A New Method for Stock Price Prediction Based on MRFs and SSVM

Trading strategies basing on both financial analysis and machine learning techniques are becoming increasingly popular due to their ability to capture micro market price movements and leverage big data. An important class of works are focusing on exploiting the structural relationships between companies for accurate stock price prediction. In this paper we develop an algorithm for learning the parameters of unary and binary potentials in binary markov random fields (MRFs) under the max-margin framework. We first show how to train unary potentials using market price data and Gaussian Mixture Models (GMMs). Then, we developed a graph-cut based algorithm to solve the inference problem exactly. We demonstrate the learning of potentials' parameters using a max-margin learning framework. Experiment is conducted by comparing performances between our formulation and conventional SVM method. Results show that our method outperforms SVM by 27.9% on train set and 40.5% on test set.