Error with suest command "xx could not be restored"

[ties jongen]

Hi everyone!

I am trying to regress the outputs from two different regressions my code looks somewhat like the following:

use alldata
drop if bottom_portfolio==1
regress x y dummy1 dummy2 dummy3 dummy4
estimates store top_portfolio


use alldata
drop if bottom_portfolio==0
regress x y dummy1 dummy2 dummy3 dummy4
estimates store bottom_portfolio

suest top_regression bottom_regression

i get the following error code :
" estimation sample of the model saved under top_regression could not be restored
r(198);"



How can I solve this?

Would really appreciate the help!

[Joro Kolev]

Do not drop data, but rather use the condition in your regression, e.g.,

use alldata
regress x y dummy1 dummy2 dummy3 dummy4 if !bottom_portfolio
estimates store top_portfolio

[ties jongen]

hmm i am also manipulating the data from all to scratch, and my do file is already quite extensive and I am writing it in a way that I can apply it across various sets eg country industry division

Is there no way to store the regression results, then manipulate again, regress store and compare thereafter?


This is the structure my do file currently follows:
Chapters:
1. Opening complete dataset
2. Correct date window and remove companies with missing observations
3. Set and drop observations to get top 20% of [Industry; Within Industry; Country]
a. set top 20% industry
b. set top 20% division
c. set top 20% within industry
d. set top 20% country
4. Calculate value weighted average of top portfolio
a. calculating daily weighted averages
b. reducing table size to only dummies and regression relevant specs
5. Regress portfolio returns against market returns and add the included events dummies
6. save/export coeffients and standard errors for every event
7. Redo ch 1 - ch 6 for bottom 20%
8. Perform two sample t test on coeficients and aggregated coefficients
9. Format nicely and create final output

[Joro Kolev]

You are asking: "Is there a way to estimate my regression, after I drop the data on which my regression is defined." The answer is very generally, No, you cannot estimate your regression if you have dropped your data beforehand.

You might be misunderstanding what -suest- does: It estimates, it does not simply combine estimation results that exist already, but it estimates. I know this because in the stored estimates the covariance of the parameters in the two regressions is not present. -suest- calculates those with additional estimation.

hmm i am also manipulating the data from all to scratch, and my do file is already quite extensive and I am writing it in a way that I can apply it across various sets eg country industry division

Is there no way to store the regression results, then manipulate again, regress store and compare thereafter?


This is the structure my do file currently follows:
Chapters:
1. Opening complete dataset
2. Remove outliers, correct date window and remove companies with missing observations
3. Set and drop observations to get top 20% of [Industry; Within Industry; Country]
a. set top 20% industry
b. set top 20% division
c. set top 20% within industry
d. set top 20% country
4. Calculate value weighted average of top portfolio
a. calculating daily weighted averages
b. reducing table size to only dummies and regression relevant specs
5. Regress portfolio returns against market returns and add the included events dummies
6. save/export coeffients and standard errors for every event
7. Redo ch 1 - ch 6 for bottom 20%
8. Perform two sample t test on coeficients and aggregated coefficients
9. Format nicely and create final output

[ties jongen]

thank you for your answer;

Would it work if I would write the script in a way that it would export the "table" after chapter 4 (and add a dummy for the portfolio it is in; eg. (1= top portfolio and 0 = bottom portfolio)
then do chapter 1-4 for the bottom portfolio; export this table as well with the corresponding dummy
thereafter append the two files;
run and store the both regressions;
then run suest?

The export format of the table would look the following:
I would regress the meanret on msciworld_exc_ret and on the event dummies after

date    msciworld_excess_return    meanret    d_e1_5    d_e2_5    d_e3_5    d_e4_5
31/08/2015    -.0076258    .00537739    0    0    0    0
01/09/2015    -.0271782    -.03196324    0    0    0    0
02/09/2015    .0093487    -.00337053    0    0    0    0
03/09/2015    .0045    .01228631    0    0    0    0
04/09/2015    -.0167791    -.0201263    0    0    0    0
08/09/2015    .0177911    .01792618    0    0    0    0
09/09/2015    .0006654    .00785138    0    0    0    0
10/09/2015    -.001703    -.011211    1    0    0    0
11/09/2015    .0009762    -.00844655    1    0    0    0
14/09/2015    -.0041251    -.0057015    1    0    0    0
15/09/2015    .0084772    .00159139    1    0    0    0
16/09/2015    .0121165    .02974624    1    0    0    0
17/09/2015    -.0005626    .0002806    1    0    0    0
18/09/2015    -.0134267    -.0190928    1    0    0    0
21/09/2015    -.0010788    -.00878104    1    0    0    0
22/09/2015    -.0172922    -.02158263    1    0    0    0
23/09/2015    -.0025614    -.01217949    1    0    0    0
24/09/2015    -.0058133    -.00806042    1    0    0    0
25/09/2015    .0046136    .00299614    0    0    0    0
28/09/2015    -.0217343    -.03789479    0    0    0    0
29/09/2015    -.0061595    -.00107269    0    0    0    0
30/09/2015    .0203085    .02136067    0    0    0    0
01/10/2015    .0038318    .00538688    0    0    0    0
02/10/2015    .0120563    .02125733    0    0    0    0
05/10/2015    .0188938    .04294595    0    0    0    0
06/10/2015    .0021308    .0238574    0    0    0    0
07/10/2015    .0076083    .03440584    0    0    0    0
08/10/2015    .0061117    .01125257    0    0    0    0
09/10/2015    .0061323    .01879123    0    0    0    0
12/10/2015    .0003658    -.00979907    0    0    0    0
13/10/2015    -.0079534    -.01669142    0    0    0    0