Closed Form Solution for Minimization involving Standard Normal CDF and PDF

163 Views Asked by At

Could someone please advice and provide detailed steps regarding any possible closed form solutions or other suggestions regarding solving a minimization problem of the type shown below? Here, $\phi$ and $\mathbf{\Phi}$ are the standard normal PDF and CDF, respectively. $\sigma$ is the standard deviation and $\theta$ is a parameter. We need to find the solution by expressing $x$ in terms of $y$ and the other parameters.

\begin{eqnarray*} =\underset{\left\{ x\right\} }{\min}\left\{ x\sigma\left[\frac{\theta x}{\sigma}+\frac{\phi\left(\frac{\theta x}{\sigma}\right)}{\Phi\left(\frac{\theta x}{\sigma}\right)}\right]+\left(y-x\right)\sigma\left[\frac{\theta\left(y-x\right)}{\sigma}+\frac{\phi\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}{\Phi\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}\right]\right\} \end{eqnarray*}

First Order Conditions (FOC) would give the below. Please point out ways to simplify or alternative approaches as well.

\begin{eqnarray*} x\left[\frac{4\theta}{\sigma}+\theta\left\{ \frac{\phi'\left(\frac{\theta x}{\sigma}\right)}{\Phi\left(\frac{\theta x}{\sigma}\right)}-\phi'\left(\frac{\theta x}{\sigma}\right)\left[\frac{\phi\left(\frac{\theta x}{\sigma}\right)}{\Phi\left(\frac{\theta x}{\sigma}\right)}\right]^{2}\right\} \right.\\ +\left.\theta\left\{ \frac{\phi'\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}{\Phi\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}-\phi'\left(\frac{\theta\left(y-x\right)}{\sigma}\right)\left[\frac{\phi\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}{\Phi\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}\right]^{2}\right\} \right] & =\\ \left(\theta y\right)\left\{ \frac{2}{\sigma}+\frac{\phi'\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}{\Phi\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}-\phi'\left(\frac{\theta\left(y-x\right)}{\sigma}\right)\left[\frac{\phi\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}{\Phi\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}\right]^{2}\right\} \\ +\frac{\sigma\phi\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}{\Phi\left(\frac{\theta\left(y-x\right)}{\sigma}\right)}-\frac{\sigma\phi\left(\frac{\theta x}{\sigma}\right)}{\Phi\left(\frac{\theta x}{\sigma}\right)} \end{eqnarray*}

1

There are 1 best solutions below

3
On BEST ANSWER

Looks like a 1-variable problem.

For analytical approach , how about taking one derivative in $x$ and setting it equal to zero and see what you can get.

For a numerical approach, since you have the derivative (and even two if you like), you can call a minimization algo, or even rootfind the derivative with Newton's method or Brent or Secant method. This will, unfortunately, only work, for fixed values of $\theta, y$.