Monitoring Rules
| Metric | Threshold |
|---|---|
| Rolling expectancy | < 0 over the latest 50 trades |
| Rolling hit rate | Below the backtest range for 3 consecutive months |
| Drawdown | Worse than 1.5x historical max drawdown |
Monitoring
How We Will Know Performance Is Still In Line
Going forward, the strategy should be monitored through rolling behavior rather than single wins and losses. We would compare live or forward results against the out-of-sample backtest using:
- rolling expectancy
- rolling hit rate
- rolling Sharpe ratio
- drawdown depth
- average holding period
- outcome mix across
Successful,Stop-loss triggered,Timed out, andBreakout failure
The key idea is that we do not expect each month to match the backtest exactly. We expect live behavior to stay inside a reasonable range around the validated backtest distribution.
How We Will Quantify When The Strategy Stops Working
The strategy will be treated as impaired when one or more of the following conditions persist long enough to suggest regime failure rather than normal noise:
Rolling expectancy < 0- using the latest
50trades
- using the latest
Rolling hit ratefalls materially below the backtest range- for several consecutive months
Drawdownexceeds a historical stress threshold- for example worse than
1.5xthe historical max drawdown
- for example worse than
ML overlay no longer improves the baseline- the scaling extension should continue to add value relative to the control strategy
Payoff shape breaks- the average winner should remain large enough relative to the average loser to justify a whale-hunting breakout approach
Baseline Versus Overlay Monitoring
The project should monitor both layers separately:
- Baseline breakout strategy: is the raw breakout rule still behaving like the original backtest?
- ML overlay: is the model-based position scaling still adding value relative to the tuned baseline?
This matters because the ML extension is intentionally small. If the overlay stops helping, the baseline may still remain tradable.
Payoff Shape To Monitor
| Metric | Value |
|---|---|
| Average winner | 1.41% |
| Average loser | -1.17% |
| Payoff ratio | 1.21 |
| Largest winner | 4.40% |
| Largest loser | -2.97% |
| Top 3 trade PnL share | 47.57% |
Final Monitoring Interpretation
The practical decision rule is:
- keep trading if behavior remains inside expected ranges
- reduce confidence if expectancy and hit rate weaken together
- pause or re-estimate the strategy if drawdown and trade-quality diagnostics both break
That gives the project a concrete answer to the two most important forward-looking questions:
- how do we know performance is still consistent with the backtest?
- how do we know when the strategy has stopped working?