Five Up, Three Down, Even Money

My trading engine attaches the same bracket to every entry. The stop sits three percent away. The take-profit sits five percent away. Written as a ratio that is a reward-to-risk of about one-point-six-seven to one, and almost everyone who sees it reads the same thing: a favorable bet, risking three to make five.

It is not a favorable bet. On a price that has no drift, that bracket is exactly break-even, and it is break-even for a reason that has nothing to do with my signal and everything to do with geometry.

Two barriers and a coin

Strip the entry down to its skeleton. Price starts at zero. There are two absorbing barriers: profit at plus five, loss at minus three. The trade ends when price touches one of them. The only question that matters for expectancy is which barrier it hits first.

For a symmetric random walk, that is the gambler's ruin problem, and it has a clean answer. The probability of reaching the far barrier first is the distance to the near barrier divided by the total span. Take-profit first: three over eight, or thirty-seven and a half percent. Stop first: five over eight, or sixty-two and a half percent.

Now multiply through. Thirty-seven and a half percent of plus five, plus sixty-two and a half percent of minus three. That is one-point-eight-seven-five minus one-point-eight-seven-five. Zero. The favorable-looking bracket has an expected value of exactly nothing.

It is not a coincidence

The clean cancellation is not a quirk of the numbers five and three. It is a theorem wearing a disguise.

A driftless price is a martingale: its expected future value is its current value. The optional stopping theorem says that if you stop a martingale at a barrier-hitting time, the expected value at that stopping time is still where you started. The bracket is exactly such a stopping rule. So the expected profit is zero no matter where I put the two barriers. Five and three, ten and two, four and four: all zero.

That is the part worth sitting with. I cannot tune my way to an edge by adjusting the ratio. Moving the take-profit out makes each win bigger and rarer; moving it in makes each win smaller and more frequent. The two effects are not approximately offsetting. They are exactly offsetting, because they are two views of the same conserved quantity.

The take-profit is the harder barrier to reach, and it is harder by precisely the factor that it pays more. The payoff ratio is five to three. The hit-probability ratio is three to five. Their product is one. That is the whole story of a bracket on a random walk in a single sentence.

Where the money actually is

If the bracket has no mean to give, then every dollar of expectancy in the system has to come from somewhere else, and there is only one place left: drift.

Drift is the edge. It is the small directional bias that makes the price more likely to move up than down after a long signal fires. Give the same two barriers even a slight upward drift and the hit probabilities tilt, the cancellation breaks, and the expectancy goes positive. The signal supplies the drift. The bracket does nothing but convert that drift into a realized distribution of outcomes.

This reframes what the risk framework is for. The stop, the target, the position size: none of them create return. They shape the distribution of a number whose average was already set the moment I chose the signal. A wider stop trades win rate for win size. A tighter one does the reverse. The framework controls variance, skew, and how often I take a loss. It does not control the mean. Only the signal does that.

It also settles an argument I keep having with the scoreboard. A win rate under fifty percent is not a failure of the engine. With a five-to-three bracket, a coin-flip signal produces a thirty-seven percent win rate by construction, and a genuinely edged signal might only lift that into the forties. Judging the engine by win rate alone measures the bracket geometry, not the edge. The wins are supposed to be rarer than the losses, because they are bigger.

The honest asterisks

Real prices are not driftless Brownian motion, and the deviations all make the live bracket worse than the clean math, not better. Prices gap, so a stop can fill below minus three, dragging the loss tail past where the geometry assumed it stopped. Barriers are checked on fifteen-minute bar closes, not continuously, which blurs both touches. Commissions and slippage turn a true random walk from zero into negative. And my trailing stop and time stop are not static barriers at all; they are moving ones, which is precisely an attempt to break the symmetry the static bracket cannot.

But the core result survives all of it. A bracket cannot manufacture an edge. It can only decide the shape of a number whose average was fixed upstream. So I stopped treating the three-and-five as a lever to optimize and started treating it as what it is: a choice about how my P&L is distributed, paid for entirely out of an edge the bracket itself will never produce.