Inference in Models With Weakly-Identified Parameters or Nearly-Integrated Variables

Harvard University, Cambridge MA

Investigators

Abstract

There are many situations, such as when parameters are weakly identified or regressors are nearly integrated, where standard asymptotic theory provides poor approximations in econometric analysis. Obtaining reliable econometric methods to deal with these increasingly important situations is a high priority need for the economics profession. This research will develop methods to conduct inference in linear models with weakly identified parameters or nearly integrated series. The research has many potential impacts on academic fields and on policy making and implementation. For example, educational reformers need accurate statistical methods to judge alternative schooling policies; financial institutions want to predict stock returns accurately; and those who decide on monetary policy would like to assess structural changes in macroeconomic variables. In practice, policy decisions are often based on inferences that use certain asymptotic approximations. However, these approximations are not satisfactory in many cases because they impose restrictive assumptions that may not hold in practice. This research will improve upon standard approximations methods using three related projects---weak identification, highly persistent variables, and structural breaks. Each project seeks to provide econometric procedures useful in different areas. The first one, on weak identification, is particularly useful to applied researchers in labor economics and macroeconomics who use non-experimental data. The research on highly persistent variables can help financial economists to predict changes in the stock market, while the third project on structural breaks is useful in determining changes in time series data, which is prevalent in macroeconomics. Besides its contribution to econometric theory, these methods could help improve policy decision-making.