Volatility: the strongest predictor of a liquidation

When we tested which metric best predicts whether an account gets liquidated, the answer was not ROI, not drawdown, not win rate. It was daily volatility - AUC 0.772. Drawdown came second at 0.715. Return itself scored 0.471, which is barely better than a coin flip.

That is the single most useful thing this data set says, and most leaderboards never say it.

What volatility means here

We take the standard deviation of the trader's daily ROI over 180 days. No annualisation, no smoothing - just how much the daily result jumps around.

• 1%/day: a quiet account. Almost no trading or a hedged book.
• 3%/day: normal for a careful swing trader.
• 5–10%/day: a normal high-return account.
• 10%+/day: a leveraged directional trader or a grid.
• 25%+/day: a grid, or a martingale that is close to failing.

Why volatility predicts liquidation

A wipeout happens when one day loses more than the margin can absorb. The size of the worst day is bounded by how much the account normally swings: an account with 5% daily volatility can have a −40% day; an account with 25% daily volatility can have a −95% day, and that is not bad luck, that is Tuesday.

Drawdown looks like the same thing but is partly a consequence: you cannot have a deep drawdown without volatility, so it correlates. Volatility is the earlier signal - it is high long before the drawdown is deep.

And ROI has no predictive power at all, which is uncomfortable if you have been sorting a leaderboard by return. High return is compatible with every outcome.

How the rating uses it

Volatility is penalised on a linear scale:

Daily volatility	Penalty
5%	0
10%	−4
15%	−8
25%	−12
40%	−15

Fifteen points is the third-largest penalty in the formula, after liquidation days and max drawdown. A trader at 40% daily volatility needs a near-perfect all-time ROI and a multi-year tenure just to reach the middle of the leaderboard - by design.

Volatility is also a follower problem

A follower does not experience the trader's volatility - they experience their own, which is the trader's volatility multiplied by copy friction:

• failed copies remove parts of the position at bad moments,
• slippage lands on the entries,
• per-order minimums change position sizes,
• a grid's re-entry sequence can trigger your daily loss limit and unsubscribe you mid-move.

So a 10% daily volatility trader is not a 10% daily volatility experience for you. It is worse.

How to use it before you subscribe

1. Sort the low-risk screen by volatility - /low-drawdown/ already requires ≤ 3%/day.
2. Compare two accounts with similar ROI. The one with lower volatility is the one whose followers are more likely to still be subscribed a year later.
3. Check volatility against tenure. High volatility with 3 years of trading days is a different object from high volatility with 200 days: the first has survived its own strategy several times.
4. Watch it move. Volatility that is doubling while the return stays flat is the classic pre-wipeout pattern. Our rating is period-independent by design (it always uses the last 180 days), so check the trend yourself across periods.

The honest limits

Volatility is calculated over 180 days. A trader who stopped trading for two months shows artificially low volatility - which is exactly why activity and tenure are also penalised in the rating, and why the review text tells you how many days were actually traded. A quiet number on an account with ten trades a year means nothing.

Past volatility does not predict future volatility. It is the best available prior for how this account behaves, and that is all.