Why your backtest lies

Every trader has seen it: a backtest with a smooth, rising equity curve, and then a live bot that does nothing of the sort. The market did not change its mind. The backtest was wrong from the start. Here are the most common ways a backtest lies, how to catch each one, and what Algonney does about them.

Lie 1: trading is free

The easiest way to make a strategy look good is to forget what trading costs. Every trade pays an exchange fee on entry and on exit. A strategy that trades many times a day with a small average gain per trade can be profitable before fees and clearly unprofitable after them.

How to catch it: compare the total fees with the total profit. If fees eat most of the gross gain, the edge is too thin to survive real conditions.

What Algonney does: fees come from the rules of the exchange you test as, and a backtest can never use a fee rate below 0.02%.

Lie 2: you always get your price

A backtest that fills a market order at the exact last price assumes perfect liquidity. In reality a market order crosses the spread and often moves through the order book. That difference is slippage, and it is larger on smaller coins and in fast markets.

What Algonney does: every market fill includes slippage that starts from the market's own value and cannot go below a floor. The defaults are:

Lie 3: funding does not exist

On perpetual futures, open positions pay or receive a funding rate at fixed times. A strategy that holds positions for days on the paying side of funding can lose a surprising share of its profit to it. A test that leaves funding out flatters every long-held position.

What Algonney does: funding is applied at each funding time on USDT perpetuals, and every trade shows the funding it paid or received.

Lie 4: the test can see the future

Look-ahead bias is using information before it existed. It hides in small details:

• entering at a candle's open on a signal that needs that candle's close;
• using a swing high or a support zone before price had turned away from it;
• assuming the take profit was hit first when both the stop and the target sit inside one candle.

What Algonney does: signals are evaluated on closed bars and fill at the next bar's open. Swing points and the structures built on them are confirmed only after price has turned away, and a strategy sees them only from that candle on. When the stop and the target fall inside one bar, the engine reads one-minute candles to see which came first.

Lie 5: the strategy was fitted to the past

This is the biggest lie, and nobody tells it on purpose. You test an idea, change a setting, test again, add a filter, test again. After enough rounds the backtest looks excellent, because the strategy now describes the past rather than a repeatable edge.

How to catch it:

• keep a period of data you never look at while building, and test the final version on it once;
• run the same rules on other symbols;
• move each setting a little and see whether the result collapses;
• prefer fewer rules and fewer settings.

AI makes this both easier and more dangerous: a model that can rerun a backtest and suggest changes can also fit noise very quickly. Jarvis, the AI assistant on Algonney, can run a backtest and propose improvements, but each change is a new revision you review, and the improvement rounds per task are limited. Treat every improvement as a hypothesis to test on new data.

Lie 6: a few trades are a proof

Ten winning trades in a row feel convincing. Statistically they prove very little. A result built on a handful of trades can be pure luck, and the maximum drawdown of a short test hides the drawdown that is still coming.

How to catch it: look at the number of trades before the return. Look at the worst losing streak and ask whether you would keep the bot running through one twice as long.

Lie 7: liquidation cannot happen

On leveraged futures, a position can be liquidated before its stop loss is reached if the stop sits too far away. A backtest that ignores liquidation can show a trade recovering from a loss it would never have survived.

What Algonney does: liquidation is checked first on every bar from the exchange's margin tiers, and a strategy whose stop does not sit inside the liquidation price is refused, in tests and live alike.

Make the backtest show its work

A trustworthy result lets you check it. On Algonney every trade shows its entry and exit reason, the indicator values at that bar, raw and executed prices, fees and funding, and it is drawn on the chart. A proof box lists the data source, the exchange rules and fees used, the engine version and a data hash, so the same inputs always give the same result.

The last test: paper trading

After the backtest, run the strategy as a paper bot on live prices with virtual USDT. It uses the same engine as a live bot. If its trades match what the backtest expected, you have a strategy worth considering. If they do not, the backtest was lying somewhere.

For a full walkthrough, read How to backtest a crypto trading strategy. The feature itself is described on Backtesting, the backtesting guide in the docs shows each screen, and the backtest limits of each plan are on Plans and pricing.

Past results do not guarantee future returns.