How to interpret results of the Empirical Distribution Test?

[kkraba]

Hi,

I have a time series that represents a demand. I need to find the distribution of that data. I am pretty sure that the best way to do it is through the Empirical Distribution Test. The problem I have is that I am not sure how to interpret the results of the test. The manual doesn't explain what we need to look for to decide whether the test against a the distributions provided (normal, exponential etc... ) passed or failed.

Do you have a pointer to that information by any chance? I know I am using the right tool, I just don't know how to interpret the results. The manual says "For a general description of empirical distribution function testing, see D’Agostino and Stephens
(1986)."
I don't have access to this book at his time and I need to come up with a solution for my assignment.

Thanks for your help!

Kamila

[trubador]

Test output should be pretty self-explanatory. The null hypothesis is that the data at hand follows the theoretical distribution you have specified. If the probability values of the empirical distribution test statistics (e.g. Anderson-Darling, Lilliefors, Cramer-von Mises, and etc.) are above a certain pre-specified significance level, say 0.05, then you cannot reject the hypothesis and assume that your data comes from that theoretical distribution ( with 95% confidence of course).

[EViews Gareth]

Note that the Empirical Distribution Test can be used to test whether a series matches a specific distribution. It should not be used to find which distribution a series does match.

[mectricsdonk]

Note that the Empirical Distribution Test can be used to test whether a series matches a specific distribution. It should not be used to find which distribution a series does match.

Why not? I know that when we do that we can give any data set very good fit, which can be false as the actual gdp might be different. But how do we in general find out what kind of distribution a dataset follows?

edit: QQplots or something?

edit: or? :eviews6:

[EViews Gareth]

Without getting too technical, if you run your data against 100 different empirical distributions, at 95% significance, you'll come up with 5 that say they match, even though they don't. In general you can't run multiple econometric/statistical tests in a row without adjusting for the fact that you have run multiple tests.