Backtest Max Drawdown vs Live: How Deep to Expect

Your backtest says max drawdown 11%. You size your live account around that number, trade the same rules, and by trade 60 you are 16% down and wondering what broke. Nothing broke. A backtest's max drawdown is one draw from a range of possib

Your backtest says max drawdown 11%. You size your live account around that number, trade the same rules, and by trade 60 you are 16% down and wondering what broke. Nothing broke. A backtest's max drawdown is one draw from a range of possible drawdowns, and live trading gets to draw again, for longer.

Max drawdown is the worst dip in one ordering of trades

Max drawdown (the largest peak-to-trough fall in account equity) sounds like a property of the strategy. It is closer to a property of the order the trades happened to arrive in. Take the same 100 trades, shuffle them, and the total profit stays identical while the worst dip changes.

TradingView reports two versions, and they are not the same number. Close-to-close max drawdown is "the largest drop in account equity from a previous closing high to a subsequent closing low." The intrabar version is "the maximum possible loss that the strategy could have incurred among all of the trades it has made," calculated for every bar spent in an open position. Check which one you are quoting before you compare it to anything.

The same strategy produces very different drawdowns

Here is the experiment. We took a strategy with a 40% win rate where winners pay 2R and losers cost 1R. R is your initial risk on a trade, so a 1R loss is whatever you would lose if your stop is hit. That strategy has a positive expectancy (average result per trade) of +0.2R. It is a genuine edge on paper.

Then we simulated 100,000 random sequences of trades for each trade count and recorded the max drawdown of each one. Outcomes are independent, sizes are fixed, there are no costs, and equity is measured on closed trades. This is our own simulation, not market data.

Read that again. Same win rate, same payoff, same positive edge. Over 100 trades the median worst dip is 10R, and one sequence in twenty reaches 19R or worse. Stretch to 1,000 trades and the median hits 20R, with a 95th percentile of 32R.

At a flat 1% risk per trade, 10R is roughly a 10% drawdown and 20R is roughly 20%. At 2% per trade those become 20% and 40%. This is why the risk of ruin math in our position sizing guide starts from the drawdown you can survive, not from the return you want.

More trades means a deeper worst case, even with no change in edge

A drawdown is the worst of many dips. Flip a coin enough times and a longer run of tails is going to show up eventually. Your strategy is no different, and the table shows it for two strategies with identical +0.2R expectancy.

Trades	40% wins, +2R / -1R: median, 95th pct	60% wins, +1R / -1R: median, 95th pct
100	10R, 19R	6R, 11R
250	13R, 24R	8R, 13R
500	16R, 28R	9R, 15R
1,000	20R, 32R	11R, 17R

The longest losing streak behaves the same way. For the 40% strategy over 100 trades the median longest streak was 7 losses in a row, and the 95th percentile was 12. If you have never sat through seven consecutive stop-outs with real money on the line, that is a number worth rehearsing, which is a big part of why a plain stop-loss plan has to be settled before the trade, not during the drawdown.

The academic version of this is not new. Magdon-Ismail and Atiya's Risk magazine paper on maximum drawdown presents analytical results relating it to the mean return and the Sharpe ratio, and proposes scaling laws for comparing track records of different lengths. The direction matches the simulation: the longer you run a noisy process, the deeper its expected worst dip.

A shallow backtest drawdown is usually a lucky one

This is the part that stings. We took 200,000 simulated 200-trade backtests of the 40% strategy and kept only the best quarter, ranked by shallowest max drawdown. Those averaged 8.7R. Then we gave each one a fresh 200 trades from the identical strategy. The live average was 13.5R, and the fresh run dug deeper than the backtest 80% of the time. The 60% strategy told the same story, 5.3R in the best-quarter backtests against 7.7R live.

Nothing about the strategy changed. The backtests simply looked good because the trades happened to land kindly, and that is the selection trick behind most pretty equity curves. It is the same problem as the sample size problem we covered in how many trades a backtest needs, just pointed at drawdown instead of win rate.

Live trading gets more trades than the test did

A 200-trade backtest compared to a live account that will run 600 trades is not like for like. In our simulation of the 40% strategy, 600 live trades were deeper than a 200-trade backtest 75% of the time, and 23% of the time they ran at least double. That is not a failure of the strategy, it is the cost of a longer track record.

Real markets are also less friendly than our coin flips. Trade outcomes cluster in volatile regimes, and fills and costs are worse than a tester assumes, which is part of why the TradingView strategy tester produces unrealistic results. Treat the simulated numbers as a floor for how bad it can get, and that is a rule of thumb from experience, not a measured statistic.

MathWorks puts the caution in plain words in its documentation for the expected-drawdown function: the expected maximum drawdown is not an upper bound on the maximum losses from a peak, but an estimate of their average. Its own worked example has the realized drawdown, 0.1813, land above the expected 0.1545.

A backtest does not tell you how deep the hole is. It tells you how deep one hole was.

— The Tradoki desk note

Run your own numbers before you size the account

You do not need a PhD for this. Export your trade list, convert each trade to R (profit divided by the amount you risked), then resample the list at random thousands of times and record the max drawdown of each sequence. Read the median and the 95th percentile, not the single backtest figure.

Then work backwards. Pick the drawdown you can live with, in dollars and in how you will behave, and find the risk per trade that keeps the 95th percentile inside it. If that risk per trade is smaller than you hoped, the account is telling you something useful before the market does. Prop firm rules make the stakes concrete, since a trailing drawdown can end an evaluation long before your own pain threshold does. The same logic applies to Kelly sizing, where half Kelly beats full Kelly once the edge is only an estimate.

● FAQ

Is the max drawdown in a backtest the worst I should expect live?

No. It is the worst dip in one specific ordering of trades, and live trading will deal you a different ordering. In our simulation of a 40% win rate strategy, the 95th percentile drawdown over 100 trades was nearly double the median. Treat the backtest figure as a median-ish number, not a ceiling.

Why does max drawdown get deeper the more trades you take?

Because a drawdown is the worst of many dips, and more trades means more chances for a bad run. In the same simulation, the median drawdown for a 40% win rate, 2R winner strategy rose from 10R over 100 trades to 20R over 1,000 trades. The edge never changed, only the number of attempts.

What is the difference between close-to-close and intrabar max drawdown on TradingView?

Close-to-close measures the drop from a previous closing equity high to a later closing equity low. The intrabar version, per TradingView's documentation, measures the maximum possible loss across every bar spent in an open position, so it is usually the deeper of the two. If your live stop gets tested inside a bar, the intrabar number is the one to watch.

How do I turn drawdown in R into a percentage of my account?

If you risk a fixed 1% of the starting account on every trade, then 1R is 1% and a 12R drawdown is roughly 12%. That is a simplification, because percentage risk compounds as equity changes. For a rough plan it is close enough, and it shows quickly whether your per-trade risk can survive the numbers.

Does a good-looking backtest drawdown predict a smaller live drawdown?

Not reliably. When we kept only the best quarter of simulated backtests, ranked by shallowest drawdown, a fresh 200 trades from the identical strategy dug deeper about 80% of the time. A lucky ordering in the test is exactly what makes the test look good.