Volatility targeting is the practice of scaling exposure up when the market is quiet and down when it is volatile, so that the strategy's own volatility stays near a chosen level. It is the portfolio-level cousin of setting stops in units of ATR, and it rests on the same observation: a fixed position size means a variable risk, because the market's movement varies.
How it works
Choose a target, say 10 per cent annual volatility for the account. Measure the recent volatility of the strategy's returns or of the markets it trades. Scale position sizes by the ratio of target to measured volatility. When volatility doubles, size halves; when it halves, size doubles, within limits.
Why it helps
Volatility clusters: high-volatility periods follow high-volatility periods, and they are also the periods in which most large losses occur. Scaling down as volatility rises reduces exposure in exactly the conditions that produce the deepest drawdowns, and scaling up in calm periods recovers the return that would otherwise be lost. Over a long run, most strategies show a better return per unit of drawdown with volatility targeting than without it.
How it goes wrong
Volatility rises during losses, so the method reduces size after a loss has already happened and can miss the recovery. In a sharp reversal, a strategy scaled down at the bottom recovers slowly. Rapid rescaling produces whipsaw, with sizes changing faster than the market; slower measurement reduces this at the cost of reacting late. And there is a ceiling on scaling up: doubling size in a calm market leaves the account exposed to the first shock, which arrives without warning. Sensible implementations cap the upward scaling and use a longer measurement window than instinct suggests.
A strategy targets 10 per cent volatility. Through a calm spring, measured volatility is 6 per cent and sizes run at 1.6 times base. In September a shock lifts measured volatility to 20 per cent within two weeks, and sizes fall to half base. The strategy's drawdown through the shock is 7 per cent; without targeting, at the 1.6 times sizing it had reached, it would have been about 22. The cost is a slower recovery in October, at half size, while the market rebounded.
An EA that reduces position size as volatility or drawdown rises is doing a form of this, and it is a reasonable design. What to check is the window and the cap: whether it reacts over days or weeks, and whether it scales up as freely as it scales down. Upward scaling without a cap is the version to be wary of, because it maximises exposure at the moment before the next shock.