Using Drawdown and Loss Streaks to Compare Trading Strategies

Summary

The discussion considers how to compare two strategies with the same number of winning and losing trades when the order of outcomes differs. It explains that there is no single performance measure that captures every criterion; a trader can instead combine the measures that reflect their priorities.

For the specific concern about clustered losses, the answer proposes calculating the maximum drawdown of the cumulative win and loss sequence, or measuring runs of consecutive losses and penalizing long runs. In the stated example, alternating outcomes produce a smaller maximum drawdown than a sequence with wins followed by losses. These measures add information about the path of outcomes, but the example treats trades as equal-sized wins or losses. It does not account for return magnitude, transaction costs, capital exposure, or the broader risk and return profile, so they complement rather than replace measures such as Sharpe or Calmar ratios.

Key ideas

• A single metric cannot represent every criterion for judging strategy performance.
• Maximum drawdown of cumulative trade outcomes can distinguish sequences with different loss clustering.
• Consecutive-loss runs can be measured and penalized when the order of trades matters.
• Path-based trade metrics complement return and risk measures rather than replacing them.
• Equal win and loss labels omit trade size, costs, and capital exposure.

Tags

# How to validate trading strategy performance


# How to validate trading strategy performance












I'm backtesting some algorithmic trading strategy based on the buy/sell signals:

To validate the strategy performance I compare it against the buy-and-hold strategy of the same asset and calculate the following metrics:

- the strategy total return

- the strategy Calmar/Sharpe ratios

- the strategy trades win/loss ratio

First of all, I want to add another metrics based on the trades win/loss positions. For example, I assume that the following trades (`1`-win, `-1`-loss): `1,-1,1,-1,1,-1` are better than these trades: `1,1,1,-1,-1,-1` Second, I would like to replace all these ratios with 1 generic metric, if such exists.

## Answer by Enrico Schumann (score 3, accepted)

https://quant.stackexchange.com/a/49984

There does not exist a single metric that encompasses all your criteria; but you could simply construct a (linear or non-linear) combination of the measures you like. For the win/loss streaks you describe, you could either look at the absolute maximum drawdown of the cumulative series (`1` in the first case, `3` in the second), or look at streaks and penalise streaks of `-1`s. For instance, in R:

```
library("NMOF")
drawdown(cumsum(c(1,-1,1,-1,1,-1)), relative = FALSE)$maximum
## [1] 1
drawdown(cumsum(c(1,1,1,-1,-1,-1)), relative = FALSE)$maximum
## [1] 3
```

For streaks, look at function `rle`.

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.