Using dynamic model averaging in state space representation with dynamic Occam's window and applications to the stock and gold market

We combine the Onorante and Raftery (2016) dynamic Occam's window approach with the Raftery et al. (2010) DMA/DMS estimator in state space representation to create forecasts using a data-rich forecasting environment. Our approach is mainly related to economic and financial time series that are subject to periods of high volatility, which increases the necessity of a time varying parameter framework. In a forecasting exercise for the stock and gold markets, we highlight the economic value-added of our approach by applying a simple trading rule to the return series. By combining both assets, we show that our approach performs better when compared to alternative forecasting models such as machine learning algorithms and standard DMA/DMS. Results for the complexity of the forecasting models highlight the advantages of high dimensional forecasting approaches in times of economic uncertainty, such as the recent financial crisis. The economic performance of the trading rule weakens when we consider transaction costs.