Price changes are not normally distributed. Moves of five or six standard deviations, which a normal distribution would produce once in thousands of years, happen every few years in most markets. The distribution has fat tails, and every risk calculation built on the assumption of normal returns underestimates what the tails can do.
Examples
The Swiss franc's rise of around 30 per cent in minutes in January 2015 when its cap was removed. The March 2020 collapse, when stock indices fell 10 per cent in a day and gold and bonds fell alongside them. Flash crashes in the pound and the yen that moved several per cent and reversed within an hour. Overnight gaps in indices on news released while the exchange was shut. None of these appears in the ordinary run of daily volatility, and any of them can arrive in a strategy's next week.
Why stops do not fully protect
A stop loss becomes a market order when triggered, and fills at the next available price. In a gap or a liquidity vacuum the next price can be far beyond the stop. A 50-pip stop on the franc in January 2015 was filled thousands of pips away. Stops limit ordinary losses. They do not limit the extreme ones, and a position size that assumes they do is sized for a world with thin tails.
Correlated tails
Extremes arrive together. Assets that are uncorrelated in ordinary conditions fall together in a crisis, so a portfolio that looks diversified on its normal correlations is, in the tail, one large position. Strategies that sell volatility or hold carry, which earn steadily in calm markets, are the ones that lose most when the tail arrives, because their edge was the tail's absence.
Sizing for it
Risk per trade should be set so that a loss several times the stop distance, across every open position at once, is survivable. Total open risk should be capped with that in mind. Leverage should be kept well below the regulatory maximum, since margin is what disappears in a gap. And the account should hold no strategy whose entire edge depends on extremes not happening.
A trader holds five positions, each risking one per cent to its stop, in markets that are uncorrelated on the last year's data. A shock hits and all five gap through their stops by three times the stop distance. The loss is 15 per cent, not 5. The trader who had capped total open risk at 3 per cent lost 9. Neither had done anything wrong by the record. One had planned for the day the record did not contain.
An EA's record, however long, has met only the tails that occurred during it. Check how it sizes total exposure, whether its drawdown controls halt new trades early, and whether any part of its strategy earns its return from calm. Then set the risk so that the account survives a day three times worse than the record's worst, because over enough years, that day is not a possibility but a schedule.