What Is A Normal Maximum Drawdown For A Trading System?

You’ve seen a 12% drawdown in testing and think that’s the worst it gets. It probably isn’t. Here’s what a normal maximum drawdown really looks like.

Will Simpson · 27 Sept 2026 · 10 min read

You’ve got the equity curve on screen. Nice slope, a few dips, worst peak-to-trough is 12%. You catch yourself thinking: “So 12% is the worst it should get.”

This is a plain-English guide to maximum drawdown.

That thought is the problem.

This whole question — what is a normal maximum drawdown for a trading system — only feels easy if you haven’t met real distribution yet.

What you think max drawdown is (and what it actually is)

First, definition, because people fudge this constantly.

Maximum drawdown is the largest peak-to-trough fall in equity within a period, expressed as a percentage of the peak.

Equity at 10,000, runs up to 12,000, then slides to 9,000 before making a new high. Peak is 12,000, trough is 9,000, so drawdown is (12,000 - 9,000) / 12,000 = 25%.

Not 10%, not “from my deposit”, not “from the last week”. Peak to trough.

Maximum drawdown is simply the worst of those events across the sample.

Key word: sample.

When you ask “what is a normal maximum drawdown for a trading system”, what you’re really asking is: how bad does this thing get, on average, over my lifetime, not just the last 3 years of data I found on a forum.

Why the historical max drawdown is lying to you

The worst historical drawdown is not a safety limit. It’s a clue.

If you’ve run 500 trades and the worst drawdown was 15%, the single most important fact is this: you have only observed 500 throws of the dice.

You haven’t seen the worst sequence the distribution can produce. You’ve just seen the worst it happened to produce in that small window. Entire careers have been ended by that understatement.

Imagine a system with these rough stats:

Win rate: 55%
Average winner: 2R (twice the risk)
Average loser: 1R
Position size: 1% of equity per trade

The expectancy per trade is positive (we walk through this properly in /blog/what-is-trading-expectancy), so in theory this is a “good” system.

Now simulate 10,000 trades and track the equity curve.

What shows up in tests like this, again and again, is that the maximum drawdown you see in the first 500 trades is usually much smaller than the maximum drawdown that appears somewhere between trades 500 and 10,000.

Because the longer you play, the more bizarre streaks the maths is allowed to throw at you.

The clean version: observed max drawdown is a minimum estimate of what the system can do to you, not a cap.

Normal maximum drawdown: a distribution, not a number

The phrase “normal maximum drawdown trading system” is already hiding the trap.

You don’t get one normal max drawdown. You get a distribution of possible max drawdowns depending on how long you run the system and how you size it.

Take that hypothetical system above and do this:

Fix the trade stats (win rate, average R, etc).
Run 1,000 different equity curve simulations of 1,000 trades each.
Record the maximum drawdown in each simulation.

You don’t get the same number each time.

You get a spread. Maybe the smallest max drawdown run is 8%. The median is around 18%. The worst simulated path has a 35% drawdown.

Same system. Same edge. Just a different order of wins and losses.

So when someone asks “what’s a normal maximum drawdown for a trading system like this?”, the honest answer is a range, tied to your time horizon.

Over 1 year, maybe you typically see 10–20%.
Over 5 years, maybe paths with 25–35% appear.
Over 10 years, 40%+ shows up in the tails.

And that’s before leverage, before panic resizing halfway through, and before your broker has a moment with spreads.

How long you trade matters more than you think

If you only test 18 months of EURUSD and then deploy, you are kidding yourself.

The longer a system is run in live markets, the more likely it is to encounter the ugly end of its own distribution — fat tails, regime shifts, clusters of volatility.

This is why “I backtested 300 trades” is not reassuring by itself; see /blog/how-many-trades-do-you-need-to-test-a-strategy for the sample size problem.

A helpful mental model:

Historical max drawdown (from backtest): Dhist
Expected live max drawdown over similar horizon: maybe 1.3–1.5 × Dhist
Expected live max drawdown over much longer horizon: 1.5–2 × Dhist, sometimes worse

These multipliers aren’t magic constants. They’re just a simple way of reminding yourself that reality tends to exaggerate the worst part of your backtest.

If you plan your risk as if Dhist is an upper bound, you are planning to be surprised.

Win rate lies, drawdown doesn’t

A system with 80% win rate can still have brutal drawdowns.

All it needs is the 20% of losers to be large and clustered. Which, by the way, is exactly how markets like to organise themselves when you’re overconfident.

This is why win rate obsession is dangerous; a smooth curve with tiny losers for two years and then one oversized series of hits can give you the same maximum drawdown as a choppy but honest trend system.

If you haven’t already, pair this with /blog/what-is-profit-factor-and-what-counts-as-good and /blog/how-many-losing-trades-in-a-row-is-normal.

Once you actually sit with the math on long losing streaks, the concept of a “normal maximum drawdown trading system” suddenly looks very different.

What different drawdown levels really mean

Forget marketing slides for a second.

Here’s a rough, hypothetical sense of what different max drawdown levels often imply for a system that’s sized sensibly and actually has an edge:

Max drawdown range	What it often signals (hypothetical)
0–5%	Very low risk, likely underleveraged or not enough data; may be overfitted.
5–15%	Common for conservative intraday or diversified systems with modest position sizing.
15–30%	Typical for many trend or swing systems with sensible sizing; uncomfortable but often necessary to capture edge.
30–50%	Aggressive risk or lumpy edge (e.g. trend-following futures, concentrated positions); survivable if planned for.
50%+	Speculative or mis-sized; may still have positive expectancy but high risk of ruin and psychological breakage.

This is not a rulebook.

A 35% max drawdown on a diversified, well-tested trend system can be more robust than a 7% max drawdown from a fragile, curve-fit scalper that simply hasn’t met its bad year yet.

The role of position sizing: you choose your drawdown

Here’s the bit most traders skip.

For a given edge, maximum drawdown is mostly a function of position sizing.

The more you bet per trade, the deeper both your average and worst drawdowns will go, on average, across the same sequence of outcomes.

If you halve your risk per trade, then in a simple model, all else equal, you roughly halve your drawdowns in percentage terms.

The system’s raw edge hasn’t changed. Just your leverage against it.

This is why “what is a normal maximum drawdown for a trading system” is incomplete without: “at what risk per trade, for how many trades, and over what market regimes?”

We cover the compounding and bet-size side more in /blog/fixed-fractional-vs-fixed-lot-sizing-that-compounds and in /blog/what-is-the-kelly-criterion-and-why-full-kelly-ruins-you.

The short version: full Kelly gives you theoretical growth with absurd, career-ending drawdowns. Fractional Kelly or plain conservative fixed-fractional sizing gives you slower compounding but max drawdowns that a human can actually sit through.

How automation changes drawdown (and how it doesn’t)

Automation solves the discipline problem, not the distribution.

If your system has a statistical expectation of 30–40% maximum drawdown over a long enough horizon, automating it does not magically cap it at 10% because an EA pressed the button.

What automation does do is remove a few classic drawdown-worsening behaviours:

Refusing to take the next signal after a losing streak.
Doubling size to “get it back” after a bad week.
Randomly switching systems when the underwater line looks ugly.

Left unchecked, those behaviours turn a mathematically normal max drawdown into “why is my account 70% down and now in a martingale death spiral like some bloke on YouTube?”.

But the core risks stay the same:

The edge might decay or disappear.
Market regimes might flip.
Your sizing might be too aggressive for the edge you actually have, not the one you hoped for.

Automation is a way to express a risk profile consistently, not to escape it.

Different markets, different “normal” drawdowns

Market behaviour matters.

A mean-reversion intraday system on a quiet FX pair might spend most of its life with shallow drawdowns and occasional spikes.

A trend-following system on indices or gold can have protracted, grinding underwater periods even when it’s profitable over a full cycle.

If you trade XAUUSD, there’s another twist: lot size granularity.

A 0.01 lot minimum on gold might look harmless, but on a small balance it forces you to take risk chunks that are large in percentage terms.

So the same system that shows a nice 12% historical max drawdown on a well-funded account can show much larger equity swings, purely because you can’t size it any smaller on a £500 account.

This is why the question “what is a normal maximum drawdown for a trading system” needs to be asked per market and per account size.

What’s normal for a trend system on indices with a six-figure account is not normal for a tiny, overgeared gold account where the minimum ticket is 5–10% of equity per trade, whether you like it or not.

How to decide what drawdown is acceptable for you

Forget what the vendor says is acceptable.

Your acceptable max drawdown is the level at which three things are simultaneously true:

You can keep trading the system without changing rules mid-drawdown.
You won’t need to withdraw capital for real-life expenses during the worst underwater period.
A hit of 1.5–2 × your backtested max drawdown still leaves your account and your head intact.

If any of those fail, then whatever “normal” max drawdown the system has on paper is irrelevant, because you won’t survive the part of its distribution that matters.

A practical way to work it:

Take your historical max drawdown from a robust backtest. Call it Dhist.
Multiply it by something conservative, say 1.5 or 2. Call that Dplan.
Ask yourself: if my equity fell by Dplan tomorrow, from peak to trough, would I still be running this system unchanged?

If the honest answer is no, you’re oversized.

And the normal maximum drawdown for you is lower than the normal maximum drawdown of the system.

Red flags disguised as “low drawdown”

A final awkward point.

Systems advertised with incredibly low maximum drawdown — 2%, 3%, always flat, never more than a tiny dip — often fall into one of three buckets:

Not traded long enough to see anything real.
Overfitted to a very specific period or parameter set.
Hidden risk structures (martingale, grid, options gamma) that haven’t blown up yet.

Ultra-smooth curves can be genuine, usually from heavy diversification and low risk per trade, but they are rare and still not immune to regime change.

So when someone asks “is a 5% max drawdown normal for this trading system?”, the sensible questions are: over how long, with what leverage, and what was hidden under the bonnet to make it look so pretty.

So what should you actually do with all this?

Your real job is not to find the system with the smallest historical max drawdown.

It’s to understand the relationship between edge, position size, and drawdown, then pick a combination you can survive.

A rough checklist around maximum drawdown looks like this:

Treat historical max drawdown as a floor, not a ceiling.
Expect live max drawdown to be at least 1.5 × the backtest over similar horizons.
Size positions so that 2 × Dhist is emotionally and financially survivable.
Be suspicious of systems with almost no drawdown unless the sample is huge and the logic is clear.
Assume long-run paths will include streaks and underwater periods you haven’t seen yet.

Once you internalise that, “what is a normal maximum drawdown for a trading system” stops being a marketing question.

It becomes what it always should have been: a risk budgeting question.

Trading and investing carry a real risk of loss; you can lose some or all of the money you commit, and no system or process removes that risk.

Start your free 14-day ArcisTrade demo →

Watch multiple automated FX, gold and index systems run side by side. See the winners, the losers, and the drawdowns in your own broker account, before you risk real capital.

P.S. Pay more attention to the underwater periods than the peaks. That’s where reality lives.

Common questions

Is a 20% maximum drawdown normal for a trading system?

A 20% maximum drawdown can be perfectly normal for many swing or trend systems, especially if they use meaningful position sizes. Whether it is acceptable depends on your risk tolerance, time horizon, and need for withdrawals. For planning, assume that live trading could see 30–40% at some point, even if backtests only show 20%.

How can I estimate future max drawdown from a backtest?

Take the historical max drawdown from a robust backtest and treat it as a minimum. A simple rule of thumb is to plan for 1.5–2 times that number in live trading, especially over longer horizons. You can also run Monte Carlo simulations that randomise trade sequences to see a distribution of possible max drawdowns and size risk around the higher end of that range.

Does a lower max drawdown always mean a better system?

No. A very low max drawdown might mean the system uses tiny position sizes, hasn’t been tested over enough data, or is overfitted to a benign period. A robust system can have a higher drawdown but a stronger, more durable edge. The key is the relationship between drawdown, expectancy, and position sizing, not the drawdown figure in isolation.

Can automation reduce my maximum drawdown?

Automation doesn’t change the underlying edge or the statistical drawdown profile. It mainly reduces behavioural errors that often make drawdowns worse, such as skipping trades, revenge trading, or randomly changing size. Your maximum drawdown is still driven by the system’s characteristics and your position sizing, not by whether an algorithm or a human clicks the button.