Diversification: What It Really Does — and What It Never Will

Key Takeaways

What it is: diversification reduces a portfolio's risk without reducing its expected return – the only "free lunch" in finance, in the phrase attributed to Harry Markowitz, who formalized the mechanics in 1952. The engine is not the number of holdings: it is the correlation between them.
What it does: it eliminates specific risk (of one company, one building, one issuer), which the market does not pay for. It also protects against an underrated danger: missing the rare big winners – across nearly a century of US data, about 4% of stocks account for all net stock market wealth creation.
What it does not do: it does not raise expected returns, does not remove market risk, and its effectiveness fades precisely in crashes, when correlations rise – a result documented by the research on extreme correlations.
The most common trap: confusing the number of holdings with real diversification. Ten products exposed to the same economic engine form a single bet; a few genuinely independent exposures diversify more than a hundred stocks in one sector.
The real cost: a diversified portfolio always underperforms its best holding and always contains one disappointing position. That cost is psychological, not financial – and it is what makes investors abandon the discipline at the worst moment.
The key reflex: judge a portfolio not by its number of positions, but by the number of independent bets it contains – and by what it would become in a scenario where correlations converge.

Introduction

It is probably the most repeated piece of financial advice in the world: "don't put all your eggs in one basket." It is also one of the most misunderstood. Many investors think diversifying means multiplying holdings; others read it as a promise of protection in all circumstances; still others suspect it of diluting performance – "diversification is for people who don't know what they're doing."

Financial research has precise answers to all three intuitions, and they surprise: diversification does not depend on the number of holdings, it does not protect in all circumstances, and it does not dilute expected returns. What it actually does – and what it never will – can be described with a rigor that owes little to intuition.

This article walks through the mechanics: where the idea comes from, what diversification really accomplishes, how many positions it takes, why it weakens in crises, and the traps that turn a seemingly diversified portfolio into a concentrated bet. It complements our article on quantitative ratings, which dealt with selecting securities; this one deals with assembling them.

Where the Idea Comes From: Markowitz and the Discovery of Covariance

Before 1952, investment theory looked at securities one at a time: find the "good" ones, avoid the "bad" ones. Harry Markowitz's "Portfolio Selection" moved the question: what matters is not each security's risk in isolation, but its contribution to the risk of the whole – and that contribution depends on the security's covariance with the rest of the portfolio far more than on its own volatility.

The intuition fits in a tiny example. Two investments with identical 10% volatility, combined in equal parts:


Correlation between the two investments	Portfolio volatility
+1.0 (perfectly correlated)	10.0%
+0.5	8.7%
0.0 (independent)	7.1%
−1.0 (perfectly opposed)	0.0%

The portfolio's expected return is the same on all four lines: the average of the two investments'. Only the risk changes. That is the precise meaning of the "free lunch": with expected return unchanged, the dispersion of outcomes shrinks – provided correlation is below 1. Diversification creates no return; it removes unnecessary randomness.

Generalizing: in an equal-weighted portfolio, as the number of positions grows, the securities' individual variances matter less and less, and the portfolio's variance converges toward the average covariance between the securities. This purely arithmetic result contains both the promise and the limit of diversification: the risk specific to each holding can be made to disappear; what the holdings have in common cannot.

What It Really Does: Eliminate Unrewarded Risk

The central distinction separates two layers of risk. Specific risk – the factory fire, the lost lawsuit, the failed product – hits one company without hitting the others: it dilutes away with numbers. Systematic risk – recession, rate shock, financial crisis – hits everyone at once: no number of holdings removes it.

That distinction carries a deep economic consequence, formalized by William Sharpe in the capital asset pricing model: since specific risk can be eliminated for free through diversification, the market does not pay for it. The concentrated investor therefore carries extra risk with no extra expected return in exchange. Concentration is not boldness compensated with higher expected returns; it is free risk in the least flattering sense of the word.

Recent research added an even more striking argument. Examining nearly every listed US stock between 1926 and 2016 (close to 26,000 securities), Hendrik Bessembinder showed that the net stock market wealth created in excess of cash came from about 4% of stocks; the remaining 96%, taken together, collectively did no better than one-month Treasury bills. More than half of all stocks returned less than cash over their entire lifetimes. The distribution of long-term returns is violently asymmetric: a few immense winners, a majority of disappointments.

The lesson for diversification is direct: a concentrated portfolio has a high probability of not containing the rare stocks that will drive the market's performance. Diversifying also means securing a share of the winners nobody can identify in advance – insurance against one's own selection.

How Many Holdings Does It Take? The Wrong Question

The classic literature long searched for the "right number" of stocks. The pioneering work of Evans and Archer concluded that about ten positions eliminated most specific risk; Meir Statman redid the math with costs included and concluded at least thirty or so. The numbers vary; the real lesson lies elsewhere: the number of holdings is a poor indicator of diversification.

What counts is the number of independent bets. Three examples make the point:

Ten Swiss residential real estate funds form, essentially, a single bet – on Swiss rents, rates, and property prices. Our article on the residential and commercial segments shows how much vehicles within one segment share the same drivers.
A hundred stocks in the same sector diversify company risk, not sector risk: the average covariance stays high, and it is what sets the risk floor.
Several funds with different names can hold the same underlying assets: stacking products creates an illusion of variety that position-level transparency dissolves – a portfolio is judged on consolidated holdings, not on wrappers.

The best-documented bias here is home bias: Kenneth French and James Poterba showed that in the late 1980s, American and Japanese investors held more than 90% of their equities in their own country's stocks – massive geographic concentration with no justification from expected returns. The phenomenon persists everywhere in attenuated form: we diversify poorly what we know well, precisely because we feel we know it.

Over-diversification exists too: beyond a point, each additional holding brings no measurable risk reduction but adds costs, complexity, and dilution of any analysis. Between ten genuinely independent bets and three hundred redundant positions, the former diversify better.

What It Will Never Do

Four limits define the flip side of the free lunch.

It does not raise expected returns. Diversification acts on the dispersion of outcomes, not on their average. A diversified portfolio is not "better" in expectation; it is more predictable around the same expectation. Anyone promising extra return from diversification alone is selling something else.
It does not remove market risk. Average covariance is a floor. In a global recession, a perfectly diversified portfolio falls – less than a portfolio concentrated in the worst sector, but it falls. Diversification is a shock absorber, not a parachute.
It weakens in crashes. François Longin and Bruno Solnik established a disturbing result by studying extreme correlations between equity markets: correlations increase in sharp downturns, but not in sharp rallies. The asymmetry is exactly the one nobody wants: markets diverge when things go well and converge when things go badly. Average correlation measured in calm times therefore overstates the protection available in a crisis.
It does not protect against regimes that hit everything at once. Some environments – inflation and rate shocks first among them – punish most nominal assets simultaneously, stocks and bonds included. Diversification across assets that depend on the same discount rates reaches its structural limit there; financial history offers several such episodes.

To these limits, add a cost of a different nature: regret. A diversified portfolio, by construction, always does worse than its best holding – and always contains a losing position you tell yourself you could have skipped. That torture of comparison is the psychological price of the free lunch; paying it without deviating may be investing's most underrated skill. Backtests capture none of it: a simulated curve does not sweat, as our article on backtest biases points out.

In Practice: Think Crisis Correlations, Not Average Correlations

Three disciplines follow from the above.

Count bets, not holdings. Group the portfolio by economic engine – Swiss growth, rates, real estate, the dollar, crypto – and count the genuinely independent groups. The resulting number is usually far smaller than the number of positions, and it is the one that describes the risk.
Test for convergence. The useful question is not "what is my average correlation?" but "what happens to my portfolio if correlations head toward 1, as in the documented extreme episodes?" That scenario, not the historical average, sizes the possible loss.
Rebalance with discipline. Diversification is not a state but an upkeep: winning positions mechanically grow until they re-concentrate the portfolio. Periodic rebalancing – selling what has risen, adding to what has fallen – maintains the intended structure, at the price of systematic discomfort: it requires doing the opposite of what instinct suggests.

None of these disciplines requires forecasting anything. That is their strength: diversification is one of the few risk management tools that works without an opinion about the future – provided its limits are accepted along with its benefits.

Conclusion

Diversification does one thing, does it well, and does only that: it eliminates the risk the market does not pay for – the risk of individual holdings – and reduces the portfolio to its genuinely shared exposures. It promises no extra return, no protection in the storms that sweep everything away, and it demands in exchange a permanent tolerance for the regret of never being fully invested in the year's best asset.

That is both less than the marketing claims and more than the prosecution alleges: a free, robust tool that depends on no forecast – in a business where almost everything else does. The question to ask of any portfolio remains the one from the start: how many independent bets does it really contain, and what happens on the day they stop being independent?