Hello,
Stata 14.1 user here! I am currently writing about the effect that sovereign credit rating actions have on the ratings of banks.
Dependent variable:
BDNn - bank is downgraded n notches (n=0,1,2,3)
Independent variables:
SDN_1 - sovereign is downgraded 1 notch (0 or 1)
SDN_2 - sovereign is downgraded 2 notches (0 or 1)
SDN_3 - sovereign is downgraded 3 or more notches (0 or 1)
SUP_1 - sovereign is upgraded 1 notch (0 or 1)
SNWQ - sovereign is on negative watch (0 or 1)
SPWQ - sovereign is on positive watch (0 or 1)
Sovereign rating - control var (from 1 to 24)
There are more variables and multiple versions but this is the 'simpler' version of the model. My initial code is:
xtset ID_number Date
xtologit BDNn SDN_1 SDN_2 SDN_3 SUP_1 SNWQ SPWQ SovereignRating
What I want to known is if SDN_1=1 what the likelihood of BDNn=1,2 or 3 or more notches is?
From what I saw, I should use the margins command. In order to do that it seems I should identify my dummy variables as categorical:
Is this correct to assume? Now onto the margins code. I first did margins, dydx(*)
but as per https://www.statalist.org/forums/for...nt-interaction the results may be meaningless. If not, then how would these be interpreted?
Would doing margins, dydx(SDN_1) predict(pu0 outcome(1)) or margins, dydx(SDN_1) - for each independent variable - be better? Any other suggestion?
Thank you very much in advance!
Stata 14.1 user here! I am currently writing about the effect that sovereign credit rating actions have on the ratings of banks.
Dependent variable:
BDNn - bank is downgraded n notches (n=0,1,2,3)
Independent variables:
SDN_1 - sovereign is downgraded 1 notch (0 or 1)
SDN_2 - sovereign is downgraded 2 notches (0 or 1)
SDN_3 - sovereign is downgraded 3 or more notches (0 or 1)
SUP_1 - sovereign is upgraded 1 notch (0 or 1)
SNWQ - sovereign is on negative watch (0 or 1)
SPWQ - sovereign is on positive watch (0 or 1)
Sovereign rating - control var (from 1 to 24)
There are more variables and multiple versions but this is the 'simpler' version of the model. My initial code is:
xtset ID_number Date
xtologit BDNn SDN_1 SDN_2 SDN_3 SUP_1 SNWQ SPWQ SovereignRating
What I want to known is if SDN_1=1 what the likelihood of BDNn=1,2 or 3 or more notches is?
From what I saw, I should use the margins command. In order to do that it seems I should identify my dummy variables as categorical:
Code:
. xtologit BDNn i.SDN_1 i.SDN_2 i.SDN_3 i.SUP_1 i.SNWQ i.SPWQ SovereignRating
Fitting comparison model:
Iteration 0: log likelihood = -4900.4703
Iteration 1: log likelihood = -4556.0509
Iteration 2: log likelihood = -4497.6561
Iteration 3: log likelihood = -4485.8624
Iteration 4: log likelihood = -4485.7416
Iteration 5: log likelihood = -4485.7374
Iteration 6: log likelihood = -4485.7371
Iteration 7: log likelihood = -4485.7371
Refining starting values:
Grid node 0: log likelihood = -4577.397
Fitting full model:
Iteration 0: log likelihood = -4577.397 (not concave)
Iteration 1: log likelihood = -4491.5689
Iteration 2: log likelihood = -4483.3882
Iteration 3: log likelihood = -4482.3499
Iteration 4: log likelihood = -4482.3486
Iteration 5: log likelihood = -4482.3486
Random-effects ordered logistic regression Number of obs = 5,822
Group variable: ID_number Number of groups = 2,016
Random effects u_i ~ Gaussian Obs per group:
min = 1
avg = 2.9
max = 21
Integration method: mvaghermite Integration pts. = 12
Wald chi2(7) = 640.84
Log likelihood = -4482.3486 Prob > chi2 = 0.0000
---------------------------------------------------------------------------------
BDNn | Coef. Std. Err. z P>|z| [95% Conf. Interval]
----------------+----------------------------------------------------------------
|
1.SDN_1 | 1.282189 .1026812 12.49 0.000 1.080937 1.48344
1.SDN_2 | 1.55724 .1463606 10.64 0.000 1.270378 1.844101
1.SDN_3 | 3.002522 .2455032 12.23 0.000 2.521345 3.4837
1.SUP_1 | -2.525022 .3708289 -6.81 0.000 -3.251833 -1.798211
1.SNWQ | -1.916259 .1896509 -10.10 0.000 -2.287968 -1.54455
1.SPWQ | -18.92281 17955.19 -0.00 0.999 -35210.45 35172.61
SovereignRating | .0262496 .0090356 2.91 0.004 .0085401 .043959
----------------+----------------------------------------------------------------
/cut1 | .989602 .049952 19.81 0.000 .8916979 1.087506
/cut2 | 2.977047 .0744732 39.97 0.000 2.831082 3.123012
/cut3 | 4.301772 .1079608 39.85 0.000 4.090173 4.513372
----------------+----------------------------------------------------------------
/sigma2_u | .1130177 .0491958 .0481531 .2652582
---------------------------------------------------------------------------------
LR test vs. ologit model: chibar2(01) = 6.78 Prob >= chibar2 = 0.0046
Code:
. margins, dydx(*)
Average marginal effects Number of obs = 5,822
Model VCE : OIM
dy/dx w.r.t. : 1.SDN_1 1.SDN_2 1.SDN_3 1.SUP_1 1.SNWQ 1.SPWQ SovereignRating
1._predict : Pr(0.BDNn), predict(pr outcome(0))
2._predict : Pr(1.BDNn), predict(pr outcome(1))
3._predict : Pr(2.BDNn), predict(pr outcome(2))
4._predict : Pr(3.BDNn), predict(pr outcome(3))
---------------------------------------------------------------------------------
| Delta-method
| dy/dx Std. Err. z P>|z| [95% Conf. Interval]
----------------+----------------------------------------------------------------
1.SDN_1 |
_predict |
1 | -.2684081 .0218655 -12.28 0.000 -.3112637 -.2255526
2 | .1566434 .0108431 14.45 0.000 .1353912 .1778955
3 | .0712929 .0080412 8.87 0.000 .0555325 .0870534
4 | .0404718 .0055584 7.28 0.000 .0295776 .051366
----------------+----------------------------------------------------------------
1.SDN_2 |
_predict |
1 | -.3235553 .0297191 -10.89 0.000 -.3818037 -.2653068
2 | .1673386 .0100742 16.61 0.000 .1475936 .1870836
3 | .0965402 .0130516 7.40 0.000 .0709596 .1221208
4 | .0596764 .0101186 5.90 0.000 .0398443 .0795086
----------------+----------------------------------------------------------------
1.SDN_3 |
_predict |
1 | -.5485901 .0277538 -19.77 0.000 -.6029866 -.4941937
2 | .1052249 .0261592 4.02 0.000 .0539539 .1564959
3 | .2171439 .0180633 12.02 0.000 .1817404 .2525474
4 | .2262213 .0396679 5.70 0.000 .1484737 .303969
----------------+----------------------------------------------------------------
1.SUP_1 |
_predict |
1 | .2702918 .0164749 16.41 0.000 .2380016 .302582
2 | -.2042009 .0137561 -14.84 0.000 -.2311623 -.1772394
3 | -.045458 .0032109 -14.16 0.000 -.0517512 -.0391648
4 | -.0206329 .0018991 -10.86 0.000 -.0243551 -.0169107
----------------+----------------------------------------------------------------
1.SNWQ |
_predict |
1 | .2471957 .0143714 17.20 0.000 .2190282 .2753631
2 | -.1850587 .0118024 -15.68 0.000 -.2081909 -.1619264
3 | -.0426171 .0030132 -14.14 0.000 -.0485229 -.0367114
4 | -.0195199 .0017978 -10.86 0.000 -.0230436 -.0159962
----------------+----------------------------------------------------------------
1.SPWQ |
_predict |
1 | .3063149 .0063064 48.57 0.000 .2939546 .3186752
2 | -.2350874 .0057033 -41.22 0.000 -.2462656 -.2239092
3 | -.0492198 .0027646 -17.80 0.000 -.0546383 -.0438014
4 | -.0220077 .0018719 -11.76 0.000 -.0256766 -.0183388
----------------+----------------------------------------------------------------
SovereignRating |
_predict |
1 | -.0048165 .0016495 -2.92 0.004 -.0080495 -.0015835
2 | .0032462 .0011103 2.92 0.003 .00107 .0054224
3 | .0010353 .0003599 2.88 0.004 .0003299 .0017407
4 | .000535 .0001895 2.82 0.005 .0001636 .0009064
---------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.
Would doing margins, dydx(SDN_1) predict(pu0 outcome(1)) or margins, dydx(SDN_1) - for each independent variable - be better? Any other suggestion?
Thank you very much in advance!
