Let $X$ be a random variable, and let $\phi_X(t)$ be its characteristic function. Let $Y = f(X)$ be a transformation of the random variable $X$ where $f$ is increasing and one-to-one. Is there a direct functional relationship between the characteristic function of the transformation $f(X)$ and the characteristic function $\phi_X(x)$? Meaning if $\phi_Y(t)$ is the characteristic function of the random variable $f(X)$, can we write $\phi_Y(t)$ in terms of $\phi_X(t)$?

  • $\begingroup$ I think if there was any relation, it'd be quite well known. fourier transforms don't have special behaviour under transformations. $\endgroup$ – Lost1 Jan 20 '14 at 20:21

If $Y=g(X)$ (let me use $g$, reserving $f$ for the densities) :

Then $$ \phi_Y(t)=E(e^{i Yt})=\int_{-\infty}^{\infty} f_Y(y) e^{i y t} dy= \int_{-\infty}^{\infty} f_X(x) e^{i g(x) t} dx=\\ =\frac{1}{2 \pi} \int_{-\infty}^{\infty} \int_{-\infty}^{\infty} \phi_X(s) e^{-i s x } ds \, e^{i g(x) t} dx $$

Of course, this is rather trivial ( $ \phi_X(t) \to f_X(x) \to f_Y(y) \to \phi_Y(t)$) and formal (the formula is not practical, it does not give a nice or simple relationship between $\phi_Y(t)$ and $\phi_X(t)$ but I don't see why you'd expect that it can be simplified.


Here's I think what would be a counterexample: if $X \sim R[0,1]$ and $Y \sim N(X, \sigma^2)$, then MGF of $X$ would be $\frac{e^t-1}{t}$ and MGF of $Y$ would be $e^{xt+\frac{\sigma^2 t^2}{2}}$ and I doubt there's a way to express the latter via the former.

  • $\begingroup$ What about something simpler. Consider $X \sim N(\mu , \sigma)$ and consider $f(x)= x/2$. (Even you could make it simpler by taking $\mu=0$ and $\sigma=1$). $\endgroup$ – kolonel Jan 20 '14 at 3:31
  • 3
    $\begingroup$ This was meant as a counterexample. I didn't say it's impossible for some functions. $\endgroup$ – Alex Jan 20 '14 at 3:35
  • $\begingroup$ Thanks. But in your case $Y=f(X)$ is not increasing and one-to-one or am I missing something? $\endgroup$ – kolonel Jan 20 '14 at 3:49

This is more of a comment than answer and is for someone stumbling into this post.

For linear transformations of the form $Y=aX+b$, we can say that $\phi_Y(t)=e^{itb}\phi_X(at)$. If we are dealing with random vectors, then for transformation $AX+B$, we have $\phi_Y(t)=e^{it^\top B}\phi_X(A^\top t)$.

Source: https://www.statlect.com/fundamentals-of-probability/joint-characteristic-function and references therein.

  • $\begingroup$ please use the comment function for comments. $\endgroup$ – Fabian Rost May 9 '17 at 7:33

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.