Forecasting Security Returns With Simple Moving Averages

Abstract

This study examines the ability of simple moving averages to forecast security returns. Five moving average variants are used to develop a forecasting model using OLS regression for the DJIA, NASDAQ, TSX and CAD-US exchange rate. The forecasting model is compared to the random-walk model without a drift and tested out-of-sample. The results suggest that the moving averages have no predictive ability on the four indices at a 1 day lag. However, the moving averages explain approximately 45% to 48% of the variation in the returns in the following 10 days and clearly outperform the random-walk model. Most of the forecasting ability is derived from the MA (5, 150). Hurst Statistic estimation is used to confirm the long-term dependencies in the lag 10 data set.