Hi
Say I have 2 batches: X(n) and Y(n), of which X is a set of sample
forecast for Y in month n and Y is the actual data in real time in month
n.
But the sample in both X and Y can contain zero. I want to see how much
error (avg and deviation) in percentage in each month n and over all
months. So this will fail me if I did a simple comparison of (X-Y)/Y for
each individual ie. (Xi - Yi)/Yi in month n, as Y can have zero.
Would this work if I estimate this by:
Average error in month n = average(Xn-Yn) / average(Yn)
and the std dev of the sample = StdDev(Xn-Yn) / StdDev (Yn)
I somehow believe that the average error in the above formula can give me
a close approx but not too sure with the std dev... any suggestion?
Also this problem develops: if you have a set of {X1(n), Y1(n)}, {X2(n),
Y2(n)},... {Xm(n), Ym(n)}
What would be the best way to approximate the error (%) of X-Y in the
general term (for all m), provided you dont have access to the raw data,
except only those averages and StdDev's described in the above formulas.
Any suggestion is a great help
Regards
Peppy Adi-Purnomo
----------------------------------------
This email was sent using Actrix Webmail
http://www.actrix.co.nz
|