Random Forest Stock Signals from Technical Indicators in South Africa

Summary

This project tests whether a Random Forest classifier using six technical indicators can generate useful long-only signals for ten South African Top40 stocks. The indicators cover trend, oscillation, volatility, and volume. The study trains separate models for each stock using the earlier 7.5 years of daily data, then evaluates daily and weekly rebalancing on the following 2.5 years. It compares those strategies with an equally weighted stock benchmark and standalone Bollinger Band, RSI, and MACD approaches.

The reported test results show both model-driven strategies outperforming the benchmark and the indicator-only comparisons; the daily strategy returned 44.69% against 21.45% for the benchmark. These are historical backtest findings from a limited universe and period, not evidence of reliable future performance. The supplied text is incomplete, omits parts of the analysis, and does not fully explain label construction, validation, or other potential sources of bias, so the results are difficult to assess independently.

Key ideas

• The study combines six technical indicators as features in Random Forest models for long-only stock signals.
• It evaluates daily and weekly rebalancing against an equally weighted benchmark and three standalone indicators.
• The test period covers the final 2.5 years of a ten-year daily dataset for ten South African stocks.
• The reported daily model strategy outperformed the benchmark and indicator comparisons in this backtest.
• The limited universe, historical sample, and incomplete methodological details constrain interpretation of the reported performance.

Tags

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