2
$\begingroup$

Let $X$ be a random variable taking positive values with the density function $f$.
Let $F(t) = P(X\leq t)$ be distribution function of $X$ and $$ h(t) = \int_0^{t} \frac{f(s)}{1-F(s)}\,ds$$ the so-called cumulative hazard function of $f$.

How could I show that $h(X)$ is exponentially distributed with parameter $\lambda = 1$?

I would be grateful for any ideas or suggestions. Thanks.

$\endgroup$
2
$\begingroup$

A standard result says that if $Y$ is a random variable whose c.d.f. is continuous, then $F_Y(Y)$, the cumulative probability distribution of $Y$ evaluate at the random variable $Y$, is uniformly distributed on $(0,1)$, so that $\Pr(F_Y(Y) \le y)=y)$ if $0\le y\le1$. $$ \Pr(F_Y(Y)\le y) = \Pr(Y\le F_Y^{-1}(y)) = F_Y\Big(F_Y^{-1}(y)\Big) = y. $$ Some difficulties arise when $F_Y$ is not one-to-one, but they can be overcome and have probably been written up in some other question posted here. I'll see if I can find it.

Now \begin{align} \Pr(h(X)\ge x) & = \Pr\left( \int_0^X \frac{f(s)}{1-F(s)}\,ds \ge x \right) = \Pr\left( \int_1^{1-f(X)} \frac{-du}{u} \ge x \right) \\[8pt] & = \Pr\left( -\log(1-F(X)) \ge x \right) = \Pr(F(X) \ge 1-e^x) \\[8pt] & = e^x. \end{align}

$\endgroup$
1
$\begingroup$

$$\int_{0}^{t}\frac{f(s)}{1-F(s)}~ds=\int_{0}^{t}h(s)~ds$$ $$\Rightarrow [-\log (1-F(s))]_{0}^{t}=\int_{0}^{t}h(s)~ds$$ $$\Rightarrow -\log[1- F(t)]+\log [1-F(0)]=\int_{0}^{t}h(s)~ds$$ $$\Rightarrow -\log[1- F(t)]=1-\exp[-\int_{0}^{t}h(s)~ds]$$ In particular, if $h(s)=\theta \forall s >0$, then $$F(t)=1-\exp[-\int_{0}^{t}\theta~ds]=1-\exp[-\theta t],t>0$$ Then the PDF of T is $$f(t)=\theta e^{-\theta t},t>0$$ which is the PDF of Exponential Distribution with mean $\frac{1}{\theta}$. Put $\theta=1$ in your problem.

Note: $$\lim_{dt \to 0}\frac{P[t<T<t+dt|T>t]}{dt}=\frac{F'(t)}{1-F(t)}=\frac{f(t)}{1-F(t)}=H(t)$$ is known as instantaneous failure rate or hazard rate at time t.

$\endgroup$

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.