1
$\begingroup$

I am solving:

$(\sigma_A^2 - 2\rho\sigma_A\sigma_B +\sigma_B^2)x^2 +2(\rho\sigma_A\sigma_B - \sigma_B^2)x +\sigma_B^2 = 0$

I need to show that a real $x$ exists if and only if $\rho = \pm 1$

Using the quadratic formula I could only get as far as

$ x = \frac{-2(\rho\sigma_A\sigma_B - \sigma_B^2) \pm \sqrt{4\sigma_A^2\sigma_B^2(\rho^2-1)}}{2( \sigma_A^2 +\sigma_B^2 - 2\rho\sigma_A\sigma_B)}$

I have a more simplified solution, which is

$x=[1- \frac{\sigma_A}{\sigma_B}(\rho\pm\sqrt{\rho^2-1})]^{-1}$

but I cannot see how to get there.

$\endgroup$
  • 3
    $\begingroup$ I think you have missed a factor of $x$ out, should it be $(\sigma_A^2 - 2\rho\sigma_A\sigma_B +\sigma_B^2)x^2 +2(\rho\sigma_A\sigma_B - \sigma_B^2) {\bf x} +\sigma_B^2 = 0$? As well, could you give some background to the equations, are the $\sigma 's$ standard deviations of some experiment and $\rho$ some correlation? $\endgroup$ – jim Mar 25 '16 at 11:02
  • 2
    $\begingroup$ If $\rho$ is a correlation then $| \rho | \le 1$. But for $| \rho | < 1$ you have no (real) solution for $x$, so you must have $| \rho | = 1$. $\endgroup$ – jim Mar 25 '16 at 11:16
  • $\begingroup$ Sorry, yes i missed a factor of $x$ out, the $\sigma's$ are standard deviations and $\rho$ is a correlation so $0\le\rho\le1$. $\endgroup$ – Kai_M Mar 25 '16 at 11:43
  • $\begingroup$ The idea behind the question is that we have a portfolio of two risky assets A and B with variance of returns equal to $\sigma_A^2$ and $\sigma_B^2$ and an amount $x$ invested in asset A and $1-x$ in asset B. The portfolio return is thus $w=(1+r_A)x+(1+r_B)(1-x)$ and the variance of the portfolio return is $V[w]= x^2\sigma_A^2 +2x(1-x)\rho\sigma_A\sigma_B +(1-x)^2\sigma_B^2$. The portfolio is riskless if $V[w] =0$, which is the equation we are trying to solve. $\endgroup$ – Kai_M Mar 25 '16 at 11:56
2
$\begingroup$

I'm assuming you've missed an $x$ factor out, as jim has suggested in a comment above.

If so, I don't think that a real $x$ exists if and only if $\rho = \pm 1$ . The discriminant is indeed $\Delta = 4\sigma_A^2\sigma_B^2(\rho^2-1)$ , so that a real solution exists if and only if $\rho^2-1 \geqslant 0$, which is true if and only if $\rho \geqslant 1$ or $\rho \leqslant -1$.

However your exercise might have extra conditions given, and I could imagine something like the condition $\left|\rho\right| \leqslant 1 $ being given here. (Edit: for instance if $\rho$ is a correlation as jim has pointed out!) If this condition is given then combining it with the conditions above ( $\rho \geqslant 1$ or $\rho \leqslant -1$ ) yields that a real solutions exists iff $\rho = \pm 1$.

Another possibility would be that the question is when a unique real solution exists, in which case we need $\Delta = 0$ and so $\rho^2 = 1$, iff $\rho = \pm 1$.

$\endgroup$
  • $\begingroup$ Thanks for your answer. What you and Jim suggested was indeed correct. $\endgroup$ – Kai_M Mar 25 '16 at 11:58
2
$\begingroup$

Hint:

The usual formula for the roots of a quadratic equation is

$$\frac{-b\pm\sqrt{b^2-4ac}}{2a}.$$

But there is a variant:

$$\frac{2c}{-b\pm\sqrt{b^2-4ac}}.$$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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