On Asymptotic Expansions: An Example

120 Views Asked by At

I am reading wikipedia's article on asymptotic expansion, and I would like it if someone can help me clarify some doubts with an example, with the inverse of the cumulative density function of the standard normal variable.

Also in wikipedia it is stated that for small $p$: $$\Phi^{-1}(p) \sim -\sqrt{\ln\frac{1}{p^2}-\ln\ln\frac{1}{p^2}-\ln(2\pi)}:=f(p)$$

A more concrete question is: what does this say about the ratio the limit of $\Phi^{-1}(p)/f(p)$ as $p \to 0^+$? Can we say this limit is 1? I understand the concept of $o$ notations, but it would be good if this claim is true and to understand why it is the case. Also, any references on this assymptotic expansion on the normal quantile function would be appreciated. Thanks!

2

There are 2 best solutions below

4
On BEST ANSWER

In terms of the inverse error and complementary error functions $$ \Phi ^{ - 1} (p) = \sqrt 2 \operatorname{erf}^{ - 1} (2p - 1) = - \sqrt 2 \operatorname{erf}^{ - 1} (1 - 2p) = - \sqrt 2 \operatorname{erfc}^{ - 1} (2p). $$ Then using the leading term and estimating the remaining ones of the known asymptotic expansion for $\operatorname{erfc}^{ - 1} $, we deduce $$ \Phi ^{ - 1} (p) = - \sqrt { \ln \frac{1}{{p^2 }} - \ln \ln \frac{1}{{4p^2 }} - \ln (2\pi )} + \mathcal{O}\!\left( {\frac{{\ln \left| {\ln p} \right|}}{{\left| {\ln p} \right|^{3/2} }}} \right) $$ as $p\to 0^+$. Now $$ \ln \ln \frac{1}{{4p^2 }} = \ln \ln \frac{1}{{p^2 }} + \mathcal{O}\!\left( {\frac{1}{{\left| {\ln p} \right|}}} \right) $$ as $p\to 0^+$, which gives \begin{align*} \Phi ^{ - 1} (p) &= - \sqrt {\ln \frac{1}{{p^2 }} - \ln \ln \frac{1}{{p^2 }} - \ln (2\pi )} + \mathcal{O}\!\left( {\frac{1}{{\left| {\ln p} \right|^{3/2} }}} \right) + \mathcal{O}\!\left( {\frac{{\ln \left| {\ln p} \right|}}{{\left| {\ln p} \right|^{3/2} }}} \right) \\ & = - \sqrt {\ln \frac{1}{{p^2 }} - \ln \ln \frac{1}{{p^2 }} - \ln (2\pi )} + \mathcal{O}\!\left( {\frac{{\ln \left| {\ln p} \right|}}{{\left| {\ln p} \right|^{3/2} }}} \right) \end{align*} as $p\to 0^+$, which agrees with the one cited in the other answer.

0
On

By chance I recently ran across the paper "Quantile function expansion using regularly varying functions" by T. Fung and E. Seneta, in Methodol. Comput. Appl. Probab. (2018) 20:1091-1103, which seems to be the same as this arXiv draft.

At the bottom of p.1092 (p.3 in the arXiv draft) is the result (in the notation of the OP) $$ \Phi^{-1}(p) = -\sqrt{-2\log(p\sqrt{-4\pi\log p})}\left(1+\mathcal O\left(\frac{\log|\log p|}{(\log p)^2}\right)\right), $$ for small $p$.

Seneta is a well trusted expert in this field, and I am sure part of the authors' motivation was exasperation at the imprecision of statements like those found in the Wikipedia article.