So assuming I am entering a trade to long A stock at 10 and short B stock at 20. And assume we want the position to be dollar neutral, so we are going to long 2 shares of A and short 1 share of B. Expected return for A is 0.3% and A's daily vol is 2%; expected return for B is 0.2% and B's daily vol is 2% too. Correlation between A and B is 95%. I got a portfolio daily std of 0.32%. The next step I guess I am suppose multiply the daily portfolio std by exposure and Z value (1.65 assuming normal), but is the exposure in this case 0? So does that mean my portfolio VaR is 0 too? Which seems wrong to me. Thank you!


1 Answer 1


But your exposure is not zero! You have an exposure to the price of stock A equal to 20, and another exposure to the price of stock B equal to -20. Since they are different stocks, albeit highly correlated, netting their exposures is too optimistic.

The expected returns aren't used in VaR.

Let $\Delta=\left(\matrix{\delta_A & \delta_B}\right)$ denote the vector of your 2 exposures; $\sigma_A$ and $\sigma_B$ denote the volatility of their returns (total returns, including dividends), $\rho$ denote the correlation of A returns and B returns, $c=\rho\sigma_A\sigma_B$ denote their covariance, $C=\left(\matrix{\sigma_{A}^2 & c\\ c & \sigma_{B}^2}\right)$ their variance-covariance matrix, then the value at risk is $\mathrm{sqrt}(\mathrm{mmult}(\mathrm{mmult}(\Delta,C),\mathrm{transpose}(\Delta)))\times\mathrm{sqrt}(\frac{d\text{ days horizon, probably 1}}{\text{trading days per year, for example 250}})\times\mathrm{normsinv}(\text{confidence level})$

In addition to VaR, you should decide what is the most money you are willing to lose under adverse scenarios, e.g. the correlation breaks, and A goes down while B goes up; and not exceed that.

  • 1
    $\begingroup$ So assuming std for both are 2%, correlation = 95%, covariance = 0.00038, exposure A is 20 and B is -20, assuming 95% confidence level with standard normal distribution (so multiply by 1.65). I am getting a number of 0.0017 for VaR. Assuming this is in dollar, so the max loss will be $0.0017 over 1 day at a 95% confidence level? And sometimes I see people just multiply by sqrt of # of days instead of dividing by 252 and it makes quite a difference. $\endgroup$
    – DLW
    Oct 13, 2022 at 13:51
  • $\begingroup$ You need to divide by sqrt(number of trading days in a year) if and only if the volatilities are annualized: so if they're alteady not annualized, then don't divide. $\endgroup$ Oct 13, 2022 at 15:35
  • $\begingroup$ Right, ok thanks! $\endgroup$
    – DLW
    Oct 13, 2022 at 15:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.