Esperanza matemática y ventaja: cómo saber si tu estrategia funciona de verdad

Every method, indicator, and trading plan eventually faces one question: does this actually work? Feelings will not answer it — memory keeps the wins and quietly archives the losses. The honest answer is a number called expectancy: what one trade of this method earns or loses on average, measured over enough trades to mean something. This sheet shows you how to compute it from a journal, why single trades and even single months are statistical noise, how many trades a track record needs before it deserves belief, and how performance claims are manufactured by people who hope you never learned any of this.

Expectancy: the formula, on a real journal

Take a trader who has finished fifty trades, all sized the same way: 1% of a $10,000 account risked per trade, so $100 behind every stop — on EUR/USD, that is 0.25 lots with a 40-pip stop at $2.50 per pip. Note the symmetry of the unit itself: a 40-pip move on that position is $100 in either direction; whether it lands as gain or loss is exactly what the method is being tested on. The journal shows 22 winners averaging +$150 and 28 losers averaging −$100. Plug those four numbers in:

Read the result in R — multiples of the amount risked — and it travels across account sizes: this method, on this sample, earned a tenth of its risk per trade. Notice what the calculation rewards and what it ignores. It ignores the win rate alone: 44% sounds poor, yet the method measured positive because winners outgrew losers. It rewards the pair of numbers together — win rate and the ▲risk-reward ratio of the average outcomes. Neither means anything without the other; their product is the whole game.

Now see how fragile the sign is. Shrink the average winner from $150 to $130 — a few exits taken slightly earlier — and expectancy falls to (0.44 × $130) − $56 = +$1.20 per trade: indistinguishable from zero, and one extra spread's width from negative. Costs live in exactly that margin, which is why expectancy must always be computed on results net of spread and swap. A method that is positive before costs and negative after them is, for you, a negative method.

One habit makes everything on this sheet lighter: keep results in R from the start. A journal in dollars confuses the method with the account size and with every sizing mistake along the way; a journal in R isolates the method itself. The fifty-trade sample reads cleanly as 22 wins averaging +1.5R against 28 losses averaging −1R, and the +0.10R expectancy says: each trade earned a tenth of whatever was risked — ten dollars at $100 of risk, five at $50, the same method either way.

Variance: why a working method has ugly months

Suppose the +0.10R figure is real — the method genuinely has that edge. What do the next twenty trades look like? On average, +2R. In practice, almost never that. Single-trade outcomes scatter widely around a small mean (in our sample they range from −1R to roughly +1.5R), and over a twenty-trade month that scatter swamps the average: results between −3R and +7R are all ordinary, with no change whatsoever in the underlying edge. A losing month is not evidence the method died; with a small true edge, it is scheduled to happen regularly.

Simulate the same method many times — same expectancy, same scatter, different random ordering — and the equity curves tell the story better than any argument. Some runs march upward; some spend forty trades underwater first; several show a ▲drawdown deeper than anything in the original fifty-trade sample. All of them share the identical edge. This is the discipline variance imposes: judge the process over many trades, never the outcome of one. The same logic, applied to losing streaks and recovery arithmetic, fills this survey's drawdown sheet — the two are companion pieces.

The third column is the point. Every verdict a trader passes on one month of results is a verdict on the luck column, not on the method — the edge was identical in all five rows, and nothing in any single row could have revealed that.

Sample size: when a record starts to mean something

How many trades before measured expectancy deserves belief? Statistics gives an unwelcome answer. The uncertainty in a measured win rate shrinks only with the square root of the number of trades:

Twenty trades cannot distinguish a coin flip from a decent edge — the measured 55% is compatible with both. A hundred trades begin to narrow it; several hundred narrow it usefully. And win rate is the easy variable: average win and average loss carry their own uncertainty, often larger, because a single outsized trade drags the mean. The practical reading is humbling and freeing at once. Humbling: your first month proves nothing, however it went. Freeing: your first month proves nothing, however it went — a bad start is as statistically empty as a good one. The verdict arrives slowly, in the journal, over hundreds of rows.

A workable ladder for your own records: under thirty trades, compute nothing — just log. From thirty to a hundred, compute expectancy but treat the sign as provisional, and let it change your sizing only downward, never up. Past a hundred trades of one setup, the number begins to deserve weight; past several hundred, comparisons between setups start to mean something. The ladder is deliberately slow. Methods are abandoned by impatience at trade fifteen far more often than they are disproven by evidence at trade three hundred.

Survivorship and cherry-picking: how claims are manufactured

Once you understand sample size, most public track records become transparent. The machinery is rarely outright forgery — it is selection. Start a hundred accounts with a hundred random methods and after a year, perhaps five look brilliant by pure chance; publish those five and the audience sees skill where there was only a wide funnel and a quiet bin of ninety-five failures. The same trick has versions at every scale:

• Survivorship: only the accounts, methods, or months that worked are shown; the rest stop being mentioned.
• Window-picking: the chart starts at the strategy's best moment and ends before its worst.
• Screenshot selection: individual winning trades, with no count of how many trades produced them.
• Demo-to-live sleight: results rehearsed on a demo, implied to be live money under live pressure.
• Backtest overfitting: a strategy tuned until it fits the past perfectly, presented as if the past were a forecast.
• Risk left unstated: a doubling that took 30% risk per trade is an advertisement for ruin, not for edge.

None of these requires a single false number — only false framing around true ones. That is why the defense is structural, not forensic: ask for the full distribution (every trade, not the good ones), the sample size, the risk per trade, and the worst drawdown. Claims built by selection dissolve under exactly those four questions.

Applying it to yourself: a journal that computes

Your own trading deserves the same audit you would run on a stranger's claims, and it needs only a journal kept with intent. Most journals fail in a predictable way: they record feelings — “entered too early, felt rushed” — without the fields the arithmetic needs, so at month's end there is a diary but no data. The fix costs one minute per trade. Five disciplined columns make expectancy computable:

• Risk taken, in money and in R — fixed by your sizing rule before entry.
• Stop distance in pips, and the size derived from it.
• Result as ▲realized P/L, net of spread and swap — the only number that is real.
• Result in R: realized P/L divided by the risk taken.
• A setup tag, so expectancy can later be computed per setup, not just overall.

Add the boring columns too — date, session, instrument, direction, a one-line reason — because expectancy questions usually arrive later as comparisons: London versus New York, longs versus shorts, one setup against another. Review monthly, but compute conclusions only at sample sizes the previous section allows. After a hundred trades on a demo, you will hold something most traders never build: an honest measurement of your own method, with the losing trades included at full weight. The survey's journal sheet covers the habit in depth; this sheet only insists on the fields that make the math possible.

Applying it to others: questions a claim must survive

Finally, point the tool outward. Anyone selling performance — a signal channel, a course, a managed account, any broker's marketing including ours — is making a claim that expectancy arithmetic can interrogate. Ask: how many trades is this based on? What was the average loss, not just the average win? What risk per trade produced these returns? What was the deepest drawdown, and when? Are the results net of all costs, on live accounts, with every trade included? Sellers with real measurements answer in rows and totals. Sellers without them answer in lifestyle photography. The questions cost nothing, and the silence they produce is itself a measurement. Notice, too, what this sheet has not told you: what expectancy to expect. That silence is deliberate — and it is the same silence you should demand from anyone who wants your money.