Diversification: What Spreading Holdings Does and Does Not Remove

Diversification means holding a number of different things rather than one. The reasoning is old and intuitive, and the useful part is more precise than the intuition: risk divides into two kinds, and diversification only addresses one of them.

The two kinds

Specific risk belongs to one company. A factory fire, a failed trial, an accounting scandal, a lost contract, a chief executive resigning. These events affect that company and are largely unrelated to what happens elsewhere.

Systematic risk belongs to the market. Interest rate changes, recessions, inflation, currency shifts, broad shocks. These affect essentially everything at once, in varying degrees.

Holding thirty companies instead of one dilutes specific risk substantially: a disaster at any single one is a thirtieth of the total rather than all of it. Holding thirty companies does approximately nothing about systematic risk, because when the market falls broadly, the thirty tend to fall together. That is the measure beta describes, and it is the residual that spreading holdings cannot remove.

How many is enough

The reduction in specific risk is steep at first and flattens quickly. Moving from one holding to ten removes a large proportion of it. Moving from ten to thirty removes a good deal more. Moving from thirty to a hundred removes comparatively little, and past that the curve is close to flat.

Studies over the decades have put the point of diminishing returns in the twenty to thirty range for broadly chosen holdings, with the caveat that the figure rises when the holdings are more volatile or more correlated with each other. It is not a threshold, it is the shape of a curve.

The correlation trap

This is where nominal diversification most often fails. Thirty holdings are only diversified if they respond to different things.

Thirty large cap US technology companies is not a diversified set in any meaningful sense. They share customers, they share supply chains, they share sensitivity to interest rates, and they respond to the same news. The count is thirty and the exposure is closer to one.

The same applies across categories that sound different but are not. Bank shares and property shares both respond strongly to interest rates. Airlines and cruise operators both respond to fuel costs and discretionary spending. Two names, one exposure.

Correlations move

The harder problem is that correlation is not a fixed property. It changes, and it changes in the direction that hurts.

During calm periods, different sectors and asset classes move somewhat independently, and a spread of holdings behaves as intended. During sharp market wide declines, historical episodes have repeatedly shown correlations rising toward one, meaning things that normally move separately fall together.

The uncomfortable implication is that diversification tends to be least effective exactly when it is most wanted. It is a real effect, well documented across multiple market stress episodes, and it does not make diversification useless. It makes it a reducer of one risk rather than a removal of risk.

What over diversification costs

Beyond the flat part of the curve, adding holdings mostly adds work. More positions to follow, more filings, more news, more transactions. A set of holdings large enough that none of them is understood is not obviously safer than a smaller set that is, and the risk of not knowing what you own is real even though it does not appear in any volatility calculation.

Checking whether holdings actually differ

A practical use of a per stock reading across a set of holdings is checking whether they are behaving as one thing or as several. If every holding carries a similar bias reading and they all move together, the spread on paper is not producing much separation in practice.

EquityBias reads four dimensions per covered stock, price behaviour, fundamentals, analyst activity and news sentiment, and reports where they disagree rather than averaging them. Related: beta for market relative movement, and drawdown for the worst case measure. Coverage by sector is in the coverage directory.

EquityBias is a market data research tool. Nothing here is financial advice. Diversification does not ensure a profit or protect against loss.