Construct Alpha Factors in Cryptocurrency Market

Factor investing holds a prominent position in the realm of financial trading. Ever since the inception of Arbitrage Pricing Theory (APT), financial scholars have persistently sought factors capable of predicting returns. These factors are seamlessly integrated into quantitative trading strategies like portfolio construction and high-frequency trading. Traditionally, the search for these factors entails a manual identification process based on logical and economic intuition. This method is not only inefficient but also incurs considerable labor costs. To address this challenge, the present study introduces a factor mining framework named Ultra Factor Optimizer (UFO), designed to efficiently produce Alpha factors on a large scale. Within this study, we integrate the Density-based Spatial Clustering of Applications with Noise (DBSCAN) clustering algorithm to group assets sharing similar characteristics. The goal is to facilitate the model's access to hidden information. Subsequently, each group undergoes Genetic Programming to create algorithmic factors through iterative training, aiming to identify factors with a high Information Ratio (IR) and low correlation with the output. Empirical analysis is conducted within the cryptocurrency market, specifically focusing on mining time-series and cross-sectional factors. The experimental outcomes demonstrate that the UFO framework can generate a significant number of Alpha factors across time-series, cross-sections, and various K-bar periods. Additionally, we confirm that pre-clustering assets with DBSCAN leads to superior results. UFO not only drastically improves the research efficiency of factor mining but also has versatile applications, significantly impacting the realms of quantitative trading and factor investing.