Choosing Adaptive Bollinger Band Parameters Without Overfitting
Summary
The document describes a daily stock strategy using moving averages and standard deviation bands. The lookback length and band multiplier determine the bands; entries are triggered when price crosses either band, with exits when price returns across the moving average. It proposes selecting these hyperparameters with rolling training and validation windows, reflecting the possibility that market behavior changes over time.
The main question is whether the multiplier should be reselected for each successive window or shared across windows. The response cautions that continually tailoring hyperparameters to recent training results can fit past noise and fail on future observations. The discussion is conceptual: it provides no performance comparison between static and adaptive settings, and it does not establish that the stated entry rules or Gaussian price assumption are reliable. Any adaptive choice would need careful out-of-sample evaluation to assess overfitting.
Key ideas
- The strategy uses a moving average and standard deviation bands, with lookback and multiplier as hyperparameters.
- A rolling validation scheme can reflect changes in the behavior of stock data.
- Reselecting the band multiplier on every window may increase the risk of fitting noise.
- The document gives no empirical evidence that dynamic or shared settings perform better.
Tags
Full text
# Value weighted Portfolio: Include Market cap of pref shares? # Value weighted Portfolio: Include Market cap of pref shares? I got Datastream MV data for my sample and i want to exclude Preference Shares. Since Datastreams definition of MV is shares*price there are two Marketvalues, one for common stock, one for preference Share. I want to calculate value weighted Portfolios: Shall i include the MV of the preference shares to weight my common stock, so its weight is based on its firm. Or shall i just use the MV for the common stock. There are firms like VW that have like 60/40 common stock/preference shares, so it definitely matters. Any recommendations ? What is the typical procedure in the Literature ? Thank you very much !
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.