Hi,
let me tell you about my problem in a bit more detail for better
understanding
I have market values of companies "MV" as observed on at the market
(stock exchange).
I try to find out which valuation model fits best for which country.
(I have 20 countries a 8 to 120 companies)
I have a model that calculates a company value "pred" (by discounting
dividends)
Now I assume that the valuation model should predict the market value
correctly.
If I plot a graph with the predicted value on the x-axsis and the
observed market value on the y-axis I would expect a line with the
slope of 1 and an intercept of 0.
I construct 2 models
1. MV = �0+�1pred + err or ln(MV) = ln�0+�1ln(pred) + err
2. MV = pred +err or ln (MV) = ln (pred)
to test the null hypostesis if �0 = 0 and �1 = 1 as this is what I
expect.
I calculate the F-test:
(ESSr - ESSu) / 2
F= --------------------------------
ESSu / (n-2)
and get F and p values.
I am looking for accepting the null hypothesis as I want to prove that
the valuation model is correct
Now THE QUESTION to the signifficance level
If the data is really good, meaning the unrestricted model shows
nearly the expected behaviour, then the restricted model would not
improve/worsen the situation very much.
HOW DO I set the signifficance level?
I would then get a very low p-value for the acceptin the null
hypothesis.
Here is a sample
country model n VARr SSEr SSEu df n df d
F p sig level
USA dre 107 0.138 14.628 13.947 2 104
2.539040654 0.0838 0.05
Canada dre 16 0.262 3.93 3.916 2 13
0.023237998 0.9771
Belgium dre 5 2.493 9.972 3.702 2 2
1.693679092 0.3712
England dre 33 0.212 6.784 6.285 2 30
1.190930788 0.3179
I KNOW that the data for the USA is very good (it shows very little
under-valuation) compared to the UK data.
NOW the tricky part: I know that the data is good WHY do I get such a
low p-value? I know that the model is very good in estimating the
value because the average undervaluation is only about 14% compared to
the average overvaluation of 192% in the UK.
What is wron here?
I would appreciate any help here.
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
