Neutralising Against Size and Liquidity

Small stocks are, on average, less efficiently priced, more volatile, and more expensive to trade than large ones. Any signal that even mildly prefers small-cap names — and many do, without anyone designing them to — will pick up a chunk of return that has nothing to do with the signal's actual insight and everything to do with small stocks being harder to arbitrage and harder to trade. Neutralising against size and liquidity is how a researcher checks whether the signal survives once that free lunch is taken off the table.

What's being controlled for

Size (usually log market cap) and liquidity (usually average daily dollar volume, or the number of days it would take to trade a target position without moving the price much) are correlated with each other but not identical — a mid-cap can be illiquid because of a concentrated shareholder base even if its market cap looks unremarkable. Both matter for a different reason: size proxies for how much the stock is followed and arbitraged, and liquidity proxies for how much of any apparent edge would survive transaction costs.

Neutralising means computing the signal's score, then removing (again, typically by cross-sectional demeaning within size/liquidity buckets, or by regression) the portion of the score that is just a restatement of "this stock is small" or "this stock is illiquid." What remains is the part of the signal that discriminates between stocks of similar size and similar liquidity — a much harder bar to clear, and a much more honest one.

Why the backtest can lie without this step

Historical backtests routinely underprice the cost of trading illiquid small caps, because standard cost models are calibrated on average costs, not on the elevated costs of trading concentrated positions in thin names at size. A signal that ranks small, illiquid names highly will show a great gross Sharpe ratio in a backtest and a much worse — sometimes negative — Sharpe ratio once realistic costs and capacity limits are applied. Neutralising against size and liquidity at the research stage catches this before capital is committed, rather than after.

Without neutralisation	With neutralisation
Signal free to prefer small/illiquid names	Signal must discriminate within a size/liquidity band
Backtest Sharpe can reflect small-cap premium, not skill	Backtest Sharpe reflects genuine cross-sectional insight
Capacity typically overstated	Capacity closer to what the book could actually trade