth. If the constraint is not satisfied, the worker will not provide effort and will have no cost of effort. Effort is provided according to the following:

 d − θ  ct et if et ≤ T 1 − s −  ct e˜ = t d − θ  ct 0 if et ≥ T 1 − s −  ct

The constraint tells us that ﬁrms must oﬀer a wage such that the ratio between

ct consumption from eﬀort to consumption for a punished shirker, s , is large enough to ct 50

induce effort. If the wages are too low workers will choose zero effort. Since workers effectively provide no labor when shirking (discussed below in the subsection on the firm), in equilibrium firms will pay wages that induce workers not to shirk.

2.4.2 Representative Family Problem

For simplicity households are assumed to have an equal number of workers per family, which results in identical wealth. Households make choices taking all possible cases of workers into account in their decision to maximize the utility of their members.

S Nt stands for the number of shirkers hired in period t and Nt stands for the workers providing eﬀort. L is the ﬁxed population size. The household maximizes its lifetime utility below subject to (1) through (4) and accounting for workers who provide effort, workers who shirk and are not caught, workers who are caught shirking and the unemployed:

    ∞  (N − dN s) ln (c ) + (L − N ) ln (cu) + dN s ln (cs)   X t  t t t t t t t  max E0 β . f ∞ {c ,Bt+1} s s t t=0  t=0  + (Nt − Nt ) θln (T − fet − ) + (L − Nt + Nt ) θln (T )  (2.10) The ﬁrst order condition for the households’ maximization problem in equilibrium with no shirkers is a standard Euler equation with the household smoothing consumption. Consumption per worker and consumption per unemployed family member in quarter t equal consumption per worker and consumption per unemployed family in quarter t + 1, adjusted for the discount factor and interest rate:

Nt L − Nt Nt+1 L − Nt+1 + u = Et β + u (1 + rt+1) . (2.11) ct ct ct+1 ct+1 51

2.4.3 Firms

The representative firm operates a Cobb-Douglas production function with α < 1 and includes effort as an input factor augmenting labor. The first deviations of firms in this model from Alexopoulos (2003) are in investment decisions and timing. Alex- opoulos (2003) leaves investment decisions to the household and firms operate in a static environment. Here firms own the capital and make forward looking decisions on investment and labor, choosing next period’s number of workers and wages this period.5 Dependent on the expected state of the economy next quarter, firms specify a contract guaranteeing a wage to the number of workers they think they will need based on expected effort levels required. However, I am allowing firms and workers to finalize contractual obligations on effort ex-post, depending on the economic conditions encountered at the beginning of the next quarter. The contract specifies a guaranteed wage next period in return for workers providing the necessary effort level required for the state of the economy. If the firms encounter a negative productivity shock driving down profits, their expectation level for worker provided effort will increase with the guaranteed wages that are now above wages firms would otherwise offer given the economic conditions. It is in the interest of the worker to change effort levels to match the firm’s adjusted expectations or risk losing a contracted share of their wages for shirking. This allows for contractual flexibility in effort while still satisfying the incentive compatibility constraint. The second and more substantial change from Alexopoulos (2003) is the introduction of separation costs, φ > 0, into the model. Separation costs will be incurred for the exogenous portion, b, of the workers who are separated each period. This construct enables me to distinguish between workers that are separated due to cir- cumstances other than changing economic conditions and those who are part of mass layoffs in a recession. There is intuition for endogenous separation costs with firms

5I move from a risk averse ﬁrm in Alexopolous (2003) to a risk neutral ﬁrm here. 52

finding it optimal to enter into contractual agreements on such things as severance payments, but here φ is exogenous in the spirit of these separation costs stemming from labor legislation that is independent of economic conditions and making it more expensive to discharge or fire an employee. In equation (2.12), the representative firm maximizes its expected value by choosing wages, capital and labor in period t for period t + 1. Dependent on the economic conditions encountered at the beginning of period t, firms will choose their control variable, effort. Firms are entering into a contract with workers for a given wage next period, independent of changes in expected economic conditions, and an effort level this period, dependent on economic conditions. The penalty for shirking applies to the agreed upon effort level chosen this period:

( ∞ ) X t α (1−α) max E0 β (AtKt (etNt) − wtNt − φbNt − It) , (2.12) wt+1,Nt+1,Kt+1,et t=0 s.t.

u u(ct, et) ≥ u(ct , 0)(IR), (2.13)

s u(ct, et) ≥ du(ct , 0) + (1 − d)u(ct, 0)(IC), (2.14)

and

It = Kt+1 − (1 − δ)Kt. (2.15)

The IR will be not binding here, due to firms maximizing (2.14),which automati- cally ensures (2.13) is satisfied. With the incentive compatibility constraint satisfied, equation (2.14) will result in: 53

d ! − θ ct e(wt) = T 1 − s − . (2.16) ct

The ﬁrst order condition on Kt+1 yields:

α−1 1−α Et βαAt+1Kt+1 (et+1Nt+1) = 1 − β(1 − δ). (2.17)

This result is standard in the ﬁrm increasing capital until the expected marginal ben- eﬁt equals the expected marginal cost.

The ﬁrst order condition on wt+1 yields:

( −d −1 ct+1 ! ) θ 1 − s s α 1−α −α T d ct+1 ct+1 Et β(1 − α)At+1Kt+1(Nt+1) et+1 s s = βNt+1 . θ ct+1 (ct+1) (2.18) Equation (2.18) states that wages will increase until the expected marginal benefit equals the expected marginal cost. Here an increase in wages benefits the firms by promoting greater effort and making each worker more productive. Again, firms and workers will contract a wage and an expected effort level, dependent on economic conditions. Firms change their expectation of effort ex-post if economic conditions are different from their forecast. With the incentive compatibility constraint satisfied, workers will not shirk in equilibrium and provide the effort firms desire. In a recession, profits for the firm will decline but wages will not adjust for one quarter. The workers receive wages above what the firm would otherwise pay if the contract allowed for wage renegotiation. Therefore firms expect greater effort in a recession. Furthermore, when profits decline in a recession, family provided consumption decreases. Looking at (2.16), a drop in profits will incentivize workers to increase their effort to match 54 the firm’s new expectations and the constraint will hold.

The ﬁrst order condition on Nt+1z yields:

α 1−α −α Et β(1 − α)At+1Kt+1(et+1) Nt+1 = β(wt+1 + φb) . (2.19)

(2.19) shows the expected marginal benefit of the firm’s labor force is equivalent to the total expected marginal cost of two different sources. The firm is looking forward and optimizing the number of expected workers needed given the effort level. The cost for each worker to provide effort is the wage the firm has agreed to pay next quarter. The additional cost is the subject of investigation in this chapter. This is the separation cost. The separation cost is paid for only the exogenous portion b of workers separated each period. Workers separated above this exogenous rate out of steady state are considered part of a mass layoff and these separation costs do not pertain to them. In recessions (booms), the total separation costs will go down (up) with changes in N. If the average separation cost increases, the effect of changing total separation costs will decrease the percentage of workers separated and slow down the hiring process following a negative productivity shock. With higher average separation costs, it will take a lesser percentage of workers to reduce total separation costs. As economic conditions improve, each worker added increases the total separation cost by a larger amount. 55

2.5 STEADY STATE

2.5.1 Parameters

d A b α β δ θ φ s T

1 .25 1 0.036 0.34 1.03 0.028 1 10 0 0.729 1369

In this model each time period is equal to one quarter. I initially calibrate the model with separation costs equal to zero for a baseline to compare steady state rates of unemployment as separation costs increase. The separation rate, b, is set to match empirical evidence of average separation rates in the Job Openings and Labor Turnover Survey (JOLTS) data from the BLS. A, α, , and T are the same as in Alexopoulos (2003). The rest of the parameters are set as close as possible to Alexopoulos (2003) while resulting in a 5.6% unemployment rate. I adjust the share of

d wages received up front, s, and the ratio, θ , to adjust the steady state unemployment d rate with a population L = 50 equal to 5.6%. I change the ratio, θ , from .056 in Alexopoulos (2003) to 1 here. I could not ﬁnd a reasonable steady state equilibrium

d with low levels of θ . As it is not the focus of this chapter, I normalized the ratio to 1, allowing for an empirically consistent steady state. I needed to choose a population, L, large enough so my impulse response functions could could be calculated. If L is too small, then the number employed N is to small so that marginal changes of N will be too large for computation of impulse response functions. The choice of L above some threshold is somewhat arbitrary. I change δ slightly to make the capital stock just below ten times output. Wage payments and capital investments in this steady state are empirically consistent at 66% and 27% of output. The other 7% goes to proﬁt, which is then used for family provided 56

consumption.6

2.5.2 Equilibrium

c In order to define the steady state, I first solve for the ratio cs . Using all of the first order conditions, (2.17), (2.18) and (2.19) along with the equation for effort, (2.16), and the utility equations, (1) through (4), I am able to define all variables in terms of parameters and the ratio. I use the variables X0 through X5 to group parameters and already determined values of steady state variables to help with the organization of the derivations. Please refer to Appendix A for more details.

From (2.17):

1 1 − β(1 − δ) α−1 X = . (2.20) 0 βαA

From (2.18):

T d X = (1 − α)AXα . (2.21) 1 0 θ

c Now I can solve for the ratio, cs , using:

−d −1 −1 c θ c c X 1 − s = (1 − s) ((1 − α)AXαe − φb) − 1 . (2.22) 1 cs cs 0 cs

c With (2.22) I can pin down cs numerically dependent on all parameters. I can then immediately solve for eﬀort using (2.16), and wages using the ﬁrst order condition on labor, (2.19):

α w = (1 − α)AX0 e − φb. (2.23) 6When separation costs are non-zero they destroy value in the economy. Proﬁt to output ratios and investment to output ratios will not change, but wage payment to output ratios will drop due to a smaller workforce. 57

With a representative household, market clearing in the bond market results in:

Π = cf . (2.24)

Deﬁning the following equations:

X3 = (1 − α)A, (2.25)

X4 = φb, (2.26)

and −1 α w + X4 X5 = 1−α , (2.27) X3e

and combining them with equation (2.24) and the two following equations from the ﬁrst order conditions for labor and capital, (2.19) and (2.17), I ﬁnd the remaining steady state values for labor and capital:

−1 α w + X4 N = 1−α K, (2.28) X3e | {z } X5 and

K = X0eN. (2.29)

Using the parameters in this model, the labor market will not clear. The result here

L−N is an endogenously determined, steady state unemployment rate, N . Starting with 0 separation costs, I set the steady state unemployment rate to 58

5.6%, which is close to historical natural rates of U.S. unemployment. For a compar- ison to the empirical data, I go back to Table 9 and perform a calibration exercise. Using “ REG7 - Compensation Following Unfair Dismissal” as a guide, I separate the 22 countries by the median possible score of three.7 Higher scores indicate greater comparative compensation. Countries with index scores of less than three have an average unemployment rate of 6.8%. In order to increase the unemployment rate from 5.6% to 6.8%, the separation cost must increase from 0 to 9, which is about 8% of the wage rate. The average index score for countries with scores at three and above is 3.65. This average index score is 2.74 times larger than the average index score of 1.33 for countries with scores below three. Multiplying a separation cost of 9 by 2.74, I ﬁnd a new separation cost of 24.66. In terms of wages, this cost is about 23% of the wage rate. Using this separation cost in the model results in an unemployment rate of 8.67%, which is very close to the average unemployment rate of 8.54% for the countries with a score of three and above.

2.6 Impulse Response Functions

I compute impulse response functions for a one percent, negative productivity shock and find I am able to replicate a number of empirical features following a recession. Looking at Figure 9 for wages and effort, we see that the forward looking decision for wages cannot be affected in the first quarter because a contract has been signed. However, with a negative productivity shock, the contractual expectation of effort changes and the incentive to provide that effort changes to match expectations. Here with firms and workers agreeing on effort ex-post (after the negative productivity shock), effort increases immediately. The increased incentive to match the firm’s new expectations is due to the negative, immediate effect on profits shown in Figure 10. Lower firm profits mean lower family provided consumption. With the penalty for

7There are 22 countries instead of 24 because there is no data for Canada and the U.S. 59 shirking applying to the ex-post expectations of effort, workers have the incentive to match expectations or risk losing further consumption from penalized wages. After the first quarter, the firm can adjust its labor force and wages will temporarily rise to about 0.6% above steady state. The central idea is that firms find it profitable to reduce their number of employees at the same time demanding more effort in this economic downturn. Wages must increase to pay for this effort. As firms begin to hire back their steady state employment levels and the shock dissipates, wages will decrease and then slowly recover as seen in the data. Looking more at Figure 10, all three types of consumption follow almost indis- tinguishable paths that mirror the movements in wages. Again, the family provided consumption follows firm profits. Profits are immediately affected by the negative productivity shock. In the second period, firms are able to adjust in the short run and increase profits by less than two percent above steady state levels. Firms are able to do this with a smaller workforce providing greater effort. Then, as the shock wanes and firms hire back their steady state labor force, profits, consumption and shirkers’ consumption will decrease quickly and recover slowly over time. Figure 11 shows the impulse response functions for labor and capital. Here we see the model exhibiting the greater role of hiring rates over separations rates in unemployment rate dynamics, as argued by Shimer (2005) and Hall (2005). In the first period neither state variable of employment level or capital stock can change. In the second period we see the number of employees drop immediately and dramatically by about 4.5%. However, the separation rate only increased by 1%, while the hiring rate went to zero from the 3.5% needed in steady state to replace separated workers. That means about 22% of the change is due to changes in separation rates and 78% to changes in hiring rates. That is more dramatic then the findings of hiring rates contributing 50% to 60% of labor dynamics by Fujita and Ramey (2009), but those findings can be matched with small changes to the exogenous separation rate. Now 60 after the second period and as the shock wears off, the number of employed workers will slowly increase. This large drop in employment and slow recovery arematch the asymmetries observed empirically in labor force dynamics. The capital stock adjusts much more slowly to the negative shock both in its decline and recovery. The main take aways from this chapter are observed in Figure 12 and Figure 13, investigating the difference in labor dynamics with changes in separation costs. Figure 12 shows the recovery of the steady state unemployment rates with zero separation costs (φ = 0) and separation costs set with φ = 23.63. These two levels of separation cost correspond to steady state unemployment rates taken from average rates in the U.S. and from countries with index scores for “ REG7 - Compensation Following Unfair Dismissal ” equal to or above three. It is immediately apparent that the increase in separation costs lead to higher steady state levels of unemployment and more sclerotic equilibrium dynamics. More importantly than the steady state unemployment rates for this chapter are the observed changes in movements out of steady state. Looking at Figure 12, we see with a one percent negative productivity shock that higher separation costs result in a smaller percentage drop in the unemployment rate and a slower recovery back to steady state. With the shock, the U.S. unemployment rate increases by 75.2%, from 5.62% to 9.85%. The average rate for countries with an index score above three increased by only 47.5%, from 8.54% to 12.6%. However, despite the much larger percentage increase in the U.S. unemployment rate, it takes less time to get back to steady state levels. In the end, I find that higher separation costs increase the marginal cost of each worker employed and therefore cause sclerosis of labor dynamics both in separations during and hirings following a recession. This sclerosis is due to decreasing (increasing) total separation costs in recessions (booms). Again, when the average separation cost increases, the effect of changing total separation costs will decrease the percentage of workers separated and slow down the hiring process following a negative productivity shock. With higher 61 average separation costs, it will take a smaller percentage of workers to reduce total separation costs. As economic conditions improve, each worker added increases the total separation cost by a larger amount. Thus one factor driving differences in the speed of hiring is the cost of firing.

2.7 Conclusions

In this chapter I use a discrete time, dynamic efficiency wage model to show that increases in the cost of firing an employee can explain observed differences in both steady state unemployment rates and dynamic movements of unemployment rates out of steady state. Using the steady state equilibrium defined in section 2.5, I find that an increase in the cost of separating an employee causes the steady state unemployment rate to rise consistent with empirical evidence from the OECD. In section 2.6, I compute impulse response functions for a one percent, negative productivity shock, incorporating separation costs and forward looking firms. With my impulse response functions, I am able to replicate dynamic features seen empirically following a recession. I find that larger separation costs not only increase the steady state unemployment rate, but they also decrease the percentage of a firm’s employees laid off during a recession and slow down the recovery time of unemployment rates back to steady state. I am also able to replicate empirical evidence finding a larger role of hiring rates over separation rates in labor dynamics. My findings are applicable to both comparisons of labor dynamics in Europe versus the United States and comparisons of labor dynamics in the United States over time. These findings give us a better understanding of the dynamics of post recession unemployment rates and have obvious policy implications. If the goal of policy makers is to have a fluid labor market with low steady state levels of unemployment and faster recoveries, then they should keep firing costs as low as possible. 62

Chapter 3: Employment-At-Will Exceptions and Jobless Recovery

In this paper I study the effects on jobless recovery of diminishing the power of an employer to fire an employee through Employment-At-Will Exceptions (EWEs). I do so by using a dynamic panel with quarterly data ranging from 1976 to 2010 for the 50 states in the United States. I test both changes in state unemployment rates and state-weighted GDP growth in single variable regressions and VAR regressions. My contribution to the literature is threefold. First, I show two of the three EWEs contribute significantly to jobless recovery in the U.S. The statistical tests in this paper show that Implied Contract Exceptions slow decreases in the unemployment rate during recovery from recession by between 0.025 and 0.033 percentage points per quarter, and Covenant of Good Faith and Fair Dealing Exceptions do so by between 0.039 and 0.055 percentage points per quarter. Second, I lend support to the predictions of theory that increased firing costs decrease the rate of hiring during recoveries. Third, I resolve differences in the various sources documenting the three types of EWEs in different states.

3.1 Introduction

In the United States, the legal relationship between employees and employers with regard to discharge has been historically guided by the Employment-At-Will doctrine. The Employment-At-Will doctrine is a common law doctrine, which gives complete discretion to the employer in discharging employees whose contracts do not expressly state the duration of the employment relationship.1 The precedent of Employment- At-Will goes back to 1884 in Payne v. Western & Atlantic Railroad, where the Supreme Court of Tennessee states that employees can be fired for, “good cause, bad cause, or no cause at all.” Beginning with California in 1959 with Petermann v. Intl Brotherhood of Teamsters and spreading to the majority of states throughout the 1980’s and early 1990’s, state court systems diminished the power of U.S. firms to fire workers at-will as governed by the doctrine of Employment-At-Will.

1Common laws come from precedent setting court cases as opposed to passage through a legisla- ture. 63

DeNicco and Laincz (2011) use the widely known Quandt-Andrews test for an unknown breakpoint and find significant structural breaks for increasingly jobless recoveries for the United States in 1959 and 1984. Here jobless recovery is a relative term comparing the relationship between GDP growth and changes in unemployment rates across states and time. The resulting structural breaks for jobless recoveries coinciding with changes in the legal framework governing employment relationships suggests a possible connection between Employment-At-Will Exceptions (EWEs) and slower recoveries in unemployment rates post recession. If EWEs increase the cost of firing a worker, DeNicco (2011) predicts they will in fact contribute to increasingly jobless recoveries through slower hiring. In this paper I examine the effects on jobless recovery of diminishing the power of an employer to fire an employee through EWEs using dynamic regression models on a state-level panel of quarterly data from 1976 to 2010. In order to do this I must test both changes in unemployment rates and GDP growth. The essence of jobless recovery is that an economy experiences growth in GDP without decreases in unemployment rates commensurate with past recoveries. First, however, I resolve differences in the various sources documenting the existence of the laws in each state. There are a number of sources that report the presence of the EWEs and not all of them agree in their timing and classification. With the updated classification of the laws, I utilize panel regressions with recovery periods isolated by using interaction terms and by splitting the data into subsamples. There are three major categories of cases in which the courts have found it proper to limit the power of firms to fire workers. These three cases, referred to as Employment-At-Will Exceptions, are Public Policy Exceptions (PPEs), Implied Contract Exceptions (ICEs) and Covenant of Good Faith and Fair Dealing Excep- tions (CGFFDEs). Muhl (2001) describes the three main categories of EWEs. PPEs are the most clear cut. They apply to wrongful discharge “when the termination is 64 against an explicit, well established public policy of the State.” Common applications of the PPEs include cases when workers have been fired due to discrimination, whistle-blowing or filing worker’s compensation claims. ICEs are “applied when an implied contract is formed between an employer and an employee, even though no express, written instrument regarding the employment relationship exists.” Common applications of the Implied-Contract exceptions are in cases when employers publish an employee handbook that leads to a “reasonable” interpretation of an altering of at-will status by the employee. Some states have gone further allowing oral statements to alter the at-will status of an employee. CGFFDEs are applied to all employee relationships regardless of the existence of implied or implicit contracts. They have been “interpreted to mean either that employer decisions are subject to a ‘just cause’ standard or that terminations made in bad faith or motivated by malice are prohibited.” Most states have applied the CGFFDE in cases when an employee has been terminated to prevent them from receiving compensation they would have otherwise been due, such as bonuses or merit pay. Court opinions explaining CGFFDEs are written loosely, leaving substantial room for legal interpretation and possible legal action against employers when an employee is discharged. All three EWEs could raise firing costs and contribute to jobless recovery. Mortensen and Pissarides (1999) suggest that higher firing costs result in slower job creation, which results in slower transitions back to steady state following the same negative shock. I provide empirical evidence of slower recovery in unemployment rates while controlling for GDP growth, following a recession when court opinions set precedent in favor of EWEs. The statistical tests in this paper show ICEs slow decreases in unemployment rates during recovery between 0.025 and 0.033 percentage points per quarter, and CGFFDEs do so by between 0.039 and 0.055 percentage points per quarter. In other words if a state adopted both laws, negative movements in un- 65 employment rates would be dampened by an average of between 0.064 and 0.088 percentage points per quarter or between 0.256 and 0.352 percentage points over the course of a year. For example, let us take the case with the larger effects for both EWEs. Starting from the 9.9% unemployment rate in 2009Q4 and controlling for actual GDP growth, my model predicts the unemployment rate would fall to 7.7% in two years without the two EWEs. However, with both EWE’s in place the unemployment rate would only fall to 8.4%. With a labor force of 150 million people, that would leave an extra 1.1 million people unemployed and possibly collecting unemployment benefits, which cost on average about $300 per worker per week. I find no evidence of PPEs contributing to jobless recovery. Through their rulings on EWEs, state court systems tried to protect those in the labor force who are already employed from wrongful discharge. However, they have made it more difficult for the unemployed to find jobs. By restricting firms in their ability to make decisions about how and why they fire their employees, my results robustly confirm the exceptions contribute to longer recovery periods in the labor market. State legislatures concerned with high unemployment should consider passing statutes clarifying the cases in which the exceptions should apply. This could reduce uncertainty for firms and help reduce ex ante firing costs. The rest of the paper is organized as follows: Section 3.2) a literature review; 3.3) the background of EWEs; 3.4) a summary look at the data and classification of case law; 3.5) results and analysis; 3.6) robustness tests; and 3.7) conclusions.

3.2 Literature Review

Theory predicts that imposing restrictions on the ability of firms to fire workers can contribute to an increased reluctance to create and fill vacancies during periods of economic recovery. Mortensen and Pissarides (1999) find that the effects of increased firing costs on steady state unemployment rates are ambiguous in a search and match- 66 ing setting, but the effects of increased firing costs clearly slow job creation. If EWEs increase firing costs, employers will be slower to hire new workers with increases in aggregate demand. Mortensen and Pissarides (1999) point out that firms have incen- tives to be more careful in their hiring practices and incur larger costs to avoid an employment match that becomes unsuitable in the future. There is also an increased marginal cost of employment due to defensive legal practices and administrative bur- dens, which can make firms more cautious in their screening process and slower to hire new workers. Hall (2005) and Shimer (2005) show there are stronger correlations between vacancies and unemployment than between separations and unemployment. Fujita and Ramey (2009) show statistical evidence dampening the findings of Shimer (2005) and Hall (2005) but still leave up to 60% of the fluctuations in the unemployment rate to vacancy creation and hiring rates. If so much of the variation in unemployment rates comes from these two factors, an increase in firing costs can result in persistence in unemployment rates through increased rigidity in the hiring process and a reluctance of firms to create vacancies. Faberman (2008) addresses jobless recovery using flow data from the relatively new Business Employment Dynamics (BED) data set from the Bureau of Labor Statis- tics (BLS) covering the period from 1990-2006. He investigates job creation and job destruction as defined by Davis, Haltiwanger and Schuh (1996), and in accordance with Hall (2005) and Shimer (2005), he attributes the jobless recovery from the 2001 recession to a persistent decline in the job creation rate.2 He links his findings to the reduction in volatility and increased persistence of job flows in the presence of

2Faberman also extends his data back to 1947 using the BED and previous estimates produced Davis, Haltiwanger, and Schuh (1996). He creates his own GMM predicted estimates of job creation and job destruction. Much in line with the structural breaks found in DeNicco and Laincz (2011 Working Paper) and the Great Moderation literature, Faberman observes the magnitude of job flows began to steadily decline in the 1960’s and the volatility of job flows dropped sharply in the mid 1980’s. He attributes the slow recovery from the 1990-91 recession to an increase in the job destruction rate. 67 aggregate shocks as seen in the Great Moderation period, first studied by Kim and Nelson (1999) and McConnell and Perez-Quiros (2000). Mortensen and Pissarides (1999) write that increased firing costs should also cause firms to be more reluctant to discharge employees, which will have a positive effect on steady state employment. This influence will cause ambiguity in the overall result for the level of the steady state unemployment rate. Gavrel and Lebon (2008) use the search and matching model and find that the overall effects of increased firing costs on unemployment rates depend on the type of firing cost. They find that if firing costs come in the form of layoff taxes that help finance unemployment benefits then unemployment will decrease due to the firm internalizing the tax on dismissals. This result is due to a decrease in payroll taxes used to finance unemployment benefits. If EWEs are not taxes on dismissal, but rather administrative and legal costs, there should be an increase in steady state unemployment rates according to this theory. I investigate this theory in two different ways, using interaction terms and subsampling my data before statistical testing to account for different phases of the unemployment and GDP growth cycles. I find no statistically significant evidence of the EWEs altering the rate of change in the unemployment rate during a contraction. However, I find strong evidence they play a significant role in slowing changes in the unemployment rate during an expansion. There are of course numerous possible causes for jobless recovery leaving room for future work. Aaronson, Rissman and Sullivan (2004) summarize and critique a number of existing arguments for the causes of jobless recovery, including self-employment, just-in-time hiring, healthcare costs, participation rates and slack aggregate labor demand. They specifically note that the high cost involved in firing workers, perhaps due to the possibility of firms facing lawsuits, may push firms to use temporary workers. With large numbers of temporary workers desiring full time work, there will be persistently high unemployment rates during recoveries with many new jobs being 68 taken by temporary workers shifting to full time. This shift leaves smaller numbers of workers moving from unemployment to employment. Workers who are part time for economic reasons are one type of temporary worker. Ideally I would like to investigate whether EWEs contribute to increases in these type of workers, but due to data limitations at the BLS this is not possible. The data for states’ part time workers only goes back to 1997 and is only available in annual averages. These limitations preclude a thorough and conclusive investigation.

3.3 Employment-at-Will Background

Three empirical papers use macro data and explore the effects of EWEs with mixed results. Dertouzos and Karoly (DK) (1992) use a difference in difference approach and find a three percent decrease in aggregate employment with the adoption of the good faith and public policy exceptions. DK find this result to be equivalent to the effect of increasing employer taxes on wages by 10%. However, Autor, Donohue, and Schwab (ADS) (2004) argue that this result from DK may be biased due to the choice of instrumental variables. Miles (2000) revisits the effects of EWE laws on employment and finds no evidence of correlations between the laws and employment, but he does not investigate the discrepancies in the findings. ADS (2006) comprehensively reevaluate the issue and find statistically significant correlations between the adoption of the EWE laws and reductions in employment. The ADS (2006) results contrast with the DK results in the magnitude of the negative effects on employment. They find an average reduction of between 0.8% and 1.6% on the employment to population ratio. ADS (2004) explain the discrepancy between ADS (2006) and Miles (2000) as due to the classification of the timing of case law. ADS (2006) consider EWEs to be internalized by firms much earlier than Miles (2000) in most instances. ADS (2006) attempt 69

to “locate the first case in a state that might trigger a client letter from attorneys warning about a change in law.” Miles (2000) does not consider the law implemented until it has worked its way through the states court system for validation. The ADS (2006) approach accounts for forward looking firms, as assumed in the theoretical papers discussed here. Therefore I follow their method for classifying the timing of case law, which is discussed in the appendix. One empirical paper more closely aligned to my goals here is from Autor, Kerr and Kugler (AKK) (2007). They use establishment level data from 1976 to 1999 to explore the effects of the EWEs on overall employment fluctuations, firm entry rates and total factor productivity (TFP). In line with my results, they find the CGFFDEs significantly reduce overall employment fluctuations. They also find CGFFDEs reduce firm entry rates and TFP, but these results are not reflected in my findings for state-weighted GDP growth. In this paper I move away from the short range difference in difference techniques used in most of the papers I have described. I use dynamic regression models on a panel with quarterly data ranging from 1976 to 2010 and isolate recovery periods through both interaction terms and split samples. While all of empirical papers except for AKK (2007) studying EWEs focus on levels of employment and unemployment, I focus on changes in the unemployment rate and GDP growth. I go beyond all of the previous papers using dynamic panel regression models to find the effects of the exceptions during recoveries on changes in unemployment rates and GDP growth. Again, jobless recovery refers to an economy that experiences growth in GDP without decreases in unemployment rates commensurate with past recoveries. Therefore it is crucial to test the effects of EWEs on both GDP growth and changes in unemployment rates. 70

3.4 Data

The first tasks necessary for this study are updating the existing EWEs through 2010 and resolving differences in the existing literature. Here I compare four different publications that independently document the existence of the exceptions in different states over time. For a further explanation of the details concerning my methodol- ogy and classifications, please refer to the appendix. This task required finding and reading hundreds of different state court opinions through Lexis Nexis, state supreme court websites and on-line databases. Documenting and categorizing differences in EWEs requires the interpretation of state court opinions and is dependent on the criteria set forth before documentation. If different sources use different criteria, discrepancies will result. In this paper I follow the criteria set forth in ADS (2006) for the timing of case law and I concur with most of their findings for the existence of precedent setting case law. I use state unemployment data from the Bureau of Labor and Statistics, and real national GDP data (in chained 2005 dollars) and state personal income data from the Bureau of Economic Analysis. In order to account for the asymmetry of both movements in unemployment rates and GDP growth over the business cycle, I separate the business cycle into the different phases of both unemployment dynamics and state-weighted GDP growth. I represent the different phases of the business cycle in my model by both using interaction terms and splitting the sample. For the unemployment cycle, I separate the data into periods of net Outflows and net Inflows. Outflows are characterized by periods of negative changes in unemployment rates. Inflows are defined as periods of positive changes in unemployment rates. For GDP growth cycles, I separate the data into two samples using changes in state-weighted real GDP. One subsample has positive or Expansion periods and the other has negative or Contraction periods. State Gross Product is only available annually, therefore 71

I use quarterly state personal income to weight national GDP and use the measure to find the effects of EWEs on state-weighted GDP. The results in the next section are robust to both the Inflows/Outflows and the Expansions/Contractions specifications. In Table 10, I show the average difference in unemployment rates and the average difference in state-weighted GDP growth conditional on the presence of the EWEs. The percent change measures the difference in the averages when the law is in existence versus when the law is not yet set in precedent. The t-tests provide summary evidence that suggests a slower recovery in unemployment rates during Outflows and Expansion periods due to all three EWEs. The effects on unemployment rates are significantly larger than on state-weighted GDP growth. The presence of the EWEs coincides with a diminished recovery of unemployment rates during Outflows and Expansion periods, and less dramatic results with conflicting signs for changes in state-weighted GDP growth. This evidence suggests that EWEs may be affecting unemployment rates more than they are affecting state- weighted GDP growth during recovery periods. This result suggests EWEs contribute to jobless recovery in the United States beyond explanations of slower GDP growth. I fail to find the same evidence on the effects of EWEs during Inflows and Con- traction periods. Comparing differences in unemployment rates between Table 10 and Table 11, I find the evidence supports the findings of Gavrel and Lebon (2008) that administrative type firing costs will have a slowing effect that is unambiguously larger on hiring than on firing.

3.5 Results and Analysis

In this section I run a benchmark specification for my empirical model and then perform robustness tests on it. My benchmark specification is modeled to test the predictions from DeNicco (2011) that increased firing costs result in slower recov- 72 eries in unemployment rates post recession. The basis for my specification comes from DeNicco and Laincz (2011), which finds structural breaks indicating increasingly jobless recoveries around the time that most of the EWEs became precedent. In subsection 3.5.1, I use interaction terms and split samples to test the relationship between EWEs and my independent variables in separate autocorrelation regressions. In subsection 3.5.2, I continue my analysis in split samples with panel VARs so that I can test changes in unemployment rates while controlling for state-weighted GDP growth. Then in section 3.6, I run robustness tests controlling for the presence of collective bargaining and a lagged implementation effect. In both sections the conclusions are the same; EWEs contribute to jobless recovery in the U.S. Working with first differences in both unemployment rates and state-weighted GDP, I expect to reject the unit root for all variables. Using the Schwarz information criteria (SIC) and looking at Table 12 for the full sample, we can reject the unit root in all cases for both a common and individual unit root process for both variables tested. All unit root tests are robust to the Akaike information criteria (AIC) standards as well. Using split samples we can also reject the unit root at the 1% threshold by all tests performed for Outflows, Expansions and Inflows. Only differences in the unemployment rate during Contraction periods fail to be rejected for a unit root, and in only one test.

3.5.1 Single Variable Autocorrelation Regressions

3.5.1.1 Diﬀerences in Unemployment Rates

Here I investigate the eﬀects of the EWEs on diﬀerences in unemployment rates. The model is represented below: 73

4 X DURs,t = α + βiDURs,t−i + β5PPEs,t + β6ICEs,t + β7CGF F DEs,t + β8EXPs,t+ i=1

β9PPEs,t ∗ EXPs,t + β10ICEs,t ∗ EXPs,t + β11CGF F DEs,t ∗ EXPs,t + fs + qt + s,t (3.1)

where DUR represents differences in unemployment rates in a cross-section of U.S. states over a quarterly time series. PPE, ICE, and CGF F DE are individual variables for the three different exceptions, and EXP represents Expansion periods. fs and qt represent fixed cross-section and quarterly time series effects, respectively. s indexes states and t indexes time. I use a four-lag representation throughout the paper due to the quarterly data, but the results are generally robust from one to twelve lags. Arellano and Bover (2005) note there is a possible bias induced in the predetermined variables by the fixed effects, but Roodman (2006) shows the cause for concern is greatest with a panel that has a large cross section, N, and a small number of time periods, T . These concerns are greatest in samples with T less than 30. In my data I have 140 time periods for the full sample, 128 time periods with an average of 73 per state for the Outflows sample, 134 time periods with an average of 106 per state for the Expansion sample, 133 time periods with an average of 62 per state for the Inflows sample and 122 time periods with an average of 29 per state for the Contrac- tion sample. I feel confident in all but the Contraction sample that my T is large enough that any bias will be as Roodman (2006) says, “insignificant and the problem will disappear.” In Table 13 we begin to see what emerges as a pattern throughout this statistical analysis. In the first regression for changes in the unemployment rate, the interaction 74 terms for EXP with ICE and CGF F DE have statistically significant coefficients.3 The negative coefficient on the EXP variable alone reflects the countercyclical nature of unemployment rates. The positive coefficients on the two interactions terms tell us the ICEs and the CGFFDEs significantly diminished the difference between Expansion and Contraction periods by 0.034 and 0.043 percentage points per quarter. Adding the coefficient of -0.012 on ICE and the coefficient of 0.034 on ICE∗EXP yields a positive sum of 0.022, which tells us that the recovery of unemployment rates post recession will be slower during Expansion periods with the exception in place. Adding the coefficient of -0.006 on CGF F DE and the coefficient of 0.043 on CGF F DE ∗ EXP yields a positive sum of 0.037, which tells us that the recovery of unemployment rates post recession when CGFDEs are in place will be even slower than when ICEs are in place. Using a Wald test, I find the difference between ICEs and CGFFDEs in the effects of EWEs on unemployment is statistically significant at the %5 threshold. However, these results are evidence of both the ICEs and CGFFDEs contributing to slower recoveries of unemployment rates during Expansion periods, controlling for lagged differences in unemployment rates and fixed state and time effects. The negative coefficients on ICE and CGF F DE tell us that as in Expansion periods there is a buffering in the movements (positive movements here) of the unemployment rate during Contraction periods. However, neither of the coefficients on the two EWEs are significant. EWEs may result in smaller positive movements in unemployment rates during Contraction periods, but their effects are much stronger during Expansion periods. Interpreting the result through the lens of Gavrel and Lebon (2008), EWEs act like destructive administrative costs. They predict that if

3One reason for the insignificant coefficients on PPE may be that many of the instances covered under PPEs, such as firing employees for race or gender, are covered with federal legislation and apply to all states. If PPEs have already been covered under federal legislation, then the timing for the state’s adoption of the law is meaningless. Beyond the timing issue, there is also no real variation across states once the federal policy is enacted. Another reason may be that 5 states have the PPEs, before my sample period starts, which again detracts from time series variation. 75

EWEs act like administrative costs, controlling for GDP growth, they will cause an increase in steady state unemployment rates by slowing job creation more than job destruction.

3.5.1.2 Log Diﬀerences In State Personal Income-Weighted Real GDP

In order to understand if the EWEs simply slowed down GDP growth causing a subsequent slow down in decreases in the unemployment rate, we need to examine the state specific measure of GDP growth. If these laws are only slowing down differences in unemployment rates during Outflow periods due to their effects on GDP growth, then we do not have a jobless recovery story. The model is represented below:

4 X DGDPs,t = α + βiDGDPs,t−i + β5PPEs,t + β6ICEs,t + β7CGF F DEs,t + β8OFs,t+ i=1

β9PPEs,t ∗ OFs,t + β10ICEs,t ∗ OFs,t + β11CGF F DEs,t ∗ OFs,t + fs + qt + s,t (3.2) where DGDP represents log differences in state personal income-weighted real GDP in a cross section of U.S. states over the quarterly time series. PPE, ICE, and CGF F DE represent indicator variables for the three different exceptions, and OF represents Outflows. fs and qt represent fixed cross section and quarterly time series effects. s indexes states and t indexes time. Again, I use a four-lag representation throughout the paper due to the quarterly data, but the results are generally robust from one to twelve lags. Examining regression 2 in Table 13, the coefficient for OF is significant and pos- itively related to changes in state personal income-weighted GDP, but none of the interaction terms are significant. This result is important because it tells us there are no significant effects through Outflows from the EWEs with respect to state-weighted 76

GDP growth that could be driving the previous results for differences in unemployment rates. The evidence here supports a story of jobless recovery as opposed to a story of diminished state-weighted GDP growth resulting in slower recovery from high unemployment rates. In order to test for consistency in these results, I rerun my regressions in split samples. I use the classification of Outflows versus Inflows and Expansion periods versus Contraction periods for the criteria in my sub-sampling. I test all four subsamples.

3.5.1.3 Outﬂows Subsample

The regression equations below examine the effects of EWEs on the recovery of unemployment rates and state-weighted GDP growth during the Outflows subsample. In 3.5.1.4 I use the Expansion periods subsample, but I put more weight on the Outflows subsample because it is more appropriate in capturing labor dynamics and unemployment cycles. The Outflows subsample better captures unemployment cycles such as the peak of unemployment rates and the subsequent recovery. However, the draw- back is that the Outflows subsample ignores periods of rising unemployment rates that take place during periods of positive state-weighted GDP growth. In fact, 40% of Expansion periods coincide with Inflows. While the Expansion periods characterize the recovery of GDP well, they miss 13.3% of Outflows. So while I put more weight on the Outflows subsample for this paper, these results are robust to either specification.4

4My results are strengthened even further with a larger magnitude of coeﬃcients when I run tests with only quarters that are part of both the Outﬂows and Expansion subsamples. 77

4 (O) X DURs,t = α+ βiDURs,t−i +β5PPEs,t +β6ICEs,t +β7CGF F DEs,t +fs +qt +s,t (3.3) i=1

4 (O) X DGDPs,t = α + βiDGDPs,t−i + β5PPEs,t + β6ICEs,t + β7CGF F DEs,t + fs + qt + s,t i=1 (3.4) where DUR(O) represents differences in the unemployment rate during Outflows and DGDP (O) represents log differences in state personal income-weighted real GDP during Outflows. PPE, ICE, and CGF F DE are individual variables for the three different exceptions. fs and qt represent fixed cross section and quarterly time series effects. s indexes states and t indexes time. In regression 1 in Table 14, the coefficient on ICE is significant at the 5% threshold and close to the 1% threshold for differences in the unemployment rate, DUR. The coefficient on CGF F DE is significant at the 1% threshold. These results are robust up to 11 lags at the 1% threshold and at the 5% threshold for the 12th lag. Consistent with the specification using interaction terms, the results indicate a slowdown in negative movements in unemployment rates with the adoption of these two exceptions. On average per quarter, changes towards lower unemployment rates experience a 0.033 percentage point slowdown with the adoption of ICEs and a 0.055 percentage point slowdown with the adoption of CGFFDEs. In regression 2 in Table 14, the coefficients indicate that EWEs play much less of a role in affecting state personal income-weighted GDP growth, DGDP , during Outflows. None of the coefficients on the EWEs are significant and the signs for the adoption of all three EWEs are positive. The results from using the Outflows subsample clearly support the findings of my specifications using interaction terms. Regression 1 and 2 in Table 14 provide evidence that despite state-weighted GDP growth, there have been slower recoveries in unemployment with the adoption of the 78

EWEs. Both ICEs and CGFFDEs contribute to jobless recovery. Moreover, my ﬁnd- ings do not just indicate that EWEs lead to slower GDP growth which in turn causes slower recoveries in unemployment. I show that EWEs act as destructive administrative costs leading to slower recovery by an amount that is over and above that of weak GDP growth.

3.5.1.4 Expansion Periods Subsample

Even with a significant slowdown in Outflows, there are enough Inflows that take place during recovery from a recession that the overall effect on changes in unemployment rates could be insignificant. Therefore, it is imperative to also test the Expansion subsample to see if the results are robust to this specification allowing for periods of rising unemployment rates during recovery from recession.5

4 (E) X DURs,t = α+ βiDURs,t−i +β5PPEs,t +β6ICEs,t +β7CGF F DEs,t +fs +qt +s,t (3.5) i=1

4 (E) X DGDPs,t = α + βiDGDPs,t−i + β5PPEs,t + β6ICEs,t + β7CGF F DEs,t + fs + qt + s,t i=1 (3.6) where DUR(E) represents differences in the unemployment rate during Expansion periods, DGDP (E) represents log differences in state personal income-weighted real GDP during Expansion periods. PPE, ICE, and CGF F DE are individual variables for the three different exceptions. fs and qt represent fixed cross section and quarterly time series effects. s indexes states and t indexes time. In regression 3 in Table 14 there are differences with respect to the Outflows subsample with slightly smaller coefficients on ICE and CGF F DE. The coefficient

5For the same reasons, I also must test both Inﬂows and Contraction periods to understand the eﬀect of EWEs during recessionary times. 79

on ICE is consistently significant close to and sometimes breaking the 5% threshold up to 12 lags. The coefficient on CGF F DE is consistently significant to the 5% threshold and sometimes at the 1% threshold up to 12 lags. Thus while using the Expansion subsample slightly dampens the significant, the results remain robust. In regression 4 in Table 14 the most noticeable difference with respect to the Outflow subsample is the consistently negative and significant coefficient on ICE for state personal income-weighted GDP. Again however, when looking at magnitude, the coefficient on ICE in the regression for state-weighted GDP growth is about 12.2% of the average for the Expansion subsample, while the coefficient on ICE in the regression for differences in the unemployment rate is about 44.6% of the average for the subsample. The results for both dependent variables in the Expansion subsample tell the same general story as the results from the interaction terms and the Outflow subsample. The PPEs play very little or no role in shaping the behavior of unemployment and an inconclusive role in affecting GDP growth. ICEs and CGFFDEs in this subsample show evidence of slowing down reductions in unemployment rates during recoveries while having little or no effect on state-weighted GDP growth. The consistency of the evidence strongly suggests that ICEs and CGFFDEs contribute to jobless recoveries in the United States.

3.5.1.5 Inﬂows and Contraction Periods Subsamples

The results in Table 15 from re-running the previous tests with subsamples from Inflows and Contraction periods again shed light on the ambiguity discussed in Mortensen and Pissarides (1999) for EWEs. For differences in unemployment rates, DUR, in regressions 1 and 3, there are no significant results for any EWEs. This 80

result clearly shows there is much less of a buffering effect on periods of increases than on periods of decreases in unemployment rates.6 EWEs seem to affect job creation more than job destruction. The results from both using interaction terms and subsample analysis lead me to interpret EWEs as destructive administrative costs along the lines of the theory proposed by Gavrel and Lebon (2008).

3.5.2 Panel VARs in Split Samples

This subsection continues the investigation of the impact of EWEs on jobless recovery in split samples using a two variable panel vector autoregression (panel VAR). The reason for this further analysis is to directly address the inextricable linkages between changes in unemployment rates and GDP growth with their recursive eﬀects on one another. Jobless recovery refers to the movements of both variables, so I investigate the eﬀects of EWEs on a system of both variables simultaneously in the model below:

4 4 X X DURs,t = α + γiDURs,t−i + βiDGDPs,t−i + β5PPEs,t + β6ICEs,t+ i=1 i=1

β7CGF F DEs,t + fs + qt + s,t (3.7)

4 4 X X DGDPs,t = ψ + θiDURs,t−i + φiDGDPs,t−i + φ5PPEs,t + φ6ICEs,t+ i=1 i=1

φ7CGF F DEs,t + fs + qt + s,t (3.8)

where again, from Roodman (2006), I am conﬁdent because of the large T in my panel regressions that any bias induced in the predetermined variables by the ﬁxed

6The results are robust up to at least 12 lags. 81

effects will be insignificant.7 The results of the panel VARs for the Outflows and Expansion subsamples are presented in Table 16. The results look very similar to the single variable analysis results. In VAR 1 for the Outflows subsample, the coefficients on PPE and CGF F DE are positive and significant for differences in unemployment rates, DUR.8 In fact, the coefficients change very little with respect to the case of the single variable analysis. This result shows us that while fixed effects in this panel VAR may introduce some bias, the results are very consistent the rest of my specifications.9 The results for state-weighted GDP growth, DGDP , for the Outflows subsample in VAR 1 are also consistent with the single variable analysis. Again, I find that while both ICEs and CGFFDEs significantly dampen the changes in unemployment rates during Outflows, they have a minimal effect on state-weighted GDP growth. The coefficient on CGF F D for state-weighted GDP growth actually indicate an increase in growth. With these results we see that running regressions on the system as a whole, as opposed to the individual parts, reinforces the evidence of EWEs contribution to jobless recovery.10 In VAR 2 in Table 16, the results also reinforce the results from the single variable analysis. Using the Expansion subsample slightly weakens the results from the Outflows subsample, but the results still indicate that EWEs contribute significantly

7Ideally I would like to run the panel VAR using the method from Love and Zicchino (2002) to remove any possible bias from fixed effects, but this is not possible with my model. The first issue is the method requires all variables to be treated as endogenous and transformed through a forward mean differencing process known as the Helmert Process. My EWE laws are exogenous, binary variables, which leave a difficult interpretation when they are forward mean differenced. 8This result is robust to all lag specifications tested, which is from 4 to 8. 9The results are generally robust to replacing fixed time effects with the structural break found in chapter one beginning in the fourth quarter of 1984. Using the Outflows subsample, the coefficient on CGF F DE continues to be significant for DUR at the 5% level from four to eight lags and for DGDP it is significant at the 5% or 1% level depending on the lag specification. The coefficient on IC for DUR is inconsistently significant at the 10% level depending on the lag length and is never significant for DGDP . The structural break is significant at the 1% level at all times. The exceptions seem to explain some but not all of the structural break. 10Again, when rerunning these VAR’s including only quarters that are in both the Expansion and Outflows subsamples, the results are robust. 82

to jobless recovery. Table 17 for Inflows and Contraction subsamples again shows no significant impact of the EWEs on changes in unemployment rates. It does show positive and significant effects for state-weighted GDP growth from both PPEs and ICEs. This result indicates that once the EWEs are in place, negative periods of growth are less severe. Again, this could be the result of EWEs influencing firms to substitute away from hiring labor towards investing in capital, which may be contributing to recessions being less deep. To illustrate the economic significance of my results, the quarterly peak for the unemployment rate was 9.9% in 2009Q4. Using VAR 1 in Table 16 for the Outflows subsample without either ICEs or CGFFDEs, the model predicts an unemployment rate two years later of 7.7%. With the ICEs and CGFFDEs the model predicts an unemployment rate of 8.4%.11 With a workforce of about 150 million that would imply an additional 1.1 million unemployed by 2011Q4. With workers receiving on average $300 per week in unemployment benefits, these results predict that if all states had adopted these EWEs they would have cost the U.S. an extra $13.3 billion in 2011 in unemployment benefits. Using the average personal income tax rate from 2010 of 11.8%, the average wage rate from the Bureau of Business & Economic Research for 2011 of $48,301 and the 1.1 million extra unemployed, there could be approximately another $6.2 billion of cost in lost tax revenue.12 Using the change in nominal GDP from the fourth quarter of 2010 to the fourth quarter of 2011 and dividing by the change in the level of employment over the same time, I calculate about $465 in value added per worker added. With 1.1 million extra unemployed, this would account for another half of a billion dollars of cost in lost output. These numbers are averages and rely on a lot of assumptions, but they make the point that the costs of adopting these exceptions can quickly add up to have a large negative economic impact.

11I use BEA estimates of GDP growth in these calculations. 122010 is the most recent year available for average personal income tax rates. 83

3.6 Robustness Tests

In this section, I present further robustness tests of the results in section 5. These tests account for the percentage of workers covered by collective bargaining agreements and the inclusion of a lagged implementation effect for firms. For both robustness checks, I rerun all of the regressions from the previous section. In both cases I find that my results are robust to these alternative specifications.13 The first robustness check addresses a potentially omitted variable by incorporating the percentage of workers covered by collective bargaining. If collective bargaining agreements make firing workers more difficult and costly, then according to Mortensen and Pissarides (1999) these agreements would have the same effect as EWEs in slowing job creation. If there is a strong correlation between states with large percentages of workers covered by collective bargaining agreements and states with EWEs, then my regression coefficients may be showing firms’ reluctance to hire new workers due to the presence of collective bargaining agreements rather than due to the EWEs. As a measure of the percentage of workers covered by collective bargaining, called union density here, I use estimated annual data of each state’s nonagricultural wage and salary employees from the Union Membership and Coverage Database.14 The percentages are estimates from 1977 to 2010 formulated using BLS methods and based on the 1983-2011 Current Population Survey (CPS), Outgoing Rotation Group (ORG) earnings files and the 1977-81 May CPS earnings files.15 I apply the annual data to all four quarters of each year. Annual variations in percentages of workers covered by collective bargaining agreements are small, making me confident that missing quarterly variations should not have a significant impact on my results. I run the following VAR for both Outflows and Expansion subsamples:

13For brevity sake I only show the tables for my OLS panel VARs, but any and all of the robustness test results are available upon request. 14www.unionstats.com 15Hirsch, Barry and David A. Macpherson, and Wayne G. Vroman (2001) 84

4 4 X X DURs,t = α + γiDURs,t−i + βiDGDPs,t−i + β5PPEs,t + β6ICEs,t+ i=1 i=1

β7CGF F DEs,t + β8UDs,ty + fs + qt + s,t (3.9)

4 4 X X DGDPs,t = ψ + θiDURs,t−i + φiDGDPs,t−i + φ5PPEs,t + φ6ICEs,t+ i=1 i=1

φ7CGF F DEs,t + φ8UDs,ty + fs + qt + s,t (3.10)

where UDs,ty stands for union density and represents the annual average of the percentage of workers under collective bargaining agreements in each state, applied to each quarter of the year. Table 18 shows some evidence in the Expansion subsample of the percentage of workers covered by collective bargaining contributing to jobless recovery. For the EWEs however, the signs, magnitudes and degrees of significance show little change when I account for union influence. These results support the previous findings that EWEs contribute to jobless recovery. The second of these two robustness tests changes the timing of the EWEs effect on jobless recovery. This check is important of two reasons. First, ADS (2006) use this method to allow for a learning effect for the firm and since I am extending their work, I want to check for consistency with their results. The second reason is that I use quarterly data, which can result in an EWE that gets credit for having taken effect for a whole quarter even if the precedent setting court case is not settled until the second or third month of the quarter. Again, I rerun all of the tests from the previous section using a lagged implementation effect of one through four quarters. 85

I only present the results for the Outflows subsample for brevity sake, but the results are generally robust to all specifications. Looking at Table 19, we see the Panel VAR’s for the Outflows subsample with a one through four quarter lagged implementation effect. The coefficient on CGF F DE is positive and significant to at least the 10% threshold for changes in the unemployment rate all the way through a four quarter lagged implementation effect. The coefficient on ICE is positive and significant up to a three quarter lagged effect for changes in the unemployment rate. The coefficient on CGF F DE is positive and significant for changes in state-weighted personal income all the way through a four quarter lagged implementation effect as well. Even with a lagged effect, we are seeing a slow down in changes in the unemployment rate with the adoption of these two EWEs. In the case of CGFFDEs this is occurring even with an increase in state-weighted GDP growth. These results bolster my confidence in the fact that ICEs and CGFFDEs are significantly contributing to jobless recovery. Controlling for state personal income-weighted GDP growth, there is a definitive slowdown in the recovery of unemployment rates within a year of these laws becoming precedent.

3.7 Conclusions

After updating and correcting the existing sources relating to Employment-At-Will Exceptions in the U.S., I use panel data from across the states in single autocorrel- lation and panel VAR testing to find these laws significantly contribute to jobless recovery. My results indicate that Implied Contract Exceptions slow decreases in the unemployment rate during economic expansion by between 0.025 and 0.033 percentage points per quarter and Covenant of Good Faith and Fair Dealing Exceptions do so by between 0.039 and 0.055 percentage points per quarter. Mortensen and Pissarides (1999) predict that increased firing costs will cause em- 86 ployers to be slower in their hiring, controlling for GDP growth. Firms become more careful in their hiring practices with the possibility of incurring larger costs if they deem a current employment match is unsuitable in the future. EWEs create a new administrative burden and a need for defensive legal practices. According to theory, that makes firms more cautious and stringent in their screening process. My results provide evidence that this effect is indeed occurring; resulting in an increased joblessness of economic recoveries. Mortensen and Pissarides (1999) also expect that increased firing costs should cause firms to be more reluctant to discharge employees, following a negative shock. However, while I find significant slowdowns in the recovery of unemployment rates, I find no significant effects on changes in the unemployment rate during either Inflows or Contraction periods as a result of Employment-At-Will Exceptions. As opposed to previous work by DK (1992) and ADS (2006), this paper focuses on changes to the unemployment rate during transition periods rather than on the steady state levels of labor market variables. Gavrel and Lebon (2008) predict that if Employment-At-Will Exceptions are administrative and legal costs, they will cause an increase in steady state unemployment rates controlling for GDP growth. The Employment-At-Will Exceptions would affect higher steady state unemployment rates by slowing job creation more than they are slowing job destruction. I leave the ex- amination of Employment-At-Will Exceptions on steady state unemployment rates to future research. There is substantial room for future work on jobless recovery examining both the demand side factors mentioned in Aaronson, Rissman and Sulli- van (2004) and supply side factors such as dramatic increases seen in social benefits, which may reduce job search intensity and increase reservation wages. In this paper, the results suggest that despite the courts’ intent to aid workers, the precedent setting rulings in favor of the Implied Contract Exceptions and Covenant of Good Faith and Fair Dealing Exceptions negatively affected some of the more vul- 87 nerable parts of the labor force. While the rulings may protect some workers, the unemployed have a more difficult time finding a job. If state legislatures want to undo the effects of these rulings, they should pass statutes that explicitly state the limitations of the Implied Contract Exceptions and the Covenant of Good Faith and Fair Dealing Exceptions. This action would allow firms to more properly adjust their firing practices and reduce the ex ante legal and administrative firing costs aimed at reducing the probability of a wrongful discharge lawsuit. 88

List of References

[1] Aaronson, Rissman and Sullivan (2004), “Assessing the Jobless Recovery.” Journal of Economic Perspectives, Q2, 2-21.

[2] Aaronson, Rissman and Sullivan (2004), “Can sectoral reallocation explain the jobless recovery?” Journal of Economic Perspectives, Q2, 36-49.

[3] Alexopoulos, Michelle (2003), “Growth and Unemployment in a Shirking Eﬃ- ciency Wage Model. Canadian Journal of Economics, V.36(3), pp.728-746.

[4] Alexopoulos, Michelle (2004),“Unemployment and the Business Cycle.” Journal of Monetary Economics, V.51(2), pp.277-298.

[5] Andrews, Donald and Werner Ploberger (1994), “Optimal Tests When a Nui- sance Parameter is Present Only Under the Alternative,” Econometrica, V.62(6), pp.1383-1414.

[6] Andrews, Donald and Werner Ploberger (1994), “Optimal Tests When a Nui- sance Parameter is Present Only Under the Alternative.” Econometrica, V.62(6), pp.1383-1414.

[7] Andrews, Donald (1993), “Tests for Parameter Instability and Structural Change with Unknown Change Point.” Econometrica, V.61(4), pp.821-856.

[8] Arellano, M. and Olympia Bover (1995), “Another Look at the Instrumental Vari- able Estimation of Error-Components Models.” Journal of Econometrics, V.68, pp.29-51.

[9] Autor, Donohue III, and Schwab (2004), “The Employment Consequences of Wrongful-Discharge Laws: Large, Small, or None at All?” Papers and Proceed- ings of the One Hundred Sixteenth Annual Meeting of the American Economic Association San Diego, CA, V.94(2), pp. 440-446.

[10] Autor, Donohue III, and Schwab (2006), “The Costs Of Wrongful-Discharge Laws.” Review of Economics and Statistics, V.88(2), pp.211-231.

[11] Autor, Kerr and Kugler (2007), “Does Employment Protection Reduce Produc- tivity? Evidence from US States.” The Economic Journal, V.117, pp.189-217.

[12] Bauman, Florian (2006), “Eﬃciency Wages and Severance Payments with En- dogenous Shocks.” Working Paper.

[13] Bentolila, Samuel and Giuseppe Bertola (1990), “Firing Costs and Labour De- mand: How Bad is Eurosclerosis?” The Review of Economic Studies, V.57(3), pp.381-402. 89

[14] Blanchard and Simon (2001), “The Long and Large Decline in U.S. Output Volatility. Brookings Papers on Economic Activity, V.32(1), pp.135-164.

[15] Blanchard, O.P. and J. Gali (2010), “Labor Markets and Monetary Policy: a New Keynesian Model with Unemployment.” American Economic Journal: Macroeco- nomics, V.2(2), pp.1-30.

[16] Burnside, A. Craig, Martin Eichenbaum and Jonas D. M. Fisher (2000),“Fiscal Shocks in an Eﬃciency Wage Model.” NBER Working Paper Series, 7515.

[17] Dau-Schmidt, Kenneth G. and Timothy A. Haley (2007), “Governance of the Workplace: The Contemporary Regime of Individual Contract.” Comparative La- bor Law & Policy Journal, V.28, pp.313-350.

[18] Davis, Steven J., John C. Haltiwanger and Scott Schuh (2006), “Job Creation and Destruction.” Cambridge: MIT Press.

[19] De Palma, Francesco (2000), “Heterogeneous Workers, Dualism and Eﬃciency Wage.” Working Papers of BETA, Bureau d’Economie Thorique et Applique, UDS, Strasbourg.

[20] DeNicco, James (2011), “The Cost of Hiring and the Speed of Hiring: Separation Costs in a Discrete Eﬃciency Wage Model.” Working Paper.

[21] DeNicco, James and Chris Laincz (2011), “Jobless Recovery: A Time Series Look at the United States.” Working Paper.

[22] Dertouzos, James N. and Lynn A. Karoly (1992), “Labor-Market Responses to Employer Liability.” Santa Monica, CA: Rand, Institute for Civil Justice.

[23] Dickey, D.A. and W.A. Fuller (1979), “Distribution of the Estimators for Au- toregressive Time Series with a Unit Root.” Journal of the American Statistical Association, V.74(366), pp.427-431.

[24] Evans, George W (1989), “Output and Unemployment Dynamics in the United States: 1950-1985,” Journal of Applied Econometrics, V.4(3), pp.213-237.

[25] Faberman, Jason (2008), “Job Flows, Jobless Recovery, and the Great Modera- tion.” Federal Reserve Bank of PhiladelphiaWorking Series Papers, 08-11.

[26] Fujita, Shigeru and Garey Ramey (2009), “The Cyclicality of Separation and Job Finding Rates.” International Economic Review, V.50(2), 415-430.

[27] Gavrel, F. and I. Lebon (2008), “ Firing Costs, Payroll Taxes and Unemploy- ment.” Annals of Economics and Statistics, V.89, 121-129.

[28] Groshen, Erica and Simon Potter (2003), “Has Structural Change Contributed to a Jobless Recovery?” Current Issues in Economics and Finance, V.9(8). 90

[29] Hall, R.E. (2005), “Job Loss, Job Finding, and Unemployment in the U.S. Econ- omy over the Past Fifty Years.” NBER Working Paper, #11678.

[30] Hirsch, Barry and David A. Macpherson, and Wayne G. Vroman (2001), “Esti- mates of Union Density by State.” Monthly Labor Review, V.124(7), pp.51-55.

[31] Ichino, Andrfea and Giovanni Maggi (2000), “Work Environment and Individual Background: Explaining Regional Shirking Diﬀerentials in a Large Italian Firm. The Quarterly Journal of Economics, V.115(3), pp.1057-1090.

[32] Kim, Chang-Jin and Charles R. Nelson (1999), “Has the U.S. Economy Become More Stable? A Bayesian Approach Based on a Markov-Switching Model of the Business Cycle.” The Review of Economics and Statistics, V.81(4), pp.608-616.

[33] Limani, Nadjia (2007), “Righting Wrongful Discharge: A Recommendation for the New York Judiciary to Adopt a Public Policy Exception to the Employment- At-Will Doctrine.” Cardoza Public Law, Policy and Ethics Journal, V.1, pp.309- 337.

[34] Love, Inessa and Lea Zicchino (2002), “Financial Development and Dynamic In- vestment Behavior: Evidence from Panel Vector Autoregression.” Policy Research Working Paper Series - The World Bank #2913.

[35] MacLeod and Malcomson (1998), ”Motivation and Markets.” American Eco- nomic Association, V.88(3), pp. 388-411.

[36] McConnell, Margaret M. and Gabriel Perez-Quiros (2000), “Output Fluctuations in the United States: What Has Changed since the Early 1980’s?” American Economic Review, V.90(5), pp.14641476.

[37] Merz, Monica (1999), “Time Series Evidence of Unemployment Flows: The Sam- ple Period Matters.” Journal of Business & Economic Statistics, V.17(3), pp.324- 334.

[38] Miles, Thomas J. (2000) “Common Law Exceptions to Employment-At-Will and U.S. Labor Markets.” Journal of Law, Economics, and Organizations, V.16(1), pp.74 101.

[39] Moosa, Imad, Mervyn Silvapulle and Paramsothy Silvapulle (2004), “Asymmetry in Okun’s Law.” The Canadian Journal of Economics, V.37(2), pp.353-374.

[40] Mortensen, D. and C. Pissarides (1994), “Job Creation and Job Destruction in the Theory of Unemployment.” Review of Economic Studies, V.61(3), pp.397-415.

[41] Mortensen, D. and C. Pissarides (1999), “New Developments in Models of Search in the Labor Market.” Handbook of Labor Economics, V.3B, Ch.39.

[42] Muhl, Charles J. (2001), “The Employment-At-Will Doctrine: Three Major Ex- ceptions.” Monthly Labor Review, V.124(1). 91

[43] Nelson, R. Charles and Charles R. Plosser (1982), “Trends and random walks in macroeconomic time series: Some evidence and implications,” Journal of Mone- tary Economics, V.10(2), pp.139-162.

[44] Okun, M Arthur (1962), “Potential GNP: Its Measurement and Signiﬁcance,” Proceedings of the business and Economic Statistics Section of the American Sta- tistical Association, pp.98-104.

[45] Olivier, Blanchard and Lawrence H. Summers (1987), “Hysteresis and the Eu- ropean Unemployment Problem,” NBER Working Paper, #1950.

[46] Peretto, P. (2006), “The Employment (and Output) of Nations: Theory and Policy Implications. Duke University Working Paper, #16.

[47] Perron, P. (1989), “The Great crash, the oil price shock and the unit root hypothesis.” Econometrica, V.57(6), pp.1361-1401.

[48] Phillips, P. C. B. and P. Perron (1988), “Testing for a Unit Root in Time Series Regression.” Biometrika, V.75(2), pp.335-346.

[49] Roodman, David (2006), “How to Do xtabond2: An Introduction to ‘Diﬀerence’ and ‘System’ GMM in Stata.” Center for Global Development Working Paper, #103.

[50] Shapiro, C. and J. Stiglitz (1984), “Equilibrium Unemployment as a Worker Discipline Device,” American Economic Review, V. 74(3), pp.433-444.

[51] Shimer, Robert (2005), “Reassessing the Ins and Outs of Unemployment.” NBER: Working Series Paper #13421.

[52] Strand, J. (1987), “Unemployment as a Discipline Device with Heterogeneous Labor.” American Economic Review, V. 77(3), pp.489-493.

[53] The National Employer (2009-2010), “50-State Survey of Wrongful Discharge Law.” Littler Mendelson, V.1, Ch 7.

[54] Valletta, R.G., (2005), “Why Has the U.S. Beveridge Curve Shifted Back? New Evidence Using Regional Data.” Federal Reserve Bank of San Francisco Working Paper Series. 92

Appendix A: Chapter Two Steady State

From Equation (2.17) :

1 K 1 − β(1 − δ) α−1 = X = eN 0 βαA

From Equation (2.18):

−d −1 T d c θ c (1 − α)AXα 1 − s = (cs) 0 θ cs cs | {z } X1

−d −1 c θ c X 1 − s = (cs) 1 cs cs From equation (2.2) and (2.3):

c = cf + w = cf + sw + (1 − s)w

cs = cf + sw

c −1 cs = (1 − s)w − 1 cs Solving for w in terms of c : cs

−1 −d −1 c c θ c (1 − s)w − 1 = X 1 − s cs 1 cs cs

−d −1 1 c θ c c w = X 1 − s − 1 1 (1 − s) cs cs cs | {z } X2 We also know from equation (2.19) that:

α w = (1 − α)AX0 e − φb The below equation allows me to deﬁne my steady state variables in terms of c exogenous parameters by using the ratio cs , which must be constant in steady state, also determined by exogenous variables. 93

−d −1 −1 c θ c c X 1 − s = (1 − s) ((1 − α)AXαe − φb) − 1 1 cs cs 0 cs We can now solve

cf = cs − sw

c = cf + w In equilibrium with no borrowing or lending:

Π = cf

I can now ﬁnd the steady state equilibrium for K and N with the following equations:

(1 − α)A Kα(e)1−αN −α = w + φb | {z } |{z} X3 X4

−1 α w + X4 N = 1−α K X3e | {z } X5

Π = (AKα(eN)(1−α) − wN − φbN − Kδ

K = X0eN 94

Appendix B: Chapter Three Case Law

In this paper I follow the criteria set forth in ADS (2006) for the timing of case law and I concur with most of their findings for the existence of precedent setting case law. I explain and cite case law for any differences in classification between this paper and ADS (2006), as well as run robustness tests for discrepancies. ADS (2006) is the first source used in this paper and covers the period of 1970 to 1999. The second source is from Muhl (2001), which describes the evolution of Employment-At-Will Exceptions over time. Muhl (2001) updates the existence of the exceptions through 2000. The third source is from Dau-Schmidt and Haley (2007), which looks more generally at the history of employment contracts in the U.S. The results from Dau- Schmidt and Haley (2007) are used in a report from the 2008 National Conference of State Legislators. The fourth source is a comprehensive 50 state survey from the 2009/2010 edition of The National Employer, which uses case law in a state by state explanation of the existence and history of Employment-At-Will Exceptions in each state. The most agreed upon category for Employment-At-Will Exceptions is the Public Policy Exception with only two states in question. Muhl (2001) included Nebraska and Dau-Schmidt and Haley (2007) included Maryland as states without the Public Policy Exception. ADS (2006) and the National Employer (2009/2010) both disagree and include the same 43 states as having the exception. I find the case law cited in ADS (2006) is clear that both states do have the Public Policy Exception.1 The discrepancies under the Implied Contract Exception are more numerous and complicated, requiring more rigorous analysis to settle. Under my criteria, I find Muhl (2001) differs with six states in my final list.2 Dau-Schmidt and Haley (2007) are closer to my findings, differing with only two states.3 My analysis agrees with the National Employer (2009/2010) and disagrees with ADS (2006) findings of Implied Contract Exceptions in two states. I find evidence in both Arizona and Pennsylvania of case law that would “trigger a client letter warning about a law change,” and “that signaled the sustained adoption of the particular at-will exception.” Dau-Schmidt and Haley (2007) and the National Employer (2009/2010) both agree with my analysis of Pennsylvania and all three other sources agree with my analysis of Arizona. In Pennsylvania I find the opinion in Martin v. Capital Cities Media, Inc., while

1Adler v. American Standard Corp., 432 A.2d 464 (Md. 1981 July).; Ambroz v. Cornhusker Square, 416 N.W.2d 510 (Neb. 1987 November). 2Indiana: Romak v. Public Service Co., 511 N.E.2d 1024 (Ind. 1987 August)., Massachusettes: Hobson v. McLean Hospital Corp., 522 N.E.2d 975 (Mass. 1988 May)., Montanta: Montana Wrong- ful Discharge from Employment Act, Mont. Code Ann. 39-2-901 to 914 (1987 June)., Pennsylvania: Martin v. Capital Cities Media, Inc., 511 A.2d 830 (Pa. Super. 1986), Texas: Johnson v. Ford Mo- tor Co., 690 S.W.2d 90 (Tex. Civ. App. 1985 April). and Virginia: Frazier v. Colonial Williamsburg Foundation, 574 F. Supp. 318 (E.D. Va. 1983 September). 3Indiana: Romak v. Public Service Co., 511 N.E.2d 1024 (Ind. 1987 August).; Tennessee: Hamby v. Genesco Inc., 627 S.W.2d 373 (Tenn. Ct. App. 1981 November). 95

rejecting the wrongful discharge claim in appeal, includes language showing the willingness of the court to enforce Implied Contract Exceptions with regard to employee handbooks.4 The court rejects the claim but shows its willingness to apply the Im- plied Contract Excpetion stating, “We do not believe a reasonable person in the appellant’s position would have read this handbook provision as converting her from an at-will employee to an employee with an indefinite contract who could never be discharged without objective just cause.” Furthermore, later cases such as Baur v. Pottsville Area Emergency Med. Servs., Inc,5 cite Martin v. Capital Cities Media, Inc. when applying the reasonable person test and finding in favor of the plaintiff’s wrongful discharge claim. In Arizona, ADS (2006) rightly show that the precedent setting case for Implied Contract Exceptions was vacated by the Arizona Supreme Court a year after the lower court opinion was rendered. However, in vacating the decision the court also stated, “We agree with Leikvold that personnel manuals can become part of employment contracts. Whether any particular personnel manual modifies any particular employment-at-will relationship and becomes part of the particular employment contract is a question of fact. Evidence relevant to this factual decision includes the language used in the personnel manual as well as the employer’s course of conduct and oral representations regarding it.”6 The decision of the lower court was overturned because as the Supreme Court writes, “Summary judgment is inappropriate where a genuine dispute exists as to material facts.”7 In other words the lower court decision was not vacated because it found employee handbooks can alter the Employment- At-Will relationship, but because they found it did alter the relationship in this case instead of letting a jury decide the question of fact. As well, in the precedent setting opinion expanding Arizona to include both Public Policy and Covenant of Good Faith and Fair Dealing Exceptions, the court expressly states its view that, “Arizona is among the jurisdictions that have recognized the implied-in-fact contract term as an exception to the at-will rule.”8 Even more than in Pennsylvania, the existence of the Implied Contract Exception in Arizona appears to be clearly settled. The Covenant of Good Faith and Fair Dealing Exception is the category with the largest number of discrepancies across the literature. My findings, which look for a Covenant of Good Faith and Fair Dealing Exception in every employment relationship, completely corroborate the findings in ADS (2006). Like ADS (2006), under my criteria I exclude the three states of New Hampshire, Vermont and Utah included in the National Employer (2009/2010) as having the exception. In New Hampshire, the court in Harper v. Healthsource N.H., Inc. finds that, “An employer violates an implied term of a contract for employment at-will by firing an employee out of malice or bad faith in retaliation for action taken or refused by the employee in consonance with public policy.” The opinion restricts the use of the exception to cases where the

4Martin v. Capital Cities Media, Inc., 511 A.2d 830 (Pa. Super. 1986) 5Baur v. Pottsville Area Emergency Med. Servs., Inc, 758 A.2d 1265 (Pa. Super. 2000) 6Leikvold v. Valley View Community Hosp., 688 P.2d 201 (Ariz. App. 1983 June), vacated, 688 P.2d 170 (Ariz. 1984). 7Washington National Trust Co. v. W.M. Dary Co., 116 Ariz. 171, 568 P.2d 1069 (1977). 8Wagenseller v. Scottsdale Memorial Hosp., 710 P.2d 1025 (Ariz. 1985 June). 96

decision to fire the employee is, “contrary to public policy.”9 The Covenant of Good Faith and Fair Dealing Exception is not a separate cause for action and is not read into every employment relationship. In Vermont, the decision cited by the National Employer (2009/2010) requires that the at will relationship has already been altered for the Covenant of Good Faith and Fair Dealing Exception to apply.10 Lastly in Utah I disagree with the National Employer (2009/2010) assertion of the existence of the exception as evidenced in Berube v. Fashion Centre, Ltd.11 Barbaresi (1990) summarizes the Berube case well in the Brigham Young University Law Review stating the opinion of Justice Durham, “allowing a breach of the implied covenant of good faith and fair dealing to serve as evidence of a wrongful discharge,” could not persuade the, “majority of the court to agree. The court in fact refused to allow the use of the Covenant of Good Faith and Fair dealing in the context of wrongful discharge.12 Dau-Schmidt and Haley (2007) by far find the largest number of states with the Covenant of Good Faith and Fair Dealing Exception at twenty-one, as compared with my final findings of twelve. The main reason for the discrepancies is that there are court findings which, using Alabama as an example, often state something along the lines of, “a covenant of good faith and fair dealing is implied in every contract.”13 How- ever, many of these states either limit the findings to certain contracts (often insurance contracts), expressly exclude employment relationships or do not address employment relationships at all.14 Since all states generally adhere to the Employment-At-Will Doctrine (besides the exceptions discussed in this paper), a state must expressly apply the exception to employment relationships. All disagreements with Muhl (2001) have already been covered in the discussion above regarding discrepancies.

9Harper v. Healthsource N.H., Inc., 674 A.2d 962, 965 (N.H. 1996). 10Ross v. Times Mirror, 665 A.2d 580, 585 (Vt. 1995). 11Berube v. Fashion Centre, Ltd., 771 P.2d 1033, 1044 (Utah 1989). 12The results of this paper are robust to including New Hampshire, Vermont and Utah individually and collectively as having the Covenant of Good Faith and Fair Dealing Exception from the dates of the cited cases. 13Grant v. Butler, 590 So. 2d 254, 256 (Ala. 1991) 14Alabama, Arkansas, Illinois, Indiana, New Hampshire, New Jersey, New York, Oklahoma, Penn- sylvania, and South Carolina. 97

Appendix C: Tables

Table 1: Historical Summary of U.S. Recoveries from Recessionary Periods

Recession Summary

It took two quarters of an average of 0.62% real GDP Q3-1953 to Q1-1954 (3 growth for the unemployment rate to peak and start quarters of negative growth) coming down after the recession ending Q1-1954.

It took one quarter of 0.61% real GDP growth for the Q4-1957 to Q1-1958 (2 unemployment rate to peak and start coming down quarters of negative growth) after the recession ending in Q2-1958.

Q2-1960 to Q4-1960 (actually It took one quarter of 1.2% real GDP growth for the two quarters of negative unemployment rate to peak and start coming down growth with one positive in after the contraction ending in Q4-1960. between)

It took two quarters of an average 0.76% of real GDP Q3-1974 to Q1-1975 (3 growth for unemployment to peak and start coming quarters of negative growth) down after the recession ending in Q1-1975. It took three quarters of an average 0.078% real GDP growth (Actually two periods of positive growth sand- Q4-1981 to Q1-1982 (2 wiching one negative quarter.) for unemployment to quarters of negative growth) peak and start coming down after recession ending in Q1-1982. It took ﬁve quarters of an average of 0.72% real GDP Q3-1990 to Q1-1991(3 growth for unemployment to peak and start coming quarters of negative growth) down after the recession ending in Q1-1991.

Q1-2001 to Q3-2001(actually It took seven quarters of 0.49% real GDP growth for two quarters of negative the unemployment rate to peak and start coming down growth with one positive in after the contraction ending in Q3-2001. between) 98

Table 2: Unit Root Tests: Null Hypothesis for a Unit Root

Test ADF t-Stat DF-GLS t-Stat Variable Lag C C+T C C+T DGDP 1 -8.10*** -8.30*** -5.28*** -7.61*** DGDP 2 -7.98*** -8.25*** -4.90*** -7.47*** DGDP 3 -7.98*** -8.25*** -4.52*** -7.33*** DGDP 4 -7.47*** -7.98*** -4.09*** -7.06*** DGDP 5 -6.44*** -6.93*** -3.27*** -5.96*** DGDP 6 -6.08*** -6.73*** -3.04*** -5.84*** DGDP 7 -5.57*** -6.06*** -2.42** -4.89*** DGDP 8 -4.79*** -5.21*** -1.91* -4.07*** DTFP 1 -9.45*** -9.61*** -3.00*** -5.69*** DTFP 2 -7.72*** -8.12*** -2.47** -4.95*** DTFP 3 -8.82*** -9.19*** -2.34** -4.87*** DTFP 4 -7.93*** -8.43*** -1.97** -4.31*** DTFP 5 -6.91*** -7.46*** -1.56 -3.63*** DTFP 6 -6.33*** -6.83*** -1.19 -2.97** DTFP 7 -5.80*** -6.42*** -1.09 -2.89* DTFP 8 -4.92*** -5.28*** -0.57 -1.95 DUR 1 -7.50*** -7.49*** -5.48*** -6.80*** DUR 2 -7.80*** -7.78*** -5.42*** -6.95*** DUR 3 -8.83*** -8.82*** -5.82*** -7.81*** DUR 4 -7.48*** -7.48*** -4.76*** -6.62*** DUR 5 -5.94*** -5.95*** -3.63*** -5.17*** DUR 6 -5.95*** -5.95*** -3.43*** -5.01*** DUR 7 -7.25*** -7.26*** -3.83*** -5.85*** DUR 8 -5.40*** -5.39*** -2.71*** -4.18*** UR 1 -3.72*** -3.80*** -2.51** -3.89*** UR 2 -3.92*** -4.26*** -2.67*** -4.16*** UR 3 -3.44*** -3.77** -2.26** -3.69** UR 4 -2.60* -3.00 -1.67* -3.02** UR 5 -2.76* -3.27* -1.87* -3.34** UR 6 -3.25** -3.76** -2.21** -3.77*** UR 7 -3.04** -3.51** -1.97** -3.51*** UR 8 -2.18 -2.62 -1.24 -2.67* ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively. ADF = Augmented Dickey- Fuller; DF-GLS = Dickey-Fuller Generalized Least Squares. C = Constant; C+T=Constant plus a time trend. 99

Table 3: OLS VARs for Outﬂows Subsample.

VAR 1 2 Speciﬁcation Structural Breaks Time Trend Variable DUR UR DGDP DUR UR DGDP -0.009 -0.009 0.012*** 0.080 0.080 0.011*** C (0.062) (0.062) (0.003) (0.063) (0.063) (0.003) -0.068*** 0.932*** 0.001* -0.067*** 0.933*** 0.001** UR(-1) (0.009) (0.009) (0.0004) (0.010) (0.010) (0.0004) 0.105*** 0.105*** -0.001 0.088** 0.088** -0.001 DUR(-2) (0.037) (0.037) (0.002) (0.039) (0.039) (0.002) -0.4287** -0.4287** 0.093 -5.228*** -5.228*** 0.088 DGDP(-1) (1.784) (1.784) (0.089) (1.914) (1.914) (0.088) 0.231*** 0.231*** -0.004** SB594 (0.043) (0.043) (0.002) 0.297*** 0.297*** -0.009*** SB844 (0.044) (0.044) (0.002) 0.001*** 0.001*** -4.5E-05*** QUARTER (0.0002) (0.0002) (1.0E-05) N 145 145 145 145 145 145 R2 0.435 0.990 0.160 0.349 0.988 0.162 Adj. R2 0.415 0.990 0.130 0.330 0.988 0.138 F-Stat 21.40*** 2734*** 5.300*** 18.74*** 2982*** 6.776*** LL 59.601 59.601 494.89 49.303 49.30 495.072 DW 2.434 2.434 1.862 2.145 2.145 1.855 ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively. ( ) contains standard errors. 100

Table 4: OLS VARs for Expansion Subsample.

VAR 1 2 Speciﬁcation Structural Breaks Time Trend Variable DUR UR DGDP DUR UR DGDP 0.239*** 0.239*** 0.009*** 0.280*** 0.280*** 0.010*** C (0.066) (0.066) (0.002) (0.065) (0.065) (0.002) -0.057*** 0.943*** 0.001*** -0.054*** 0.946*** 0.0008*** UR(-1) (0.011) (0.011) (0.0003) (0.011) (0.011) (0.0003) 0.120*** 0.120*** 0.001 0.118*** 0.118*** 0.001 DUR(-2) (0.043) (0.043) (0.001) (0.043) (0.043) (0.001) -14.257*** -14.257*** 0.215*** -13.954*** -13.954*** 0.198*** DGDP(-1) (1.886) (1.886) (0.055) (1.924) (1.924) (0.054) 0.159*** 0.159*** -0.004** SB601 (0.049) (0.049) (0.001) 0.164*** 0.164*** -0.007*** SB813 (0.049) (0.049) (0.001) 0.0006** 0.0006** -4.0E-05*** QUARTER (0.0002) (0.0002) (6.7E-06) N 217 217 217 217 217 217 R2 0.360 0.981 0.203 0.338 0.980 0.223 Adj. R2 0.345 0.980 0.184 0.326 0.980 0.210 F-Stat 23.71*** 2149*** 10.738*** 27.068*** 2610*** 15.174*** LL 12.666 12.666 780.47 9.059 9.059 783.19 DW 2.128 2.128 2.005 2.052 2.052 2.016 ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively. ( ) contains standard errors. 101

Table 5: Robustness Test 1: OLS VARs for Outﬂows Subsample.

VAR 1 2 Speciﬁcation Structural Breaks Time Trend Variable DUR DGDP DUR DGDP -0.411*** 0.015*** -0.413*** 0.021*** C (0.061) (0.003) (0.073) (0.003) 0.081 -0.009*** 0.090 -0.011*** DUR(-1) (0.060) (0.003) (0.063) (0.003) 0.141** -0.001 0.151** -0.003 DUR(-2) (0.054) (0.002) (0.058) (0.002) 0.097 0.001 0.104 -0.001 DUR(-3) (0.060) (0.003) (0.063) (0.003) -0.189*** 0.006*** -0.182*** 0.006*** DUR(-4) (0.050) (0.002) (0.052) (0.002) -1.122 -0.106 -1.028 -0.199** DGDP(-1) (2.238) (0.100) (2.396) (0.100) 2.726 -0.056 3.055 -0.140 DGDP(-2) (2.239) (0.100) (2.407) (0.101) 4.964** -0.066 5.611** -0.132 DGDP(-3) (2.375) (0.106) (2.486) (0.104) 2.344 0.082 3.008 0.028 DGDP(-4) (2.327) (0.104) (2.422) (0.101) 0.104** -0.002 SB602 (0.042) (0.002) 0.214*** -0.007*** SB951 (0.048) (0.002) 0.001*** -4.8E-05*** QUARTER (0.000) (1.0E-05) N 145 145 145 145 R2 0.391 0.270 0.347 0.312 Adj. R2 0.346 0.216 0.303 0.266 F-Stat 8.618*** 4.963*** 7.971*** 6.800*** LL 54.216 505.085 49.111 509.348 DW 2.313 1.783 2.200 1.775 ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively. ( ) contains standard errors. 102

Table 6: Robustness Test 2: OLS VARs for Outﬂows Subsample.

VAR 1 2 Speciﬁcation Structural Breaks Time Trend Variable DUR UR DTFP DUR UR DTFP -0.060 -0.060 2.478** 0.027 0.027 1.894** C (0.058) (0.058) (1.096) (0.059) (0.059) (1.041) -0.067*** 0.933*** 0.310** -0.066*** 0.934*** 0.348** UR(-1) (0.009) (0.009) (0.167) (0.010) (0.010) (0.171) 0.130*** 0.130*** 0.963 0.117*** 0.117*** 1.024 DUR(-2) (0.035) (0.035) (0.661) (0.038) (0.038) (0.664) -0.005 -0.005 -0.037 -0.009* -0.009* -0.018 DTFP(-1) (0.005) (0.005) (0.085) (0.005) (0.005) (0.085) 0.240*** 0.240*** -1.806** SB594 (0.044) (0.044) (0.832) 0.316*** 0.316*** -3.071*** SB844 (0.045) (0.045) (0.846) 0.001*** 0.001*** –0.013*** QUARTER (0.0002) (0.0002) (0.004) N 145 145 145 145 145 145 R2 0.417 0.990 0.132 0.330 0.988 0.116 Adj. R2 0.396 0.989 0.101 0.311 0.988 0.091 F-Stat 19.92*** 2650*** 4.224*** 17.24*** 2898*** 4.582*** LL 57.377 57.377 -367.83 47.254 47.254 -369.16 DW 2.343 2.343 1.852 2.076 2.076 1.849 ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively. ( ) contains standard errors.

Table 7: Forecast Model (Outﬂows: UR as Dependent Variable)

Variable C UR(-1) DUR(-2) DGDP(-1) SB594 SB844 UR 0.056 0.916 0.146 -3.569 0.256 0.310 Stand Error 0.070 0.011 0.039 1.915 0.045 0.048 Probability *** *** * *** ***

Statistics N R sqr. Adj. R sqr. Log Likelihood F Stat DW 136 0.988 0.988 46.947 1843.251 2.098 103

Table 8: Percent Industry Composition: Bureau of Labor and Statistics

Year 1950 1960 1970 1980 1990 2000 2010

Goods-Producing 37.34 35.91 31.93 27.48 21.97 18.84 13.79

Mining & Logging 2.02 1.44 0.97 1.15 0.70 0.45 0.53

Construction 5.09 5.57 5.08 5.09 4.97 5.16 4.35

Manufacturing 30.23 28.90 25.89 21.24 16.31 13.22 8.92

Service-providing 49.03 48.78 50.51 54.68 61.40 65.43 68.86

Financial Activities 4.12 4.60 4.90 5.46 6.04 5.87 5.89

Profes. & Bus. 6.54 6.76 7.38 8.22 9.87 12.55 12.74

Education & Health 4.84 5.33 6.37 7.64 9.84 11.44 14.95

Leisure & Hosp. 6.24 6.32 6.68 7.39 8.51 8.96 10.03

Other 1.89 2.09 2.49 2.98 3.87 3.93 4.10

Government 13.63 15.31 17.56 17.84 16.63 15.73 17.34 104

Table 9: Average Index Scores and Unemployment Rates from 1990 to 2010

Country REG5 REG7 REG8 UR Australia 0.00 1.00 3.00 7.05 Austria 2.00 2.50 6.00 4.27 Belgium 0.00 3.00 0.00 8.31 Canada* 0.00 NA 2.00 8.38 Denmark 0.00 2.00 2.00 5.85 Finland 3.00 3.00 0.00 9.05 France* 4.08 3.00 0.00 9.63 Germany* 4.00 3.00 3.00 8.47 Greece 1.00 1.00 4.00 8.96 Hungary 0.00 2.00 4.00 8.33 Ireland 0.00 4.00 2.00 10.33 Italy* 0.00 3.00 4.00 9.09 Japan* 2.00 1.00 6.00 3.66 Korea 3.68 1.00 6.00 3.43 Mexico 6.00 3.00 2.00 3.66 Netherlands 3.00 1.75 2.00 4.91 New Zealand 0.00 1.00 2.00 6.19 Norway 5.00 2.00 4.00 4.03 Poland 0.00 0.00 2.00 13.77 Portugal 4.48 3.96 4.48 6.55 Slovak Republic 0.63 1.38 5.00 14.59 Spain 3.50 4.50 0.00 14.25 Sweden 4.00 6.00 2.00 6.07 Switzerland 0.00 1.00 0.00 2.72 United Kingdom* 0.00 1.00 2.00 7.30 United States* 0.00 NA 1.00 5.90 Average 1.79 2.54 4.96 7.05 * denotes G7 countries; Scores range between 0 and 6 REG5 - Deﬁnition of Unfair Dismissal REG7 - Compensation Following Unfair Dismissal REG8 - Possibility of Reinstatement Following Unfair Dismissal UR - Average Annual Unemployment Rate 105

Table 10: Average Changes in Unemployment Rates and State-Weighted GDP Growth, Conditional on the Presence of EWEs: Outﬂows and Expansion Periods.

Subsample Outﬂows Expansion EWE Presence of EWE DUR DGDP DUR DGDP No -0.2970 0.0112 -0.0750 0.0135 Public Policy Yes -0.2082 0.0091 -0.0482 0.0107 % Change -30%*** -19%*** -36%*** -21%*** No -0.3025 0.0117 -0.0761 0.0137 Implied Contract Yes -0.2058 0.0089 -0.0474 0.0106 % Change -32%*** -24%*** -38%*** -23%*** No -0.2411 0.0096 -0.0607 0.0115 Good Faith Yes -0.2056 0.0107 -0.0341 0.0119 % Change -15%*** 11%** -44%** 3% ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively for a two tailed t-test.

Table 11: Average Changes in Unemployment Rates and State-Weighted GDP Growth, Conditional on the Presence of EWEs: Inﬂows and Contraction Periods.

Subsample Inﬂows Contraction EWE Presence of EWE DUR DGDP DUR DGDP No 0.3416 0.0031 0.2486 -0.0127 Public Policy Yes 0.2865 0.0041 0.2626 -0.0081 % Change -16%*** 32% 06% -36%*** No 0.3495 0.0028 0.2890 -0.0131 Implied Contract Yes 0.2809 0.0043 0.2404 -0.0078 % Change -20%*** 54%** -17%* -41%*** No 0.3074 0.0036 0.2578 -0.0099 Good Faith Yes 0.2876 0.0045 0.2575 -0.0082 % Change -6% 25% 0% -17% ** ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively for a two tailed t-test. 106

Table 12: Panel unit root tests: Summary For Diﬀerences in Unemployment Rates and State Personal Income-Weighted GDP, includes individual eﬀects and a linear trend.

Test Statistics Null: Common Unit Root Null: Individual Unit Root Sample Levin, Breitung Im, Pesaran & ADF-Fisher PP-Fisher Lin Chu t* t-stat Shin W-stat χ2 χ2 -23.9*** -19.9*** -30.2*** 1036.6*** 1040.3*** DUR, Full (6794) (6744) (6794) (6794) (6900) -64.8*** -4.8*** -76.2*** 1459.6*** 1633.4*** DUR, OF (3726) (3676) (3726) (3726) (3741) -20.0*** -4.9*** -35.0*** 1221.7*** 1245.9*** DUR, Exp. (5351) (5301) (5351) (5351) (5426) -25.9*** -10.8*** -32.0*** 981.4*** 970.7*** DUR, IF (3136) (3086) (3136) (3136) (3159) -1.8** -2.7*** -0.7 194.6*** 160.4*** DUR, Cont. (1448) (1399) (1448) (1448) (1448) -61.7*** -40.3*** -61.6*** 2493.1*** 3141.1*** DGDP, Full (6875) (6825) (6875) (6875) (6900) -47.6*** -8.8*** -59.7*** 1641.8*** 1626.7*** DGDP, OF (3741) (3691) (3741) (3741) (3741) -70.1*** -2.9*** -95.9*** 3248.2*** 3468.6*** DGDP, Exp. (5417) (5367) (5417) (5417) (5426) -26.0*** -9.3*** -34.6*** 1037.8*** 1271.1*** DGDP, IF (3149) (3099) (3149) (3149) (3159) -36.7*** -7.9*** -49.7*** 2246.3*** 2295.3*** DGDP, Cont. (1448) (1399) (1448) (1448) (1448) ***,**,* denote significance rejecting the unit root at the 1%, 5%, and 10% threshold respectively. ( ) contains the number of observations in the test. Cross sections = 50 for all samples except the Contraction subsample, which has 49 due to a lack of data points in one state. OF=Outflows, IF=Inflows, Exp.=Expansion, Cont.=Contraction 107

Table 13: Single Variables Analysis with Interaction Terms.

Regression 1 2 Dependent Variable DUR DGDP Statistics Coefficient Stand. Error Coefficient Stand. Error C 0.052*** (0.018) 0.0042*** (0.0007) DUR(-1) 0.404*** (0.012) DUR(-2) -0.129*** (0.013) DUR(-3) 0.030** (0.013) DUR(-4) -0.002 (0.012) DGDP(-1) -0.0595*** (0.0122) DGDP(-2) 0.0358*** (0.0122) DGDP(-3) 0.0158 (0.0121) DGDP(-4) 0.0821*** (0.0121) PPE 0.005 (0.020) 0.0014** (0.0007) ICE -0.012 (0.021) 0.0002 (0.0007) CGFFDE -0.006 (0.024) 0.0006 (0.0008) EXP -0.060*** (0.017) PPE*EXP -0.025 (0.020) ICE*EXP 0.034* (0.020) CGFFDE*EXP 0.043** (0.022) OF 0.0027*** (0.0006) PPE*OF -0.001 (0.0007) ICE*OF -0.0003 (0.0007) CGFFDE*OF 0.0006 (0.0008) R2 0.613 0.613 Adj. R2 0.602 0.602 F-Stat 53.545*** 53.545*** AIC 0.112 0.112 DW 2.017 2.017 ***,**,* denote significance at the 1%, 5%, and 10% threshold respectively. ( ) contains standard errors. 108

Table 14: Single Variable Analysis with Outﬂows and Expansion Subsamples.

Sample Outﬂows Expansion Regression 1 2 3 4 Dependent Variable DUR DGDP DUR DGDP -0.2591*** 0.0099*** -0.0466*** 0.0122*** C (0.0136) (0.0008) (0.0129) (0.0005) 0.0868*** 0.5892*** DUR(-1) (0.0125) (0.0152) -0.0663*** -0.264*** DUR(-2) (0.012) (0.0146) 0.0022 0.0462*** DUR(-3) (0.0131) (0.0130) -0.0264* -0.0036 DUR(-4) (0.0137) (0.0119) -0.0973*** 0.0032 DGDP(-1) (0.0165) (0.0120) -0.0315*** -0.0148 DGDP(-2) (0.0152) (0.0106) 0.0449** -0.0035 DGDP(-3) (0.0155) (0.0104) 0.0097 0.0511*** DGDP(-4) (0.0163) (0.0113) 0.0028 0.0003 -0.0133 0 PPE (0.0136) (0.0007) (0.0133) (0.0005) 0.0328** 0.0003 0.0251** -0.0014*** ICE (0.0133) (0.0007) (0.0129) (0.0005) 0.0551*** 0.0015 0.0386** 0.0004 CGFFDE (0.0182) (0.001) (0.0166) (0.006) T 128 128 134 134 N 50 50 50 50 Obser 3669 3669 5304 5304 R2 0.373 0.333 0.564 0.363 Adj. R2 0.3402 0.298 0.5482 0.339 F-Stat 11.33*** 9.52*** 35.04*** 15.39*** AIC -0.4137 -6.242 -0.173 -6.75 DW 1.4141 1.784 1.836 1.466 ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively. ( ) contains standard errors. 109

Table 15: Single Variable Analysis with Inﬂows and Contraction Subsamples.

Sample Inﬂows Contraction Regression 1 2 3 4 Dependent Variable DUR DGDP DUR DGDP 0.2436*** 0.0011 0.2437*** -0.0137*** C (0.0175 ) ( 0.0009 ) (0.0339) (0.0012) 0.5238*** 0.1620*** DUR(-1) (0.0219 ) (0.0246) -0.2595*** -0.0128 DUR(-2) (0.0261 ) (0.0318) 0.1459*** 0.1161*** DUR(-3) (0.0220 ) (0.0404) -0.0132 -0.0304 DUR(-4) (0.0158 ) (0.0372) -0.0402** -0.1381*** DGDP(-1) (0.0185 ) (0.0209) 0.0967*** 0.1154*** DGDP(-2) (0.0207 ) (0.0260) -0.0345* 0.0889*** DGDP(-3) (0.0197 ) (0.0268) 0.1469*** 0.0791*** DGDP(-4) (0.0185 ) (0.0209) -0.0222 0.0022** 0.0047 0.0024** PPE (0.0178 ) (0.0009 ) (0.0344) (0.0012) -0.0242 -0.0002 -0.0266 0.0029** ICE (0.0186) (0.0009 ) ( 0.0380) (0.0013) 0.0138 0.0003 -0.0543 -0.0010 CGFFDE (0.0225 ) (0.0011) (0.0487) (0.0017) T 133 133 122 122 N 50 50 50 50 Obser 3081 3081 1446 1146 R2 0.615 0.438 0.682 0.45 Adj. R2 0.59 0.401 0.638 0.33 F-Stat 24.59*** 11.97*** 15.37*** 5.076*** AIC -0.08 -6.05 0.659 -6.08 DW 1.43 2.25 1.94 1.797 ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively. ( ) contains standard errors. 110

Table 16: OLS Panel VAR with Fixed State and Time Eﬀects for Outﬂows and Expansion Subsamples.

VAR 1 2 Sample Outﬂows Expansion Dependent Variable DUR DGDP DUR DGDP -0.257*** 0.0098*** -0.049*** 0.0122*** C (0.014) (0.0007) (0.014) (0.0005) 0.087*** -0.0009 0.590*** -0.0025*** DUR(-1) (0.012) (0.0007) (0.0152) (0.0006) -0.065*** -0.0015** -0.264*** -0.0002 DUR(-2) (0.012) (0.0007) (0.0147) (0.0005) 0.001 -0.0005 0.0452*** -0.0005 DUR(-3) (0.013) (0.0007) (0.013) (0.0005) -0.024* -0.0024*** -0.003 -0.0015*** DUR(-4) (0.014) (0.0007) (0.012) (0.0004) -0.487 -0.1018*** 0.146 -0.0037 DGDP(-1) (0.304) (0.0165) (0.315) (0.0117) -0.398 -0.0382** -0.779*** -0.0205* DGDP(-2) (0.281) (0.0152) (0.286) (0.0106) 0.198 0.0356** 0.124 -0.0108 DGDP(-3) (0.287) (0.0155) (0.282) (0.0105) 0.707** 0.0019 0.858*** 0.0469*** DGDP(-4) (0.302) (0.0163) (0.304) (0.0113) 0.001 0.0003 -0.0137 -0.0001 PPE (0.014) (0.0007) (0.013) (0.0005) 0.033** 0.0004 0.0256** -0.0013*** ICE (0.013) (0.0007) (0.013) (0.0005) 0.055*** 0.0017* 0.0386** 0.0006 CGFFDE (0.018) (0.0010) (0.017) (0.0006) T 128 128 134 134 N 50 50 50 50 Obser 3669 3669 5304 5304 R2 0.375 0.338 0.566 0.368 Adj. R2 0.341 0.302 0.549 0.344 F-Stat 11.16*** 9.504*** 34.472*** 15.430*** AIC -0.414 -6.134 -0.174 -6.761 DW 1.419 1.787 1.834 1.473 ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively. ( ) contains standard errors. 111

Table 17: OLS Panel VAR with Fixed State and Time Eﬀects for Inﬂows and Con- traction Subsamples.

VAR 1 2 Sample Intﬂows Contraction Dependent Variable DUR DGDP DUR DGDP 0.259*** 0.0034*** 0.263*** -0.0125*** C (0.018) (0.0009) (0.035) (0.0012) 0.516*** -0.006*** 0.154*** -0.0019** DUR(-1) (0.022) (0.0011) (0.025) (0.0008) -0.268*** -0.0018 -0.023 -0.0036*** DUR(-2) (0.026) (0.0013) (0.032) (0.0011) 0.141*** -0.0003 0.101** 0.0008 DUR(-3) (0.022) (0.0011) (0.041) (0.0014) -0.02 -0.0026*** -0.041 -0.0021 DUR(-4) (0.016) (0.0008) (0.038) (0.0013) -1.022*** -0.074*** -1.999*** -0.1535*** DGDP(-1) (0.375) (0.0187) (0.618) (0.0212) -0.917** 0.0709*** -1.808** 0.0978*** DGDP(-2) (0.415) (0.0207) (0.769) (0.0264) -0.403 -0.0521*** 0.082 0.0759*** DGDP(-3) (0.393) (0.0196) (0.792) (0.0272) -0.199 0.1348*** 0.271 0.0724*** DGDP(-4) (0.368) (0.0184) (0.608) (0.0209) -0.022 0.0019** 0.004 0.0021* PPE (0.018) (0.0009) (0.034) (0.0012) -0.025 -0.0002 -0.029 0.0031** ICE (0.019) (0.0009) (0.038) (0.0013) 0.015 0.0004 -0.047 -0.0006 CGFFDE (0.022) (0.0011) (0.049) (0.0017) T 133 133 122 122 N 50 50 50 50 Obser 3081 3081 1446 1446 R2 0.617 0.453 0.686 0.424 Adj. R2 0.591 0.417 0.641 0.342 F-Stat 24.227*** 12.471*** 15.263*** 5.148*** AIC -0.082 -6.078 0.652 -6.092 DW 1.423 2.237 1.945 1.824 ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively. ( ) contains standard errors. 112

Table 18: OLS Panel VAR with Fixed State and Time Eﬀects for Outﬂows and Expansion Subsamples with Union Density.

VAR 1 2 Sample Outﬂows Expansion Dependent Variable DUR DGDP DUR DGDP -0.238*** 0.0099*** -0.122** 0.0180*** C (0.035) (0.0019) (0.032) (0.0012) 0.087*** -0.0009 0.590*** -0.0025*** DUR(-1) (0.012) (0.0007) (0.015) (0.0006) -0.065*** -0.0015** -0.264*** -0.0002 DUR(-2) (0.012) (0.0007) (0.015) (0.0005) 0.001 -0.0005 0.0449*** -0.0005 DUR(-3) (0.013) (0.0007) (0.013) (0.0005) -0.024* -0.0024*** -0.003 -0.00150*** DUR(-4) (0.014) (0.0007) (0.012) (0.0004) -0.491 -0.1019*** 0.144 -0.00372 DGDP(-1) (0.305) (0.0165) (0.315) (0.0117) -0.400 -0.0383** -0.757*** -0.0203* DGDP(-2) (0.281) (0.0152) (0.286) (0.0106) 0.197 0.0356** 0.148 -0.0106 DGDP(-3) (0.287) (0.0155) (0.282) (0.0105) 0.707** 0.0019 0.878*** 0.0471*** DGDP(-4) (0.302) (0.0163) (0.304) (0.0113) -0.001 -0.0000 0.004** 0.0000 UD (0.002) (0.0001) (0.002) (0.0001) 0.002 0.0003 -0.0161 -0.0001 PPE (0.014) (0.0007) (0.013) (0.0005) 0.033** 0.0004 0.026** -0.0013*** ICE (0.013) (0.0007) (0.013) (0.0005) 0.054*** 0.0017* 0.043** 0.0007 CGFFDE (0.018) (0.0010) (0.017) (0.0006) T 128 128 134 134 N 50 50 50 50 Obser 3669 3669 5304 5304 R2 0.375 0.338 0.566 0.368 Adj. R2 0.341 0.302 0.55 0.344 F-Stat 11.104*** 9.451*** 34.361*** 15.351*** AIC -0.414 -6.247 -0.175 -6.76 DW 1.419 1.787 1.836 1.474 ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively. ( ) contains standard errors. 113

Table 19: Outflows: OLS Panel VAR with Fixed State and Time Effects and a Lagged Implementation Effect

Variable/Learning Eﬀect PPE ICE GFFDE AIC SIC DUR/1Q 0.0008 0.0349*** 0.0516*** -0.4144 -0.0963 DGDP/1Q 0.0007 0.0002 0.0017* -6.2477 -5.9296 DUR/2Q 0.0003 0.0284** 0.0451** -0.4132 -0.0951 DGDP/2Q 0.0008 0.0004 0.0018* -6.2481 -5.9300 DUR/3Q 0.000 0.0229* 0.0411** -0.4124 -0.0944 DGDP/3Q 0.0009 0.0005 0.0019** -6.2484 -5.9304 DUR/4Q -0.0007 0.0137 0.0385** -0.4117 -0.0937 DGDP/4Q 0.0006 0.0003 0.0022** -6.2484 -5.9303 ***,**,* denote signiﬁcance at the 1%, 5%, and 10% threshold respectively. 114

Appendix D: Figures

Figure 1: US Unemployment (1948-2012)- BLS

10.00

8.00

6.00

4.00 Unemplymnet Rate

2.00

0.00 March-48 March-50 March-52 March-54 March-56 March-58 March-60 March-62 March-64 March-66 March-68 March-70 March-72 March-74 March-76 March-78 March-80 March-82 March-84 March-86 March-88 March-90 March-92 March-94 March-96 March-98 March-00 March-02 March-04 March-06 March-08 March-10 March-12 115

Figure 2: Out of Sample Forecast Results for Recovery from the 2001 Recession

Recovery Forecast for 2001 Recession

7.00

6.00

5.00

4.00

3.00 UR Predicted 2.00 Unemployment Rate UR 1.00

0.00

April-03 June-03 April-04 June-04 April-05 June-05 April-06 June-06 August-03 August-04 August-05 August-06 February-03 October-03 February-04 October-04 February-05 October-05 February-06 October-06 February-07 December-02 December-03 December-04 December-05 December-06 DATE 116

Figure 3: Forecast Results from 2012

Unemployment Rate Forecast from 2012: Oulows

10.00

9.00

8.00

7.00

6.00

5.00

UR Before 1959Q4

UR Between 1959Q4 and 1984Q4 4.00 UR Aer 1984Q4

Long Run Average 3.00

2.00 July-08 July-09 July-10 July-11 July-12 July-13 July-14 July-15 July-16 March-08 March-09 March-10 March-11 March-12 March-13 March-14 March-15 March-16 November-08 November-09 November-10 November-11 November-12 November-13 November-14 November-15 November-16 117

Figure 4: Forecast Results from 2012

Unemployment Rate Forecast from 2012: Expansion

10.00

9.00

8.00

7.00

6.00

5.00

4.00

UR Before 1960Q1 3.00

UR Between 1960Q1 and 1981Q3 2.00 UR Aer 1981Q3

1.00 Long Run Average

0.00 July-15 July-26 May-17 May-28 June-16 June-27 April-18 April-29 March-19 March-30 March-08 August-14 August-25 January-10 January-21 January-32 October-12 October-23 October-34 February-09 February-20 February-31 December-10 December-21 December-32 November-11 November-22 November-33 September-13 September-24 September-35 118

Figure 5: JOLTS: Job Openings and Labor Turnover Survey (2001-2012) - BLS

Hiring, Separaon, Job Opening, Layoﬀs & Discharge Rates (3 month moving averages)

4.5

4.0

3.5

3.0

2.5

Percent 2.0

1.5

1.0

0.5 Hiring Separaon Job Opening Layoﬀs & Discharges

0.0 Jun-01 Jun-02 Jun-03 Jun-04 Jun-05 Jun-06 Jun-07 Jun-08 Jun-09 Jun-10 Jun-11 Jun-12 Oct-01 Oct-02 Oct-03 Oct-04 Oct-05 Oct-06 Oct-07 Oct-08 Oct-09 Oct-10 Oct-11 Oct-12 Feb-01 Feb-05 Feb-11 Feb-02 Feb-03 Feb-04 Feb-06 Feb-07 Feb-08 Feb-09 Feb-10 Feb-12 119

Figure 6: Montly Participation Rate from (1948-2012) - BLS

U.S. Parcipaon Rate

68.0

66.0

64.0

62.0

60.0 Percent

58.0

56.0

54.0

52.0 Jul-53 Jul-64 Jul-75 Jul-86 Jul-97 Jul-08 Jan-48 Jan-59 Jan-81 Jan-70 Jan-92 Jan-03 Sep-51 Sep-95 Sep-62 Sep-73 Sep-84 Sep-06 Nov-49 Nov-60 Nov-71 Nov-82 Nov-93 Nov-04 Mar-57 Mar-68 Mar-79 Mar-90 Mar-01 Mar-12 May-55 May-77 May-88 May-99 May-66 May-10 120

Figure 7: Median Usual Weekly Earnings of Full-Time Wage and Salary workers - BLS

Median usual weekly earnings - in constant (1982-84) dollars, seasonally adjusted 350

340

330

320

310

300

290 Jul-82 Jul-87 Jul-92 Jul-97 Jul-02 Jul-07 Jul-12 Jan-80 Jan-85 Jan-90 Jan-95 Jan-00 Jan-05 Jan-10 Sep-81 Sep-86 Sep-91 Sep-96 Sep-01 Sep-06 Sep-11 Nov-80 Nov-85 Nov-90 Nov-95 Nov-00 Nov-05 Nov-10 Mar-79 Mar-84 Mar-89 Mar-94 Mar-99 Mar-04 Mar-09 May-83 May-88 May-93 May-98 May-03 May-08 121

Figure 8: Industry Unemployment Rates: BLS

6 Month Moving Average Unemployment Rate

Service Average (SA) 20 Construcon (SA)

15 Goods Producing without Construcon Average (SA)

10 Unemployment Rate

0 Jul-01 Jul-05 Jul-07 Jul-08 Jul-09 Jul-00 Jul-02 Jul-03 Jul-04 Jul-06 Jul-10 Jan-00 Jan-01 Jan-02 Jan-03 Jan-04 Jan-05 Jan-06 Jan-07 Jan-08 Jan-09 Jan-10 Oct-00 Oct-01 Oct-02 Oct-03 Oct-04 Oct-05 Oct-06 Oct-07 Oct-08 Oct-09 Oct-10 Apr-01 Apr-05 Apr-07 Apr-08 Apr-09 Apr-00 Apr-02 Apr-03 Apr-04 Apr-06 Apr-10 DATE 122

Figure 9: Impulse Response Functions for Eﬀort and Wages

0.8 Eﬀort Wages 0.6

0.4

0.2

-0.2

-0.4 Percent Deviaon From Steady State

-0.6

-0.8 0 10 20 30 40 50 60 70 80 Quarter 123

Figure 10: Impulse Response Functions for Consumption Types

0.8

C Cf Cs 0.6

0.4

0.2

-0.2

-0.4 Percent Deviaon From Steady State

-0.6

-0.8 0 9 18 27 36 45 54 63 72 Quarter 124

Figure 11: Impulse Response Functions for Labor and Capital

2 Labor Capital

-1

-2

-3 Percent Deviaon From Steady State

-4

-5 0 10 20 30 40 50 60 70 80 Quarters 125

Figure 12: Recovery of Unemployment Rate Following a Negative Productivity Shock

13 10.1

Unemployment Rate 2 ; PHI=23.67 ; LEFT AXIS 12.5 9.6

Unemployment Rate 1 ; PHI=0 ; RIGHT AXIS 12 9.1

11.5 8.6 Rate

11 8.1

10.5 7.6 Unemployment 10 7.1

9.5 6.6

9 6.1

8.5 5.6 0 1 2 3 4 5 6 7 8 9 10 Quarter 126

Figure 13: Percent Deviations From Steady State

0.8

Steady State Unemployment Rate = 5.62 0.7

Steady State Unemployment Rate = 8.54 0.6

0.5

0.4

0.3

Percent Deviaon From Steady State 0.2

0.1

0 1 2 3 4 5 6 7 8 9 Quarter