Forecasting Gold Return Using Wavelet Analysis

This article shows the effect of wavelet analysis on forecasting performance from two perspectives of directional accuracy and the magnitude of the forecast errors. Wavelet analysis is used for decomposing gold return series into some subseries with distinct properties and data generating processes. Then each subseries is forecasted by a recursive ARIMA-GARCH in which the parameter estimates are updated recursively in light of new observations and its regressors are chosen recursively based on Akaike's Information Criterion regarding statistical significance of all coefficients. Next, the predicted values of each subseries are added to get the final forecast.