Do Extreme Falls Help Forecasting Stock Returns? International Evidence
CUBS Faculty of Finance Working Paper No. 06
18 Pages Posted: 22 Oct 2001
Date Written: October 2001
We report international evidence for the presence of stock return rebounds following extreme falls in market indices. The data consists of weekly national index returns for 21 world markets. A non-linear time series model is used to capture part of the variation in return autocorrelations across countries and over time. A third order polynomial model PAR(3,1) on lagged returns, coupled with GARCH residuals, is capable of generating a time varying auto-correlation structure. In all of the national markets, the return forecasts from the PAR(3,1) models are above those from linear alternative s in weeks following extreme falls. For emerging markets, where the non-linearity is more pronounced, a trading rule test is also implemented, in addition to the traditional likelihood ratio test of model specification. The overall results indicate the presence of non-linearity in the mean equation of stock return processes.
Keywords: Stock return rebounds, predictability, autocorrelations, non-linear models, model selection, trading-rule tests, emerging markets, market indices, weekly return dynamics
JEL Classification: G14, G15
Suggested Citation: Suggested Citation