OPTIMIZING PREDICTIVE ACCURACY: DECOMPOSITION STRATEGIES FOR STOCK INDEX FORECASTS

Abstract

This study examines two distinct approaches to forecasting stock exchange index values: the conventional method of modeling past index values and an alternative approach that involves forecasting individual stock prices and aggregating them with corresponding weights. The viability and relevance of the latter method are questioned, considering the principles of market efficiency. Market efficiency suggests that even the direct method of forecasting index values may not yield substantial success, making the indirect approach, which lacks the inertia of the forecasted time series, appear even less promising. In turbulent market conditions, decomposing the index may reveal hidden movements that would otherwise go unnoticed when considering the index as a whole. This study conducts a comparative analysis between the two forecasting approaches using AR(1) and AR(1) models with ARCH(1) enhancements. Furthermore, the study explores differences in results between recessionary and pre-recession periods, revealing that the alternative approach proves more successful in forecasting changes in index values during economic downturns.