Measuring Volatility in Rolling and Walk-Forward Strategy Tests
Summary
The document considers how to estimate strategy volatility when performance is evaluated in overlapping rolling windows. One response recommends modeling conditional volatility with a chosen GARCH specification or using an average volatility estimate; exponential smoothing in the style of RiskMetrics is offered as another option. These are suggestions rather than a comparison of methods, and the response notes that the question is underspecified.
A second response argues that volatility should be measured from the strategy’s daily return series, even when the strategy is evaluated with rolling or walk-forward windows. It recommends standard deviation as a conventional risk-profile measure, while cautioning that volatility is not the same as overall risk. If the goal is to assess return stability, it suggests calculating volatility, value at risk, and drawdown over separate, non-overlapping periods. The document gives no data or tests establishing which estimator performs best, so the appropriate choice depends on whether the aim is conditional volatility, overall dispersion, or stability across periods.
Key ideas
- Rolling evaluation windows do not require calculating volatility from overlapping window returns.
- Daily strategy returns can be used to estimate volatility with standard deviation.
- GARCH and exponential smoothing are proposed as alternatives for estimating conditional volatility.
- Non-overlapping periods can help assess stability using volatility, value at risk, and drawdown.
Tags
Full text
# Volatility of a rolling window strategy # Volatility of a rolling window strategy What methods can be applied to determine the volatility of strategy using a rolling window? Using normal standard deviation would bias the results as the returns will be highly correlated. Although, after reading several papers where a rolling window is applied, I've not seen one mention anything about it. Is it because it is not necessary or is it "too simple"? By rolling window I simply mean that, for instance, the the strategy is tested over a 6 month period, then rolled forward 1 month and from there tested over a 6 month period, etc. ## Answer by not.so.quanty (score 1) https://quant.stackexchange.com/a/12910 One could use a GARCH of his choice to estimate the volatility. A mean over your period would be a good indicator, otherwise the instant conditional sd is as good as it gets. Another way could be via an exponential smoothing of the risk-metrics type. Your question is not so clear is to be honest. ## Answer by Simon (score 1) https://quant.stackexchange.com/a/14605 I've analysed numerous strategies, and have never encountered problem similar to yours. Your approach of volatility measurement may be a bit deviating from the conventional thinking. Essentially, why would you measure volatility of overlapping returns? It's not sensible. No matter what rolling or walk-forward schemes you are adopting, one can always derive the daily return, and measure their standard deviation as a proxy for volatility. If risk profile is what you are concerning with, then standard deviation is a good enough and pretty standard metric, although I admit that volatility does not equal risk. If stability of returns is your main concern, then you might want to measure the volatility, VaR and draw down, etc., in each non-overlapping periods.
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.