The investigator studies four distinct topics, with the first somewhat larger in scope than the second and third, which in turn are somewhat larger in scope than the fourth. The first topic is methodological, concerning statistical bias in certain commonly studied economic relations. "Bias" is a precise statistical term that refers to a tendency to systematically estimate a value that is wrong, on average. In economic data, biased estimation leads to forecasts and economic advice that are misleading, and possibly very misleading if the bias is large. Examples of the relevant economic relations include those between stock returns and dividend yields, exchange rate changes and forward premiums, and interest rates and inflation. Estimates of these important relations are used in forecasting economic activity, making financial decisions and deciding economic policy.
Disparate pieces of previous evidence have suggested bias in estimates of these relations produced using a technique called regression. The investigator develops a general, unifying framework that allows systematic study, interpretation and, if necessary, correction for bias in such regressions. The advantage of the general framework is that it lets researchers understand whether biases are likely to be large or small under a given scenario. The results can be used to interpret estimation and inference in the relevant regressions, to improve estimation and inference, and to understand simulation results from economic models. Importantly, the approach allows for complications that are common in economic data; econometric terms for these complications are serial correlation and conditional heteroskedasticity. The investigator applies the results to regressions of exchange rate returns on the forward premium.
The second topic that the investigator studies, is related to how monetary policy affects floating exchange rates between countries with roughly similar inflation rates. This research is empirical, and relies on a technique called vector autoregressions. Standard macroeconomic models focus on a direct link from monetary policy, measured as the difference between U.S. and foreign interest rates, to the value of the U.S. dollar. These models treat any other factors that influence exchange rates as evolving exogenously to monetary policy. The investigator proposes and develops an approach that allows the researcher to evaluate whether surprise movements in monetary policy in fact affect exchange rates solely through interest rates or also via an impact on excess returns or risk premia. This research improves our understanding of monetary policy.
The third topic that this project investigates is how to compare the accuracy of forecasts when there are complicated relationships among the forecasting models being evaluated. This research is methodological. In economic research, some forecast competitions compare a simple model and complicated models, with the simple model a special case of the more complicated models (called "nested" model comparisons). Other forecast competitions evaluate complicated models where no model is a special case of other models ("nonnested"). Finally, there are some forecast competitions that simultaneously involve both types of comparisons (nested and nonnested). That part of the project develops and evaluates a technique that can be used to evaluate accuracy when the comparison simultaneously involves nested and nonnested models. This research helps economists and policy makers better identify models that lead to lasting improvements in prediction. The methodological technique is applied to U.S. inflation and output growth.
The fourth topic of this study concerns the production of forecasts of a set of commodity prices. This research is empirical. Most forecasts of prices of commodities such as oil, tin or rubber focus on a single commodity. The investigator considers how movements in the price of one commodity help forecast movements in the price of another commodity.