Forecasting the Next Renko Bar with CatBoost and Volume Features

Summary

The article presents a binary classifier for predicting the direction of the next Renko bar on EURUSD. It converts five-minute price data into Renko bars using an ATR-based block size, then builds features from recent bar directions, directional streaks, and volume measures. A CatBoost classifier estimates both the next-bar direction and its probability; the author describes using a probability threshold before treating a forecast as a signal.

The reported test accuracy is 59.27%, and the feature-importance results place recent and average volume measures ahead of price-pattern features. These are results from one described sample, rather than evidence of durable profitability: the article gives no trading-return, transaction-cost, or broader out-of-sample analysis. It also makes strong claims about technical analysis and model performance that the experiment as presented does not establish generally. The method is best read as an example of feature construction and a limited forecasting experiment, not as a validated trading system.

Key ideas

• Renko bars encode price movement using a fixed price-distance threshold rather than regular time intervals.
• The example chooses the block size from average true range to adapt it to volatility.
• Features combine recent bar directions, streak statistics, and volume measures.
• CatBoost predicts the next bar's direction and supplies a probability used with a signal threshold.
• The reported test accuracy and feature rankings come from one EURUSD experiment and do not establish profitability.

Tags

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