Accounting for Autocorrelation in Trade Return Quality Measures

Summary

The document asks whether a Sortino style measure calculated from closed trade returns should be scaled by the square root of the number of trades when returns are autocorrelated. It defines the proposed measure as average trade return divided by the standard deviation of losing returns, multiplied by the square root of the trade count. The concern is that this scaling assumes independent, identically distributed observations.

The question mentions lag one autocorrelation as a suggested adjustment, but provides no answer, formula, data, or empirical evidence. It therefore frames a statistical issue rather than resolving it. Any adjustment would also depend on the return dependence structure and the precise performance measure; the document alone does not establish whether lag one autocorrelation is sufficient or how downside deviation should be treated.

Key ideas

• The proposed trade return measure scales a Sortino style ratio by the square root of the trade count.
• That scaling raises concerns when trade returns are autocorrelated rather than independent.
• The question mentions lag one autocorrelation as a possible adjustment input.
• No adjustment formula or evidence is supplied, so the question remains unresolved.

Tags

# Adjust a quality measure (Sortino Ratio) to account for Autocorrelation in Trade Returns


# Adjust a quality measure (Sortino Ratio) to account for Autocorrelation in Trade Returns












Given a series of closed trade profits and loses, a quality measure that could be applied is Sortino Ratio multiplied by the square root of N (N being the number of trades).

```
( average trade return / standard deviation of negative trade returns, i.e. losing trades ) * sqrt( N )
```

However, according to an online AI chatbot, the N value should be adjusted to account for the trade returns not being i.i.d, i.e. autocorrelated. Is this actually true, and if so, what is the formula for making this adjustment? The chatbot suggests using lag-1 autocorrelation.

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.