Free Trading Bot Backtesting: How to Test One Honestly
Free trading bot backtesting is possible with historical exchange data, open-source tools or a spreadsheet, and the results can be useful. The catch is that a free backtest is easy to make look better than it is. Most of the damage comes from a few predictable mistakes: ignoring fees, using prices the bot could not have seen, and tuning settings until the chart looks good.
This guide shows how to run a backtest on your own strategy or a bot you are evaluating, and what to check before you trust the output.
What a free backtest can and cannot tell you
A backtest replays a strategy on historical candles and reports what would have happened. It can help you answer a few questions: does the rule keep a positive expectancy after costs, how often does it lose, how deep does the drawdown get, and how many trades does it take before the result means anything?
It cannot tell you what the market will do next. A backtest is evidence about one past period. Treat it as a filter, not a forecast.
Where free backtests usually go wrong
- Fees and slippage left out. A scalping rule that wins 55% of the time can still lose money after round-trip fees and worse-than-expected fills.
- Look-ahead bias. The signal uses a candle's close, but the trade is "filled" at that same close or earlier.
- Survivorship. Only coins or pairs that still exist are tested.
- Overfitting. The settings were chosen after looking at the results.
- Too few trades. Twenty trades can look excellent by chance alone.
Step 1: Fix the data and the costs first
Before you run anything:
- Use the exchange's own candles for the same market you plan to trade, for example a USDT perpetual on Bybit. Check for gaps and missing candles.
- Include taker fees and a slippage assumption. Test at least two levels, such as 0.05% and 0.10% per side.
- Include funding if positions can stay open across funding intervals. Funding is small each time but adds up.
- Fix the position size in USDT before the run and keep it constant.
Step 2: Split the history
Do not test on all of the data at once.
- In-sample period: where you build or adjust the rules.
- Out-of-sample period: kept untouched until the rules are final. Run the strategy on it once, with no edits afterwards.
- Walk-forward: roll the window forward, re-check the rules on each new slice, and see whether the edge holds across many slices rather than one.
If the strategy only works in the in-sample window, it is probably overfit.
Step 3: Check the trade count
A result with few trades is mostly noise. As a rough guide, aim for at least 100 trades before reading much into the win rate, and treat anything under about 30 as a sketch.
Also check how the trades are spread over time. Fifty trades inside one week of one trend are one market regime, not a strategy.
Step 4: Read the drawdown and the equity curve
Look past the final return:
- Max drawdown: the largest fall from a peak to the next trough. Ask whether you could hold through it with your real capital.
- Profit factor: gross profit divided by gross loss. Above 1 means winners outweigh losers after costs.
- Losing streaks: the longest run of consecutive losses. A strategy that wins 60% of the time can still have a ten-trade losing run.
- Market beta: does the strategy just ride the market? If it only makes money when the coin rises, it has not shown an edge of its own.
Worked example (illustration only)
The numbers below are made up to show the arithmetic. They are not BOTLAB results.
Suppose an illustrative backtest has 120 trades, each with 100 USDT notional:
- Gross profit before costs: +120 USDT (about 1 USDT per trade)
- Fees at 0.10% per side on 100 USDT notional, about 0.2 USDT per trade including both sides: -24 USDT
- Slippage at an assumed 0.05% per side: about -12 USDT
- Net on paper, before funding: +84 USDT
That looks good. Now apply the checks:
- Remove the 10 best trades: the net falls to about +30 USDT. The result depends on a handful of winners.
- Run the out-of-sample period: -30 USDT over 40 trades.
- Conclusion: the rule is not yet convincing. The honest answer is "needs more testing", not "profitable".
How BOTLAB reports results
We publish paper-trading results. These are simulated with real market prices and no real money. Every result states its period and trade count. As an example of the format, our paper1m/live bot, from 2026-09-12 to 2026-11-21, shows 485 simulated trades over 70 days with a 42.5% win rate. It passes 3 of our 6 statistical gates (G1–G6), defined on the site, and is marked NOT_READY, with the verdict "market beta, no proven edge". A bot is marked ready for micro live only after passing all 6 gates. As of 21/11/2026, none of the 270 bots we track has passed all six.
Paper results are a starting point for a forward test, not proof of future performance. You can read how we run and report paper trading at https://botlab.org.uk/platform.
A simple checklist before you trust a backtest
- Costs included, tested at more than one level
- No look-ahead: signals use only data that existed at the time
- Parameters fixed before the out-of-sample run
- At least 100 trades, spread across different market conditions
- Result compared with simply holding the market
- A drawdown you could actually live with
- Then a forward test on paper, using a trade-only API key with withdrawals off if you ever go further
FAQ
Can I backtest a trading bot for free?
Yes. Free historical candle data from exchanges and open-source backtesting libraries cover most needs. The real limits are your time and how carefully you handle fees, data quality and out-of-sample testing.
Why does my backtest look great but live trading fails?
Usually one of four things: costs were understated, the signal used future data, the settings were overfit, or the market regime changed. Compare live or paper trades with the backtest assumptions, trade by trade.
How many trades does a backtest need?
There is no exact number, but under about 30 trades is too few to judge. Around 100 or more, spread across different conditions, is a more useful starting point.
Is a backtest enough to decide whether to use a bot?
No. A backtest is only the first filter. Forward testing on paper, with costs tracked, comes next, before any real capital is at risk.
Key takeaways
- A free backtest is only useful if fees, slippage and funding are included.
- Split the data into in-sample and out-of-sample periods, and keep the out-of-sample test untouched.
- Judge trade count, drawdown, profit factor and market beta, not just the final return.
- A few winning trades can make a weak strategy look strong, so test again without the best trades.
- Forward test on paper before making any real-money decision.
Paper trading results are simulated. Past results don't predict future results. Crypto trading can lose money, including all capital. Not financial advice.
Not available in restricted jurisdictions (see the Service Agreement). You trade with your own exchange account; BOTLAB never holds funds.
Want these checks run on your own strategy? See the strategy and backtest audit.
More guides
- Wagyu XMR1 Exploit: Refund Status and Scam Warning
- Trading Strategy Backtest Results: How to Read Them
- Polkadot Frozen Balance Explained: Why DOT Is Locked
- Free Multi-Chain Wallet Scanner (Read-Only): How to Use One
Check your bot free