A1681
Title: Forecasting cryptocurrency returns with a sparse dynamic factor model
Authors: Tatsuma Wada - Keio University (Japan) [presenting]
Akihiko Noda - Meiji University (Japan)
Abstract: The predictability of cryptocurrencies is assessed using a sparse dynamic factor model (DFM). The motivation for using this model stems from the categorization of cryptocurrencies into two groups: stablecoins and non-stablecoins. Stablecoins are pegged to stable assets, such as fiat currencies, making them less susceptible to speculative transactions, while non-stablecoins are not, leading to more volatile prices. Given these distinctions, it is reasonable to consider several factors that influence the prices of various cryptocurrencies. Our findings indicate that the sparse DFM outperforms the random walk model. However, the comparison with the vector autoregressive (VAR) model is largely inconclusive. Notably, however, when additional financial variables are incorporated into the VAR, the sparse DFM demonstrates superior performance. The model also outperforms the VAR when the rolling window is relatively small and the forecasting horizon is moderate. These findings suggest that the sparse DFM is robust in small sample sizes and is particularly well-suited for forecasting cryptocurrency prices in the near to mid-term future.
