Risk management for an expert advisor comes down to one decision made before you attach it: how large a position, against how much deposit, and at what point you stop. The EA will not make that decision. It will keep trading its rule through the worst stretch in its record and through whatever comes after.
Four numbers do the work, and only one of them is usually quoted:
- The worst trough, in money. Not the percentage — the amount the balance actually gave back.
- The worst losing run. Median 8 trades across these records. Worst: 38 in a row.
- The worst single trade, which sits a median of 1.89× the average loss and up to 10.2× it.
- How long you would have waited. Median 422.5 days without a new equity high — that one across the fourteen listings the parent experiment measured, not all 23.
The first three are measured below across all 23 listings, from the published closed-trade lists.
| Test conditions | |
|---|---|
| Experiment ID | EXP-EA-RISK-BUDGET-001 |
| Parent experiment | EXP-CATALOGUE-PROFITABILITY-001 |
| Source | src/content/ea/en/*.md frontmatter btTrades |
| Population | 23 EN catalogue listings, all with a closed-trade list |
| Model | MetaTrader 5 strategy tester, published backtests |
| Deposit | 10,000 |
| Last verified | 2026-08-25 |
What to look for: the four numbers that set your size
Read them off the EA’s own page before you decide anything. Three of the four are not in the summary box.
| Number | Why it sets your size | Where the summary box misleads |
|---|---|---|
| Worst trough in money | It is the amount your deposit has to absorb without you intervening | The box shows a percent of a 10,000 account, not of yours |
| Worst losing run | It is how many times in a row you have to not intervene | Not shown at all — you count it from the trade list |
| Worst single loss | One trade can cost 10× the average one | The average loss looks reassuring and is the wrong input |
| Longest time underwater | It decides whether you will still be running the EA at the recovery | Not shown; a profitable record can hide two years of nothing |
The spread across the catalogue is the reason this matters. Sorted by worst trough as a share of the recommended minimum:
| EA | Risk score | Worst trough | Share of the 10,000 minimum | Worst losing run |
|---|---|---|---|---|
| Chrysalis | 7 | 2,649.54 | 26.50% | 35 |
| Cairn | 6 | 2,479.13 | 24.79% | 14 |
| Nautical | 7 | 2,464.64 | 24.65% | 38 |
| Thunderhead | 5 | 1,359.02 | 13.59% | 9 |
| Iridescence | 4 | 965.88 | 9.66% | 8 |
| Almanac | 3 | 575.34 | 5.75% | 10 |
| Orrery | 3 | 67.43 | 0.67% | 4 |
Median across all 23: 8.67%. So the typical listing gives back roughly a twelfth of the recommended minimum at some point — and the deepest gives back more than a quarter.
The risk score works. The deposit figure does not.
It is worth separating the two, because one of them is doing its job. We checked our own data rather than assuming. Grouping all 23 listings by the riskScore printed on each page, the median worst trough rises with every step:
| Risk score | EAs | Median worst trough | Range |
|---|---|---|---|
| 3 | 5 | 3.24% | 0.67% – 5.75% |
| 4 | 9 | 7.48% | 4.36% – 11.92% |
| 5 | 4 | 10.05% | 8.46% – 13.59% |
| 6 | 2 | 22.20% | 19.60% – 24.79% |
| 7 | 3 | 24.65% | 21.43% – 26.50% |
The ordering is real — the medians are monotone. Use the score to sort candidates. Do not use it to size, because the ranges overlap: a score of 4 covers everything from 4.36% to 11.92%, which is a factor of nearly three inside one label.
The deposit figure is the opposite case. It is identical on all 23 pages while the thing it is supposed to cover varies fortyfold. That is not a rounding problem; it is a number that has not been asked to do any work.
Check the record was produced at the size it recommends
One more check, and it takes ten seconds. The recommendation is only meaningful if the published record was produced at it.
Of the 23 listings, 23 ran on the recommended 10,000 deposit and 22 used the recommended 0.1 lots. The exception is Orrery, whose record was produced at 3.0 lots while the page recommends 0.1 — a thirtyfold difference on a JP225 index instrument. Its 0.67% trough is the trough of the 3.0-lot record. Follow the recommendation instead and you are running a position the published record does not describe, in either direction.
Fixed lots vs percent risk vs sizing from the record
Three approaches, and the one most articles recommend is the one that fits an EA worst.
| Approach | What it does | Where it breaks with an EA |
|---|---|---|
| Fixed lots | One size, never changes | Safe and honest, but the size is arbitrary unless you derived it from a trough |
| Percent risk per trade | Size scales to a stop-loss distance | Assumes a fixed stop. Many EAs exit on a condition, not a price, so “1% per trade” has nothing to attach to |
| Sizing from the record’s worst trough | Pick a size where the worst observed trough is money you can sit through | Needs the trade list, and gives you a bigger number than you wanted |
The middle row is the trap. Position sizing by percent-of-account is sound advice for a discretionary trader who sets a stop on every entry. An EA that manages its exit dynamically does not give you the distance the formula needs. Substituting the average loss understates the tail badly: across these records the worst trade is a median of 1.89× the average one. Iridescence is the extreme — its worst single trade cost 545.00 against an average loss of 53.40, 10.2 times larger.
The third row produces uncomfortable numbers, which is the point. If Chrysalis’s worst trough is 26.5% of a 10,000 account at 0.1 lots, then running it on 5,000 at the same size means the same sequence takes 53% — and no plan survives being half wrong before it is right.
Running the plan in practice
- Open the EA’s page and read its worst trough in money, its worst losing run and its longest underwater stretch from the record — not from the summary box.
- Double the worst trough. The record bounds what has happened, not what will. Six of these 23 listings already exceed a quarter of the recommended minimum once you double them; one exceeds half.
- Set your deposit so that doubled figure is money you can watch disappear without switching the EA off. If it is not, halve the lot size rather than hoping.
- Add the floating-equity margin. Every one of the 23 records has a deeper open-position drawdown than its closed-trade one — a median of 1.08× and up to 1.76× worse — because the balance only moves when a trade closes.
- Write down two stop conditions in numbers before you enable Algo Trading: a money figure that ends the test, and a number of weeks underwater that ends it. Decide them while you have nothing at stake.
Step 4 is the one people skip. Everything measured on this page comes from a closed-trade equity curve, and a closed-trade curve is blind between entries. The account that receives a margin call is looking at floating equity, which in these same records runs deeper in every single case. Treat the trough figures here as a floor.
The losing run matters more than the drawdown depth
A trough that recovers and relapses is survivable in a way that an uninterrupted one is not, because each winning trade in between gives you a reason to keep going. Nautical is the case worth studying: its worst losing run is 38 trades, and the money lost in that run — 2,464.64 — is exactly its whole worst trough. Every one of those 38 trades lost. Nothing interrupted it.
That is 38 consecutive opportunities to conclude the EA is broken, at a point in the record where the correct answer was to do nothing. Compare Cairn, which reached a similar trough of 2,479.13 with a worst run of 14 — a materially easier thing to sit through for the same money.
Risk, and the numbers that mean stop
Four failure modes, in the order they actually occur.
Sizing to the average. The average loss across these records is 0.66% of the recommended minimum at the median, which sounds like a comfortable 150 losing trades of runway. The worst single trade reaches 5.45% of the deposit. An account sized on the average meets the tail with a tenth of the buffer it assumed.
Intervening inside the drawdown. This is the expensive one, and the numbers above are the reason it happens: 38 losses in a row is long enough for any explanation to become plausible. The defence is not willpower. It is deciding the exit number in advance and sizing small enough that the number does not arrive early.
Running several EAs on one account and adding the troughs. Every figure above describes one EA measured alone. Nothing in these records says two of them cannot be underwater in the same week, and the naive assumption — that diversifying shrinks the drawdown — is the one worth testing before you rely on it. The only measurement we have of combining rules comes from a different population, the Builder template gallery in EXP-MACD-RECIPE-BASELINE-001: a portfolio of three rules reached a drawdown 3.76 times deeper than the single rule inside it, while its profit factor moved by 0.02. Three rules bought no improvement and nearly quadrupled the hole. Size a multi-EA account against the sum of the individual troughs until you have measured your own combination, not against the average of them.
Treating the backtest as the worst case. It is the best-documented case, not the worst. None of these records is a live-account run — the parent experiment counts zero verified live runs across the listings it measured — and a rolling backtest extends the record without ever testing execution, slippage or a broker outage. Doubling the trough is not pessimism. It is the minimum correction for measuring the future with a ruler built from the past.
Nothing here is financial advice, and no past drawdown bounds a future one.
Next steps
Read one record end to end before you size anything: browse the catalogue and open a listing whose risk score matches what you can sit through, because the trade list on the page is what these four numbers come from and you can check every one of them yourself.
From there:
- To see what the same records say about returns rather than sizing, are Expert Advisors profitable reads the catalogue for what “profitable” turns out to mean across a full record.
- If the EA you are sizing came from somewhere else, how to use a free EA covers the checks that come before this one — starting with whether anyone independent measured the numbers at all.
- To produce a trough figure for a rule of your own, run a backtest and read the closed-trade list rather than the summary line.
- How we test and score the listings referenced here is set out in our testing methodology.