Diversification

Portfolio expected return is a simple weighted average of component returns, but risk is not. For two assets, portfolio variance equals the first weight squared times the first variance, plus the second weight squared times the second variance, plus twice the product of the two weights, the two standard deviations and the correlation coefficient. The final cross term is where diversification lives: only it depends on how the assets move together, and only it can be small or negative.

Because that term shrinks as correlation falls, the same two assets produce a whole range of portfolio risks depending on their co-movement. At a correlation of one the portfolio standard deviation collapses to the weighted average of the individual standard deviations and no benefit exists. At a correlation of minus one, a specific weighting drives variance to zero. Every intermediate case delivers partial risk reduction with no sacrifice of expected return, which is why diversification is described as the only free improvement in investing.

Extend the two-asset case to an equally weighted portfolio of N securities and the variance decomposes into two pieces: the average individual variance divided by N, plus the average pairwise covariance multiplied by one minus one over N. The first piece is asset-specific risk, and it is destroyed rapidly as N grows. The second piece converges on the average covariance among the holdings and does not disappear at any portfolio size.

That average covariance is the systematic floor. Adding the twenty-first stock helps far less than the second did, and beyond roughly two or three dozen well-spread names the marginal reduction becomes economically trivial. The practical implication is that the quality of diversification is determined by the average correlation of what is held, not by the count. A portfolio of one hundred regional bank shares has a high average covariance and therefore a high floor, however many line items appear on the statement.

Correlations are estimated from history and are not stable. They tend to rise in stress episodes, particularly among risk assets that share funding, liquidity and growth sensitivities, so measured diversification is weakest precisely when it is needed. Tail dependence, forced deleveraging and common ownership all compress the differences between labels such as equities, high-yield credit, listed real estate and cyclical commodities.

Structural exposures create the same problem quietly. Home-country bias, concentration in employer stock alongside employer-dependent human capital, and portfolios of funds that share the same factor tilts all leave a single dominant driver in place. Illiquid holdings mark slowly, so reported correlations understate true economic co-movement. The professional response is to diversify across risk factors and economic drivers rather than across product names, and to test exposures under stressed correlation assumptions rather than long-run averages.

Expect to compute a two-asset portfolio standard deviation from weights, standard deviations and either a correlation or a covariance, and to convert between the two using the relationship that covariance equals correlation times the product of the standard deviations. You should be able to show that a lower correlation bends the risk-return combination line further to the left, generating the curvature of the efficient frontier, and to identify the minimum-variance combination in simple cases.

Conceptually, you must separate total risk into systematic and non-systematic components and connect that split to asset pricing: because investors can eliminate asset-specific risk cheaply, equilibrium models compensate only non-diversifiable risk, measured by beta. From this follow the market portfolio, the capital allocation line, and the reasoning behind index-based cores. Be ready to judge a described portfolio's true diversification from its correlation structure and factor exposures rather than its holding count.