Backtesting Trading Strategies for a Funded Crypto Account

Your strategy tested clean. Two years of data, a win rate you can live with, an equity curve that climbs from left to right. You buy the challenge, trade the same rules with the same discipline, and the account is gone in three weeks with the strategy still working exactly as designed.

That is not bad luck, and the backtest was not broken. It measured the wrong thing.

What backtesting proves, and what it does not

Backtesting is running a strategy's rules against historical price data to measure how it would have performed. It proves the rules had an edge on the data you tested. That is the whole of what it proves.

It does not prove the edge survives live execution, where your fill is worse than the candle's close suggested. It does not prove the conditions still resemble your sample. And it says nothing at all about whether you will follow the rules at 3am when the position is underwater and the exit is still ninety seconds away.

Those limits are not reasons to skip the test. They are reasons to know what you are holding when you finish it: evidence about rules, not a forecast about outcomes. A strategy that already survived its own results record is a different kind of claim from one that only survived a spreadsheet.

Why a good backtest still loses money

Most backtests fail in live trading for one of two reasons, and neither one shows up in the profit total. The first is fitting. You move the stop, then the timeframe, then add a filter, and you keep going until the curve looks right. What you have built by then is a description of that particular stretch of history rather than a rule that travels. A strategy tuned through twenty variations has learned the sample by heart.

The second is the depth of the dip. A strategy can be genuinely profitable across three years and still fall 30 percent from its peak somewhere in the middle of them. On a chart that recovers in a few months. In an actual account it often does not, because a trader 30 percent down cuts size, skips the next signal, or stops altogether, and the recovery the backtest promised belongs to someone who kept trading the rules through it.

Why a profitable backtest still fails a funded account

On a funded account that second problem stops depending on your nerve and becomes arithmetic. A prop firm puts up its own capital for you to trade once you pass an evaluation, and in exchange it sets loss limits that close the account when you cross them. Nobody talks themselves into holding on, because the decision is not theirs to make.

So the drawdown limit, not the profit target, is what ends evaluations. Our maximum loss limit is 6 percent of the initial balance, and the daily limit is 4 percent on the one-step model and 5 percent on the two-step. That profitable strategy with the 30 percent dip never reaches its good year here.

The daily limit has a second edge that catches good traders. By default it trails from the highest equity point reached during the day, unrealized profit included, so an open winner that gives half of itself back has already raised the floor underneath you. Understand how a trailing limit moves against you before assuming your equity curve fits inside it.

Per-trade caps turn that account-level limit into a number you can act on: on our accounts the realized loss on any single position, commissions included, cannot exceed 3 percent of the initial balance.

The costs a crypto backtest has to carry

A crypto backtest that leaves out funding and fees reports a profit the strategy never earned. Perpetual futures settle funding every eight hours, at 00:00, 08:00 and 16:00 UTC, so a swing position held four days crosses twelve of those timestamps and pays at every one of them if the rate runs against it, winning or losing. Spot and forex carry no equivalent charge, which is why funding is the cost that goes missing when a strategy arrives from either of those playbooks.

Fees compound the same way. On our Bybit routes the standard rate is 0.0200 percent maker and 0.0550 percent taker, and CLEO simulates Binance rates at 0.0200 percent maker and 0.0500 percent taker. A scalping strategy taking forty round trips a week pays that spread eighty times.

Slippage belongs in the same line item. Model it as a fixed cost per trade if you have nothing better, because assuming you filled at the printed price is the single most flattering lie a backtest tells.

Choosing data that will not lie to you

Crypto price history is exchange-specific, which means the wick that triggers your stop on one venue may not exist on another. Your backtest inherits whichever venue's data you fed it, so test on the exchange you will actually trade, not the one with the convenient CSV.

Two more habits separate a useful sample from a flattering one. Cover at least one full regime change rather than a single directional run, because a trend-following strategy tested only on 2023 to 2024 has been graded on the easiest possible exam. And remember that delisted tokens quietly remove their own failures from the record, so a universe built from today's top fifty coins has already deleted every coin that died.

One detail specific to funded accounts: the daily limit resets at UTC, so your backtest's day boundary has to be UTC too. A strategy that looks fine on exchange-local days can breach a limit that a UTC clock would have caught.

How to backtest a strategy, step by step

Write the rules down, pick the venue and period, run every trade with costs attached, log each result against both profit and the rule limits, then re-run on data you never looked at. Five steps, and the fourth is the one that decides whether the answer is usable on a funded account.

Write the rules until they are boring

Your entry, exit, stop, and size, written so another trader could execute them without asking you a question. If a rule contains the word "usually", it is not a rule yet, and you will find that out later as an argument with yourself over whether a trade counted.

Pick the venue and the period

Choose the exchange you will trade and a window holding at least one trend, one range, and one violent liquidation event. Shorter than that and you are measuring a season.

Run every trade with costs attached

Funding, fees, and a slippage assumption on each fill, applied per trade rather than as a haircut on the total. The cost falls hardest on strategies that trade most, and averaging hides exactly that.

Log results against the limits, not just the total

Record each trade's profit and the running peak-to-trough equity dip beside it, then read the worst dip against the account you plan to trade. This is the step that turns a backtest into something a funded trader can use.

Re-run on data you never touched

Hold back a period while you build, then test on it once. If the result falls apart, you fitted the rules to the sample rather than the market. One clean attempt is the whole value, and re-testing until the held-back period agrees with you turns it into more fitted data.

How many trades before the number means anything

Thirty trades tells you almost nothing, and a hundred is a floor rather than a comfortable answer. The rarer your setup, the more of the result rides on a handful of outliers, and a strategy whose profit comes from three exceptional trades in two years has not been tested so much as sampled.

Count the losing streak your sample contains rather than the win rate alone. A 55 percent win rate feels safe until you notice the sample held eight consecutive losses. Then take that streak, multiply it by your per-trade risk, and check the answer against your daily and maximum limits. That single arithmetic step tells you more about whether you will pass an evaluation than the profit factor ever will.

Backtesting for free, and what free actually costs

You do not need to buy anything built for backtesting. Bar replay on a charting platform plus a spreadsheet, one row per trade, is a complete method, and most traders should start there rather than reaching for software. The obvious objection is correct. It is slower. It is also the only version where you see the setups you would have talked yourself out of, and a coded backtest run before you have hand-tested fifty trades mostly teaches you that your code works.

The real cost is hours, and the real risk is self-deception. A manual tester who lets the next candle load before deciding has not run a test, and nobody catches that but you. Log the entry before you scroll.

Can an AI model backtest your strategy?

No, and the reason is worth knowing, because a language model can write the backtest and cannot be the backtest. Ask one to evaluate your rules against last year's price action and it will produce numbers with the shape and confidence of results, assembled from nothing, because it holds no price history to check them against.

Used the other way it earns its place: describe your rules in plain language, have it write the script, then run that script against real data you supply. The model is a translator from your rules into code. The data has to come from somewhere real, and the verdict has to come from the run.

The step after the backtest

Forward testing runs the same rules on live data going forward, and it is the only stage that tests execution, funding drag, and your own discipline at once. A backtest tells you the rules had an edge. A forward test tells you whether you can trade them.

We run a demo challenge you can use at no cost, built for exactly this stage. It takes no card, comes in sizes up to 200,000 USDT, and runs on exchange-based environments using real market data, which is what separates it from paper trading on a simulator, where the price feed is the sandbox's own. Traders who start there show a 30 percent higher success rate once funded.

Put the rules in front of a live market before the money is real: start a free trial account and forward-test your strategy on a demo challenge built for evaluation practice.

Note: the rule figures above describe our evaluation and funded accounts at the time of writing and are here for information, not as the binding version. Check the trading rules page for the rules currently in force on your account.