Risk of Ruin and Position Sizing for Expert Advisor Traders: The Maths That Decides Whether You Survive

What risk of ruin means for an EA trader

Risk of ruin is the probability that a sequence of losses takes your account below the point where you can no longer trade it, either because the money is gone, because a broker or prop firm closes you out, or because you switch the EA off in a panic. It is not about the average outcome. It is about the worst realistic path, and every trading system has one. A profitable EA can still have a high risk of ruin if its position size is too large, and a mediocre EA can have a low one if the size is small enough to survive the bad stretches.

Three quantities decide it. The first is the edge: how much the system makes per unit of risk on average, which is its expectancy. The second is the variability of results around that average, which shows up as streaks and drawdowns. The third is the fraction of the account staked on each trade. The first two belong to the strategy. The third is yours, and it is the only one you can change today. That is why the same EA can make one buyer a steady return and another a blown account in a month.

Why this matters: most retail accounts do lose

National regulators have measured what happens to ordinary clients of online forex and CFD brokers. ESMA summarised their analyses in 2018: typically 74% to 89% of retail CFD accounts lose money, with average losses per client between 1,600 and 29,000 euros. The French regulator AMF studied 14,799 active clients over 2009 to 2012 and found that 89% lost money, with an average loss of about 10,900 euros, and that the most active traders lost the most. These are base rates for retail clients in general, not for EA users, but they show where the typical account ends up. The common ingredients of those losses are well known: too much leverage, too many trades and no fixed limit on loss. Each of them is a position-sizing problem.

Losing streaks are not a malfunction

The most common reason people switch off a sound EA at the worst moment is a losing streak that felt impossible. It was not. Streaks are a mathematical property of sequences of wins and losses, and they are longer and more frequent than intuition suggests. The table below gives the exact chance of at least one run of consecutive losses of a given length somewhere in a stretch of trades. It assumes each trade is independent with a fixed win rate, which is a simplification, but it shows the right order of magnitude. We calculated it by dynamic programming, not by simulation.

Read the first table slowly. A trend-following EA with a 40% win rate and big winners, which is a perfectly respectable design, will run into ten losses in a row somewhere in a thousand trades more than nine times in ten. A scalper at 50% will do it more than a third of the time. Neither is broken when it happens. What matters is what the streak costs your account, and that is entirely a function of risk per trade.

Put the two tables together. A 40% win-rate EA has a 91.5% chance of a ten-loss streak in 1,000 trades. At 1% risk per trade that streak costs about 9.6% of the account, which is uncomfortable and survivable. At 5% it costs 40.1%, and at 10% it costs 65.1%. The strategy is identical in all three cases. Only the size differs. When a buyer says an EA 'blew up their account', the first question to ask is what the risk per trade was, because the answer usually explains the outcome better than the strategy does.

The recovery arithmetic that punishes big drawdowns

A loss needs a larger gain to undo, and the gap widens quickly. The gain required to return to the starting balance equals 1 divided by (1 minus the loss) minus 1. This is simple algebra, and it is why deep drawdowns are so damaging: they do not just cost money, they cost time and psychological capital while the account climbs back.

The curve is why experienced traders think of the first 20% of drawdown as the zone to stay in. Past it, recovery depends on an edge that has to be larger, working for longer, with fewer mistakes. You can check any drawdown you face with the calculator below and the drawdown calculator, and the guide on what to do when an EA is in drawdown covers the decision itself.

How risk per trade changes the odds of a deep drawdown

Streak tables show single events. To see how position size changes the whole picture we simulated two systems with fixed-fraction sizing (each trade risks the same percentage of the current balance) and measured how often the account fell 50% from a peak at any point in 1,000 trades. Each figure comes from 20,000 simulated runs with a fixed random seed, so you can reproduce it. Trades are independent, there are no costs and no slippage, so real results will be worse. The point is the shape, not the decimals.

Two lessons stand out. First, a system with a genuine edge becomes almost unbreakable at a small size and almost certain to wreck you at a large one, with a cliff somewhere between 2% and 5% in this example. The edge did not change between rows. Second, a system with no edge at all is not safe even at a small size, because over a thousand trades random variation alone produces a 50% drawdown in about one run in six at 1% risk and four in five at 2%. This is the reason to be sceptical of any EA that has not shown an edge over many trades, and the reason to size every untested system as if it had none.

Kelly sizing: the right answer to the wrong question

The Kelly criterion, published by John L. Kelly Jr. in 1956 and popularised in finance by Edward Thorp, gives the fraction of capital to stake on a repeated bet to maximise the long-run growth rate. For a bet that wins with probability p and pays b times the stake, the fraction is p minus (1 minus p) divided by b. It is mathematically elegant and, taken literally, dangerous.

Look at the first row. Kelly says stake 16.7% of the account on every trade of a system that wins half the time with 1.5R winners. The simulation above shows that 10% a trade wrecked the account in every run, so 16.7% would do worse. Kelly maximises growth if your probabilities are exactly right, and in trading they never are. You estimate the win rate and payoff from a finite sample, the true values drift, and overestimating the edge pushes Kelly sizing past the point where growth turns negative. Practitioners therefore use a fraction of Kelly, commonly half or a quarter, and many traders on small accounts use far less, which is what the 0.5% to 2% range in the tables reflects. If you remember one thing: Kelly is a ceiling that tells you what not to exceed, not a target.

How EAs actually size positions

Different EAs offer different sizing rules, and the label on the input matters less than what it does to your risk per trade. Our guide on lot sizing methods for EAs goes through them. The short version:

The recovery-style modes deserve special attention. If an EA doubles its lot after each loss, the streak table tells you it will meet a long run of losses, and the loss table tells you what that run costs when the lot is no longer small. Even a 12-loss run, which has an 11.4% chance in 1,000 trades for a 50% win-rate system, requires 4,096 times the base lot at the last step of a pure doubling sequence. Our guide on the risks of martingale EAs shows how that plays out.

A worked example: setting the lot size from a risk limit

Here is a method that works for any instrument. Choose the risk you will accept on one trade as a percentage of the account. Convert it to money. Divide by the money lost per lot if the stop is hit. The result is the lot size. The calculation below uses round numbers; use your own broker's contract specification, or the lot size calculator and the pip value calculator.

Now compare that with the common beginner choice of a fixed 1.00 lot on the same 10,000 account. On EURUSD with a 25-pip stop, one lot risks 250 per trade, 2.5%, and ten stops in a row cost about 22.4% of the account. On gold with a 5-unit stop, one lot risks 500, or 5% a trade, and ten stops cost about 40.1%. The identical lot is two different risk levels on two instruments, which is why lot size should always be derived from the stop and the account, never copied from a screenshot.

Many EAs have no fixed stop or use a stop that moves. In that case, estimate the worst plausible loss per trade from the backtest report (the largest losing trade) and size so that this worst case is no more than twice your intended risk. Our guide on reading the MT5 backtest report shows where to find it. An EA that opens a basket of positions needs the same treatment applied to the basket as a whole.

The three limits to set before switching an EA on

1. 1

   Step 1: A risk per trade

   Often 0.5% to 1% for a new EA on a real account, rising only after weeks of verified behaviour. If the EA uses fixed lots, convert the lot into a percentage of the account using the method above and write the number down.

2. 2

   Step 2: A daily loss limit

   A loss, for example 2% to 3% of the account, at which the EA stops opening trades for the day. Streaks cluster in time, and a daily limit turns a bad session into a bad session rather than a bad week. If you trade a prop firm account, set it comfortably inside the firm's own daily limit.

3. 3

   Step 3: An equity stop

   A level of account equity at which you switch the EA off and review. It is the number you decided on while calm. Writing it down in advance removes the choice at the moment you are least able to make it. The guide on when to turn off your EA helps you choose the trigger.

These limits are not a prediction that the EA will fail. They are the price of staying in the game while it proves itself. Notice also what they do to the numbers above: with 1% risk, a 3% daily limit and a 20% equity stop, even a bad streak leaves you with an account you can still recover.

More than one EA: risks add up faster than you expect

Running several EAs on one account feels like diversification, and sometimes it is. But if two EAs trade the same instrument in the same direction at the same time, their losses arrive together and their risks add. Two EAs that each risk 1% and tend to lose on the same days behave like one EA risking close to 2%. Three correlated gold EAs risking 1% each behave like one EA at nearly 3%, and the streak and drawdown tables should be read at that figure. Before adding an EA, check how correlated its trades are with the existing ones, set a combined risk limit for the whole account and keep each EA on its own magic number so you can measure them separately. Our guides on portfolio risk for several EAs and running multiple EAs with magic numbers cover the practical side.

Leverage limits do not size your trades for you

Retail clients in the European Union face leverage limits set by ESMA's product intervention measures: 30 to 1 on major currency pairs, 20 to 1 on non-major pairs, gold and major indices, with a 50% margin close-out rule across the account and negative balance protection. These limits cap the largest position you can open for a given margin. They do not stop you from opening a position that risks 10% of your account inside the limit. A trader using 3 to 1 of effective leverage with a 5% stop-out risk is safer than one using 25 to 1 with a wide stop, and neither is constrained by the rule. Position size, not the leverage ceiling, determines how much a bad streak costs, which is why the maths on this page applies equally to a regulated European account and to an offshore one. Our guide on EA trading in Europe under ESMA rules covers the regulatory side.

Size for volatility, not for the symbol name

A lot is a quantity of a contract, not a unit of risk. What one lot risks depends on how far the instrument moves against you before your stop, and instruments differ enormously. A 25-pip stop on EURUSD and a 5-unit stop on gold are both reasonable on their charts, yet the example above showed that one lot of gold carries twice the money at risk of one lot of EURUSD for those stops. When volatility rises, as it does around central bank decisions and inflation data, the stop a strategy needs widens too. If you keep the lot fixed through that change, your risk per trade quietly rises. If you hold the risk percentage fixed and let the lot shrink, the risk stays where you set it.

This is the case for either a risk-percent mode or a volatility-scaled lot. A simple check works for any EA: take the average true range of the instrument on the EA's timeframe, multiply by the stop multiple the EA uses, and convert to money for your lot. If that figure is bigger than your risk limit when volatility doubles, the lot is too large for a quiet market and dangerous in a busy one. Gold is the instrument where this matters most for EA buyers, and our guide on XAUUSD EA settings shows how stop and lot interact there.

Funded accounts: sizing against someone else's drawdown rule

On a prop firm challenge your risk of ruin is not the point where the money is gone. It is the point where a rule is breached, and the rules are tighter than an ordinary account's. Many firms apply a daily loss limit and a maximum overall loss, measured on balance or on equity, sometimes trailing from a high-water mark. The arithmetic of the tables still applies, but the line you must not cross is much closer. If a firm's daily limit is 5% and your risk per trade is 1%, five stops in a row end the day. The streak table says a run of five losses is close to certain over a few hundred trades even for a 50% system (it appeared 81% of the time in just 100 trades in our calculation). A sensible approach is to size so that the firm's daily limit would need at least ten full-risk losses in a row, which for a 5% daily limit means 0.5% risk per trade or less, and to set your own daily limit at around half of the firm's.

Measure the rule the way the firm does. A limit based on equity counts floating losses, which an EA with open positions can run up during a spike even if it later recovers. Our guides on using an EA with a prop firm and on EAs for FTMO cover how firms treat automated trading, and the prop firm challenge calculator lets you test a plan against the numbers.

Test your own EA's risk of ruin with a simple simulation

You do not need specialist software to put numbers on the risk of your own setup. Export the list of closed trades from a backtest or a live history and convert each to R, its result divided by the money risked. Then resample: draw trades at random, with replacement, from that list to build many alternative histories of the same length, apply your intended risk percentage to each, and count how often the equity curve falls below your equity stop. That is a Monte Carlo analysis, the same method behind the table above, and it can be done in a spreadsheet with a few thousand rows or in a short script. The Strategy Tester in MetaTrader 5 also offers a robustness view that reshuffles trades, and the guide on reading the MT5 backtest report explains the statistics it reports.

Resampling has two limits you should respect. It assumes future trades look like the past ones, so a backtest that was over-fitted gives an optimistic distribution, and it throws away the order of trades, so it cannot capture streaks created by market regimes. Use it to rank sizing choices against each other and to see how fragile the system is, not to produce a precise probability. If the analysis shows a meaningful chance of breaching your limit at the size you wanted, the answer is a smaller size, not a better-looking chart.

When and how to increase size

Sizing is not a one-off decision. The sensible path starts small and earns its way up, with each step conditioned on evidence rather than on mood. The table sets out one possible ladder. The figures are examples, not recommendations, and the principle is what matters: a larger size is justified only by a longer, cleaner live record at the previous size.

Notice what the ladder does not do: it never scales on the basis of a few good weeks. The sample-size problem that makes a short provider record unreliable, discussed in our guide to EA, copy trading and signals, applies equally to your own EA. A good month after a dozen trades is mostly noise. The other direction is as important. If the live record turns materially worse than the backtest after costs, step down a stage rather than waiting for it to recover.

Why people break their own limits

Limits written in advance fail for human reasons, so it helps to know the usual ones. The first is loss aversion in the middle of a drawdown, the urge to raise the lot to win it back faster, which the recovery table shows is exactly the wrong direction. The second is overconfidence after a good run, the temptation to double the size because the system has been right. Research on individual investors by Barber and Odean found that the households that traded most earned markedly less than those that traded little, and attributed the gap to overconfidence and trading costs. The third is switching the EA off at the bottom of a drawdown that the numbers said was normal, and back on after the recovery has gone. The remedy for all three is the same: decide the numbers while calm, write them where you will see them, and automate the limits where the platform allows it, for example with an equity stop in the EA or a daily loss setting at the broker or prop firm.

Common sizing mistakes

• Sizing from the backtest's best period. A backtest that risked 3% a trade and made a spectacular return also had a drawdown to match. Look at the drawdown, not the return, when choosing the size.
• Raising the lot after a win. A winning streak is the least informative moment to scale up. It is exactly when sizing mistakes are cheapest to make and costliest to repay.
• Treating the maximum historical drawdown as a ceiling. It is a floor for what to expect, as our guide on drawdown explains.
• Ignoring costs. Spread, commission and slippage widen losses and shrink wins, which pushes a marginal system into negative expectancy. See the guide on backtest versus live results.
• Using the same lot on every instrument. One lot of gold carries far more risk than one lot of EURUSD for a typical stop.
• Moving the stop or the limits during a drawdown. The limits are worth having only if you keep them.
• Starting too large because the account is large. A bigger account does not make the system safer; it only makes the same percentage risk a bigger sum of money to lose.

A short plan you can apply today

1. Find the largest losing trade and the maximum drawdown in the EA's backtest and, if available, its live record. Our guide on verifying a track record shows what to trust.
2. Choose a risk per trade that makes a 15-loss streak survivable. Use the loss table: at 1% that streak costs 14.0%, at 2% it costs 26.1%, at 5% it costs 53.7%.
3. Convert it to a lot size for each instrument with the worked example, or switch the EA to a risk-percent mode.
4. Set the daily loss limit and the equity stop and write them where you will see them.
5. Run on demo or at the smallest live size for the first weeks, and compare actual drawdown with the expected one. The routine in the guide on the first 30 days with an EA gives a schedule.
6. Review monthly. Increase size only if the live results, after costs, are in line with the backtest and the drawdown stayed within your limit.

None of this makes an EA profitable. It cannot create an edge that is not there. What it does is make sure that if the edge exists, you are still trading when it shows up, and that if it does not, the loss is a number you chose in advance rather than one the market chose for you.