Bad Trade Versus Bad System: When to Cut the Position and When to Cut the Method

A single losing trade, or even three in a row, is usually noise. Markets are probabilistic. A method with a positive expectancy still produces losing periods. Treating every loss as evidence the system is broken leads to constant switching and the destruction of whatever edge existed.

The opposite error is worse. The method’s assumptions have changed—regime, competition, costs, or the investor’s own execution—and the losses are no longer noise. They are signal. Most people attribute the wins to skill and the losses to bad luck or external events. The result is that genuine edge erosion is noticed only after the damage is large.

What you actually need to measure

You need a minimum sample, a defined metric, and the willingness to treat the process as a statistical object rather than a personal narrative. Expectancy (average win times win rate minus average loss times loss rate), average R-multiple, maximum drawdown relative to the historical distribution, and win rate are the usual candidates. The exact metric matters less than consistency and honesty about costs.

A journal that records the thesis at entry, the invalidation level, the size rationale and the subsequent path makes the review possible. Without it, every review becomes a story. With it, you can compare realised expectancy against the numbers you assumed when you started.

Annie Duke’s framing is useful here: judge the quality of the decision by the information available at the time, not by the outcome. A good process can produce a bad result. A bad process can produce a good result. Over a large enough sample the process shows through.

When the system is the problem

Flag the method for reduction or retirement when realised performance falls outside the statistical bounds you set in advance, when the market regime has changed in a way the method cannot handle, or when your own execution has drifted (missed stops, oversized positions, skipped reviews). The 2015 episode where months of gains disappeared in hours was a sizing failure, not a thesis failure. The method still had an edge. The size did not match the edge. Cutting the method would have been the wrong response. Cutting the size was the correct one.

For longer-horizon methods the sample takes longer to accumulate. Value underperformance from 2017 to 2020 did not prove value was dead. It proved that the investor who could not sit through it would not capture the later recovery. The review question is whether the underperformance is inside the historical range or outside it.

The practical cadence

Quarterly for most satellite methods. After a pre-defined number of trades for active trading. The review asks three questions: is the expectancy still positive after costs, is the drawdown inside the range I planned for, and have I followed the rules I wrote? If the answer to the third is no, fix the execution before you change the method. If the first two are no and the sample is large enough, reduce size or stop.

What to do with this

Open the last twenty closed positions or the last four quarters of a satellite. Calculate net expectancy after costs. Compare the largest drawdown with the number you wrote down before you started. If you do not have the number written down, write it now for the next period. The measurement is the filter. Without it every loss feels like a crisis and every win feels like proof.