Ready to go further
Next: turn win rate and sizes into one number
The expectancy guide shows how accuracy, average win and average loss combine to decide whether a rule makes or loses money.
What is included
- What it is, in plain words
- The formula or mechanism, with exact arithmetic
- How it hits grid, martingale and DCA bots
- A check you can run before you switch a bot on
Who it is for
- People who run or plan to run a trading bot
- Anyone reading a backtest or a bot sales page
- Readers who want the numbers, not the pitch
- Not advice, and not a signal service
Key facts
- AI stock prediction means using a fitted model to forecast something about a price, such as tomorrow's direction or return. The forecast is a number. A trade is a separate decision with a cost.
- Accuracy counts how often the forecast was right. Profit depends on how big the wins and losses were and on costs. The two can point in opposite directions.
- Worked example (made-up): a model right 55% of the time, with equal wins and losses of $50, still loses $5.00 per trade after a $10 cost (calculated by us on 29 Oct 2026, exact).
- With those sizes and that cost, the model needs 60% accuracy just to break even. Every extra point of cost raises the bar.
- A published accuracy figure with no cost, no move size and no out-of-sample period tells you little. Ask for all three.
What a prediction is, and what a trade is
A prediction model takes past data and outputs a forecast. The forecast might be a price, a return, or a simple up or down. The model is fitted by minimising its error on past examples. See machine learning trading.
A trade is different. To trade on a forecast you choose a size, a stop, an entry and an exit, and you pay the spread and fees. The forecast being good is one input. Whether the trade makes money is a second question that the forecast alone cannot answer.
Why accuracy is not profit
There are three reasons.
- Size of the moves. A model that is right on many small days and wrong on a few large ones can have high accuracy and a loss. A model right on only a few days, which are large ones, can have low accuracy and a gain. This is the same point as in expectancy.
- Costs. Each trade pays the spread, any fee and some slippage. A small edge per trade is eaten by these.
- Almost no edge to start with. Daily price direction is close to a coin flip. A real accuracy of 51% or 52% is already hard to find and is easy to produce by accident in a test. See overfitting.
The break-even rule for equal sized wins and losses is simple. If a win is W, a loss is W, and the cost per trade is C, the accuracy p needed satisfies (2p - 1) x W = C, so p = (W + C) / (2W).
Worked example: 55 percent right and still losing
Everything here is made up. A model calls the direction of a stock each day and a trade follows each call with $10,000. We run 1000 trades. The round-trip cost (spread, fee and slippage together) is $10 per trade. Three versions of the model are compared.
| Model | Accuracy | Average win | Average loss | Per trade before cost | Per trade after cost | Over 1000 trades |
|---|---|---|---|---|---|---|
| A: 55% right, wins and losses equal | 55% | $50 | $50 | $5.00 | -$5.00 | -$5,000 |
| B: 55% right, wins smaller than losses | 55% | $40 | $60 | -$5.00 | -$15.00 | -$15,000 |
| C: 45% right, wins bigger than losses | 45% | $90 | $40 | $18.50 | $8.50 | $8,500 |
Calculated by us on 29 Oct 2026, exact. Model A is right on 550 trades and wrong on 450. With wins and losses of $50, that is $27,500 won and $22,500 lost, a gross $5,000. Costs are $10 on 1000 trades, which is $10,000, so the result is -$5,000. The model looked good by accuracy and lost money. It needs 60% to break even, and at 55% it is five points short.
Model B has the same 55% accuracy but its wins average $40 and its losses $60. It loses even before costs. Model C is right only 45% of the time, but its wins average $90 against $40 for its losses, and it ends ahead by $8.50 per trade after costs. Ranked by accuracy, C came last, yet it finished first. Note that C is only an arithmetic case: it assumes the big wins really occur, and nothing says a model can be built to do that.
How it hits bots
- A bot that trades every forecast pays the cost every time. A small forecasting edge on a bot with many trades is eaten faster. See high frequency trading.
- Stops and targets change the win and loss sizes, so they change the break-even accuracy. A tight stop with a wide target lowers accuracy and can change the sign of the result.
- A published accuracy from a backtest is usually measured without costs, and sometimes on data the model saw. See look-ahead bias.
How to check a prediction claim
- Ask for accuracy, average win, average loss, number of trades and costs used, all on data the model did not train on.
- Compute the break-even accuracy yourself with the break-even calculator or the formula above.
- Check for a test period that was sealed until the end, and for how many model versions were tried.
- Paper trade the forecasts with real costs before using money. See paper trading.
What this page does not tell you
It does not say a model cannot find a real edge, and it does not measure any product. It shows that accuracy by itself is not enough to judge one, using made-up numbers.
Keep Going
Related Tools and Guides
Keep exploring
FAQ
Frequently Asked Questions
Can AI predict stock prices?
Not reliably. A model can find patterns in past data, but price direction is close to a coin flip and any pattern has to hold on new data and after costs. Treat claims of high accuracy with care.
What accuracy do I need for a prediction to be profitable?
It depends on win size, loss size and cost. With equal wins and losses of $50 and a $10 cost, the example on this page needs 60 percent. Compute it for your own numbers.
Why can a less accurate model make more money?
Because profit depends on the size of the wins and losses, not only how often the model is right. A model right less often with larger wins can come out ahead.
Do costs matter that much?
Yes, when the edge per trade is small. A cost on every trade turns a small positive result into a loss, and the more trades a bot makes the more it adds up.
How should I test a prediction model?
Fit it on early data, test on later data it never saw, include spread and slippage, and then paper trade. Count how many versions were tried.
Are the numbers on this page real?
Example numbers are made up to show how the maths works, and the page says so. A figure only counts as exact when the page says it was calculated on a stated date. Nothing you enter in a calculator leaves your browser.
Can I use this to decide whether to run a bot?
It can show whether a bot's numbers hold together and how much an account can lose. It cannot say what the market will do, and it is not advice. Use it to ask better questions, then read the terms of your own broker or exchange.
Is this financial advice?
No. It is general education. The example numbers are made up and nothing here is a recommendation.
Know the Risk Before You Run It
Free calculators and guides, added every week. Leave your details and we will email you when new ones go up.
Trading carries a risk of loss