Denoising the Equity Premium

Previous studies have shown that a variety of economic variables fails to deliver consistently accurate out-of-sample forecasts for the equity premium. In this study we propose a wavelet denoising framework in the context of equity premium forecasting. First, we decompose the time-series using wavelet analysis and then we remove the noise in different frequencies. Our results show that the proposed method improves the forecasting ability of linear models indicating that wavelet denoising can successfully identify the underlying persistent signal in the equity premium time-series. Motivated by this, we apply various linear and nonlinear models such as wavelet networks and neural networks to further improve the accuracy of our forecasts.