Proof of likelihood ratio ordering implying hazard rate ordering

56 Views Asked by At

This question was originally asked in stats.stackexchange (the link is given below). I'm reposting it here as I did not get any answer to the said post. https://stats.stackexchange.com/questions/423833/proof-of-likelihood-ratio-ordering-implying-hazard-rate-ordering


I'm studying stochastic orders and am stuck in the proof of likelihood ratio ordering implying hazard rate ordering.

Definitions: Let $X$ and $Y$ be two random variables with respective densities $f(t)$ and $g(t)$, and respective survival functions $\bar{F}(t):=\int_t^{\infty} f(u)\,du$ and $\bar{G}(t):=\int_t^{\infty} g(u)\,du$. Then we say that: $(i)$ $X \leq_{HR} Y$ if $\bar{G}\left(t\right)/\bar{F}\left(t\right)$ is nondecreasing in $t$ and $(ii)$ $X \leq_{LR} Y$ if $g\left(t\right)/f\left(t\right)$ is nondecreasing in $t$.

The Problem: To show that $X \leq_{LR} Y \implies X \leq_{HR} Y$.

My Progress: We assume that $X \leq_{LR} Y$. Then by definition, for every $t<t'$, we have $$\frac{g(t)}{f(t)} \leq \frac{g(t')}{f(t')},$$

i.e. $f(t)g(t') \geq f(t')g(t)$. Let $x \leq y$. Integrating over $(x, y) \times (y, \infty)$, we have $$\int_x^y \int_y^{\infty} f(t)\,g(t')\,dt'\,dt \geq \int_x^y \int_y^{\infty} f(t')\,g(t)\,dt'\,dt \tag{1}$$

I want to achieve: $$\int_x^{\infty} \int_y^{\infty} f(t)\,g(t')\,dt'\,dt \geq \int_x^{\infty} \int_y^{\infty} f(t')\,g(t)\,dt'\,dt, \tag{2}$$

so that we can separate out the integrals and write $$\int_x^{\infty} f(t)\,dt \int_y^{\infty} g(t')\,dt' \geq \int_y^{\infty} f(t')\,dt' \int_x^{\infty} g(t)\,dt,$$

i.e. $\bar{F}(x)\bar{G}(y) \geq \bar{F}(y)\bar{G}(x)$ for $x \leq y$. Since, the choice of $x$ and $y$ are arbitrary, $\bar{G}/\bar{F}$ is increasing, hence $X \leq_{HR} Y$.

Question: How do I come to $(2)$ from $(1)?$ Thank you.

P.S. If there is any other way to prove this result, please do share the source where I can find it.