A0792
Title: Prediction of non-negative variables under misspecification and nonstationarity
Authors: Genaro Sucarrat - BI Norwegian Business School (Norway) [presenting]
Christian Francq - CREST and University Lille III (France)
Abstract: In empirical practice, it is commonly assumed that the entertained model is equal to the conditional expectation. While this simplifies theory and interpretation, it is unlikely to hold in reality. The purpose is to consider a broad class of predictive specifications for non-negative variables (e.g., volatility, spreads, volume, and unemployment). The predictive specifications need not equal the (unknown) conditional expectation, and the non-negative variables can be nonstationary. Examples of predictive specifications in the class include ARMA specifications and nonstationary seasonality predictors (and combinations thereof). Consistent and asymptotically normal estimation of the parameters of the predictive specifications is established for the quasi maximum likelihood estimator (QMLE). The finite sample properties of the estimator is studied by simulation, and an empirical application illustrates the results.
