A Transductive Learning-Based Early Warning System for Housing and Stock Markets With Off-Policy Optimization

Abstract

This research examines the link between the housing and stock markets, particularly emphasizing bubbles in the housing market. Typically, the fluctuations in the stock market have a more substantial effect on the movement of housing prices than the influence of housing market changes on the stock market. Nonetheless, when information from the housing market is used as an indicator, the changes in housing prices can serve as predictors of stock market volatility. Accordingly, we propose a novel early warning system (EWS) tailored to the housing market. This system is powered by an advanced spatial attention-based transductive Long Short-Term Memory (TLSTM) model, enhanced with an off-policy proximal policy optimization (PPO) algorithm for effective feature selection. The TLSTM model stands out for its ability to detect subtle temporal shifts in data, a significant improvement over the conventional LSTM model. It emphasizes transductive learning that prioritizes data points near the test sample for increased accuracy. The off-policy PPO algorithm further strengthens our model by identifying complex patterns and relationships that might elude standard models, thereby improving the EWS’s predictive capabilities. Additionally, our methodology includes a finely tuned Differential Evolution (DE) algorithm that optimizes the model’s hyperparameters for peak performance. We validated our EWS using data from the Korean market, demonstrating its exceptional ability to forecast with an impressive accuracy of 85.422. By providing a more robust analytical framework, our study contributes significantly to the field of financial analytics, particularly in understanding and predicting the interactions between housing and stock markets during periods of market bubbles.

Keywords

• Housing market bubble
• early warning system
• long short-term memory
• feature selection
• hyperparameter optimization