Macroeconometrics

Econometric models, large and small, have played an increasingly important role in macroeconomic forecasting and policy analysis. However, there is a wide range of model types used for this purpose, including simultaneous-equation models in either reduced or structural form, vector autoregressive models (VAR), autoregressive distributed-lag models, autoregressive integrated moving-average models, leading-indicator models, and error-correction models (ECM). Hendry, Pagan, and Sargan (1984) discuss a typology for dynamic single-equation models for time-series variables, and Hendry (1994) presents a typology for the various types of dynamic model used in the analysis of systems of equations. There is also a wide range of views about the appropriate way to develop and evaluate models. Sims (1980, 1992) advocates the use of VAR models, which can accurately represent the time-series properties of data, while eschewing the reliance on “incredible dentifying restrictions” that characterizes the use of simultaneous equation models of the structural or Cowles Commission type. The potential value of structure (loosely defined) within the context of VAR models has led to the development of structural VAR models, and Canova (1995) provides a recent review of this literature. Leamer (1978, 1983), on the other hand, has been critical of the use of non-Bayesian models that do not analyze formally the role and value of a priori information, especially when there is no checking of model sensitivity. Summers (1991), though aware of the important developments made in theoretical statistics and econometrics in this century, argues that too much emphasis is placed on the technical aspects of modeling and not enough on the real issues that are concerned with the analysis of well-established and fundamental relationships between economic variables. One approach to modeling that does not overemphasize the role of model evaluation and statistical technique is that associated with real business cycle analysis and the calibration of economic theory, rather than its evaluation. Kydland and Prescott (1982, 1995) have been pioneers in this field, and Canova, Finn, and Pagan (1994) provide a critique.

Lucas (1976, p. 41) proposes an explanation for why coefficients in econometric equations might be nonconstant when policy rules change: “[G]iven that the structure of an econometric model consists of optimal decision rules of economic agents, and that optimal decision rules vary systematically with changes in the structure of series relevant to the decision maker, it follows that any change in policy will systematically alter the structure of econometric models.” Lucas’s critique of econometric models focuses on how parameters in policy rules may enter parametrically into economic agents’ optimization rules. Lucas (1976) considers examples where agents’expectationsof policy behavior enter into their optimization problem, and so parameters relating to policymakers’ rules appear in the agents’ first-order conditions. In essence, the issue is whether an econometric model isolates “invariants” of the economic process. Such invariance or autonomy is a topic with a lengthy and contentious history in the econometrics literature: see Haavelmo (1944) and Frisch (1948) for early discussions and Aldrich (1989) for an extensive historical perspective.