Diversification

Diversification is the one genuinely free lunch in finance. If you spread your money across many holdings that don't all rise and fall together, the ups in one tend to cancel the downs in another, so your total portfolio jumps around less than any single holding does. The remarkable part: you get this smoother ride without accepting a lower average return. You are not trading away reward for safety, you are simply deleting risk that you were never being paid to take.

Why spreading out cuts risk

The whole effect runs on one word: correlation, how tightly two holdings move together. (Here "risk" means volatility, the size of a holding's swings.) If two stocks always move in perfect step (correlation $+1$), owning both is no different from owning a double dose of one, the swings just add up. But if they move somewhat independently, a bad day for one is often an ordinary day for the other, and the bad news gets partly washed out in the average. The less correlated your holdings, the more the individual shocks cancel, and the calmer the portfolio.

The lever behind that curve is correlation. Take just two assets and slide the correlation $\rho$ between them: the lower it goes, the further the combined portfolio bows to the left — same return, less risk. That leftward bow is diversification made visible.

The math in one formula

For a portfolio of two holdings with weights $w_1, w_2$ (fractions of your money), the portfolio variance is

$$\sigma_p^2 = w_1^2\sigma_1^2 + w_2^2\sigma_2^2 + 2\,w_1 w_2\,\rho\,\sigma_1\sigma_2,$$

where $\sigma_1, \sigma_2$ are the two volatilities and $\rho$ is their correlation. The correlation $\rho$ sits in the last term, and it is the whole story: lower $\rho$ means a smaller last term, which means smaller portfolio variance.

Now stack $n$ identical holdings, each with variance $\sigma^2$ and the same pairwise correlation $\rho$, in equal weights. The formula collapses to a clean one:

$$\sigma_p^2 = \frac{\sigma^2}{n} + \left(1 - \frac{1}{n}\right)\rho\,\sigma^2.$$

As you add names ($n$ grows), the first term $\sigma^2/n$ melts toward zero, that is the diversifiable, company-specific risk vanishing. But the second term marches toward $\rho\sigma^2$ and stays there. That leftover, $\rho\sigma^2$, is the systematic floor in the diagram: the shared market risk you cannot escape by adding more of the same.

Worked example

Take two stocks, each with volatility $\sigma = 20\%$, split 50/50.

• Perfectly correlated ($\rho = 1$): $\sigma_p^2 = 0.25(0.04) + 0.25(0.04) + 2(0.25)(1)(0.2)(0.2) = 0.04$, so $\sigma_p = 20\%$. No benefit, you doubled up on the same bet.
• Uncorrelated ($\rho = 0$): the last term is zero, so $\sigma_p^2 = 0.25(0.04) + 0.25(0.04) = 0.02$, giving $\sigma_p = \sqrt{0.02} = 14.1\%$. Same expected return, nearly a third less risk.
• Perfectly opposed ($\rho = -1$): $\sigma_p^2 = 0.02 - 0.02 = 0$, so $\sigma_p = 0\%$. The two exactly hedge each other.

Now the $n$-name version with a realistic $\rho = 0.3$: a single stock has $\sigma = 20\%$. With $n = 10$ names, $\sigma_p^2 = 0.04/10 + 0.9(0.3)(0.04) = 0.004 + 0.0108 = 0.0148$, so $\sigma_p = 12.2\%$. Push $n \to \infty$ and the floor is $\sqrt{0.3}\times 20\% = 11.0\%$. So ten names already capture almost all the benefit, and the last stubborn $11\%$ is pure market risk.

Systematic vs unsystematic risk

Splitting risk into two buckets is the single most useful idea here. Unsystematic (or idiosyncratic) risk is specific to one company, a factory fire, a bad earnings call, and it is diversifiable: enough independent names average it away. Systematic risk is market-wide, a recession or a rate shock hits everything at once, and it is undiversifiable. This is exactly why the efficient frontier and Markowitz portfolio theory are built on covariances rather than individual variances, and why only systematic risk earns a return: nobody pays you to carry a risk you could have deleted for free. That link between risk you cannot diversify and reward is the heart of the risk-return tradeoff.

Common pitfalls

• Correlations spike in a crash. The $\rho$ you measured in calm markets is not the $\rho$ you get in a panic, when almost everything falls together. Diversification tends to fail exactly when you need it most.
• More names is not more diversification. Adding a fifth tech stock that moves with the four you own barely helps. What matters is low correlation, not the raw count of holdings.
• Diworsification. Past a couple dozen well-chosen names the marginal risk reduction is tiny, while trading costs and diluted conviction pile up. Owning everything is a way to guarantee mediocrity, not to manage risk.
• Hidden concentration. Thirty stocks that are all energy names, or all funded by the same factor, are one bet wearing thirty costumes. Diversify across drivers, not just tickers.