SBR-9422873 Charles Whiteman and Beth Ingram In conducting empirical research, guidance from theory is extremely valuable. Yet in macroeconomics, as theoretical models have become increasingly complex and stylized, such guidance has become difficult to incorporate in a formal statistical sense. This project develops and then implements a coherent statistical framework for combining theoretical and empirical information. The framework is Bayesian and enables the formal yet probabilistic incorporation of theoretical restrictions (e.g., cross-equation restrictions) in working with reduced-form models. A distinctive feature of the approach proposed here is that theoretical models are viewed as sources of prior information which can then be formally combined with data information via Bayes' Rule. This is particularly useful in working with reduced-form models such as vector autoregressions (VARs), where researchers frequently have strong views about the parameterization of the theoretical model but only weak views about the parameterization of the VAR. The new framework also enables economists to incorporate prior information about parameters values in the estimation of structural models. This stands in marked contrast with the current practice of incorporating such information only informally and implicitly. The result of this practice is inference without a firm statistical basis. The methodology is then used in a variety of empirical projects: an analysis of the source of shocks which have driven the U.S. post-war business cycle; an analysis of competing sources of economic growth; construction and estimation of a fully specified nominal business cycle model (which includes monetary and fiscal policy variables) for the purpose of conducting policy analysis; an analysis of the empirical viability of simple heterogeneous agent models with incomplete markets; and an analysis of the usefulness of theoretical asset pricing models in the modeling of financial time series.
|Effective start/end date
|3/15/95 → 5/31/99
- National Science Foundation: $238,748.00