How to run out-of-sample test with clogit?

[Max Immonen]

Hi all,

I have ran conditional logistic regression with my estimation sample (data from 2001-2014) and based on it selected the most relevant and least multicollinear variables. Now I want to run conditional logistic regression within my prediction sample (2015-2018). Therefore, I want to have all the same coefficients generated from the model when running it within estimation sample as I am now running it within the prediction sample.

How do I save these coefficients and run the regression again with different(prediction sample) dataset?

Estimation sample and prediction sample are two different excel data sets, they are not combined.

Thanks in advance

BR
Max Immonen

[Clyde Schechter]

-help estimates store-
-help estimates restore-
-help estimates save-
-help estimates use-

For predicted probabilities pc1 or pu0, or for xb, -predict- will provide out-of-sample predictions.

Also, if it makes sense in other respects, you could -append- the two data sets together before doing the original conditional logistic regression. In that case, you wouldn't even need to store or save the estimates. Just use -if inrange(year, 2001, 2014)- for the estimation sample regression, and then apply -predict- with no -if- clause.

[Max Immonen]

Hi Clyde,

Just to be sure I did this right way I copied the commands from STATA below.

1) I imported the estimation sample to STATA
2) I ran clogit within the estimation sample
3) I used ''estimates store'' command
4) I imported the prediction sample to STATA
5) I used ''estimates restore'' command
6) I used ''predict, pc1'' command

So now I should have the takeover probabilities calculated with coefficients from estimation sample but within prediction sample data?

Thank you very much!



import excel "/Users/omistaja/Desktop/STATA.READY.DATA.xlsx", sheet("Estimation sample") firstrow clear

. clogit Target1NonTarget0 LNMARKETCAP LNEV LNNETSALES LNTOTALASSETS PE PB PS SalesGrowth ROE ProfitMargin EBITDAMargin Operating
> CashFlowMargin AssetTurnover Gearing CurrentRatio Liquidity InvestmentBehavior Leverage IndustrydisturbanceDUMMY GRDUMMY, group
> (YEARSIC3)
note: multiple positive outcomes within groups encountered.
note: 2,497 groups (8,780 obs) dropped because of all positive or
all negative outcomes.

Iteration 0: log likelihood = -873.87402
Iteration 1: log likelihood = -856.33418
Iteration 2: log likelihood = -856.26146
Iteration 3: log likelihood = -856.26145

Conditional (fixed-effects) logistic regression

Number of obs = 5,015
LR chi2(20) = 582.64
Prob > chi2 = 0.0000
Log likelihood = -856.26145 Pseudo R2 = 0.2539

------------------------------------------------------------------------------------------
Target1NonTarget0 | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------------------+----------------------------------------------------------------
LNMARKETCAP | -1.447227 .1717374 -8.43 0.000 -1.783826 -1.110628
LNEV | 2.433571 .1948839 12.49 0.000 2.051606 2.815536
LNNETSALES | -1.446167 .228276 -6.34 0.000 -1.89358 -.9987539
LNTOTALASSETS | .6060187 .2335801 2.59 0.009 .1482101 1.063827
PE | .0017869 .0011036 1.62 0.105 -.0003761 .0039499
PB | -.0015023 .022008 -0.07 0.946 -.0446372 .0416327
PS | -.0242994 .0337597 -0.72 0.472 -.0904671 .0418684
SalesGrowth | -.0264502 .0404047 -0.65 0.513 -.1056419 .0527416
ROE | .0087524 .199407 0.04 0.965 -.3820782 .3995831
ProfitMargin | 1.229701 .6304049 1.95 0.051 -.0058699 2.465272
EBITDAMargin | -1.621412 .4868985 -3.33 0.001 -2.575716 -.6671084
OperatingCashFlowMargin | .3656425 .3313136 1.10 0.270 -.2837202 1.015005
AssetTurnover | .9327297 .1788173 5.22 0.000 .5822542 1.283205
Gearing | -.1243681 .0518317 -2.40 0.016 -.2259563 -.0227799
CurrentRatio | -.4329885 .0678735 -6.38 0.000 -.5660181 -.299959
Liquidity | 6.982205 .5560583 12.56 0.000 5.89235 8.072059
InvestmentBehavior | .4625813 .5579098 0.83 0.407 -.6309019 1.556065
Leverage | -1.910667 .5446221 -3.51 0.000 -2.978107 -.8432278
IndustrydisturbanceDUMMY | .5266789 .1982566 2.66 0.008 .1381032 .9152547
GRDUMMY | .1721919 .1326558 1.30 0.194 -.0878088 .4321925
------------------------------------------------------------------------------------------

. estimates store
name required
r(100);

. estimates store esamplecoeffi

.
. import excel "/Users/omistaja/Desktop/STATA.READY.DATA.xlsx", sheet("Prediction sample") firstrow clear

. estimates restore esamplecoeffi
(results esamplecoeffi are active now)

. estimates dir esamplecoeffi

-------------------------------------------------------
name | command depvar npar title
-------------+-----------------------------------------
esamplecoe~i | clogit Target1Non~0 20
-------------------------------------------------------

. predict pc1
(option pc1 assumed; probability of success given one success within group)
(1 missing value generated)

. brow

. rename pc1 takeoverprobability

[Clyde Schechter]

Yes, except for your glitch with -estimates store- the first time, which you immediately corrected, this looks right to me.