Comparing Bollinger Zone Forecasts with Direct Price Direction Models
Summary
The article compares two forecasting targets for a Bollinger Bands trading approach: whether price will rise or fall over a future horizon, and which of four zones around the bands price will occupy. It describes labeling historical GBPUSD daily data, generating Bollinger features, and training equivalent Linear Discriminant Analysis models. The zone-based version interprets predicted transitions as directional signals, while exploratory plots examine price outcomes and transitions across zones. The reported comparison uses time-series cross-validation and finds direct price-direction prediction more effective than predicting zone transitions.
The zone approach can make classifications more accurate in the author’s experiments, but it is harder to interpret when the predicted state remains unchanged and produces fewer signals because it waits for a zone change. The article also states that Bollinger parameters were not optimized, and its observations come from the described dataset and setup. It offers a comparison of modeling targets, not evidence that either method will generalize or be profitable in live trading.
Key ideas
- The study compares direct up-or-down price labels with labels for transitions among four Bollinger Band zones.
- It uses Linear Discriminant Analysis models and time-series cross-validation for the comparison.
- In the reported experiment, direct price-direction forecasts perform better than forecasts of Bollinger zone transitions.
- Zone predictions can be less transparent when the model forecasts that price will remain in its current zone.
- Waiting for zone changes produces fewer signals, and the indicator parameters were not optimized.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.