Dimson Beta Calculation

[louisgwie]

Hello, I would like to know if there is a simple way to estimate this equation (dimson beta) on eviews 6

rit = c + b1*rm(t-1) + b2*rm(t) + b3*rm(t+1)

it's a dimson beta equation. ex. If I want to have beta for year 2006, then it would be

ri(2006) = c + b1*rm(2005) + b2*rm(2006) + b3*rm(2007)

how should I manage the structure of the data? should it be time series daily or panel data?
I actually have made a beta for one company in which I put the series (2005,2006,2007,etc) in a different object. But this won't be helpful because I'd have to work with 277 Companies and 2005-2010 data span. Any suggestion?

[EViews Glenn]

The general part of the answer is

Code: Select all

equation eq01.ls rit c rm(-1) rm rm(1)

The other part depends on how you want to handle the panel nature of your data. Do you want the coefficients to be restricted across cross-sections? Different intercepts?

[louisgwie]

Thanks Glenn. But I actually have separated my data into "unstructured data". So, I have different series for different company and different year. (ex. comp1_2005, comp1_2006, company2_2005, company2_2006, etc)
Is it possible to process all data with the same equation at once?

I want to process the data like this:

"genr rcomp1_2005=d(log(comp1_5))"

but I want to apply that equation for all series. Is there any way to do that? thanks a lot

[EViews Glenn]

I'm very confused about the structure of your data. You have

. 277 companies observed over the period 2005-2010
. your workfile is unstructured
. you have different series for companies and year (eg., comp1_2005, comp1_2006)

These things don't seem to add up. Are these data non-annual? How many observations are in your unstructured workfile? What does the COMP1_2005 series look like?

[louisgwie]

My data is indeed a time-series data (could be panel data). It consist of 277 companies within 2005-2010 period (daily stock price). But the problem is, if I use time-series for my workfile structure (which is 1 series for 1 company, 2005-2010) it would be impossible for me to apply the equation, wouldn't it?

Rit = c + b1*Rm(t-1) + b2*Rm(t) + b3*Rm(t+1)

ex. Ri(2006) = c + b1*Rm(2005) + b2*Rm(2006) + b3*Rm(2007)

How could I recall only "some" part of the data (2005 only, 2006 only, etc) for the equation, meanwhile each of the series consist of daily price from 2005-2010? (if I used time-series data). Or there is a way to do that?

Therefore, I treat the data as an unstructured data so that I could separate the series base on their company AND the year..

Is my explanation clear enough? Sorry if it makes you confuse, glenn. Really appreciate your feedback..

[EViews Glenn]

That helps.

Do you want to estimate this for each firm, assuming the same relationship across firms? Or different relationships for each firm?

[louisgwie]

Yes,exactly. I want to estimate the beta for each company (2006-2009)

[EViews Glenn]

To confirm: a separate relationship (different coefficients) for each company?

[louisgwie]

"ex. Ri(t) = c + b1*Rm(t-1) + b2*Rm(t) + b3*Rm(t+1)"

i: Company
t: period (year)

so I'll have different coefficients (b1,b2,b3) for EACH company and EACH year. Is it also what you mean?

[EViews Glenn]

I'm still confused, you write

"ex. Ri(t) = c + b1*Rm(t-1) + b2*Rm(t) + b3*Rm(t+1)"

i: Company
t: period (year)

so I'll have different coefficients (b1,b2,b3) for EACH company and EACH year.

and

ex. Ri(2006) = c + b1*Rm(2005) + b2*Rm(2006) + b3*Rm(2007)

but then say that your data are daily. Are the lags in your equation really over years? How does the daily nature of the data come into play? How can you have different coefficients for each firm and year if the lags are over year?

Let's make this easier. Suppose you have R as the economic variable of interest. There are two firms A and B, and daily data for 2010 and 2001. What series do you have?

R_A_2010
R_A_2011
R_B_2010
R_B_2011

or something different? How many observations are in the workfile?

[jasimalajmi]

were you able to calculate Dimson beta for more one company

[jasimalajmi]

I have 150 columns of rate of return of 150 companies and one column of market rate of return
I would like to know if it possible to run regression for each company (that is running 150 regressions) using the same independents variables but different dependents (company's rate of return.
If yes, how?
Thanks

[EViews Glenn]

There are a number of examples in the programming section of this kind of loop. Basically, you'll loop over the dependent variables and estimate specifications with the same independent variables. See the programming reference under loops or perhaps even better, look at Gareth's introduction to programming posts in the forum...

viewtopic.php?f=5&t=1638

[jasimalajmi]

Thanks a lot.
It looks that the examples cover different situations. I want regress each y against one or more x.
It would be highly appreciated if you can direct to a sample suitable for my case.

[EViews Glenn]

Not exactly what you want, but it should be enough...

viewtopic.php?f=5&t=5333&hilit=loop+regressions