The popularity of commodities and cryptocurrencies in trading has rapidly increased recently, and a large number of studies examined their price dynamics. However, due to their unique characteristics, it is very challenging to forecast the time series of these asset classes. Previous studies have suggested that both commodity and cryptocurrency prices are prone to contain a considerable speculative component and bubble-like behavior. Technical indicators that are based on a security’s past returns are often used to capture these characteristics. Therefore, in this paper time series momentum (TSMOM) and moving average indicators are used to forecast commodity and cryptocurrency returns. However, unlike in Moskowitz, Ooi, and Pedersen (2012)’s and Huang et al. (2020)’s study, several MA and TSMOM indicators for different time horizons are applied simultaneously. Multiple indicators are applied in three different models, in a simple multivariate ARX, in an ANN, and in a hybrid method in order to find out whether each of these models can capture a time series momentum effect that is more complex than a one-year TSMOM signal in a linear model. The results suggest that whereas the predictive power of the ARX and the hybrid model decreases when more than just a 1-year time series momentum is used as input, on some of the assets the ANN can uncover more complex TSMOM effect and hence provide a better predictive ability than a simple univariate time series regression.
