Why My Five Percent Take Profit Is the Most Expensive Rule I Run

Most of the rules I have written into my stock engine lose me money. That is fine. A risk gate that fires when it is supposed to is doing its job whether or not the trade it blocked would have made money. The point of a rule is to reduce variance to a level I can sleep with, which is usually different from the level that maximises expected return.

The five percent take-profit on the stock engine is the rule I would defend the loudest if you asked me to put a number on it. It is also the rule that costs me the most. This week I sat with the data and quantified that cost.

The rule

Every stock bracket order opens with three children: a stop loss three percent below entry, a trailing stop that activates at plus three percent and trails by two percent, and a take-profit at plus five percent. The take-profit is unconditional. When the position trades through plus five percent, the broker fills at or above that level and the position closes. No regime check, no momentum check. The order just fires.

I added the rule in week one of paper trading because I did not trust myself to take profit on green days. I have not touched it since.

What the audit looked at

I pulled every stock-engine exit from the last sixty paper-trading sessions. Of those, thirty-eight were take-profit fires at plus five percent. For each one I asked a counterfactual. If the take-profit had not fired, what would the position have done over the next five trading days, capped at the existing time stop?

I priced the counterfactual two ways. First, a naive hold to the time stop, no trailing, no stop loss. Second, a hold with the existing trailing stop still active. The naive hold tells me the theoretical ceiling. The trailing hold tells me what a small structural change to the engine would actually have produced.

The numbers

Across the thirty-eight take-profit fires, the median position continued advancing for two more sessions before either reversing or stalling. The median additional move past the take-profit, naive hold, was about one point eight percent. The mean was higher, around three point one percent, dragged up by a long right tail in trending names.

Translated into dollars on the equity base, the take-profit rule cost roughly nine hundred dollars of expectation across the thirty-eight trades versus the naive hold. Against the trailing-hold counterfactual the cost was closer to four hundred and twenty dollars. The trailing version captured most of the additional move while still taking back the gain when momentum turned.

If those were the only numbers I had, I would change the rule. Replace the fixed take-profit with the trailing stop alone, recover four hundred dollars of expectation, move on.

The variance side

The numbers I do not love show up when I plot the equity curve under each policy. The take-profit rule, current behaviour, produces a daily P&L distribution with a standard deviation of about seventy-three dollars across the sixty sessions. The trailing-only counterfactual, replayed on the same trades, produces a daily standard deviation closer to one hundred and twelve dollars. The maximum five-day drawdown moves from negative four hundred and ten dollars to negative six hundred and ninety. The longest drawdown duration moves from eleven days to seventeen.

The trade-off is legible. Four hundred dollars of expected return, paid as a roughly fifty percent increase in daily volatility and a forty-five percent deeper drawdown. The rule is buying drawdown variance and it is buying it at a price I would not pay if I were optimising for return alone.

I am not optimising for return alone. I am running a paper account whose realistic ceiling is the rule discipline I can carry to a live account, and the rule discipline I can carry to a live account is the one whose drawdown shape lets me sleep. The take-profit rule loses money in expectation. It buys me a drawdown shape I will actually keep using. The math says keep it.

The thing that almost tricked me

The audit produced a clean number for the cost of the rule. It did not produce a clean number for the value of the rule. The value sits on the side of the equation that does not get logged, which is the number of trades I would have taken off the rails myself if I were managing the exits by hand. That number is unknowable, because the counterfactual never happens.

The trailing-stop replay assumes I would have honoured the trailing stop on every trade. I would not have. The first big winner that pulled back fifteen percent before the trailing fired, I would have closed manually somewhere in the middle of the pull-back and called it discipline. The next one, I would have held a session too long because I was sure it was the one. Both of those interventions are invisible to the counterfactual.

The honest read is that the trailing-stop counterfactual is generous to the alternative policy. The actual policy I would have run, by hand, is some weighted mix of the clean trailing-stop replay and a much worse policy where I intervene at the wrong moment on roughly a third of the trades. I cannot quantify that mix from the data I have. The conservative read is that the take-profit rule is even cheaper than the audit said, because the audit was comparing it to a counterfactual I would not have executed cleanly.

What I changed

Nothing in the engine. The five percent take-profit stays.

What I changed is what I quote when someone asks why the rule exists. I used to say it was the rule that took profit before greed reversed it. That framing is wrong. The rule is a variance trade. It costs four hundred dollars of expectation and buys me a drawdown shape I can hold through.

The first framing made the rule look like a return rule. It is not. It is a sleep rule, and sleep rules are the ones I am most willing to keep paying for.