Suppose X ~ Normal(0; 1) and Y = X + epsilon, where epsilon ~ Normal(0; sigma^2) and epsilon is independent of X. Conditional on observing Y , what is the distribution of X?
SBs for 1st correct answer
Edit - this is in response to //www.wallstreetoasis.com/forums/the-hardest-brainteaser-that-you-ever-did-see
Edit - solved below so scrolling down is cheating if you want to solve.
Well, by my calculations, the correct answer is -theta --- +epsilon(X-Y^epsilon+1)
Well, by my calculations, the correct answer is -theta --- +epsilon(X-Y^epsilon+1)
Bivariate Normal Distribution FTW
Zy,Zx f(x, y)dxdy = 1 f(x, y) >= 0
Where Z is the integral
the distribution of X is the same as the distribution of Y.
Y = X + epsilon, so X = Y - epsilon. Since epsilon is an unbiased normal variate, X and Y are equivalent under permutation. By symmetry, X and Y follow the same distribution.
lol
then i will go with my second choice: X ~ Normal(0; 1).
i hope this isn't a gag question.
11
X~N(y, sigma^2)
where Y observed is Y=y
This puzzle cannot be solved due to chaos theory . . .
normal (Y, sigma^2) ?
What kind of interview are they asking you this in?
Hey,
Is the answer:
P(X|Y) ~ N(rho*y/(sigma^2+1),(1-rho)^2)
where rho is the correlation coefficient of X and Y? (i.e. rho = Cov(X,Y)/(Var(X)Var(Y)) = Cov(X,Y)/(sigma^2+1))
I had to use paper and pencil for this [x has been edited to y, since y is observed...]
Ok what about this:
P(X|Y) ~ N(rho*y/A, 1-rho^2)
where A = sqrt(sigma^2+1)
I spent some time on this now, is it wrong? Again rho is the correlation coeff. AFAIK all Gaussians used here should be 1-D.
The answer is 42.
Can you pm me the answer? I'm kinda dying here.
Your answer is looking better. Hint - there is no rho in the answer. Let me know if you still want the answer.
Dr ... this isn't that hard.
(X|Y) ~ N(Y/(1+sig^2), (sig^2)/(1+sig^2))
Also ML, you can compute rho explicitly given independence of X and eps...you know the formula for covariance and you know the variances of X and Y....
Also, I never said this was that hard. But I think it is a very good question to test overall comfort with conditional probability. Just having the insight to say that E[X|Y] =!= Y, and E[X|Y] is between 0 and Y is sufficient here for most purposes.
Shouldnt the E(X|Y)=Y/(1+sigma) without the square power?
Also, a note to the above who fell into the trick of thinking the mean was Y (which is a gut instinct), an easy demonstration of the fallacy occurs when we allow
Y = X1 + X2, with X1 and X2 iid. (this is generalizable to Y = Sigma(X_i))
Then, clearly, E(X1|Y) = E(X2|Y) and by linearity E(X1|Y) + E(X2|Y) = Y, so E(X1|Y) = Y/2.
A more general form of this reasoning is used to give the result here.
Ok, I guessed I rushed it, yes rho = 1/sqrt(1+sigma^2) with X and epsilon independent. I don't know if this counts as a brainteaser though - its not involved but I doubt this will be asked at general analysts' interview, or maybe I'm wrong.
Autem sed qui cupiditate error hic. Animi sit sequi soluta et. Natus voluptatem ut facere et. Voluptatem temporibus quas repellat nobis.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...
Quae voluptatem aut eum suscipit voluptas voluptas et commodi. Enim consectetur ut modi. Numquam et id aspernatur iure hic doloremque ea. Neque sint qui est.
Quae ducimus qui vero dolorem quasi. Nihil nobis alias rerum ut excepturi. Voluptas dolor porro laborum quasi quam minus adipisci quas. A aperiam inventore cum eum aut totam fugit ut. Voluptas ut consequatur optio.
Facere voluptas consequatur voluptate molestiae doloribus nobis. Ut repellendus at qui nisi. Neque et iste voluptatum adipisci. Cum et et voluptate in.
Vitae voluptatem porro voluptatem quis. Iure omnis impedit doloremque fugit aut. Fugiat placeat non omnis qui nihil nisi accusamus. Similique dolore dolore excepturi velit quia placeat qui.