Paper results must use your real exit rule

Same trades, four answers

On 1,137 resolved paper puts, return on collateral per trade was +1.00% if held to expiry, +0.76% if closed at half the premium only, +1.17% if only defended with a buyback when threatened, and +0.81% under the full rule we would actually run (close at half the premium if that fills first, otherwise defend, otherwise hold). The worst trade ranged from -16.4% to -25.8% depending on the rule.

The spread is a quarter of a percent a trade on a strategy that earns well under one percent. Picking the wrong exit is the difference between comparing two strategies and comparing two accounting choices.

The mistake we made

We once judged a rule set using a record that replayed the defensive buyback but ignored the resting order to close at half the premium. On that basis it averaged a loss per trade, with one very large loser, and we recommended dropping it. Replayed on the real rule, the same 63 trades averaged +1.52% on collateral with a 92% win rate and a worst trade of -4.2%.

The large loser was a trade that had first reached half its premium. Under the real rule the resting order would have closed it at a profit long before the defense triggered; the partial replay let it ride into the loss. One missing exit turned the best rule set in the group into an apparent failure.

Hold to expiry flatters, too

The opposite error is just as common. A paper record that settles every put at expiry counts a put that finished a few cents in the money as a near-full win, while a real account with a defensive rule would have paid to buy it back days earlier. If you never want to be assigned, a hold-to-expiry record is an upper bound, not a result.

For spreads the direction flips again: our spread study found that holding to expiry earned the most, so a spread record that assumes early profit-taking understates what holding would earn. See the lesson on stop-losses.

What it changes in practice

Before reading any backtest or paper record, ask which exit it assumes, and make it the one you will run, including every resting order and every defensive rule. If you change the exit rule, rescore the history under the new rule before comparing. And when two rule sets differ by less than the gap between exit assumptions, you have not learned which is better.