Direct answer: divergence in Average Win Loss
Divergence in Average Win Loss means the “average win” and the “average loss” stop moving together in a consistent way. Instead of the win side and loss side behaving similarly to what you expected from an earlier period, one side’s typical outcome grows, shrinks, or changes character while the other does not.
In plain terms: Average Win Loss is a comparison of two averages from trade outcomes. Divergence happens when that comparison looks different across time segments, datasets, or calculation methods.
Mechanism or definition: how the averages are built
To explain divergence, start with how “Average Win” and “Average Loss” are constructed:
- Average Win is usually computed as the arithmetic mean of the sizes of winning trades (e.g., profit amounts) in a chosen sample.
- Average Loss is usually computed as the arithmetic mean of the sizes of losing trades (e.g., loss amounts) in a chosen sample.
Divergence then appears when, for example:
- In one time period, average wins are relatively large while average losses are relatively small, but later the pattern reverses.
- Or average losses become less negative/more negative while average wins remain about the same.
A simple model of “what changed” is that either the distribution of outcomes changed, or the way you sampled and measured outcomes changed. Those two drivers matter because they lead to different interpretations.
Common measurement choices that can change the averages (without anything “mysterious” happening) include:
- the exact time window used,
- whether you include all trades or only a subset,
- whether averages are computed on raw amounts or after certain costs,
- and whether the dataset includes partial fills or adjustments.
Evidence or example: why divergence can show up
Consider a hypothetical backtest-like setup with clear assumptions:
- Assume you compute average win and average loss from the same strategy’s trade history.
- Assume you split the history into two equal time windows.
- Assume execution quality and spreads/costs remain constant inside each window but can differ across windows.
If average wins and average losses move differently between Window 1 and Window 2, you observe divergence.
This divergence can be explained by at least three non-exclusive causes:
- Market regime change: the typical distance price moves before a win or loss changes.
- Cost or execution change: if costs rise during Window 2, losses may worsen more than wins (or vice versa), altering the averages differently.
- Sampling and calculation differences: if the later window contains more extreme outliers on one side, the average shifts more for that side.
A key point is that “divergence” is a description of the data you computed, not an automatic explanation of a single cause.
Limitations and risks: confirmation limits and hindsight bias
There are material limitations and failure modes in interpreting divergence:
- Divergence does not automatically imply an indicator is “working” or “failing.” It only tells you that historical average win and loss characteristics differ across the segments you compared.
- Hindsight bias: if you look at the largest divergence first, you may “fit” an explanation after seeing the outcome, even if the prior expectation was unclear.
- Overfitting to one period: divergence can reflect a temporary condition. Historical relationships do not guarantee future results.
- Uncertainty from costs and execution: averages are sensitive to how outcomes are recorded. Two datasets that differ slightly in transaction costs or execution modeling can show different divergence.
Verification or next question: what you can check independently
To verify what divergence means in your context, you can check whether it is stable under reasonable, predefined variations:
- Recompute Average Win and Average Loss using consistent rules (same definition of win/loss, same inclusion criteria).
- Compare multiple time splits (not only the period where divergence looks biggest).
- Separate effects where possible by examining outcome amounts before versus after stated costs, if your records support that split.
- Check for calculation sensitivity by repeating the averages with and without obvious data anomalies (like missing trades or non-comparable execution records).
If divergence remains across these checks, it suggests that the underlying outcome distributions changed. If it disappears, it suggests measurement/sampling sensitivity rather than a stable change in performance characteristics.