Skip to content
All library documents

Backtesting and Evaluating a Bollinger Bands Mean-Reversion Strategy

Article Quant Q&A · Author: Caeta
Read the original

Summary

The document describes an exploratory intraday strategy that buys near the lower Bollinger Band and sells near the upper band. The author reports testing it on E-mini S&P 500 data from 1999 and observing a stretch of profitability, while also noting that the strategy was not consistently successful and that sharp price moves were problematic. The post asks whether parameter tuning, machine learning, or statistical methods could improve it.

The response recommends testing parameter choices across multiple historical years and judging results with a risk-adjusted measure such as the Sharpe ratio. It also mentions momentum as a strategy to investigate and points toward machine-learning study and existing implementations as references. The document gives no detailed performance statistics, transaction-cost assumptions, or out-of-sample results. Its brief result is therefore only a starting observation; tuning on historical data can select parameters that fit past behavior without establishing future profitability.

Key ideas

  • The proposed strategy buys near the lower Bollinger Band and sells near the upper band on an intraday timeframe.
  • The author reports a period of profitability in a historical E-mini S&P 500 backtest, but not consistent success.
  • The response suggests comparing parameter settings across multiple years.
  • Risk-adjusted performance measures can help evaluate backtest results.
  • Historical optimization alone does not establish that the strategy will remain profitable.

Tags

FuturesMean reversionTechnical indicatorsBacktestingRisk management
Full text
# Optimize Bollinger Bands Strategy


# Optimize Bollinger Bands Strategy












I was proving a very simple strategy with Bollinger Bands for a intraday timeframe (1 minute) that buy on lower band and sell in a higher band (Very common strategy), but in backtesting in E-Mini SP 500 at 1999 I think that it generated interesting results. This the graphic:

It's not winner but for a long period of the year it was profitable. First, considering that I'm a newbie i could like that you tell me if it can achieve to something profitable with some optimization, and if it so, what kind of optimization can I implement? or if it's a rubbish and I should try with other stuff

I was thinking that maybe I could implement some kind of machine learning for detect abrupt raises or falls prices which are the main problem as show in the image below

Or something with statistics that I should learn (my statistic knowledge it's not very widely) and prove

Any advice it would be greatly appreciated

## Answer by F0l0w (score 2, accepted)

https://quant.stackexchange.com/a/59559

This might not be a direct response, but just some general advise / ideas.

I think giving you "optimal" parameters is not a straight realistic response. It depends on the security you are analizing. For instance, what you can do is grab a set of historical years, run multiple simulations on different parameters, and check which ones provide you the higher return (you can use a risk-adjusted performance metric, e.g., sharpe ratio).

There are other strategies you can also try, such as the Momentum one. If you want to explore machine learning on it, you can check out Marcos Lopez de Prado's book on advances in financial machine learning. You can also backtest using Quantopian.

Bollinger Bands is a common strategy, thus you might find some implementations of it already, to use as reference.

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.