Conditional expectation of sum of two uniform random variables

398 Views Asked by At

I want to compute the conditional expectation of the sum of two independent uniform random variables, with two conditions. Is there anything wrong with my approach below?

Formally, let $X,Y$ both be independent draws from $U(0,1)$. I want the expectation of $X+Y$, conditional on $X\ge c_1$, where $c_1$ is a constant, and conditional on $X+Y\ge c_2 \iff Y\ge c_2-X$, where $c_2$ is another constant.

Edit: Via double integrals: $$E[X+Y|X\ge c_1,X+Y\ge c_2]=\frac{\int_0^1 \int_0^1 (x+y) 1\{x\ge c_1\}\cdot 1\{x+y\ge c_2\} dy dx}{\int_0^1 \int_0^1 1\{x\ge c_1\}\cdot 1\{x+y\ge c_2\} dy dx}$$ $$=\frac{\int_{c_1}^1 \int_0^1 (x+y) 1\{x+y\ge c_2\} dy dx}{\int_0^1 \int_0^1 1\{x\ge c_1\}\cdot 1\{x+y\ge c_2\} dy dx}$$ $$=\frac{\int_{c_1}^1 \int_{\max\{0,c_2-x\}}^1 (x+y) dy dx}{\int_0^1 \int_0^1 1\{x\ge c_1\}\cdot 1\{x+y\ge c_2\} dy dx}$$ $$=\frac{\int_{c_1}^1 \left[ xy+y^2/2\right]_{\max\{0,c_2-x\}}^1 dx}{\int_0^1 \int_0^1 1\{x\ge c_1\}\cdot 1\{x+y\ge c_2\} dy dx}$$ $$=\frac{\int_{c_1}^1 x(1-\max\{0,c_2-x\})+\frac{1-\max\{0,c_2-x\}^2}{2} dx}{\int_0^1 \int_0^1 1\{x\ge c_1\}\cdot 1\{x+y\ge c_2\} dy dx}$$ Now, for $x\ge c_2$, $\max\{0,c_2-x\}=0$, and $\max\{0,c_2-x\}=c_2-x$ otherwise. Hence, for $c_2> c_1$, $$=\frac{\int_{c_1}^{c_2} x(1-c_2+x)+\frac{1-(c_2-x)^2}{2} dx+\int_{c_2}^1 x+1/2 dx}{\int_0^1 \int_0^1 1\{x\ge c_1\}\cdot 1\{x+y\ge c_2\} dy dx}$$ $$=\frac{\left[ \frac{(1-c_2)x^2}{2}+\frac{x^3}{3}+\frac{x(1-c_2^2)}{2}-\frac{x^3}{6}+\frac{c_2x^2}{2} \right]_{c_1}^{c_2}+\left[ \frac{x^2+x}{2} \right]_{c_2}^1}{\int_0^1 \int_0^1 1\{x\ge c_1\}\cdot 1\{x+y\ge c_2\} dy dx}$$ where I will not simplify the numerator further since it will remain relatively messy, with cubic terms for $c_1$ and $c_2$. Instead of integrating in the denominator, I use basic geometry to calculate the area, which for $c_2>c_1$ is a square minus a triangle. Hence, $$=\frac{\left[ \frac{(1-c_2)x^2}{2}+\frac{x^3}{3}+\frac{x(1-c_2^2)}{2}-\frac{x^3}{6}+\frac{c_2x^2}{2} \right]_{c_1}^{c_2}+1-\frac{c_2^2+c_2}{2}}{1-c_1+c_1c_2-\frac{c_1^2+c_2^2}{2}}.$$ Now, for $c_1,c_2=0$, the two conditions become non-binding, and this expression simplifies to 1, which is the correct unconditional expectation.

Is this approach correct, and can you spot any mistakes in the computations?

1

There are 1 best solutions below

0
On BEST ANSWER

(Moving comment to answer as OP suggested.)

The approach and most of it looks fine — you’re definitely on the right track. However, I think the lower limit of $\int^1_{\max(0,c_2-x)}$ should be $\min(1,\max(0,c_2-x))$. For example, sketch the region of possible $(x,y)$ in the case $c_1=0.1,c_2=1.2$. This is surprisingly tricky to do in general!