Risk: what actually ends trading accounts

The four risks that matter

Textbooks call risk the standard deviation of returns. That definition is good for math and useless for survival. Accounts end in four ways.

Drawdown. The drop from your equity peak. This is the honest measure of pain. A 20 percent drawdown is twice as hard to hold as a 10 percent one, whatever the annual return. And traders do not quit because the math broke. They quit in the middle of a drawdown, usually near its bottom, which converts a temporary loss into a permanent one.

Ruin. Losing enough that recovery is unrealistic. The arithmetic is cruel and most beginners have never run it. A 50 percent drawdown needs a 100 percent gain to get back to even. A 75 percent drawdown needs 300 percent. The deeper the hole, the harder the climb, and the climb is being attempted by a trader with less money and less confidence than the one who dug the hole. Ruin does not mean zero. It means enough.

Correlation. Five “different” strategies that all lose in the same week were one strategy with five names. The diversification was an illusion, and the illusion is revealed at the worst possible time, because correlated losses are what reveal it. See portfolio construction.

Tails. The event your backtest never contained. Crypto produces them on a schedule: flash crashes, exchange failures, funding spikes, liquidation cascades. You cannot price them precisely. You can size so that they hurt instead of kill. That is the whole game with tails.

The metrics, in one breath each

• Max drawdown. Worst peak-to-trough drop. The most useful single number.
• Calmar. Annual return divided by max drawdown. Reward per unit of pain.
• Volatility. How bumpy the ride is. Says nothing about surviving it.
• Sharpe and Sortino. Return per unit of bumpiness. Useful, gameable. See statistical tools.
• VaR. “95 percent of days lose less than X.” Fine. The account is ended by the other 5 percent, about which VaR says nothing.

No single number is enough. A strategy can pass four of these and fail the fifth, and the fifth is the one that gets you.

Sizing is where risk is actually controlled

Entries and exits decide whether the strategy makes money. Sizing decides whether you are still present when it does. The standard methods:

• Fixed notional. Same dollars every trade. Simple, and blind. The same $1,000 is a different bet on a quiet day and a violent one.
• Fixed fractional. Risk a set percentage of current equity per trade. The account breathes. Ruin gets mathematically harder.
• Volatility-targeted. Size down when the market is wild, up when it is quiet, so each trade carries similar risk. The professional default.
• Kelly. The optimal size if you know your edge exactly. You do not. You have an estimate from a backtest, and estimates flatter. Full Kelly on a flattering estimate is a fast road to ruin. Half or quarter Kelly is the honest version, and even that assumes more than most backtests can deliver.

Leverage and liquidation

Leverage adds no edge. It multiplies the outcome of whatever edge or noise is already there. A losing strategy levered loses faster. A winning strategy with deep drawdowns gets liquidated before the winnings arrive, which makes it a losing strategy with better marketing.

On perps the floor is hard. A drawdown can recover. A liquidation cannot. Size against your worst realistic drawdown, not your average one, because the liquidation engine only cares about the worst one. And remember funding: a levered position pays or collects every few hours, and the direction of that payment can flip in a week. See spot vs. perpetual futures.

Crypto is its own animal

Coming from stocks or FX, recalibrate three things. Volatility is several times higher, so imported sizing rules run hot. Venues themselves are a risk: outages, frozen withdrawals, depegs, none of which appear in backtests, all of which appear in life. And funding regimes rotate in months, so a strategy's cost structure can invert while its rules stay the same.

How Edgecraft handles this

Edgecraft backtests include funding, slippage, and partial fills by default, so the drawdown you see in testing resembles the one you will feel. Portfolio tools check correlation and concentration before deployment, when fixing them is cheap. Live strategy health analysis compares your real drawdown against the backtest's envelope and flags the moment reality leaves the range the evidence promised. The flag arrives by statistics, not by panic, which is the only way it is useful.