Decision-Tree Signals for a Single-Stock Long-Only Strategy

Summary

This project describes a machine-learning system that uses a decision tree to generate binary signals for trading individual stocks. Indicator buy triggers are used as inputs, while indicator sell rules are omitted to keep the model focused; a zero signal exits an existing position and a one signal initiates a long position when no stock is held. The author emphasizes explainability and reports using adjusted daily closing prices for a set of Indian large-cap stocks, with historical data sourced from a retail-accessible provider.

The project reports backtest outcomes including annualized returns, Sharpe ratios, trade counts, and drawdowns for example stocks, while acknowledging that results varied and that some drawdowns were substantial. The author found that combining many stocks or indicators produced implausible or unreliable results, and recommends beginning with a single stock and a focused prediction goal. These results are preliminary: data quality and corporate-action treatment are concerns, broker data was unavailable, paper trading remained unfinished, and the system was not yet adapted to common broker frameworks. The evidence is historical backtesting, not live validation.

Key ideas

• The model uses a decision tree to produce binary signals from indicator buy triggers.
• A zero signal closes an existing position, while a one signal opens a long position when no holding exists.
• The author reports varied backtest returns and drawdowns across individual stocks.
• Combining many securities or loosely related predictors produced unreliable results in the project.
• Data quality, unfinished paper trading, and limited framework compatibility constrain the evidence.

Tags

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