RiskMiner: Discovering Formulaic Alphas Via Risk Seeking Monte Carlo Tree Search

The formulaic alphas are mathematical formulas that transform raw stock datainto indicated signals. In the industry, a collection of formulaic alphas iscombined to enhance modeling accuracy. Existing alpha mining only employs theneural network agent, unable to utilize the structural information of thesolution space. Moreover, they didn't consider the correlation between alphasin the collection, which limits the synergistic performance. To address theseproblems, we propose a novel alpha mining framework, which formulates the alphamining problems as a reward-dense Markov Decision Process (MDP) and solves theMDP by the risk-seeking Monte Carlo Tree Search (MCTS). The MCTS-based agentfully exploits the structural information of discrete solution space and therisk-seeking policy explicitly optimizes the best-case performance rather thanaverage outcomes. Comprehensive experiments are conducted to demonstrate theefficiency of our framework. Our method outperforms all state-of-the-artbenchmarks on two real-world stock sets under various metrics. Backtestexperiments show that our alphas achieve the most profitable results under arealistic trading setting.