Given $X$ ~ $LogN(\mu, \sigma^2)$, I want to prove that the distribution function of $X$ is given by $\Phi\Big(\frac{\ln(x)-\mu}{\sigma}\Big)$, where $\Phi()$ is the distribution function of a $N(0,1)$-distributed r.v.

I know that the density of $X$ is given by

$$ f(x) = \frac{1}{\sqrt{2\pi}\sigma x}\exp\{-\frac{1}{2} \Big(\frac{\ln(x)-\mu}{2\sigma}\Big)^2\}$$

so we get

$$ F(x) = \frac{1}{\sqrt{2\pi}\sigma}\int_{-\infty}^x\exp\{-\frac{1}{2} \Big(\frac{\ln(t)-\mu}{2\sigma}\Big)^2\}\frac{1}{t}dt$$ $$= \frac{1}{\sqrt{2\pi}}\int_{0}^{\frac{\ln(x)-\mu}{\sigma}}\exp\{-\frac{1}{2} z^2\}dz = \Phi\Big(\frac{\ln(x)-\mu}{\sigma}\Big)$$

Is it correct, that the lower bound transforms to $0$?

Furthermore, how can one estimate the parameters $(\mu, \sigma^2)$ for given data using the Maximum-Likelihood-method for the r.v. $\ln(X)$?

For the second question I think we can deduce this to the ML-estimators of the normal distribution, since

$$ L(x; \mu, \sigma^2) = \prod \Big(\frac{1}{x_i}\Big) N(\ln(x); \mu, \sigma^2) $$

where $L$ is the density of the log-normal distribution and $N$ is the density of the normal distribution. For the log-lieklihood this means

$$ l_L(x; \mu, \sigma^2) = -\sum_k \ln(x_k) +l_N(\ln(x); \mu, \sigma^2) $$

If we differentiate this, the sum of the logarithms always cancels, since it is constant, and therefore it is enough to know the ML-estimators for the normal distribution. Therefore we get in the end that

$$ \hat{\mu}=\frac{1}{n}\sum_k \ln(x_k), \text{ } \hat{\sigma}^2=\frac{1}{n}\sum_k (\ln(x_k)-\hat{\mu})^2.$$


Given $X \sim \operatorname{LogNormal}(\mu, \sigma^2)$, then by definition, $X = e^Y$ for $Y \sim \operatorname{Normal}(\mu,\sigma^2)$, hence $X = e^{\sigma Z + \mu}$ for $Z \sim \operatorname{Normal}(0,1)$. It immediately follows that $$F_X(x) = \Pr[X \le x] = \Pr[e^{\sigma Z + \mu} \le x] = \Pr\left[Z \le \frac{\log x - \mu}{\sigma}\right] = \Phi\left(\frac{\log x - \mu}{\sigma}\right).$$ It also follows from the above that the density of $X$ is $$f_X(x) = F'_X(x) = \frac{d}{dx} \left[\Phi\left(\frac{\log x - \mu}{\sigma}\right)\right] = \frac{1}{\sigma x} f_Z\left(\frac{\log x - \mu}{\sigma}\right) = \frac{1}{\sqrt{2\pi} \sigma x} e^{-(\log x - \mu)^2/(2\sigma^2)}.$$

  • $\begingroup$ True, that would be the short way without specifiying the integrals. Thank you! $\endgroup$ – ducks17 Mar 28 '17 at 17:27
  • $\begingroup$ And what about the secon question? $\endgroup$ – ducks17 Mar 28 '17 at 19:18
  • $\begingroup$ @ducks17 Show what you have tried or what you know about maximum likelihood estimation, and I may address that part accordingly. As it stands, that part of your question is too broad for me to furnish a succinct answer. $\endgroup$ – heropup Mar 28 '17 at 19:20
  • $\begingroup$ See the post for my way of solving the problem. Do you think it's correct? $\endgroup$ – ducks17 Mar 28 '17 at 19:51
  • $\begingroup$ @ducks17 Yes, it is correct. You can from the very beginning note that $\ln(X_i)$ have Normal distribution and take likelihood function for it. $\endgroup$ – NCh Mar 29 '17 at 1:29

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.