Validate a trading strategy without hindsight · 3 / 5
Measure what an extra trading confirmation adds
More agreement on a chart is not automatically more independent evidence. Test the incremental effect of a filter on the same opportunity set before crediting it with an edge.
Athenum7 minUpdated:
Freeze the base opportunity set
Define the entry rule, exits, sizing and opportunity timestamps before assessing the additional confirmation. Indicators derived from the same price sequence may select overlapping observations. The relevant comparison is a policy with the filter versus the same policy without it, not a count of coloured signals.
Retain rejected opportunities so you can reconstruct what changed. If the filter was chosen after inspecting winners and losers, these observations belong to development, not independent validation. Freeze the selected policy before evaluating later untouched opportunities.
Compare more than average return per trade
Report trade count, net total, exposure time and missed opportunities alongside average result. A filter can improve the average per trade while reducing calendar-period returns, or concentrate risk in one regime. Position overlap and capital constraints can also change what the policy could execute.
Our example assumes fixed additive risk, identical exits, no overlapping capital constraints and equal cost per trade. These assumptions make the arithmetic transparent; they do not turn summed R into a compounded portfolio return. A later evaluation must account for the actual implementation.
The filter changes costs without changing gross total
The base rule creates twenty trades: twelve earn +1R and eight lose −1R, giving +4R gross. A confirmation accepts eight of them, with six winners and two losers, also +4R gross. The twelve rejected opportunities contain six winners and six losers, or zero gross.
At 0.1R cost per executed trade, the base earns +2R net and the filtered policy +3.2R. That is a 1.2R improvement on this hypothetical sample, not proof of future predictive value. The accepted gross average rises from 0.20R to 0.50R per trade, while the sample becomes smaller.
| Policy | Trades | Gross R | Cost R | Net R |
|---|---|---|---|---|
| Base | 20 | 4 | 2 | 2 |
| Base plus filter | 8 | 4 | 0.8 | 3.2 |
| Rejected subset if traded | 12 | 0 | 1.2 | −1.2 |
- Base
- 2 R
- Filtered
- 3.2 R
- Rejected subset
- -1.2 R
Identical selections do not double the evidence
If a second indicator accepts exactly the same eight opportunities, their intersection still contains eight trades. It is not a sixteen-trade sample. The indicators need not be identical in every market; they simply add no selection difference here. Test each filter's incremental contribution on frozen alternatives.
Before acting
- Freeze the base signal.
- Keep rejected opportunities.
- Compare with and without each filter.
- Report trade count and calendar exposure.
- Evaluate the selected rule on unseen observations.
Check your understanding
An additional filter removes one accepted +1R winner and no losing trade. What is the new net total at 0.1R cost per trade?
Show the explained answer
Seven trades remain: five winners and two losers give +3R gross. Costs are 0.7R, leaving +2.3R. That is above the base's +2R in this sample but below the first filter's +3.2R. More confirmations did not improve this result.