Win rate inflation from partial closes

[Fairman]

Win rate inflation from partial closes

A strategy that closes half the position at a small profit and lets the rest run can show a very high win rate. The first half closes green most of the time, and each partial counts as a winning trade.

The statistics look excellent: 80 percent winners. Then the second halves get stopped at breakeven or a loss, and the real picture is modest.

The issue is how trades are counted. If each partial close is a separate record, the win rate counts the easy wins and dilutes the losses. Averaging them together hides the true distribution.

The fix is to treat the entire position, from entry to the final exit, as one trade, and compute the win rate and the profit per trade on that basis.

Platforms that split positions on partial closes make this harder, since the history shows separate tickets. Merging them requires linking by the original entry.

A related distortion happens with scale-in strategies. Several entries that exit together count as several winners or several losers, when they were a single decision.

Comparing strategies, or comparing against published results, requires the same counting rules. Without it, a high win rate says almost nothing.

To see the damage in a real record, rebuild the journal in two ways. First, as the platform shows it, with every partial close as its own line. Second, merged by original entry, so each decision appears once with its total profit. Put the two summary tables side by side: win rate, average win, average loss, and profit factor.

The merged table is nearly always less flattering. The gap between the two is the amount of self-deception the first table allowed. It also changes which parameter settings look best, since a setting that earns an extra small partial win on every trade looks excellent in the first view and merely decent in the second.

How are partial closes counted in the journal, and did switching to a per-position view change the numbers?