Quant Memo
Advanced

What Neutralisation Costs You

Every neutralisation removes an unwanted exposure and, almost always, some real signal along with it. The judgment call isn't whether to neutralise but how much signal you're willing to sacrifice for how much cleanliness.

Prerequisites: Residualise the Signal or Constrain the Portfolio?

Neutralisation is usually framed as pure risk reduction — take out the exposure you don't want, keep the exposure you do. In practice it is never that clean, because the thing being removed and the thing being kept are rarely orthogonal. Sector, size, beta and the raw signal are all correlated with each other to some degree, and stripping one out drags some of the other along with it.

Where the cost comes from

If a value signal is genuinely more likely to flag stocks in cyclical, capital-intensive sectors — because that's where deep discounts to book value actually occur — then demeaning by sector doesn't just remove a market-timing bet, it also removes some of the reason the signal worked in the first place. The unwanted exposure and the real information were never fully separable; they were tangled together in the same raw number, and any neutralisation blunt enough to remove one will dull the other.

This is not an argument against neutralising. An un-neutralised sector or size bet is a real, uncompensated risk that most research desks are not mandated to take, and it makes the strategy's true stock-selection skill impossible to measure cleanly. It is an argument for treating the amount of neutralisation as a dial, not a switch, and checking what it costs before turning it up.

Measuring the cost, not just assuming it

The practical test is to compute the signal's information coefficient (or backtest Sharpe) before and after neutralisation, at a few different strengths — say, no neutralisation, partial (shrink the exposure toward zero rather than fully removing it), and full. A signal that loses almost nothing going from none to full neutralisation had very little of its power tied up in the unwanted exposure to begin with; a signal that collapses under full neutralisation was leaning on that exposure more than the researcher realised.

Neutralisation strengthTypical effect on ICTypical effect on risk cleanliness
NoneHighest raw IC, includes unwanted exposureBook carries sector/size/beta bets
Partial (shrink toward zero)Modest IC lossMost of the unwanted exposure removed
FullLargest IC loss if signal and exposure are entangledCleanest, but may over-correct

Neutralisation is a trade, not a free cleanup: it exchanges some amount of raw predictive power for a cleaner risk profile. The right amount is whatever a desk can defend, not automatically "as much as possible."

Partial neutralisation — shrinking an unwanted exposure toward zero rather than fully to zero — is often a better trade than the extremes. It captures most of the risk-cleanliness benefit while keeping more of the entangled signal, and the shrinkage strength itself becomes a tunable parameter worth testing rather than guessing.

Related concepts

Practice in interviews

Further reading

  • Chincarini & Kim, Quantitative Equity Portfolio Management
ShareTwitterLinkedIn