Why reversing a losing strategy usually still loses
A common idea: if a rule loses money, do the opposite and win. We checked this on rules CMI tested and rejected. In almost every case the opposite loses too. Three reasons.
1. You pay the costs twice
A rule's loss is partly the market and partly fees and slippage. Reverse it and the market part flips sign — but you still pay the costs, again. A rule that loses mostly because of costs becomes a rule that loses because of costs in the other direction.
2. The payoff flips with the win rate
A rule that wins often but small, and loses rarely but big, reversed becomes a rule that loses often but small and wins rarely but big. Swapping which side you are on does not change the shape that made it lose; with stops and targets placed for the original direction, the reversed trade usually has a worse plan, not a better one.
3. Our numbers
| Rule | Trades | After costs | Costs per trade | Before costs | Reversed (rough) |
|---|---|---|---|---|---|
| ICT1 | 27,527 | −0.115 R | 0.070 R | −0.044 R | −0.026 R |
| OF1 | 162,425 | −0.027 R | 0.041 R | +0.015 R | −0.056 R |
| LD1 | 116,857 | −0.023 R | 0.011 R | −0.012 R | +0.001 R |
| FD1 | 498 | −0.025 R | 0.017 R | −0.008 R | −0.009 R |
| LL1 | 592 | −0.116 R | 0.043 R | −0.073 R | +0.030 R |
R = the result in units of the amount risked per trade. “Reversed (rough)” = minus the before-costs result, minus the costs paid again — an estimate, not a tested rule; real reversed trades also need their own stops and targets. Source: docs/research/reverse-screen.md (item 41a, scripts/research/reverse_screen.py); reversed = −(before costs) − cost per trade, a rough symmetric estimate, not a test; LD1 with its full ladder cost from the cost-only re-export (docs/reports/ld1-reexport, 2026-10-29).
The two small positive estimates are within a few thousandths of an R of zero — less than the uncertainty of the costs. None is offered as a rule, and none has been tested.
The one exception we found
Not a reversal of direction, but of a filter. “Only take trend trades on coins that beat Bitcoin” (RS1) lost: −0.192 R per trade over 1,703 trades. The complement — coins that lagged Bitcoin (LG1) — did well in its first check (+0.740 R, 104 trades) and on 459 trades on other coins (+0.252 R), but only +0.052 R without its best year. It is an information label, being checked on live prices before it may change any rule.
Past results of tests; they do not tell what will happen next. Education, not advice.