9th Symposium on Finance, Banking, and Insurance
Universität Karlsruhe (TH), Germany, December 11 - 13, 2002

Abstract

Econometric Model Specification Based on Theories of Financial Intermarket Links to Predict Turning Points

Thorsten Poddig and Claus Huber

Universität Bremen

When developing a model to forecast financial markets, a crucial question is which explaining variables to include. In this paper, we base variable selection on the theoretical approaches of Pring (The All-Season Investor, 1992) and Murphy (Intermarket Technical Analysis, 1991). Their general idea is that financial markets are mutually dependent: Murphy assumes international interrelationships, whereas Pring focuses on the domestic level. We specify econometric models in the spirit of either Pring or Murphy and compare their power to predict turning points with simple technical models (=ARMA-models) and a naive benchmark with respect to economic performance. To predict turning points, we use financial time series of monthly periodicity and implement a Monte-Carlo-based regression approach introduced by Wecker (1979) and enhanced by Kling (1987). An important feature of this method is to explicitly incorporate the uncertainty inherent in any kind of forecast in probabilistic statements for turning points. We judge the performance of the models by an out-of-sample backtesting simulation. To incorporate the possibility of structural change within the backtesting period we conduct rolling regressions and fit a multitude of models to the data. The best model is selected by an out - of-sample selection procedure. Hence this kind of model development is purely data driven, it can be regarded as data mining. We find that the international models disappoint, whereas the simpler domestic and ARMA-models seem to be valuable forecasting tools.