Diversification reduces portfolio risk by combining assets whose returns are not perfectly correlated, so that losses in some positions are offset by gains in others. As more assets are added, idiosyncratic (firm-specific) risk averages away, but systematic (market-wide) risk that affects all assets simultaneously cannot be diversified away. The benefit of adding another asset depends on its correlations with existing holdings, not on its standalone volatility. This distinction between diversifiable and non-diversifiable risk is fundamental: rational markets should only compensate investors for systematic risk, since idiosyncratic risk can be cheaply eliminated through diversification.
Simulate adding randomly chosen stocks to a portfolio and plot how the portfolio's standard deviation decreases with N, eventually flattening at the systematic risk floor. Compare portfolios that are diversified across industries vs. concentrated in a single sector to see the limits of naive diversification.
From your study of expected return and variance, you know that holding a single risky asset exposes you to all of its variance. The key insight of diversification is that when you combine assets, the portfolio's variance depends not just on each asset's individual variance but critically on how their returns move together — the covariance, or equivalently the correlation coefficient.
Recall that the variance of a two-asset portfolio is Var(portfolio) = w₁²σ₁² + w₂²σ₂² + 2w₁w₂σ₁σ₂ρ₁₂, where ρ is the correlation between the two assets. If ρ = 1 (perfect positive correlation), the portfolio variance is just the weighted average of the individual variances — no risk reduction. If ρ < 1, the cross term is smaller, and the portfolio variance falls below the weighted average — diversification is working. The lower the correlation, the more dramatic the risk reduction. If ρ = -1 (perfect negative correlation), you can theoretically reduce portfolio variance to zero.
As you add more assets, something systematic happens: idiosyncratic risk — the firm-specific fluctuations that affect one company but not others — averages away. A drug trial failure at one pharmaceutical company is unrelated to a software bug at a tech firm; when you hold both, their idiosyncratic shocks tend to cancel. The mathematics shows that as N grows, the contribution of idiosyncratic variance to the portfolio falls roughly as 1/N. However, the covariance terms — which capture how assets move together in response to economy-wide forces — do not average away. What remains after diversifying fully is systematic risk: the component of return variance driven by market-wide factors like recessions, interest rate changes, or geopolitical shocks that affect every asset simultaneously.
This is the critical distinction that drives all of asset pricing theory. Idiosyncratic risk can be eliminated cheaply by any investor willing to hold a diversified portfolio. A rational market should therefore offer no reward for bearing it — why pay for insurance you could have gotten for free? Systematic risk, by contrast, cannot be avoided by any investor who wants to participate in capital markets; it must be borne, and so markets compensate investors for it with a risk premium. This logic leads directly to the Capital Asset Pricing Model: expected return should be a function of systematic risk (beta), not total risk (volatility).
One subtle but important implication: it is the correlation with your *existing* portfolio that determines whether adding an asset reduces risk, not the asset's standalone volatility. A highly volatile asset with low correlation to your holdings might reduce portfolio risk more than a low-volatility asset that is highly correlated with what you already own. This counterintuitive result is what makes correlation, not variance, the central concept in portfolio construction.