What is Diversification?

By EC Assets Research Team · Published 17 August 2026 · Updated 17 August 2026

Diversification: Combining assets whose returns do not move together, so that portfolio risk falls below the weighted average risk of the holdings. The benefit is governed by correlation rather than by the number of positions, and it stops at a floor set by the risk common to everything held.

What Diversification Actually Measures

Diversification is the reduction in portfolio risk that comes from holding assets whose returns are imperfectly correlated. It is often described as not putting everything in one place, which is true and unhelpful: the mechanism is not the count of holdings but the degree to which they fail to move together.

The effect is unusual in finance because it is not a trade. No premium is paid, no counterparty takes the other side, and the expected return of the portfolio is simply the weighted average of its parts. Only the risk falls. That asymmetry is why diversification is routinely described as the one free lunch available to investors — and why the conditions under which it stops working deserve more attention than the benefit itself.

How It Works

For two assets with weights $w$ and $1-w$, volatilities $\sigma_A$ and $\sigma_B$ and correlation $\rho$:

$$\sigma_p = \sqrt{w^2\sigma_A^2 + (1-w)^2\sigma_B^2 + 2w(1-w)\rho,\sigma_A\sigma_B}$$

The correlation term is the entire mechanism. At $\rho = 1$ the expression collapses to the weighted average of the two volatilities and there is no benefit at all. As $\rho$ falls, the third term shrinks and portfolio risk drops below that average. At $\rho = -1$ a particular weighting eliminates volatility altogether.

For $N$ equally weighted assets sharing a common volatility $\sigma$ and pairwise correlation $\rho$, the expression simplifies to:

$$\sigma_p = \sigma\sqrt{\rho + \frac{1 - \rho}{N}}$$

As $N$ grows the second term vanishes and portfolio volatility approaches $\sigma\sqrt{\rho}$. That limit is the systematic risk floor: the portion of risk shared by every holding, which no amount of further diversification removes.

Portfolio volatility by number of holdings

Each holding carries 25 percent volatility with pairwise correlation of 0.3. About 85 percent of the achievable benefit arrives by the tenth position (and about three-quarters by the fifth), and the floor at 13.7 percent is the systematic component that cannot be diversified away.

Worked Example

An allocator combines an equity sleeve at 18 percent volatility with a short-duration credit sleeve at 8 percent, correlation 0.2, equally weighted.

$$\sigma_p = \sqrt{0.25(324) + 0.25(64) + 2(0.25)(0.2)(18)(8)} = \sqrt{111.4} \approx 10.55%$$

The weighted average of the two volatilities is 13.0 percent. The blend carries 10.55 percent — 2.45 percentage points less risk than the average of its parts, for no reduction in expected return.

The same pair at different correlations

The benefit is entirely a function of correlation. At perfect correlation the mix has the average risk of its parts, which is the state markets converge toward in a crisis.

When It Applies (and Limitations)

Correlation is not stable. The single most important limitation is that the input governing the whole benefit moves, and it moves against the investor. Correlations across risk assets rise in stressed markets, so the diversification measured in calm conditions is smallest exactly when it is needed. October 2008 and March 2020 both produced episodes in which equities, credit, commodities and listed real assets fell together.

Tail dependence differs from average correlation. Two series can show a modest full-sample correlation while being strongly dependent in their left tails. A correlation matrix estimated on all observations understates joint downside risk for assets that share a common funding or liquidity exposure.

Diversification does not raise expected return. Portfolio expected return is the weighted average of its components. Adding a low-correlation asset with a poor expected return lowers risk and lowers return; whether that is an improvement depends on the resulting risk-adjusted profile, not on the diversification alone.

Counting is not diversifying. Thirty holdings drawn from one sector share that sector's systematic exposure. The floor is set by common risk, so a long list of similar positions delivers the appearance of breadth without the substance.

Estimation error can outweigh the theory. DeMiguel, Garlappi and Uppal found that naive equal weighting frequently outperformed optimised allocations out of sample, because the covariance and mean estimates that optimisation depends on are noisy enough to destroy the advantage they promise.

It is not free of implementation cost. Wider portfolios carry more transactions, more rebalancing turnover, more custody relationships and more operational surface. The lunch is free in theory and modestly priced in practice.

Why It Matters for Institutional Investors

The compounding benefit. Because volatility drag scales with portfolio variance, a rebalanced portfolio compounds faster than the weighted average of its components' compounded returns. Booth and Fama documented this diversification return, which is the same arithmetic as volatility drag read in reverse: the portfolio suffers less of it than its parts do.

Risk budgeting rather than capital budgeting. Equal capital weights do not produce equal risk contributions. Risk parity and related approaches allocate by contribution to portfolio variance instead, which is a direct application of the same formula.

Manager overlap. Ten managers running similar factor exposures form a concentrated portfolio wearing ten names. Look-through analysis of underlying exposures, rather than counting mandates, is what establishes whether diversification exists.

Diminishing returns discipline. With most of the achievable benefit captured within roughly ten to twenty genuinely distinct exposures, adding positions beyond that point buys little risk reduction while adding monitoring cost and dilution of conviction.

Correlation as a monitored input. Because the entire benefit rests on one unstable parameter, rolling correlation is a risk-management variable in its own right. A portfolio whose diversification is quietly eroding looks unchanged in every other report.

References

1. Markowitz, H. (1952). Portfolio Selection. Journal of Finance, 7(1), 77–91.
2. Booth, D. G., & Fama, E. F. (1992). Diversification Returns and Asset Contributions. Financial Analysts Journal, 48(3), 26–32.
3. DeMiguel, V., Garlappi, L., & Uppal, R. (2009). Optimal Versus Naive Diversification. Review of Financial Studies, 22(5), 1915–1953.
4. Ang, A. (2014). Asset Management: A Systematic Approach to Factor Investing. Oxford University Press.