How to differentiate $E[X\vert Y\geq y](1-F(y))$ w.r.t $y$

68 Views Asked by At

It is known that $$E[X\vert Y\geq y](1-F(y)) = \int_y^\infty E[X\vert Y=t]f(t)dt$$

To take the derivative, I think this is just Leibniz rule, but I don't know how to handle the conditional expectation on the RHS in the integral.

Do I need to write the conditional Expectation as an integral and then take the derivative of the double integral?

I don't see double integrals in the answer though, which is why I am confused. The answer is

$$\frac{d}{dy}E[X\vert Y\geq y] = f(y)\frac{E[X\vert Y\geq y] - E[X\vert Y=y]}{1-F(y)}$$

Also, if I use Leibniz rule I don't get an $E[X\vert Y\geq y ]$ term

1

There are 1 best solutions below

2
On BEST ANSWER

We just need to differentiate $\int_y^\infty f(\tau) \,d\tau$: $$\frac d {dy} \operatorname{E}(X \mid Y \geq y) = \frac d {dy} \frac {\int_y^\infty \operatorname{E}(X \mid Y = \tau) \,f_Y(\tau) \,d\tau} {\operatorname{P}(Y \geq y)} = \\ -\operatorname{E}(X \mid Y \geq y) \,\frac {\frac d {dy} \operatorname{P}(Y \geq y)} {\operatorname{P}(Y \geq y)} + \frac {\frac d {dy} \int_y^\infty \operatorname{E}(X \mid Y = \tau) \,f_Y(\tau) \,d\tau} {\operatorname{P}(Y \geq y)} = \\ \operatorname{E}(X \mid Y \geq y) \,\frac {f_Y(y)} {1 - F_Y(y)} - \frac {\operatorname{E}(X \mid Y = y)\,f_Y(y)} {1 - F_Y(y)}.$$