Adaptive Feature Subset and Dynamic Trend Indicators for Medium-Term Stock Market Predictions: A 70 Trading Days Forecasting Approach

Abstract

This study stands out for its focus on medium-term stock market forecasting over a horizon of 70 trading days, approximately 3.5 months, targeting major indices such as NASDAQ100, Dow Jones, and DAX. Diverging from prevalent short-term, often next-day prediction research, it recognizes the need for adaptable predictive models in dynamic market conditions. Our approach employs a diverse array of sub-models, each comprising various traditional and novel input features, systematically paired with machine learning techniques like ANN and SVM. These features are processed in parallel to optimize configurations for different market phases. A key innovation is the development of new relative indicators and trend forecast indicators, derived from methods like SMA, EMA, LOWESS, and linear regression. Utilizing rolling windows and techniques like exponential smoothing, we project trends 70 trading days ahead, enhancing our models’ ability to predict significant market movements. Preliminary results show promising predictability across the evaluated indices. Beyond mere predictions, our approach identifies periods of high efficacy and those of reduced predictability, where investors should avoid the market. This research advances medium-term market forecasting, offering investors valuable insights and aiding informed investment strategies in fluctuating market conditions.

Keywords

• Adaptable predictive models
• machine learning
• medium-term stock market forecasting
• relative indicators
• trend forecast indicators