wilmott.com

Hi there,A question about tracking error. The usual definition of (ex post) tracking error is the standard deviation of the difference between the benchmark and the actual portfolio. Someone did me the suggestion that a better and the correct formulation is that tracking error is SQRT[( Mean^2 )+(Var)]. Where you need to know the sign of the mean to have meaningful information.Can anyone confirm this expression and eleborate on it?Has this also consequences for ex ante tracking error which I usually approach with w'SIGMAw. With w a vector of weighting differences between the portfolio and the benchmark en SIGMA the variance covariance matrix.

This just follows from the definition of a variance doesn't it? The measure you first propose is the sample analogue of E(X-Y)^2. The variance of (X-Y) is equal to E(X-Y)^2 - (E(X-Y))^2 so rearranging tells us that E(X-Y)^2 = Var(X-Y) + (E(X-Y))^2 i.e. the variance of the tracking error plus the square of the mean.richg.

Hi the Red Sniper,I have a little answer for your tracking error problem.Let R be the difference between your portfolio return and the benchmark return R=Rp-Rblet X = [R-Avg(R)]² with Avg(R) is the mean of RSo the Tracking Error becomes TE = sqrt(Avg(X))In a world of normally ditributed returns, the portfolio would not deviate from its benchmark by more than TE in two out of any three given years, and would not underperform by more than TE in 8.4 out of every 10 years.Yours sincerely.FABIUS

hi, it seems that there r many ways to measure tracking error. Standard deviation of the difference betw the return of tracking portfolio and the benchmark is the one, the standard deviation of the risudual error( = sigma (RB)* sqrt(1- correl(RB,RP)) is another way. RB- the return of benchmark index, RP- the return of tracking portfolio. I am wondering the background of the 2nd formular and in what situation it is being used. I assume that the risudual error term is from the CAPM. RP= RB*Beta+Alpha+ ErrorCan anybody help?