Diversification: How Many Stocks Does a Portfolio Need?

The evidence on diversification does not produce a magic number. What it consistently shows is a curve of diminishing returns: the reduction in stock-specific risk is steep across the first several holdings, noticeably slower after that, and eventually close to flat. Where the curve flattens depends on how correlated the holdings are and how evenly the capital is spread, which is why published estimates range from a couple of dozen names to well over a hundred.

The more useful framing is that the number of holdings is a poor summary of how diversified a portfolio is. It is one input among several, and on its own it can be badly misleading.

The two risks being separated

A holding’s price movement can be split, conceptually, into two parts.

Market risk, sometimes called systematic risk, is the part that moves with everything else. A rate decision, a global growth scare, a currency shock or a broad change in risk appetite affects most listed companies at once, in the same direction, to varying degrees. No amount of adding more stocks from the same market removes this.

Stock-specific risk, also called idiosyncratic or diversifiable risk, is the part attributable to the individual company. A plant fire, a promoter dispute, a failed product launch, an accounting problem, a regulatory action against one firm. These events are largely unrelated across companies, which is exactly why they average out as holdings are added.

Diversification is the mechanism that reduces the second category. Everything the number-of-holdings question is really asking is: how many holdings does it take before the second category stops being the dominant source of portfolio volatility.

What the shape of the evidence actually says

Studies going back to the 1960s have measured this by building portfolios of increasing size at random and observing how portfolio volatility behaves as holdings are added. The consistent finding across markets and decades is the shape rather than the level.

Going from one holding to two removes a great deal of risk. Going from two to five removes a lot more. Going from five to fifteen still helps substantially. Somewhere after that the marginal contribution of each additional name becomes small, and the curve asymptotes toward the volatility of the market itself, which is the floor that diversification within one equity market cannot go below.

The disagreement in the literature is about where “small enough” sits, and that disagreement is genuine. Later work argued that early studies understated the number needed, for reasons worth understanding rather than memorising:

• The benchmark for “diversified enough” is a judgement. Reducing diversifiable risk to a level most researchers would call acceptable takes far fewer holdings than reducing it to near zero, and different papers pick different thresholds.
• Average outcomes hide the spread. A portfolio of twenty randomly chosen stocks has an expected volatility close to a larger portfolio’s, but the range of outcomes across such portfolios is much wider. Some twenty-stock draws end up much worse than the average, and an investor experiences one draw, not the average.
• Correlation assumptions drive everything. The faster diversifiable risk disappears, the lower the average correlation between holdings. Markets and periods differ on this.
• Return distributions have fat tails. When individual stock outcomes include rare, very large moves, more holdings are needed before those tails stop dominating.

None of these debates change the practical takeaway, which is that the first several holdings do most of the work and additional holdings past a point buy progressively less.

Why the count is a weak measure

Three portfolios can hold thirty stocks each and be diversified to completely different degrees.

Correlation. Thirty holdings across public sector banks, private banks and non-bank lenders share one dominant driver: the credit and rate cycle. Thirty holdings spread across unrelated business models share far less. What matters is how much the holdings move together, not how many there are. Reading that structure is what a correlation matrix is for, along with the awkward fact that correlations between holdings tend to rise in exactly the stressed conditions where diversification is most wanted.

Weight distribution. A portfolio with thirty names where three positions hold half the capital behaves much more like a three-stock portfolio than a thirty-stock one. The remaining twenty-seven are, in risk terms, decoration.

This second point has a standard fix. Concentration measures such as the Herfindahl index square each position weight and sum them, and the reciprocal of that sum gives an effective number of holdings. An equally weighted thirty-stock portfolio has an effective number close to thirty. The lopsided version above might have an effective number in the single digits. The gap between the headline count and the effective count is often the most informative single number about a portfolio’s real spread, and how to compute and read it is covered in concentration risk in portfolios.

Shared factor exposure. Even holdings in different sectors can share a common driver: all expensive growth names, all leveraged balance sheets, all exporters exposed to one currency pair. Sector labels can conceal this entirely.

The costs on the other side

Adding holdings is not free, and the costs are mostly not the trading kind.

Attention is finite. Every holding requires reading results, following filings, tracking guidance and reassessing the thesis. A team that can genuinely follow twenty businesses cannot genuinely follow ninety, and holdings nobody follows are risk positions nobody has evaluated. The practical mechanics of keeping up with a book are covered in how to monitor a portfolio of holdings.

Dilution of research conviction. If a process produces a ranked list, positions far down the list get the same benefit of diversification as positions near the top while contributing less expected value. Adding holdings past the point where the research has anything distinctive to say converts an active portfolio into an expensive approximation of an index.

Operational and liquidity friction. More holdings means more rebalancing trades, more corporate actions to track, more tax lots, and in smaller companies, more positions that cannot be exited quickly.

Mandate constraints cut both ways. Many professional mandates impose minimum diversification, single-name caps or sector caps. Those constraints often decide the count before any risk analysis does.

The tension is straightforward to state and impossible to resolve in general: diversification reduces the damage any single mistake can do, and concentration is what makes correct judgement pay. Different investors sit at different points, and both ends of the range are defensible when the choice is deliberate. How the count interacts with sizing and constraints is part of portfolio construction basics.

What a holdings count does not tell you

• It says nothing about market risk. Diversification within one equity market removes company-specific risk and leaves market risk fully intact. A broadly spread Indian equity portfolio still falls in a broad Indian market fall.
• It says nothing about quality. Fifty poorly researched holdings are diversified and still poor. The count measures spread, not merit.
• Randomly built portfolios are not real portfolios. The classic studies assemble holdings at random. Real portfolios are assembled by a process that deliberately selects similar characteristics, which usually raises correlation and means more holdings are needed to reach the same effect.
• The evidence is period dependent. Correlations, sector composition and the market’s own concentration all change over time, so a conclusion drawn from one decade’s data does not automatically carry to another.
• It ignores what sits outside the equity book. Diversification across asset classes, geographies and currencies is a separate question that a single-market stock count cannot address.
• Effective diversification drifts. A portfolio built evenly becomes uneven as prices move, so today’s effective number of holdings is not the one it was built with. Tracking that change is the subject of measuring portfolio drift.

The most defensible position the evidence supports is a modest one. There is a steep early benefit to spreading capital across genuinely different businesses, a rapidly shrinking benefit after that, and a hard floor set by market risk that no count reaches past. Beyond that, the count is far less informative than the correlation structure, the weight distribution and the honest question of how many businesses can actually be followed properly.

Related reading

• Portfolio metrics explained: the hub covering the full set of portfolio measures and how they connect.
• Concentration risk in portfolios: measuring concentration by position, sector and factor, including effective number of holdings.
• Reading a correlation matrix: why correlation, not count, decides how much diversification a portfolio really has.
• Portfolio construction basics: how sizing, constraints and diversification fit together when a portfolio is built.
• How to monitor a portfolio of holdings: the practical limit on how many businesses can be followed properly.

This article is educational. Altys Labs is not a registered research analyst or investment adviser, and nothing here is investment advice or a recommendation to buy, sell, or hold any security.