This project develops econometric methods that do not rely on the correct specification of the structure of economic models for their validity. This is an extremely important line of research because these methods permit empirical research to concentrate on drawing economic inferences from data without risking potentially large specification errors. The basic idea is to replace the structural model with an approximation that sequentially improves as more information becomes available. The methods developed by this project permit reliable inference at each intermediate stage of model evolution. The procedures are general enough to encompass most econometric inference procedures. Under a previous grant the investigator succeeded in developing statistical methods that have certain advantages particularly relevant to econometric applications: simplicity, ease of implementation, and ease of extension to nonlinear, multivariate, and time series applications. Under this grant these methods will be extended to a dynamic setting and used to analyze financial markets. As in the past, algorithms implementing the theoretical and empirical work will be coded and documented to current standards for professionally written, scientific software and put in the public domain. The methods developed under this project are termed seminonparametric (SNP) methods because they are parametric yet have nonparametric properties. The procedures replace the structure models with a truncated series expansion, the error density with a truncated expansion, or both. By letting the truncation grow adaptively with sample size, the approximation is accurate enough at each intermediate stage to permit reliable inference and ultimate convergence to the underlying data generating mechanism. Applications are made to conditionally heterogeneous time series such as occur in finance and macroeconomics. Bayesian methods are also studied.
