STRUCTURAL ECONOMETRIC MODELING: …

1 Chapter 64 STRUCTURAL ECONOMETRIC MODELING: RATIONALES AND EXAMPLES FROM INDUSTRIALORGANIZATIONPETER C. REISSG raduate School of Business, Stanford University, Stanford, CA 94305-5015, A. WOLAKD epartment of Economics, Stanford University, Stanford, CA 94305-6072, Introduction42812. STRUCTURAL models defined42823. Constructing STRUCTURAL Sources of Why add structure? Evaluating structure single equation Evaluating structure simultaneous equation The role of nonexperimental data in STRUCTURAL modeling43014. A framework for STRUCTURAL ECONOMETRIC models in The economic The stochastic Unobserved heterogeneity and agent Optimization Measurement Steps to STRUCTURAL model epilogue43145. Demand and cost function estimation under imperfect Using price and quantity data to diagnose The economic Environment and primitives4317 Handbook of econometrics , Volume 6 ACopyright 2007 Elsevier All rights reservedDOI: (07) Reiss and Behavior and The stochastic Summary43246.

2 Market power models more Estimating price cost Identifying and interpreting price cost Summary43337. Models of differentiated product Neoclassical demand Micro-data A household-level demand Goldberg s economic The stochastic A product-level demand The economic model in The stochastic More on the ECONOMETRIC Functional form assumptions for Distribution of consumer Unobserved product quality Cost function Summary43608. Games with incomplete information: Auctions Descriptive STRUCTURAL Nonparametric identification and Further Parametric specifications for auction market Why estimate a STRUCTURAL auction model? Extensions of basic auctions models43819. Games with incomplete information: Principal-agent contracting Observables and Economic models of regulator utility Estimating productions functions accounting for private Symmetric information Asymmetric information ECONOMETRIC Estimation Further extensions439810.

3 Market structure and firm Overview of the issues4399Ch. 64: STRUCTURAL ECONOMETRIC Airline competition and An economic model and modeling profits and The ECONOMETRIC Epilogue441011. Ending Reiss and WolakAbstractThis chapter explains the logic of STRUCTURAL ECONOMETRIC models and compares themto other types of ECONOMETRIC models. We provide a framework researchers can useto develop and evaluate STRUCTURAL ECONOMETRIC models. This framework pays particu-lar attention to describing different sources of unobservables in STRUCTURAL models. Weuse our framework to evaluate several literatures in industrial organization economics,including the literatures dealing with market power, product differentiation, auctions,regulation and ECONOMETRIC model, market power, auctions, regulation, entryJEL classification: C50, C51, C52, D10, D20, D40Ch.

4 64: STRUCTURAL ECONOMETRIC Modeling42811. IntroductionThe founding members of the Cowles Commission definedeconometricsas: a branchof economics in which economic theory and statistical method are fused in the analysisof numerical and institutional data [Hood and Koopmans (1953, p. xv)]. Today econo-mists refer to models that combine explicit economic theories with statistical models asstructural ECONOMETRIC chapter has three main goals. The first is to explain the logic of STRUCTURAL econo-metric modeling . While STRUCTURAL ECONOMETRIC models have the logical advantage ofdetailing the economic and statistical assumptions required to estimate economic quan-tities, the fact that they impose structure does not automatically make them sensible. Tobe convincing, STRUCTURAL models minimally must be: (1) flexible statistical descriptionsof data; (2) respectful of the economic institutions under consideration; and, (3) sen-sitive to the nonexperimental nature of economic data.

5 When, for example, there islittle economic theory on which to build, the empiricist may instead prefer to use non- STRUCTURAL or descriptive ECONOMETRIC models. Alternatively, if there is a large body ofrelevant economic theory, then there may significant benefits to estimating a structuraleconometric model provided the model can satisfy the above second goal of this chapter is to describe the ingredients of STRUCTURAL models andhow STRUCTURAL modelers go about evaluating them. Our discussion emphasizes that theprocess of building a STRUCTURAL model involves a series of related steps. These steps areby no means formulaic and often involve economic, statistical and practical compro-mises. Understanding when and why STRUCTURAL modelers must make compromises, andthat STRUCTURAL modelers can disagree on compromises, is important for understandingthat STRUCTURAL modeling is in part art.

6 For example, STRUCTURAL modelers often intro-duce conditioning variables that are not explicitly part of the economic theory as away of controlling for plausible differences across third goal is to illustrate how STRUCTURAL modeling tradeoffs are made in , we examine different types of STRUCTURAL ECONOMETRIC models developed byindustrial organization ( IO ) economists. These models examine such issues as: theextent of market power possessed by firms; the efficiency of alternative market alloca-tion mechanisms ( , different rules for running single and multi-unit auctions); andthe empirical implications of information and game-theoretic models. We should em-phasize that this chapter is NOT a comprehensive survey of the IO literature or evena complete discussion of any single topic. Readers interested in a comprehensive re-view of a particular literature should instead begin with the surveys we cite.

7 Our goal isinstead to illustrate selectively how IO researchers have used economic and statisticalassumptions to identify and estimate economic magnitudes. Our hope is that in doing so,we can provide a better sense of the benefits and limitations of STRUCTURAL begin by defining STRUCTURAL ECONOMETRIC models and discussing when one wouldwant to use a STRUCTURAL model. As part of this discussion, we provide a Reiss and Wolakfor evaluating the benefits and limitations of STRUCTURAL models. The remainder of thechapter illustrates some of the practical tradeoffs IO researchers have STRUCTURAL models definedIn STRUCTURAL ECONOMETRIC models, economic theory is used to develop mathematicalstatements about how a set of observable endogenous variables,y, are related to an-other set of observable explanatory variables,x.

8 Economic theory also may relate theyvariables to a set of unobservable variables, . These theoretical relations usually arein the form of equalities:y=g(x, , ), whereg( )is a known function and asetof unknown parameters or functions. Occasionally, economic theory may only deliverinequality relations, such asy g(x, , ).Economic theory alone usually does not provide enough information for theeconometrician to estimate . For this reason, and because the economic modely=g(x, , )may not be able to rationalize the observed data perfectly, the econo-metrician adds statistical assumptions about the joint distribution ofx, and otherunobservables appended to the model. Taken together, these economic and statisti-cal assumptions define an empirical model that is capable of rationalizing all possibleobservable outcomes.

9 In order to estimate the underlying primitives of this model, re-searchers use statistical objects based on the model, such as a log-likelihood functionfor the data, (y, x| ), or conditional moments, such asE(y|x, ).Nonstructural empirical work in economics may or may not be based on formalstatistical models. At one end of the spectrum are measurement studies that focus onconstructing and summarizing data, such as labor force participation and unemploymentrates. At the other end are those that use formal statistical models, such as autoregres-sive conditional volatility models. Both types of nonstructural empirical work have along and respected tradition in economics. An excellent early example isEngel s (1857)work relating commodity budget shares to total income. Engel s finding that expenditureshares for food were negatively related to the logarithm of total household expendi-tures has shaped subsequent theoretical and empirical work on household consumptionbehavior [seeDeaton and Muellbauer (1980)and Pollak and Wales (1992)].

10 A some-what more recent example of descriptive work is the Phillips (1958)documented an inverse relationship between United Kingdom unemployment rates andchanges in wage rates. This work inspired others to document relationships betweenunemployment rates and changes in prices. In the ensuing years, many economic theo-ries have been advanced to explain why Phillips curves are or are not stable empirical models usually are grounded in economics to the extent thateconomics helps identify which variables belong inyand which inx. This approach,however, ultimately estimates characteristics of the joint population density ofxandy,f (x, y), or objects that can be derived from it, such as:Ch. 64: STRUCTURAL ECONOMETRIC Modeling4283f(y|x), the conditional density ofygivenx;E(y|x), the conditional expectation ofygivenx;Cov(y|x), the conditional covariances (or correlations) ofygivenx;or,Q (y|x)the conditional quantile most commonly estimated characteristic of the joint density is the best linear pre-dictor (BLP(y|x)) recently researchers have taken advantage of developments in nonparamet-ric and semiparametric statistical methods to derive consistent estimates of the jointdensity ofyandx.