To compute the [[Variance]] of a portfolio of assets, we can extend the [[Variance of Sum of Random Variables#Multiple Random Variables|Variance of Sum of Random Variables]] by a weighted sum, where each asset has a weighting.
No assumptions on the distributions of individual [[Random Variable|Random Variables]] need to be taken, as long as a mean and a variance exist.
Assume we have a portfolio with $n$ assets, and each asset has:
- $R_i$: Return of the asset (random variable)
- $\mu_i$: [[Expectation]] of $R_i$.
- $w_i$: Relative weight in the portfolio
The portfolio return is defined as:
$ R_p = \sum_{i=1}^n w_iR_i$
The portfolio variance can be derived as follows
$
\begin{align}
\mathrm {Var}(R_p)&=\mathbb{E}\left[\Big(\sum_{i=1}^n w_i(R_i-\mu_i) \Big)^2\right] \tag{1}\\[4pt]
&=\mathbb E\left[ \sum_{i=1}^n w_i^2(R_i-\mu_i)^2\right] + \mathbb E\left[\sum_{(i,j): i \neq j}^n w_i w_j(R_i-\mu_i)*(R_j-\mu_j) \right] \tag{2}\\
&=\sum_{i=1}^n w_i^2*\mathbb E\Big[ (R_i-\mu_i)^2\Big] + 2*\sum_{(i,j): i < j}^nw_i w_j* \mathbb E\Big[(R_i-\mu_i)*(R_j-\mu_j) \Big] \tag{3}\\
&=\sum_{i=1}^nw_i^2*\mathrm{Var}(R_i) + 2*\sum_{i < j}^n w_i w_j\mathrm{Cov}(R_i, R_j) \tag{4}\\
&=\sum_{i=1}^nw_i^2*\mathrm{Var}(R_i) + 2*\sum_{i < j}^n w_i w_j \rho \,\sigma_i\sigma_j \tag{5}
\end{align}
$
where:
- (2) Expanding the quadratic formula into the quadratic terms and the cross terms.
- (3) Factoring out constants out of the expectation (e.g. $w_i$)
- (4) Recognizing the variance and [[Covariance]].
- (5) Express the covariance in terms of [[Correlation]] $\rho$ and standard deviations $\sigma_i$ and $\sigma_j$.
## Special Cases
Correlation of $0$:
$ \mathrm{Var}(R_p)=\sum_{i=1}^nw_i^2*\mathrm{Var}(R_i)$
Correlation of $1$:
$
\mathrm{Var}(R_p)=\sum_{i=1}^n w_i^2*\mathrm{Var}(R_i) + 2*\sum_{i < j}^n w_i w_j \,\sigma_i\sigma_j=\left(\sum_{i=1}^n w_i \sigma_i \right)^2
$
Correlation of $0$ (with equal weighting and variance):
$
\begin{align}
\mathrm{Var}(R_p)
&=\sum_{i=1}^n w_i^2*\mathrm{Var}(R_i) \\[4pt]
&=\sum_{i=1}^n \left(\frac{1}{n}\right)^2*\sigma^2 \\[4pt]
&=\left(\frac{1}{n}\right)^2n*\sigma^2\\[8pt]
&=\frac{\sigma^2}{n}
\end{align}
$