Test · Backtests that lie
Why crypto backtests fail with real money.
Most backtests skip fees, slippage and funding, peek at candles that hadn't closed, and get tuned until the curve looks perfect.
- 01Notice
- 02Check
- 03Test
- 04Size & practice
- 05Review
- Foundations
- Learn the mechanics
- Nothing to steal
Most crypto backtests fail for the same three reasons: costs left out, information used before it was available, and settings tuned until the curve looks perfect. A backtest answers one narrow question, how these rules would have done on this history under these assumptions, and changing the assumptions can flip the answer. SuperQuant's Research Lab makes the assumptions honest by default, then actively tries to break your result.
What it looks like
A backtest is easy to fool, and the person fooled is you.
- Fees, slippage and funding are left out, so the edge is imaginary.
- Signals use candles that hadn't closed yet: accidental time travel.
- Settings get tweaked until the curve fits one slice of history.
Illustration · each line adds one honest assumption
Why it happens
The three ways a backtest lies
- CostsThe silent killerEvery perp trade pays fees and spread, and positions pay or receive funding.
- Look-aheadPeeking at the futureUsing a candle before it closes makes any rule look clairvoyant.
- OverfitTuned to the pastEnough tweaks will fit any history, and predict none of the future.
In depth
Costs: what most crypto backtesting leaves out
A strategy that trades 300 times a year pays fees and spread on every entry and exit. At a combined 0.1% per side, that adds up to about 60% of the position size in costs over the year, before funding. Perpetual futures also charge or pay funding for every interval a position is open, which can help or hurt depending on the side.
A backtest without these costs measures a market that doesn't exist. SuperQuant models fees, slippage and hourly funding by default and shows the costs you would have paid next to the equity curve.
In depth
Look-ahead bias and overfitting
Look-ahead bias creeps in when a signal uses data that wasn't final yet, such as a candle's close while the candle is still open. In SuperQuant, signals use closed candles only and fill at the next open.
Overfitting is subtler. Try enough moving-average lengths and stop distances and one combination will look excellent by chance. The Disproof Desk reruns your result with higher costs, perturbed settings, later data, walk-forward windows and resampled trade order. A result that survives most of those is worth a closer look; one that doesn't was probably luck.
Worked example
Worked example: the same rules, measured honestly
Illustrative figures for one strategy.
| As tuned, no costs | +184% |
|---|---|
| Fees and slippage included | +71% |
| Funding included | +38% |
| Last 30% of history only | −6% |
Each line adds one honest assumption. The strategy's real edge, if it has one, is the last number, not the first.
Before and after
What changes with SuperQuant
| Without SuperQuant | With SuperQuant |
|---|---|
| Fees, slippage and funding left out | Modeled fees, slippage and hourly funding |
| Signals peek at unfinished candles | Closed candles only, fills at the next open |
| Tuned until it looks perfect | Nine diagnostics plus a reserved holdout |
| Results that quietly change | Evidence Passports and Time Capsules record what you tested |
The SuperQuant solution
The Research Lab: built to prove you wrong.
Write the rules without code, test them fairly, then try hard to break the result before you trust it.
/backtestDecide candle by candle before the reveal; describe a strategy in plain English and Quant drafts it.Step by step
How it works
- Write the rulesStart from a recipe, or type "BTC 20/50 trend with a 2% stop".
- Backtest with costsNet return, drawdown and every trade, after modeled fees, slippage and funding.
- Try to break itRun the nine diagnostics and see what survives.
- Practice itReplay the market with the future hidden, then paper trade it.
Survives costs and settings · depends on a few big trades
No hype
What the Lab is, and what it is not
It does
- Runs in your browser; your strategies stay private
- Shows which data source every chart uses
- Records what you tested so results can't quietly change
It won't
- Promise a backtest predicts the future
- Use tick data; it tests on hourly candles
- Place real orders
SuperQuant is research and education software. It never places trades and is not financial advice. Trading perpetual futures carries a high risk of loss.
FAQ
Questions people ask
Do I need to code?
No. Start from one of six recipes, edit the rules in simple fields, or describe the strategy in plain English and Quant drafts it for you.
What data does it use?
Hourly candles: built-in data for BTC, ETH, SOL and HYPE perps, read-only fetches from Hyperliquid and Binance, or your own hourly CSV. Every chart labels its source.
If a strategy survives the Disproof Desk, will it make money?
Not necessarily. Surviving the diagnostics makes a result worth a closer look; it doesn't predict the future. That's why the next step is paper trading.
Why does my crypto backtest work but fail live?
Usually costs (fees, spread, slippage, funding), look-ahead bias, or overfitting to one period. Test with costs in, signals on closed candles only, and a holdout you never tune on before trusting a result.
Learn
Further reading
Free, source-cited guides from the SuperQuant Learn library.
Know the odds before you take the trade.
Run a backtest that tries to prove you wrong.
Six recipes, no code, and nine ways to break your result. Start with a free account.