Volatility Forecasts as a Sizing Input, Explained

Two names can carry the same signal score and be completely different bets, because one is twice as volatile as the other. A fixed-dollar position in each contributes very different amounts of risk to the book. Volatility forecasts fix this by feeding directly into position size, so that "same conviction" translates into "same risk contribution," not "same dollar amount."

Sizing by forecast risk, not by fiat

The standard approach scales a position inversely with forecast volatility. If a name's forecast volatility doubles, its position size roughly halves, so that the position's expected contribution to portfolio risk stays about the same as before. A simple version of this is

$w_i = \frac{s_i}{\sigma_i}$

where $w_i$ is the position weight for name $i$ , $s_i$ is the raw signal score, and $\sigma_i$ is the forecast volatility for that name. In words: divide conviction by forecast risk, so a high-conviction call on a wild stock ends up sized similarly, in risk terms, to a similarly high-conviction call on a calm one.

This matters because volatility forecasts are, as a rule, considerably more reliable than return forecasts. Volatility clusters and is strongly autocorrelated, so a forecast built from recent realized volatility captures most of what's predictable about it. That reliability is exactly why it's a good sizing input: you're using the part of the problem you can actually forecast well to control the part — how much to bet — that has the biggest, most mechanical effect on risk, while leaving the shakier return forecast to do only the job it's suited for, deciding direction and relative conviction.

Volatility forecasting and return forecasting play different roles in the same strategy: the return forecast decides which names to bet on and roughly how hard; the volatility forecast decides how many dollars that conviction should actually translate into, so risk — not dollar exposure — is what's evenly spread across the book.

Same conviction, different volatility — inverse-vol sizing shrinks the position on the noisier name so both contribute similar risk.

A worked example

Two names both score $s = 1$ on a signal. Stock A has forecast annualized volatility of 20%; stock B has forecast volatility of 40%. Using $w_i = s_i / \sigma_i$ : stock A gets a weight proportional to $1/0.20 = 5$ , stock B gets a weight proportional to $1/0.40 = 2.5$ — half as large. Stock A ends up with twice the dollar position of stock B, but each contributes roughly the same amount of risk to the portfolio, because stock B's larger per-dollar volatility is offset by its smaller position size.

Inverse-volatility sizing assumes the volatility forecast is reasonably accurate and doesn't itself lag a genuine regime shift. Sizing off a stale, too-low volatility estimate right before a spike is a classic way this approach turns a small position into an accidentally large risk contributor.

Volatility Forecasts as a Sizing Input

Sizing by forecast risk, not by fiat

A worked example

Related concepts

Practice in interviews

Further reading