If $X_1, \ldots, X_n \sim N(0,1)$, which each of them independent, then a common result from probability theory is that:

$$ E\left(\max_{1 \leq i \leq n}X_i\right) = O\left(\sqrt{\log\ n}\right) $$

The notation on the left is big-Oh notation. Now, I read in a paper that this signifies that while the normal distribution can be good for fitting data, it is hard for the normal distribution to predict extreme events. However, I am not sure why the above result necessarily shows this. Is it because the term $\sqrt{\log\ n}$ increases very slowly and hence if we have a very large $X_i$, we would need a large number of samples before we can predict (via the expectation) largely? Thanks.

  • $\begingroup$ The maximum, although it naturally tends to grow with the number of samples, grows quite slow compared to, say, a power law. $\endgroup$ – Ian Sep 17 '16 at 3:15
  • $\begingroup$ Are the random variables $X_i$ independent? $\endgroup$ – i707107 Sep 17 '16 at 3:26
  • $\begingroup$ Hi, yes the random variables are independent $\endgroup$ – user321627 Sep 17 '16 at 3:28

A random variable from $N(0,1)$ has the expectation $0$. But, for large values of $n$, $\max_{1 \leq i \leq n}X_i$ (maximum of normally distributed rv's) goes to infinity (in expectation) as your theory says. This means that the normal distribution is not good at extreme events.


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.