6
$\begingroup$

What is the distribution of $\min\{0, X\}$ when $X$ follows some general normal distribution?

$\endgroup$
  • $\begingroup$ Maybe it can help to know that $\forall x,y\in\Bbb R,\;\min(x,y)= \dfrac{x+y-|x-y|}2$. $\endgroup$ – paf Aug 2 '18 at 10:37
2
$\begingroup$

Suppose $X\sim N(\mu, \sigma^2)$ with CDF $\Phi(\cdot; \mu, \sigma^2)$. Then you can calculate the probability $p$ that $X>0$, by taking

$$ p = 1- \Phi(0; \mu, \sigma^2) $$

Then $Y:=\min\{0,X\}$ is a between

  • a point mass at $0$ with weight $p$
  • and a truncated normal distribution with parameters $\mu$ and $\sigma^2$, truncated from above at $0$, with weight $(1-p)$.

This is also known as a "censored normal distribution". You may be interested in previous threads tagged both "censoring" and "normal-distribution". We have a thread on the mean and variance in the multidimensional case.

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ +1. However, censoring may be a little more refined than suggested here. You describe a "delta (truncated) normal" distribution or "zero inflated" truncated normal, rather than a censored normal. The distinction is that in the present case the zeros will be evaluated as zeros whereas with censoring the zeros are considered to be the intervals $[0, \infty).$ This isn't a superficial difference: it changes likelihood calculations for instance. $\endgroup$ – whuber Aug 2 '18 at 12:02
1
$\begingroup$

We have a simple answer for the distribution of $\min(0,X)$ for any random variable $X$ in terms of its distribution function.

Suppose $Y=\min(0,X)$. That is, $$Y=\begin{cases}X&,\text{ if }X\le 0\\0&,\text{ if }X>0\end{cases}$$

Trivially, a plot of $y=\min(0,x)$ for real $x$ would look like

enter image description here

Then distribution function (DF) of $Y$ is simply

$$P(Y\le y)=\begin{cases}P(X\le y)&,\text{ if }y<0\\1&,\text{ if }y\ge 0\end{cases}$$

For $X\sim\mathcal N(\mu,\sigma^2)$, we get

\begin{align} P(Y\le y)=\begin{cases}\Phi\left(\frac{y-\mu}{\sigma}\right)&,\text{ if }y<0\\1&,\text{ if }y\ge 0\end{cases} \end{align}

, where $\Phi$ as usual is the DF of standard normal distribution.

As already mentioned, $Y$ has a mixed distribution as it contains the mass point $0$ for $X>0$ as well as having a continuous part for $X\le 0$.

| cite | improve this answer | |
$\endgroup$

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.