Learn what separates a backtest from a wish
Most strategies fail for a short list of well-understood reasons, and almost none of them are exotic. These guides cover the list — what goes wrong, how to catch it, and what the honest version of the result usually looks like.
Nothing here sells a strategy or promises a return. The methods are the same whether you build in Wise Apple, write Python, or work in a spreadsheet, because the failure modes belong to the method, not the tool. Where a guide names a Wise Apple setting, it is one paragraph near the end, and the rest of the page stands without it.
Start here If you are new to this, read how to backtest a trading strategy — it is the whole loop, and every other page here is one step of it in depth. If you have a backtest you already suspect, start with data leakage : the one flaw that makes results better instead of worse.
Why backtests lie The gap between a backtest and a live account has causes, and they are findable.
How to backtest a trading strategy without fooling yourself The complete loop: rules, data, fills, costs, splits, leaks, sample size, the report, and a rehearsal before money. Why backtests fail in live trading Seven causes ranked by damage, the fingerprint each leaves, and a diagnostic order that starts with the trades. Data leakage: six ways your strategy sees the future The flaw that inflates results instead of degrading them, the routes it takes, and a test for each. Repainting indicators, explained Three kinds of repaint, the five-minute bar-replay test, and why a chart is not a backtest. Survivorship bias: why your universe is lying to you Dead coins, delisted stocks, and the strategies that look brilliant only on the survivors. Is my strategy overfit? Seven signs and four tests Parameter cliffs, too-smooth curves, profit in three trades — and the tests that settle it. Realistic backtest costs Spread, slippage, fees, impact, borrow and funding, and the multiplier test that catches a fake edge. How many trades before you can trust a backtest? The error bars at 30, 100, 500 and 1,000 trades, trades per regime, and the multiple-testing discount. Why trading strategies stop working Regime change versus the four things that look like it, and how to tell them apart. Is there an edge here? Before the backtest: what an edge is, whether the simple rules still have one, and what beating the market has to mean.
Testing that holds up Validation methods that survive contact with a time series, and the ways each one can still be fooled.
Reading the results The numbers in a report, in the order that catches the most problems soonest.
Machine learning, honestly What the machine-learning side is really doing, and which parts of it earn their keep on market data.
Which machine-learning models actually work on market data Six families compared, the benchmark every model must beat, and why deep learning usually disappoints. Ensemble voting: why agreement beats tuning Fewer trades at higher precision, weighting by skill not profit, and when an ensemble is worse. Meta-labeling and the triple-barrier method Labels that describe the trade you would take, and a second model that decides whether to take it. Lorentzian Classification: what it actually does A nearest-neighbour classifier with an unusual metric — is it ML, does it repaint, and what to ask of it. LSTM stock prediction: why the tutorial curve is a lie The one-step-behind copy, the scaler that leaked the future, and what works instead. Can ChatGPT backtest a trading strategy? What language models can genuinely do, the four ways they fabricate results, and a workflow that keeps them useful. Feature engineering for trading models Why raw prices are the worst feature, the transforms that work, and how to judge a feature without fooling yourself. Why your ML trading model predicts 50% The honest failure: no signal at all. The diagnostic order, what a realistic result looks like, and why accuracy is the wrong lens. An AI wrote your trading bot. Now what? Seven checks, in order, before you trust code you did not write — with a five-minute test for each. Sizing, exits, and going live The half of a strategy that decides whether a good rule becomes a blown account, and the gate between a backtest and real money.
Position sizing: fixed fraction, Kelly, and volatility scaling The arithmetic of each, why half-Kelly, how long a drawdown lasts, and the sizing failure modes. Stop loss, take profit, and time exits The same entry with four exits is four strategies. What each exit does to the trade distribution, and how to test one honestly. When to go live with a trading strategy A gate, not a feeling: what a backtest proves, what paper trading proves, the checklist, and the kill rule. Getting started For the trader who did not study computer science and does not intend to.
Tool comparisons Fair read-outs of where the popular platforms are strong, where they stop, and who each one suits.
Test the idea instead of trusting it
Wise Apple is a private research lab that runs on your own machine. Build a strategy without writing code, test it with the guardrails already on, and read every trade candle by candle.
Load the app →
