Nonparametric Regression Techniques in Economics Author(S): Adonis Yatchew Source: Journal of Economic Literature, Vol

Total Page:16

File Type:pdf, Size:1020Kb

Nonparametric Regression Techniques in Economics Author(S): Adonis Yatchew Source: Journal of Economic Literature, Vol American Economic Association Nonparametric Regression Techniques in Economics Author(s): Adonis Yatchew Source: Journal of Economic Literature, Vol. 36, No. 2 (Jun., 1998), pp. 669-721 Published by: American Economic Association Stable URL: http://www.jstor.org/stable/2565120 . Accessed: 19/08/2011 15:48 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to Journal of Economic Literature. http://www.jstor.org Journal of Economic Literature Vol. XXXVI (June 1998) pp. 669-721 NonparametricRegression Techniques in Economics ADONIS YATCHEW' 1. Introduction on an infinite number of unknown pa- rameters. (Nonparametric regression 1.1 Setting the Stage typically assumes little else about the shape of the regression function beyond 1.1.1 Benefits of Nonparametric Esti- some degree of smoothness.) Yet, we mation. If economics is the dismal sci- will argue that it is because such tools ence, then econometrics is its ill-fated lead to more durable inferences that offspring. The limited number of strong they will become an enduring-even in- implications derived from economic dispensable-element of every econo- theory, the all but complete inability to mist's tool kit, in much the same way perform controlled experiments, the that linear and nonlinear regression are paucity of durable empirical results, the today. After all, reliable empirical re- errant predictions, the bottomless res- sults are essential to the formulation of ervoir of difficult to measure but poten- sound policy. tially relevant variables-these do not From the point of view of a pure data set a pleasant working environment for analyst, the added value of nonparamet- the empirical analyst in economics. ric techniques consists in their ability to When one contrasts the overall qual- deliver estimators and inference proce- ity of economic data and empirical in- dures that are less dependent on func- ferences with the plethora of sophisti- tional form assumptions. They are also cated econometric theory already useful for exploratory data analysis and available, it would appear difficult to as a supplement to parametric proce- justify learning (or teaching) techniques dures. If one has reservations about a where the regression function depends particular parametric form, specifica- 1 Department of Economics, University of tion tests against nonparametric alter- Toronto. The author is grateful to Andrey Feuer- natives can provide reassurance. verg;er, Zvi Griliches, James MacKinnon, Angelo From the point of view of the econo- Me ino, John Pencavel and particularly to Frank Wolak for their patient reading and insightful mist/econometrician, such techniques comments. The statistics and econometrics litera- possess additional appeal in that most ture on nonparametric regression is massive and implications of economic theory are inevitably many interesting and important results have been given little or no exposure in this intro- nonparametric. Typically, theoretical duction to the subiect. The author regrets this arguments exclude or include variables, state of affairs and hopes that his lacunae will be they imply monotonicity, concavity, or corrected in complementary and more artful re- views by other writers. The author may be con- homogeneity of various sorts, or they tacted at [email protected]. embody more comDlex structure such as 669 670 Journal of Economic Literature, Vol. XXXVI (June 1998) the implications of the maximization hy- because the bound on the first derivative pothesis. They almost never imply a implies that If(xt)-f(xti1) I < L Ixt - xt- I .) specific functional form (the pure quan- As long as z is not perfectly correlated tity theory of mioney equation being one with x, the ordinary least squares estima- exception). This paper will therefore fo- tor of f3using the differenced data, that is cus some considerable attention on con- strained nonparametric regression esti- X(yt - yt-i) (Zt - Zt-i) = Xzzt)2 (1.2) and Pdiff mation testing. y(Zt - Zt_1)2 In some cases, the researcher may feel comfortable with a particular para- has the approximate sampling distribu- metric form for one portion of the re- tion gression function, but less confident about the shape of another portion. fdiff NNf(Vj PI 1 ) 1.3 Such varying prior beliefs call for com- bining parametric and nonparametric where y2 is the conditional variance of z techniques to yield semiparametric re- given x.2 gression models (these have been stud- We have applied this estimator to ied extensively). Inclusion of the non- data on the costs of distributing elec- parametric component may avoid tricity (Figure 1). Factors which can inconsistent estimation which could re- influence distribution costs include cus- sult from incorrect parameterization. tomer density (greater distance be- 1.1.2 An Appetizer for the Reluctant tween customers increases costs), the Palate. For those who have never con- remaining life of physical plant (newer sidered using a nonparametric regres- assets require less maintenance), and sion technique, we suggest the follow- wage rates. These are among the "z" ing elementary procedure, much of variables which appear parametrically which can be implemented in any stan- in the model. However, the nature of dard econometric package. Suppose you scale economies is unclear-the effect are given data (yi,zi,x1). .. (yT,zT,xT) on the on costs could be constant, declining model y = zf3 +f(x) + e where for simplic- to an asymptote or even U-shaped. ity all variables are assumed to be sca- Indeed, important regulatory decisions lars. The Ct are i.i.d. with mean 0 and (such as merger policy) may involve variance (y2 given (z,x). The x's are determining the minimum efficient drawn from a distribution with support, scale. say, the unit interval. Most important, In the left panel, we estimate the cost the data are rearranged so that function parametrically, incorporating a XI < ... < XT. All that is known about f is quadratic term for the scale variable "x" that its first derivative is bounded by a which we measure using the number of constant, say L. Suppose that we first customers of each utility. In the right difference to obtain panel, we report the coefficient esti- mates of the z variables after applying yt - yt-1 = (Zt - Zt-1)f the differencing estimator (1.2), suit- + (f(Xt) -f(xt-I)) + ?t-t - (1.1) ably generalized to allow for vector z. There is moderate change in coefficients. As sample size increases-packing the unit interval with x's-the typical difference 2 Throughout the paper, the symbol - will de- note that a random variable has the indicated ap- xt - xt-1 shrinks at a rate of about IIT so proximatedistribution and the symbol_ will indi- that f(xti1) tends to cancel f(xt). (This is cate approximateequality. Yatchew: Nonparametric Regression Techniques in Economics 671 Figure 1. Returns to Scale in Electricity Distribution Model: Restructuring,vertical unbundling, and deregulation of electricity industries have led to reexaminationof scale economies in electricity distribution.The efficiency of small (usually municipal) distributorsthat exist in the U.S., Norway, New Zealand, Germany and Canadais at issue. The objective is to assess scale economies. The model is given by y = z i + f(x) + ? where y is variable costs of distributionper customer (COST PER CUST). The vector z includes: customer density (ruraldistribution is more expensive) as measured by average distance between customers (DIST); load per customer (LOAD); local generation per customer (GEN); remaining life of assets (LIFE) - older plants require more maintenance; assets per customer (ASSETS);and a proxy for labour wage rates (WAGE). The scale of operation, x, is measured by the number of customers served (CUST). The scale effect may be nonlinear, (e.g., decreasing to an asymptote or U-shaped), hence the inclusion of a nonparametriccomponentf (x). Data: 265 municipal distributorsin Ontario, Canadawhich vary in size from tiny ones serving 200 customers to the largest serving 218,000 (a 'customer' can be a household, or a commercial, industrialor governmental entity). ESTIMATEDMODELS PARAMETRIC:Quadratic scale effect SEMIPARAMETRIC:Smooth scale effect y= a + Z3 +y1x +y2x2 + ? Y = z 3 +f(x) + ? OLS Differencing Estimator1 Coefficient StandardError Coefficient StandardError oc 15.987 37.002 DIST 3.719 1.248 ADIST 2.568 1.560 LOAD 1.920 .661 ALOAD .437 .912 GEN -.051 .023 AGEN .0005 .032 LIFE -5.663 .798 ALIFE -4.470 1.070 ASSETS .037 .005 AASSETS .030 .0072 WAGE .003 .0007 AWAGE .003 .00086 71 -.00154 .00024 72 .1x10-7 .14x10-8 R2 .45 ESTIMATEDIMPACT OF SCALEON DOLLARCOSTS PER CUSTOMERPER YEAR , 200 Kernel w34 - - - - - - Quadratic 0 E- 150- Q ~~~~~~~~~~0 = 100 o 0 C0 05 DCo ? E C/) 0%00Qx 800 0 80 0 00 0~~~~~~~ 500 1000 5000 0 1000500 100 1 aarodrds htx < .t. XT,.0 othe al-0 vaibe difeecedAw= w0 - , foloe by0 orinr lesqurs Standarderor ~ mutple by XTS as per eqato (13) DeenenA- vaial is COS PE CUST. 500 1000 5000 10000 50000 100000 CUSTOMERS ? ? 1Data reordered so that x, ... XT, then all variables differenced A wt = Wt- Wtl1, followed by ordinaryleast squares. Standarderrors multiplied by Vfi75, as per equation (1.3). Dependent variable is A COST PER CUST. 672 Tournal of Economic Literature. Vol. XXXVI (Tune 1998) Though the "differencing estimator" vations. For the approach described of f3 does not require estimation of f, here, the implicit estimate of f(xt) is one can visually assess whether there is f(xt-i).
Recommended publications
  • Nonparametric Regression 1 Introduction
    Nonparametric Regression 10/36-702 Larry Wasserman 1 Introduction Now we focus on the following problem: Given a sample (X1;Y1);:::,(Xn;Yn), where d Xi 2 R and Yi 2 R, estimate the regression function m(x) = E(Y jX = x) (1) without making parametric assumptions (such as linearity) about the regression function m(x). Estimating m is called nonparametric regression or smoothing. We can write Y = m(X) + where E() = 0. This follows since, = Y − m(X) and E() = E(E(jX)) = E(m(X) − m(X)) = 0 A related problem is nonparametric prediction. Given a pair (X; Y ), we want to predict Y from X. The optimal predictor (under squared error loss) is the regression function m(X). Hence, estimating m is of interest for its own sake and for the purposes of prediction. Example 1 Figure 1 shows data on bone mineral density. The plots show the relative change in bone density over two consecutive visits, for men and women. The smooth estimates of the regression functions suggest that a growth spurt occurs two years earlier for females. In this example, Y is change in bone mineral density and X is age. Example 2 Figure 2 shows an analysis of some diabetes data from Efron, Hastie, Johnstone and Tibshirani (2004). The outcome Y is a measure of disease progression after one year. We consider four covariates (ignoring for now, six other variables): age, bmi (body mass index), and two variables representing blood serum measurements. A nonparametric regression model in this case takes the form Y = m(x1; x2; x3; x4) + .
    [Show full text]
  • Nonparametric Regression with Complex Survey Data
    CHAPTER 11 NONPARAMETRIC REGRESSION WITH COMPLEX SURVEY DATA R. L. Chambers Department of Social Statistics University of Southampton A.H. Dorfman Office of Survey Methods Research Bureau of Labor Statistics M.Yu. Sverchkov Department of Statistics Hebrew University of Jerusalem June 2002 11.1 Introduction The problem considered here is one familiar to analysts carrying out exploratory data analysis (EDA) of data obtained via a complex sample survey design. How does one adjust for the effects, if any, induced by the method of sampling used in the survey when applying EDA methods to these data? In particular, are adjustments to standard EDA methods necessary when the analyst's objective is identification of "interesting" population (rather than sample) structures? A variety of methods for adjusting for complex sample design when carrying out parametric inference have been suggested. See, for example, SHS, Pfeffermann (1993) and Breckling et al (1994). However, comparatively little work has been done to date on extending these ideas to EDA, where a parametric formulation of the problem is typically inappropriate. We focus on a popular EDA technique, nonparametric regression or scatterplot smoothing. The literature contains a limited number of applications of this type of analysis to survey data, usually based on some form of sample weighting. The design-based theory set out in chapter 10, with associated references, provides an introduction to this work. See also Chesher (1997). The approach taken here is somewhat different. In particular, it is model-based, building on the sample distribution concept discussed in section 2.3. Here we develop this idea further, using it to motivate a number of methods for adjusting for the effect of a complex sample design when estimating a population regression function.
    [Show full text]
  • Nonparametric Bayesian Inference
    Nonparametric Bayesian Methods 1 What is Nonparametric Bayes? In parametric Bayesian inference we have a model M = ff(yjθ): θ 2 Θg and data Y1;:::;Yn ∼ f(yjθ). We put a prior distribution π(θ) on the parameter θ and compute the posterior distribution using Bayes' rule: L (θ)π(θ) π(θjY ) = n (1) m(Y ) Q where Y = (Y1;:::;Yn), Ln(θ) = i f(Yijθ) is the likelihood function and n Z Z Y m(y) = m(y1; : : : ; yn) = f(y1; : : : ; ynjθ)π(θ)dθ = f(yijθ)π(θ)dθ i=1 is the marginal distribution for the data induced by the prior and the model. We call m the induced marginal. The model may be summarized as: θ ∼ π Y1;:::;Ynjθ ∼ f(yjθ): We use the posterior to compute a point estimator such as the posterior mean of θ. We can also summarize the posterior by drawing a large sample θ1; : : : ; θN from the posterior π(θjY ) and the plotting the samples. In nonparametric Bayesian inference, we replace the finite dimensional model ff(yjθ): θ 2 Θg with an infinite dimensional model such as Z F = f : (f 00(y))2dy < 1 (2) Typically, neither the prior nor the posterior have a density function with respect to a dominating measure. But the posterior is still well defined. On the other hand, if there is a dominating measure for a set of densities F then the posterior can be found by Bayes theorem: R A Ln(f)dπ(f) πn(A) ≡ P(f 2 AjY ) = R (3) F Ln(f)dπ(f) Q where A ⊂ F, Ln(f) = i f(Yi) is the likelihood function and π is a prior on F.
    [Show full text]
  • Nonparametric Regression and the Bootstrap Yen-Chi Chen December 5, 2016
    Nonparametric Regression and the Bootstrap Yen-Chi Chen December 5, 2016 Nonparametric regression We first generate a dataset using the following code: set.seed(1) X <- runif(500) Y <- sin(X*2*pi)+rnorm(500,sd=0.2) plot(X,Y) 1.0 0.5 0.0 Y −0.5 −1.5 0.0 0.2 0.4 0.6 0.8 1.0 X This is a sine shape regression dataset. Simple linear regression We first fit a linear regression to this dataset: fit <- lm(Y~X) summary(fit) ## ## Call: ## lm(formula = Y ~ X) ## ## Residuals: ## Min 1Q Median 3Q Max ## -1.38395 -0.35908 0.04589 0.38541 1.05982 ## 1 ## Coefficients: ## Estimate Std. Error t value Pr(>|t|) ## (Intercept) 1.03681 0.04306 24.08 <2e-16 *** ## X -1.98962 0.07545 -26.37 <2e-16 *** ## --- ## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 '' 1 ## ## Residual standard error: 0.4775 on 498 degrees of freedom ## Multiple R-squared: 0.5827, Adjusted R-squared: 0.5819 ## F-statistic: 695.4 on 1 and 498 DF, p-value: < 2.2e-16 We observe a negative slope and here is the scatter plot with the regression line: plot(X,Y, pch=20, cex=0.7, col="black") abline(fit, lwd=3, col="red") 1.0 0.5 0.0 Y −0.5 −1.5 0.0 0.2 0.4 0.6 0.8 1.0 X After fitted a regression, we need to examine the residual plot to see if there is any patetrn left in the fit: plot(X, fit$residuals) 2 1.0 0.5 0.0 fit$residuals −0.5 −1.0 0.0 0.2 0.4 0.6 0.8 1.0 X We see a clear pattern in the residual plot–this suggests that there is a nonlinear relationship between X and Y.
    [Show full text]
  • Nonparametric Regression
    Nonparametric Regression Statistical Machine Learning, Spring 2015 Ryan Tibshirani (with Larry Wasserman) 1 Introduction, and k-nearest-neighbors 1.1 Basic setup, random inputs Given a random pair (X; Y ) Rd R, recall that the function • 2 × f0(x) = E(Y X = x) j is called the regression function (of Y on X). The basic goal in nonparametric regression is ^ d to construct an estimate f of f0, from i.i.d. samples (x1; y1);::: (xn; yn) R R that have the same joint distribution as (X; Y ). We often call X the input, predictor,2 feature,× etc., and Y the output, outcome, response, etc. Importantly, in nonparametric regression we do not assume a certain parametric form for f0 Note for i.i.d. samples (x ; y );::: (x ; y ), we can always write • 1 1 n n yi = f0(xi) + i; i = 1; : : : n; where 1; : : : n are i.i.d. random errors, with mean zero. Therefore we can think about the sampling distribution as follows: (x1; 1);::: (xn; n) are i.i.d. draws from some common joint distribution, where E(i) = 0, and then y1; : : : yn are generated from the above model In addition, we will assume that each i is independent of xi. As discussed before, this is • actually quite a strong assumption, and you should think about it skeptically. We make this assumption for the sake of simplicity, and it should be noted that a good portion of theory that we cover (or at least, similar theory) also holds without the assumption of independence between the errors and the inputs 1.2 Basic setup, fixed inputs Another common setup in nonparametric regression is to directly assume a model • yi = f0(xi) + i; i = 1; : : : n; where now x1; : : : xn are fixed inputs, and 1; : : : are still i.i.d.
    [Show full text]
  • Significance Redux
    The Journal of Socio-Economics 33 (2004) 665–675 Significance redux Stephen T. Ziliaka,∗, Deirdre N. McCloskeyb,1 a School of Policy Studies, College of Arts and Sciences, Roosevelt University, Chicago, 430 S. Michigan Avenue, Chicago, IL 60605, USA b College of Arts and Science (m/c 198), University of Illinois at Chicago, 601 South Morgan, Chicago, IL 60607-7104, USA Abstract Leading theorists and econometricians agree with our two main points: first, that economic sig- nificance usually has nothing to do with statistical significance and, second, that a supermajority of economists do not explore economic significance in their research. The agreement from Arrow to Zellner on the two main points should by itself change research practice. This paper replies to our critics, showing again that economic significance is what science and citizens want and need. © 2004 Elsevier Inc. All rights reserved. JEL CODES: C12; C10; B23; A20 Keywords: Hypothesis testing; Statistical significance; Economic significance Science depends on entrepreneurship, and we thank Morris Altman for his. The sympo- sium he has sparked can be important for the future of economics, in showing that the best economists and econometricians seek, after all, economic significance. Kenneth Arrow as early as 1959 dismissed mechanical tests of statistical significance in his search for eco- nomic significance. It took courage: Arrow’s teacher, the amazing Harold Hotelling, had been one of Fisher’s sharpest disciples. Now Arrow is joined by Clive Granger, Graham ∗ Corresponding author. E-mail addresses: [email protected] (S.T. Ziliak), [email protected] (D.N. McCloskey). 1 For comments early and late we thank Ted Anderson, Danny Boston, Robert Chirinko, Ronald Coase, Stephen Cullenberg, Marc Gaudry, John Harvey, David Hendry, Stefan Hersh, Robert Higgs, Jack Hirshleifer, Daniel Klein, June Lapidus, David Ruccio, Jeff Simonoff, Gary Solon, John Smutniak, Diana Strassman, Andrew Trigg, and Jim Ziliak.
    [Show full text]
  • Stolper-Samuelson Time Series: Long Term US Wage Adjustment
    Stolper-Samuelson Time Series: Long Term US Wage Adjustment Henry Thompson Auburn University Review of Development Economics, 2011 Factor prices adjust to product prices in the Stolper-Samuelson theorem of the factor proportions model. The present paper estimates adjustment in the average US wage to changes in prices of manufactures and services with yearly data from 1949 to 2006. Difference equation analysis is based directly on the comparative static factor proportions model. Factors of production are fixed capital assets and the labor force. Results have implications for theory and policy. Thanks for comments and suggestions go to Farhad Rassekh, Charles Sawyer, Roy Ruffin, Ron Jones, Manoj Atolia, Santanu Chaterjee, and Svetlana Demidova. Contact: Henry Thompson, Economics, Auburn University, AL 36849, 334-844-2910, fax 5999, [email protected] 1 Stolper-Samuelson Time Series: Long Term US Wage Adjustment The Stolper-Samuelson (SS, 1941) theorem concerns the effects of changing product prices on factor prices along the contract curve in the general equilibrium model of production with two factors and two products. The result is fundamental to neoclassical economics as relative product prices evolve with economic growth. The theoretical literature finding exception to the SS theorem is vast as summarized by Thompson (2003) and expanded by Beladi and Batra (2004). Davis and Mishra (2007) believe the theorem is dead due unrealistic assumptions. The scientific status of the theorem, however, depends on the empirical evidence. The empirical literature generally examines indirect evidence including trade volumes, trade openness, input ratios, relative production wages, and per capita incomes as summarized by Deardorff (1984), Leamer (1994), and Baldwin (2008).
    [Show full text]
  • User Pays Applied to Pro-Life Advocates
    A Service of Leibniz-Informationszentrum econstor Wirtschaft Leibniz Information Centre Make Your Publications Visible. zbw for Economics Pope, Robin Working Paper The Big Picture of Lives Saved by Abortion on Demand: User Pays Applied to Pro-Life Advocates Bonn Econ Discussion Papers, No. 14/2009 Provided in Cooperation with: Bonn Graduate School of Economics (BGSE), University of Bonn Suggested Citation: Pope, Robin (2009) : The Big Picture of Lives Saved by Abortion on Demand: User Pays Applied to Pro-Life Advocates, Bonn Econ Discussion Papers, No. 14/2009, University of Bonn, Bonn Graduate School of Economics (BGSE), Bonn This Version is available at: http://hdl.handle.net/10419/37046 Standard-Nutzungsbedingungen: Terms of use: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise use the documents in public. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open gelten abweichend von diesen Nutzungsbedingungen
    [Show full text]
  • Testing for No Effect in Nonparametric Regression Via Spline Smoothing Techniques
    Ann. Inst. Statist. Math. Vol. 46, No. 2, 251-265 (1994) TESTING FOR NO EFFECT IN NONPARAMETRIC REGRESSION VIA SPLINE SMOOTHING TECHNIQUES JUEI-CHAO CHEN Graduate Institute of Statistics and Actuarial Science, Feng Chic University, Taiehung ~072~, Taiwan, ROC (Received June 8, 1992; revised August 2, 1993) Abstract. We propose three statistics for testing that a predictor variable has no effect on the response variable in regression analysis. The test statistics are integrals of squared derivatives of various orders of a periodic smoothing spline fit to the data. The large sample properties of the test statistics are investigated under the null hypothesis and sequences of local alternatives and a Monte Carlo study is conducted to assess finite sample power properties. Key words and phrases: Asymptotic distribution, local alternatives, nonpara- metric regression, Monte Carlo. 1. Introduction Regression analysis is used to study relationships between the response vari- able and a predictor variable; therefore, all of the inference is based on the as- sumption that the response variable actually depends on the predictor variable. Testing for no effect is the same as checking this assumption. In this paper, we propose three statistics for testing that a predictor variable has no effect on the response variable and derive their asymptotic distributions. Suppose we have an experiment which yields observations (tin,y1),..., (tn~, Yn), where yj represents the value of a response variable y at equally spaced values tjn = (j - 1)/n, j = 1,...,n, of the predictor variable t. The regression model is (1.1) yj = #(tin) + ey, j --- 1,...,n, where the ej are iid random errors with E[el] = 0, Var[r ~2 and 0 K E[e 41 K oe.
    [Show full text]
  • The Causal Direction Between Money and Prices
    Journal of Monetary Economics 27 (1991) 381-423. North-Holland The causal direction between money and prices An alternative approach* Kevin D. Hoover University of California, Davis, Davis, CA 94916, USA Received December 1988, final version received March 1991 Causality is viewed as a matter of control. Controllability is captured in Simon’s analysis of causality as an asymmetrical relation of recursion between variables in the unobservable data-generating process. Tests of the stability of marginal and conditional distributions for these variables can provide evidence of causal ordering. The causal direction between prices and money in the United States 1950-1985 is assessed. The balance of evidence supports the view that money does not cause prices, and that prices do cause money. 1. Introduction The debate in economics over the causal direction between money and prices is an old one. Although in modern times it had seemed largely settled in favor of the view that money causes prices, the recent interest in real *This paper is a revision of part of my earlier working paper, ‘The Logic of Causal Inference: With an Application to Money and Prices’, No. 55 of the series of working papers in the Research Program in Applied Macroeconomics and Macro Policy, Institute of Governmental Affairs, University of California. Davis. I am grateful to Peter Oownheimer. Peter Sinclair, Charles Goodhart, Thomas Mayer, Edward Garner, Thomas C&iey, Stephen LeRoy, Leon Wegge, Nancy Wulwick, Diran Bodenhom, Clive Granger, Paul Holland, the editors of this journal, an anonymous referee, and the participants in macroeconomics seminars at the Univer- sity of California, Berkeley, Fall 1986, the University of California, Davis, Fall 1988, and the University of California, Irvine, Spring 1989, for comments on this paper in its various previous incamatio;ts.
    [Show full text]
  • Chapter 5 Nonparametric Regression
    Chapter 5 Nonparametric Regression 1 Contents 5 Nonparametric Regression 1 1 Introduction . 3 1.1 Example: Prestige data . 3 1.2 Smoothers . 4 2 Local polynomial regression: Lowess . 4 2.1 Window-span . 6 2.2 Inference . 7 3 Kernel smoothing . 9 4 Splines . 10 4.1 Number and position of knots . 11 4.2 Smoothing splines . 12 5 Penalized splines . 15 2 1 Introduction A linear model is desirable because it is simple to fit, it is easy to understand, and there are many techniques for testing the assumptions. However, in many cases, data are not linearly related, therefore, we should not use linear regression. The traditional nonlinear regression model fits the model y = f(Xβ) + 0 where β = (β1, . βp) is a vector of parameters to be estimated and X is the matrix of predictor variables. The function f(.), relating the average value of the response y on the predictors, is specified in advance, as it is in a linear regression model. But, in some situa- tions, the structure of the data is so complicated, that it is very difficult to find a functions that estimates the relationship correctly. A solution is: Nonparametric regression. The general nonparametric regression model is written in similar manner, but f is left unspecified: y = f(X) + = f(x1,... xp) + Most nonparametric regression methods assume that f(.) is a smooth, continuous function, 2 and that i ∼ NID(0, σ ). An important case of the general nonparametric model is the nonparametric simple regres- sion, where we only have one predictor y = f(x) + Nonparametric simple regression is often called scatterplot smoothing, because an important application is to tracing a smooth curve through a scatterplot of y against x and display the underlying structure of the data.
    [Show full text]
  • Professor Robert F. Engle," Econometric Theory, 19, 1159-1193
    Diebold, F.X. (2003), "The ET Interview: Professor Robert F. Engle," Econometric Theory, 19, 1159-1193. The ET Interview: Professor Robert F. Engle Francis X. Diebold1 University of Pennsylvania and NBER January 2003 In the past thirty-five years, time-series econometrics developed from infancy to relative maturity. A large part of that development is due to Robert F. Engle, whose work is distinguished by exceptional creativity in the empirical modeling of dynamic economic and financial phenomena. Engle’s footsteps range widely, from early work on band-spectral regression, testing, and exogeneity, through more recent work on cointegration, ARCH models, and ultra-high-frequency financial asset return dynamics. The booming field of financial econometrics, which did not exist twenty-five years ago, is built in large part on the volatility models pioneered by Engle, and their many variations and extensions, which have found widespread application in financial risk management, asset pricing and asset allocation. (We began in fall 1998 at Spruce in Chicago, the night before the annual NBER/NSF Time Series Seminar, continued in fall 2000 at Tabla in New York, continued again in summer 2001 at the Conference on Market Microstructure and High-Frequency Data in Finance, Sandbjerg Estate, Denmark, and wrapped up by telephone in January 2003.) I. Cornell: From Physics to Economics FXD: Let’s go back to your graduate student days. I recall that you were a student at Cornell. Could you tell us a bit about that? RFE: It depends on how far back you want to go. When I went to Cornell I went as a physicist.
    [Show full text]