🤖 AI Summary
This paper addresses the problem of directional forecasting for Bitcoin price movements. Methodologically, it proposes a lightweight, price-driven ensemble framework that integrates multi-dimensional OHLC time-series features; notably, it systematically validates— for the first time—that the low price (Low) exhibits the highest contribution to directional prediction, establishing it as the core feature. The framework combines a Gated Recurrent Unit (GRU) with benchmark models into an ensemble predictor and enhances interpretability via temporal feature importance analysis. Experimental results demonstrate that using only raw price sequences achieves state-of-the-art accuracy (>62%), while non-price features—including trading volume and sentiment—exhibit negligible impact. This work delineates the effective boundary of pure time-series price modeling and establishes a novel paradigm for cryptocurrency directional forecasting: concise, robust, and interpretable.
📝 Abstract
Bitcoin is one of the cryptocurrencies that is gaining more popularity in recent years. Previous studies have shown that closing price alone is not enough to forecast stock market series. We introduce a new set of time series and demonstrate that a subset is necessary to improve directional accuracy based on a machine learning ensemble. In our experiments, we study which time series and machine learning algorithms deliver the best results. We found that the most relevant time series that contribute to improving directional accuracy are Open, High and Low, with the largest contribution of Low in combination with an ensemble of Gated Recurrent Unit network and a baseline forecast. The relevance of other Bitcoin-related features that are not price-related is negligible. The proposed method delivers similar performance to the state-of-the-art when observing directional accuracy.