Mean, StDev, N, t-value in one command

[Anthony Kira]

Hello Guys

I have a dataset which shows the following:

Firm	date	return	DayofWeek
AARGKB	January 3, 2008	-0.001	4
AARGKB	January 4, 2008	-0.007	5
AARGKB	January 7, 2008	-0.013	1
AARGKB	January 8, 2008	-0.0121	2
.....

I have a dataset with over 200 different firms. Now i want to have this result:

DayofWeek	Mean	St. Dev	N	t-value
1
2
3
4
5

So im trying to collapse the mean of every monday in one, the mean of every tuesday in one and this for every day of the week.

collapse (mean) adjretadd1, by(dayofweek)

With this code i managed to get the mean, which i think is correct. But when i now try to get the other ones (StDev, N and t-value), i don't know how i can do it with one command.

correlate dailyreturns, means cov

With this, i get then the mean of the 5 days in one, but that's not what i wanted. I also get only N of 5, which is also not correct (the problem is that i then take the results i get from my first command and calculate with them, but i want to calculate with the initial dataset). Is there a command with which i get the results i wanted at once?

Thank you very much for every help.

Best,
Anthony

[Clyde Schechter]

-collapse- is not restricted to calculating one statistic.

collapse (mean) Mean = adjretadd1 (sd) Std_Dev = adjretadd1 (count) N = adjretadd1, by(dayofweek)

That will get you everything but "t-value" in one command. I'm not sure I grasp what kind of "t-value" you are looking for here. If you mean a t-statistic for the hypothesis that Mean = 0, you can calculate that easily after the above with:

gen t_stat = Mean/(Std_Dev/sqrt(N))

[Oscar Ozfidan]

bysort DayofWeek: summarize return
would give you everything but t-value
if you want to get the t value of the coefficient estimate for the mean of returns you need to
reg return i.DayofWeek
those would be the t values to test if mean return by day of week is different than the mean of the left out category
if you want the tvalue for the test if the mean is different from 0 then you can do

reg return i.DayofWeek, noconstant

[Anthony Kira]

Hello Clyde and Oscar

Thank you very much, both ways worked and you helped me a lot!
I have one question remaining; If i do either our your ways, i get for N a value of 364 for every day (only a sample, not the whole data). Is this N the number of Mondays/Tuesdays..../Fridays i have in my data?

[Oscar Ozfidan]

Yes each N would be the number of data points from which the stats are calculated for each day. If you had data missing for one Friday lets say N would be one less than the other Ns for the Friday stats.

[Anthony Kira]

Hello Clyde and Oscar

Thank you very much, both ways worked and you helped me a lot!
I have one question remaining; If i do either our your ways, i get for N a value of 364 for every day (only a sample, not the whole data). Is this N the number of Mondays/Tuesdays..../Fridays i have in my data?

To be more specific: Lets say i take the return of a firm for 10 years. So i have approx. 500 mondays for this firm. Only in 400 Mondays i have actually a return, in the other 100 mondays the firm was not listed or illiquid (return empty).
The N should be 400 and not 500. Does the command from Clyde or Oscar take this into account?


UPDATE: My post crossed with Oscar's.
Thank you very much Oscar and Clyde!

[Oscar Ozfidan]

Yes if return is missing for a specific day, that day is not included in as an observation. So, N you would see would correspond to 400 in your example in either approach.