Is US Real GNP Chaotic? On Using the BDS test to Decide Whether an ARMA Model forthe US GNP Genreates I.I.D. Residuals

In this paper we use the BDS test developed by Brock-Dechert-Scheinkman (1987) to investigate whether ARMA Models for the US real GNP generate i.i.d. residuals. The second step,after reviewing some results from Brock-Sayer (1988) and Scheinkman-LeBaron (1989), SL, we will use a different kind of specifications for the US real GNP such as a model with different volatility pre and post World War II as in SL (1989), and a threshold autoregressive specification as in Potter (1990). The last point consists of analysing a modified threshold model that takes into account the observation made by SL (1989) and next we evaluate the forecast performance of the "best" models among those examined.