⚖ Compare + SPRT — does each strategy pass, from its own trades?

The question it answers

"Is this strategy winning often enough to make money, or not?" It answers with the trades as they come out, so we do not have to wait for the end of the month, and it stops as soon as the evidence is strong enough either way.

Imagine a strategy

Pick one and press play. Each closed trade is a step: a win pushes the blue line up, a loss pushes it down. The test stops the moment the line enters a coloured band.

1 · The bar a strategy has to clear: reward:risk (R)

Every trade has a stop and a target. We measure money in R: 1R is what you lose if the price hits the stop. Reward:risk says how far the target is compared with the stop. With R = 1.7, the target is 1.7 times as far as the stop:

Each trade also pays 0.1R of costs, win or lose. So one trade is worth:

• a win: +1.7R − 0.1R = +1.6R
• a loss: −1R − 0.1R = −1.1R

The strategy breaks even when its wins pay for its losses: win rate × 1.6 = (1 − win rate) × 1.1, which gives a win rate of 40.7%. Win more often than that and it makes money; less, and it loses money, even if it wins "a lot".

The bigger the target, the fewer wins you need. Breakeven win rate for each reward:risk (costs 0.1R):

Important: this page uses R = 1.7, the reward:risk the desk's books are backtested with. The R in Test settings must be the strategy's real reward:risk: assume a bigger R than it really has and the test asks for too few wins (too easy on it); assume a smaller one and it asks for too many (too hard).

So the test compares two worlds:

• "No edge" (H0): it wins 40.7%, the breakeven.
• "Real edge" (H1): it wins 50.7%, the breakeven plus 10 points.

2 · The blue line: every trade is a vote

The line starts at 0 and each closed trade moves it:

• a win (net P&L > 0 after costs) moves it up by +0.22;
• a loss moves it down by -0.18.

The height of the line (log-likelihood ratio, Λ) is how much more the trades so far look like "real edge" than "no edge". Up = evidence of edge, down = evidence of none.

3 · The two lines decide

• Green line, ACCEPT (A = +2.94): enough evidence of an edge. Keep trading it.
• Red line, REJECT (B = -2.94): enough evidence of no edge. Stop trading it.
• Between them, CONTINUE: not enough trades yet; keep collecting.

The lines are set so that a strategy with no edge wrongly reaches ACCEPT at most 5% of the time (α), and a strategy with the real edge wrongly reaches REJECT at most 5% of the time (β).

Reading a card

"Crossed the REJECT line (first crossed at trade #27)" means: after 27 closed trades the strategy had lost often enough that the line fell below the red band. A strategy that wins about as often as the breakeven drifts slowly down; one that wins clearly more climbs into the green band.

What it does not say

• It counts wins and losses, not their size: two strategies with the same win rate look the same here even if one makes much more per win. Read it next to profit factor and net R.
• The bar depends on R and costs in Test settings (see section 1): with the wrong R the verdict is too easy or too hard on the strategy.
• A strategy with only a few trades always reads CONTINUE: that is the test saying "too early", not "it works".