Comparing Kalman Filters and XGBoost for Stock Price Forecasting

Summary

The document describes a stock-price forecasting study that compares a Kalman filter, XGBoost, and a hybrid that uses XGBoost to select features before passing them to a Kalman filter. Inputs include price and volume data, accounting measures, and quarterly market sentiment. The evaluation covers stocks from India’s NSE and the NYSE, with separate experiments on price-only data and data augmented with sentiment.

Reported average accuracy favors XGBoost across the two exchanges, while Kalman filter results vary more by exchange. The hybrid does not lead on average across the full samples, although the authors report better results for certain individual stocks and higher-priced stock data. Adding sentiment improved the Kalman results in the illustrated HDFC example, while XGBoost metrics reportedly changed little. The evidence is limited by the study’s sample and its accuracy-based evaluation; the document gives little detail on trading returns, transaction costs, or whether the reported measures translate into profitable out-of-sample strategies.

Key ideas

• The hybrid pipeline uses XGBoost feature importance to select inputs for a Kalman filter.
• The study compares price-based inputs with inputs that also include market sentiment.
• XGBoost has higher reported average accuracy than the hybrid across the NSE and NYSE samples.
• Kalman filter performance is reported as less consistent across exchanges.
• Forecast accuracy alone does not establish that a model can support profitable trading.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.