A portfolio holding twenty stocks reads as diversified to the eye. Whether it is diversified in any meaningful sense depends entirely on the correlation structure among the twenty. A portfolio of twenty US large-cap technology stocks is a concentrated bet on one sector, held twenty ways. A portfolio of twenty stocks selected from ten uncorrelated industries carries substantially more diversification than the raw count would suggest. Understanding this distinction is one of the most consequential portfolio-construction lessons.
The mathematics of diversification
The variance of a portfolio of N assets depends on two things: the average variance of the individual assets, and the average covariance among them. As N grows, the individual-variance term shrinks (weighted by 1/N), but the covariance term does not — it depends on the average correlation among the holdings.
This has a specific implication. In a portfolio where all assets are perfectly correlated (correlation = 1), adding more assets does not reduce portfolio volatility at all. The portfolio variance equals the individual variance regardless of N. In a portfolio where all assets are uncorrelated (correlation = 0), the portfolio variance shrinks as 1/N — a portfolio of 100 uncorrelated assets has 1% of the variance of the individual assets.
Real portfolios sit somewhere between these two extremes. The correlation among typical US large-cap equities averages around 0.4 in normal periods. This means a 20-stock portfolio of typical large-caps has meaningfully less variance than any individual stock, but the reduction plateaus quickly — going from 20 to 100 stocks removes very little additional variance, because the average correlation is doing most of the work.
The number that matters
The concept of "effective number of independent bets" captures the diversification question better than the raw count. A 20-stock portfolio with an average correlation of 0.5 has roughly 2 effective independent bets. A 20-stock portfolio with an average correlation of 0.1 has roughly 8. The difference in real diversification is dramatic.
Most retail portfolios calculated this way have very few effective independent bets. Five megacap tech names have an effective count near 1. Twenty US large-caps split across sectors might have an effective count of 3 to 5. A global multi-asset portfolio (stocks + bonds + international + real assets) can have an effective count of 8 to 12, depending on the specific allocation.
The insight is not that count is worthless — more positions do help — but that the marginal diversification benefit of adding a position depends heavily on how correlated it is to what you already have.
Correlations shift under stress
The complicating pattern is that correlations are not stable. They rise sharply during periods of market stress, precisely when the diversification is most needed. Assets that show a 0.3 correlation in normal periods often show correlations of 0.7 or higher during crises.
The 2008 financial crisis is the classic example. In late 2008, correlations across virtually every risky asset class rose toward 1. US stocks, international stocks, emerging markets stocks, high-yield bonds, commodities, and even some traditionally uncorrelated alternative asset classes all moved together. The diversification benefit that the portfolios had shown in the pre-crisis period evaporated in the moment when it was most needed.
The March 2020 pandemic sell-off produced a similar though shorter pattern. Bonds initially rose (providing a genuine hedge), but as forced liquidations spread, even Treasury bonds sold off along with stocks for a period, breaking a correlation pattern that had held for decades.
The lesson is that portfolio diversification analysis based on long-run average correlations understates the risk during crises. Stress-period correlations are higher, and any risk analysis that ignores this understates portfolio vulnerability.
Sources of genuine diversification
Given that correlations shift, what actually provides diversification that survives stress?
Different asset classes with different fundamental drivers. Bonds and stocks are driven partly by shared macro factors but partly by different ones (bonds by rates and inflation, stocks by earnings and multiples). Their long-run correlation has been positive in some periods and negative in others, but in most stress periods, high-quality bonds have moved differently from stocks.
Different geographies with different macro exposures. US, European, and emerging-market equities have different fundamental drivers even as global capital flows tend to correlate them in the short term. Over long horizons, the correlation is meaningful but not overwhelming.
Different factors within equities. Value and momentum, quality and small-cap — these factor exposures often produce different returns even when applied to the same underlying universe.
Real assets. Real estate, infrastructure, commodities, and inflation-linked bonds have historically produced different return patterns from equities and nominal bonds during specific macro regimes (particularly inflation shocks).
Not all of these hold up equally in every stress scenario. Bonds provided diversification in the 2008 crisis but were less helpful in 2022. Commodities helped in the 2022 inflation shock but hurt in the 2008 deflationary shock. The pattern that emerges is that genuine diversification requires multiple sources of independent return, each of which may fail in some scenarios, with the aggregate producing more consistent performance than any single source.
The concentrated-portfolio question
For investors who prefer concentration — a smaller number of high-conviction positions — the correlation question becomes more urgent, not less. A 10-stock portfolio with an average correlation of 0.7 is effectively a 3-bet portfolio. If the conviction on those bets is genuinely strong, the concentration may be justified; if not, the return distribution is much wider than a naive read of the count suggests.
The best framework for a concentrated portfolio is to identify the specific independent bets it represents rather than counting positions. If your concentrated portfolio represents 3 independent economic bets held with high conviction, that is a coherent portfolio structure. If it represents 3 independent bets held with modest conviction, plus 7 positions that correlate closely with them, that is a portfolio with an unclear risk structure.
The rule to internalise
Position count is a weak proxy for diversification. What matters is the correlation structure among the positions and the number of effective independent bets they represent. Most retail portfolios have far less effective diversification than their holder assumes, and the diversification they do have often deteriorates during exactly the periods when it is most needed. Building portfolios around genuine sources of independent return — not just around adding more names — is the discipline that produces diversification that holds up under stress.
Educational content only. Not investment advice.