Understand diversification benefits — Stock-Market notes

WHAT is diversification?

The key statistical fact: risk is NOT additive. Expected returns add up linearly, but standard deviations do not — because when one asset zigs, another may zag, and those movements partially cancel.

WHY does it work? (First-principles derivation)

Let's build the two-asset portfolio risk formula from scratch. This is the heart of everything.

Setup. Weights $w_1$ and $w_2$ with $w_1+w_2=1$ . Random returns $R_1, R_2$ . Portfolio return:

$R_p = w_1 R_1 + w_2 R_2$

Step 1 — Expected return (why linear?) Expectation is a linear operator, so:

$E[R_p] = w_1 E[R_1] + w_2 E[R_2] = w_1\mu_1 + w_2\mu_2$ Why this step? $E[aX+bY]=aE[X]+bE[Y]$ always — no assumptions needed. Return just averages.

Step 2 — Variance (why NOT linear?) Variance of a sum expands with a cross term:

$\text{Var}(R_p) = E\big[(R_p - E[R_p])^2\big]$

Substitute $R_p - E[R_p] = w_1(R_1-\mu_1) + w_2(R_2-\mu_2)$ and square:

\sigma_p^2 = w_1^2\,E[(R_1-\mu_1)^2] + w_2^2\,E[(R_2-\mu_2)^2] + 2w_1w_2\,E[(R_1-\mu_1)(R_2-\mu_2)]

Why this step? Squaring a binomial gives $a^2 + b^2 + 2ab$ . The first two terms are variances; the last term is where the magic hides.

Step 3 — Name the pieces. Define $\text{Cov}(R_1,R_2)=\sigma_{12}=E[(R_1-\mu_1)(R_2-\mu_2)]$ and $\rho_{12}=\dfrac{\sigma_{12}}{\sigma_1\sigma_2}$ (correlation, always in $[-1,1]$ ).

HOW the correlation drives the benefit

Look at the cross term $2w_1w_2\rho_{12}\sigma_1\sigma_2$ (weights positive):

$\rho_{12}$	Cross term	Result
$+1$	maximum positive	No benefit: $\sigma_p = w_1\sigma_1 + w_2\sigma_2$ (just a weighted average)
$0$	zero	Meaningful benefit: $\sigma_p = \sqrt{w_1^2\sigma_1^2 + w_2^2\sigma_2^2}$
$-1$	maximum negative	Maximum benefit: risk can hit zero at the right weights

Proof that $\rho=+1$ gives no benefit. Set $\rho=1$ : $\sigma_p^2 = w_1^2\sigma_1^2 + w_2^2\sigma_2^2 + 2w_1w_2\sigma_1\sigma_2 = (w_1\sigma_1 + w_2\sigma_2)^2$ So $\sigma_p = w_1\sigma_1 + w_2\sigma_2$ — a perfect square, exactly the weighted average. Nothing cancels.

Proof that $\rho=-1$ can kill all risk. Set $\rho=-1$ : $\sigma_p^2 = (w_1\sigma_1 - w_2\sigma_2)^2$ This is $0$ when $w_1\sigma_1 = w_2\sigma_2$ , i.e. $w_1 = \dfrac{\sigma_2}{\sigma_1+\sigma_2}$ . A perfectly hedged, riskless combo.

Figure — Understand diversification benefits

Many assets: systematic vs unsystematic risk

For an equally-weighted portfolio of $N$ assets ( $w_i=1/N$ ), each with variance $\sigma^2$ and average pairwise covariance $\bar{c}$ :

$\sigma_p^2 = \frac{1}{N}\sigma^2 + \left(1-\frac{1}{N}\right)\bar{c}$

Why this step? There are $N$ variance terms each weighted $(1/N)^2$ giving $\sigma^2/N$ , and $N(N-1)$ covariance terms each weighted $(1/N)^2$ giving $\frac{N-1}{N}\bar c$ .

80/20 takeaway: Most diversification benefit is captured with the first ~20–30 stocks; adding the 500th stock barely helps because $\sigma^2/N$ is already tiny. Focus energy on picking low-correlation assets, not on the sheer count.

Worked Examples

Worked example Example 1 — Two stocks, moderate correlation

$\mu_1=10\%,\ \sigma_1=20\%$ ; $\mu_2=8\%,\ \sigma_2=15\%$ ; $\rho=0.3$ ; weights $w_1=w_2=0.5$ .

Return: $E[R_p]=0.5(10)+0.5(8)=9\%$ . Why? Returns are linear.

Variance: $\sigma_p^2 = 0.5^2(0.20^2) + 0.5^2(0.15^2) + 2(0.5)(0.5)(0.3)(0.20)(0.15)$ $= 0.01 + 0.005625 + 0.0045 = 0.020125$ Std dev: $\sigma_p=\sqrt{0.020125}=14.19\%$ . Why the win? The naive weighted average of risk is $0.5(20)+0.5(15)=17.5\%$ . We got 14.19% — same-ish return, less risk. That gap (17.5→14.19) is the diversification benefit.

Common Mistakes (Steel-manned)

Recall Feynman: explain to a 12-year-old

Imagine you sell ice cream and umbrellas. On sunny days ice cream sells; on rainy days umbrellas sell. You almost always earn something — your income is steady even though each business alone is wild. That's diversification: pick things that don't have bad days at the same time, and your total is smoother. But you can never remove the risk that hits everyone at once (like a whole town-wide power cut) — that's the "market risk" you're stuck with.

Active Recall Flashcards

Why do portfolio expected returns add linearly but risks don't?

Expectation is a linear operator so

E[R_p]=\sum w_i\mu_i

; variance has a cross (covariance) term that isn't linear.

Where does the diversification benefit live in the two-asset variance formula?

In the cross term

2w_1w_2\rho_{12}\sigma_1\sigma_2

— lower correlation shrinks it.

What is

\sigma_p

when

\rho=+1

The plain weighted average

w_1\sigma_1+w_2\sigma_2

— zero benefit.

\rho=-1

, what weight makes a two-asset portfolio riskless?

w_1=\sigma_2/(\sigma_1+\sigma_2)

, giving

\sigma_p=0

N\to\infty

for equal weights, what does

\sigma_p^2

approach?

The average pairwise covariance

\bar c

(the systematic/market risk floor).

Which risk can be diversified away and which cannot?

Unsystematic (company-specific) can be removed; systematic (market) cannot.

Why is "own 200 stocks" often wasteful?

Benefit is bounded below by

\bar c

; past ~20–30 low-corr names the marginal reduction is tiny (80/20).

Two stocks 50/50,

\sigma_1=20\%,\sigma_2=15\%,\rho=0.3

: what's

\sigma_p

\approx14.19\%

vs naive average

17.5\%

Is diversification a free lunch? Why?

Yes — it lowers risk without lowering the linear weighted-average expected return.

Connections

Correlation and Covariance — the engine of the benefit
Efficient Frontier — plotting these risk/return combos
Capital Asset Pricing Model (CAPM) — prices only systematic risk (beta)
Systematic vs Unsystematic Risk
Portfolio Variance and Standard Deviation
Modern Portfolio Theory (Markowitz)

Concept Map

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho, diversification ka matlab simple hai: apna saara paisa ek hi stock mein mat daalo. Agar tum alag-alag stocks lo jinke ups-and-downs ek saath nahi hote (yaani unka correlation kam hai), toh jab ek girta hai tab dusra sambhal leta hai. Sabse mazedaar baat — expected return toh sirf weighted average hota hai (linear), lekin risk (standard deviation) linear nahi hota. Isliye tum return same rakhte hue bhi total risk kam kar sakte ho. Yahi "free lunch" hai finance mein.

Poora khel is cross term par tika hai: $2w_1w_2\rho\sigma_1\sigma_2$ . Jitna kam $\rho$ , utna chhota yeh term, utni kam portfolio risk. Agar $\rho=+1$ ho toh koi fayda nahi — risk bas weighted average ban jaata hai. Agar $\rho=-1$ ho toh perfect hedge ban sakta hai aur risk zero tak ja sakti hai (right weights par). Isliye asli kaam hai low-correlation waale assets dhoondhna, sirf zyada stocks jodna nahi.

Ek important baat: risk do type ki hoti hai. Unsystematic risk (ek company ki apni problem) ko tum diversify karke khatam kar sakte ho. Par systematic risk (poora market, recession, interest rate) sab stocks ko ek saath maarti hai — usko kabhi diversify nahi kar paoge. Formula bolta hai jaise $N$ badhta hai, $\sigma^2/N$ zero ki taraf jaata hai par $\bar c$ (average covariance) reh jaata hai.

80/20 ka funda: pehle 20-30 acche, alag-alag sector ke stocks se lagbhag saara diversification benefit mil jaata hai. 200 tech stocks lene ka koi matlab nahi kyunki woh sab ek jaise chalte hain (high correlation). Toh yaad rakho — CORR is the boss, aur "unique risk hatao, universe risk nahi."

Test yourself — Portfolio Theory

Connections