Market volatility prediction based on long- and short-term memory retrieval architectures

Abstract

Predicting market volatility is a critical issue in financial market research and practice. Moreover, in natural language processing, how to effectively leverage long- and short-term event sequences to predict market volatility is still a challenge. Especially, applying traditional recurrent neural networks (RNNs) on an extremely long event sequence is infeasible due to the high time complexity and the limited capability of the memory units in RNNs. In this paper, we propose a new deep neural network-based architecture named Long- and Short-term Memory Retrieval (LSMR) architecture to forecast short-term and mid-term volatility. LSMR architecture consists of three separate encoders, a query extractor, a long-term memory retriever, and a volatility predictor. The query extractor and the long-term memory retriever compose a long-term memory retrieval mechanism that enables the LSMR to handle the extremely long event sequences. Experiments on our novel news dataset demonstrate the superior performance of our proposed models in predicting highly volatile scenarios, compared to existing methods in the literature.