Extreme numbers of standard deviations

19 Views Asked by At

How would you approximate the magnitude of the probability of an event with an extremely large number of standard deviations? All numerical approximations that I have found have far too large relative errors or do not handle large σ well.

My first rough estimate would be that the value of in 10^- increases by half the value of σ for each additional σ.

For example, for a 670,000σ event, would approximately be half the sum of 1 to 670,000 = 224,450,335,000/2 ≈100,000,000,000.

I'd like to know if this would be roughly correct, and if there is a better approximation.

1

There are 1 best solutions below

0
On

It depends on the distribution, of course. If $X$ has a normal distribution with mean $\mu$ and standard deviation $\sigma$, then as $n \to \infty$ $$ \mathbb P(X \ge \mu + n \sigma) = \frac{\exp(-n^2/2)}{\sqrt{2 \pi}} \left(\frac{1}{n} - \frac{1}{n^3} + \frac{3}{n^5} + O(1/n^7)\right)$$ Asymptotic results for other distributions are also available.