Pvarsoc command not producing results for lag determination

[Jesse Nooijen]

Dear reader,

I am trying to determine the optimal number of lags for my reduced form pvar model, but it will not produce the intended results. My sample ranges from 2010-2019. I am replicating Kang et al (2016), and improving their approach by incorporating the methods of Abrigo & Love (2016).

My data set descriptive statistics are as follows:


TABLE 3

Descriptive Statistics

Variables	CSR	CSI	Tobin’s q	Firm Size	FL	ROA	Advertising Intensity	R&D Intensity
CSR	1.000
CSI	0.457	1.000
Tobin’s q	0.016	-0.047	1.000
Firm Size [log(emp)]	0.456	0.334	-0.146	1.000
Financial Leverage	0.035	0.013	-0.016	0.149	1.000
ROA	0.105	0.104	-0.158	0.236	-0.078	1.000
Advertising Intensity	-0.013	-0.021	0.085	-0.046	0.001	-0.171	1.000
R&D Intensity	-0.019	-0.021	0.030	-0.052	-0.008	-0.083	0.401	1.000
M	.816	.515	1.518	1.378	.252	-.079	.046	7.774
SD	1.978	1.185	1.666	1.36	.273	.495	.381	143.486
Note: The means, standard deviations, and pairwise correlations reported here are based on unstandardized CSR and CSI scores. Note that the CSR and CSI scores are standardized for the empirical analysis.

I coded the following

xtset company Year //declaring panel data
pvarsoc FP stdCSR stdCSI, maxlag(5) //determining optimal lag

which produces

Panel variable: company (unbalanced)
 Time variable: Year, 2010 to 2019, but with gaps
         Delta: 1 unit

. pvarsoc FP stdCSR stdCSI, maxlag(5) //determining optimal lag 
Running panel VAR lag order selection on estimation sample
.....

 Selection order criteria
 Sample:  2015 - 2018                              No. of obs      =      8210
                                                   No. of panels   =      2507
                                                   Ave. no. of T   =     3.275

  +--------------------------------------------------------------------------+
  |   lag |    CD          J      J pvalue     MBIC       MAIC       MQIC    |
  |-------+------------------------------------------------------------------|
  |     1 |  .9870712          .          .          .          .          . |
  |     2 |  .9863182          .          .          .          .          . |
  |     3 |  .9874768          .          .          .          .          . |
  |     4 |  .9850972          .          .          .          .          . |
  |     5 |  .9895069          .          .          .          .          . |
  +--------------------------------------------------------------------------+

As you can see I used the standardized CSR and CSI scores, as Kang et al uses those in their reduced form pvar model as well, but I cannot get results. I tried working with 'pinst' option as well, but I could not get that command to run. I am wondering what I am doing wrong as I cannot get it to produce the intended results, and it does not use my full data range. Does anyone familiar with pvar models and the pvarsoc command have any information on what I could try to resolve this?

Thanks for taking the time,
Jesse

[George Ford]

You are not alone. I've had the same problem, and never found a solution.

[Jesse Nooijen]

You are not alone. I've had the same problem, and never found a solution.

With some help of my professor I managed to get it to work. The number of instruments specified must be larger than the number of endogenous variables. So if you want to incorporate 4 lags in the VAR specification you need > 5 instruments. E.g.

pvarsoc FP stdCSR stdCSI, maxlag(4) pvaropts(instlags(1/5))

The instlags command can be used to specify the number of lags of the endogenous variables to be used as instruments in the GMM estimation, in this case 5.

This yields;

Running panel VAR lag order selection on estimation sample
....

 Selection order criteria
 Sample:  2015 - 2018                              No. of obs      =      8210
                                                   No. of panels   =      2507
                                                   Ave. no. of T   =     3.275

  +--------------------------------------------------------------------------+
  |   lag |    CD          J      J pvalue     MBIC       MAIC       MQIC    |
  |-------+------------------------------------------------------------------|
  |     1 |  .9875516   172.4863   9.89e-20  -151.9856   100.4863   14.18138 |
  |     2 |  .9869475   150.2329   4.66e-19  -93.12098   96.23294   31.50422 |
  |     3 |  .9882275   135.9303   3.88e-20  -26.30565    99.9303   56.77782 |
  |     4 |  .9903285    36.1288   .0000376  -44.98917    18.1288  -3.447438 |
  +--------------------------------------------------------------------------+

[George Ford]

Good to know. Thanks for posting.

[Jesse Nooijen]

I must say I am still rather confused as to how to interpret the varsaoc command. I want to look for high MMSC scores, low J value and a J P-value of >0.01 (given large sample this significance level was OK according to my professor). Now I am wondering which options to specify because they influence my results so much. By specifying different maxlags and instlags I get very different results. Should I simply look at the results of the varsoc command specifiying maxlags of 4 and instlags of 5, as that tests the whole model? I think so right? See my complete list of attempts below

pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(4) pvaropts(instlags(1/5))
pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(3) pvaropts(instlags(1/4))
pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(2) pvaropts(instlags(1/3))
pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(1) pvaropts(instlags(1/2))

pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(3) pvaropts(instlags(1/5))
pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(2) pvaropts(instlags(1/5))
pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(2) pvaropts(instlags(1/4))
pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(1) pvaropts(instlags(1/5))
pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(1) pvaropts(instlags(1/4))
pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(1) pvaropts(instlags(1/3))

I tested various maximum lags with various combinations of instrument lags, but now that I am contemplating about it, I believe I should only run the first command as that tests up to all 4 lags, while the other tests are incomplete because they don't test up to four lags, is that line of thought correct? Therefore I would be using

[pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(4) pvaropts(instlags(1/5))

which results in the following output

. pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(4) pvaropts(instlags(1/5))
Running panel VAR lag order selection on estimation sample
....

 Selection order criteria
 Sample:  2015 - 2018                              No. of obs      =      2160
                                                   No. of panels   =       704
                                                   Ave. no. of T   =     3.068

  +--------------------------------------------------------------------------+
  |   lag |    CD          J      J pvalue     MBIC       MAIC       MQIC    |
  |-------+------------------------------------------------------------------|
  |     1 |  .9851308   123.5912   1.56e-11  -152.8119    51.5912  -23.16938 |
  |     2 |  .9867257   109.3719   7.09e-12  -97.93044   55.37188  -.6985561 |
  |     3 |   .830628   27.09561   .0772196  -111.1059  -8.904392  -46.28468 |
  |     4 |  .9724388   8.252158   .5089483  -60.84861  -9.747842  -28.43799 |
  +--------------------------------------------------------------------------+

The model with 4 lags and 5 instrument lags would be preferred if I am not mistaken. Anyone that could extend their knowledge surrounding this to me would be greatly appreciated!

[George Ford]

Spit balling here. TS is not my specialty.

Are these variables stationary? They need to be to do any of this. I think Kang may have used first differences.

You've only got data for 2010-2019, so the fewer lags the better. The sample may be large, but in N not T.

And you've got pvarstable to run, which may give different results at different lags.

I'd set maxlag(3) and no more than 4. See

https://journals.sagepub.com/doi/pdf/10.1177/1536867X1601600314

. The 2nd example has a similar dataset to yours, and maxlag(3) is used (though you can play with the instrument setup).

There's no stability in those pvarsoc results, which is worrisome. Play with the instruments lags to see if you can smooth it out (2/4 or some such).

Personally, I'd run pvar separately with 1, 2, or 3 lags, check pvarstable, and then see if the results are comparable.

And don't be shy about using the same lag structure as the Kang paper. By BIC and AIC, the best is one lag if you set a short maxlag.

[Jesse Nooijen]

Spit balling here. TS is not my specialty.
Are these variables stationary? They need to be to do any of this. I think Kang may have used first differences.
.

Thank you so much for taking the time to answer and help!

Kang et al. used first differences indeed, my professor told me to skip stationarity testing and unitroot tests and move to the pvar part. I actually just mailed him if I should backtrack to incorporate them, however, I also found the option 'fd' (see page 784 Abrigo and Love) which explains "fd specifies that the panel-specific fixed effects be removed using first difference instead of forward orthogonal deviations. By default, the first 7. This version of the software corrects the implementation of forward orthogonal deviation used in the earlier version of the program. # lags of transformed (that is, differenced) depvarlist in the model are instrumented by the (#+1)th to (2#)th lags of depvarlist in levels (that is, untransformed)." therefore I believe this should accommodate stationarity.

You've only got data for 2010-2019, so the fewer lags the better. The sample may be large, but in N not T.

And you've got pvarstable to run, which may give different results at different lags.
.

I agree, I noticed that using a maxlag of 2 or 3 with my pvarsoc command increases my observation count from 2160 to 3898 and 2972 respectively, which is preferred. What I am unsure of is, that there are so many options available in terms of maxlag combinations with instlags. If I run a maxlag of 4 with 5 instrument lags, I get very different results for lag 2 for example. I tried all combinations I could envision a possibility for me, and must say I am rather overwhelmed with the possibilities. I'll concisely post a few of them below. I was also wondering if you are familiar with the transition from pvarsoc to pvar, I need to run an instrumental pvarsoc setup of >3 to test for maxlag 2, but afterwards when running pvar I could use either pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(2) pvaropts(instlags(1/3)) or pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(2) - thus removing the instrumental part.

I do believe that I should specific the fd option in my pvar command to account for first differncing, similar to Kang et al. Given that they applied a lag of 2, with the first differencing I believe my model should be; pvar FP stdCSR stdCSI, lags(2) exog(FS FL ROA AI RDI) instlags(1/3) fd

The 2nd example has a similar dataset to yours, and maxlag(3) is used (though you can play with the instrument setup).

There's no stability in those pvarsoc results, which is worrisome. Play with the instruments lags to see if you can smooth it out (2/4 or some such).

Personally, I'd run pvar separately with 1, 2, or 3 lags, check pvarstable, and then see if the results are comparable.

.

Regarding this, I tried several options, maxlag(3) with instrumental(4) yielded no acceptable p-values. With 2 and 4 lags it did. I ran the following code to analyze further;

pvar FP stdCSR stdCSI, lags(2) exog(FS FL ROA AI RDI) instlags(1/3) fd
pvarstable
pvargranger

pvar FP stdCSR stdCSI, lags(4) exog(FS FL ROA AI RDI) instlags(1/5) fd
pvarstable
pvargranger

which resulted in the output

. pvar FP stdCSR stdCSI, lags(2) exog(FS FL ROA AI RDI) instlags(1/3) fd

Panel vector autoregresssion



GMM Estimation

Final GMM Criterion Q(b) =     .0215
Initial weight matrix: Identity
GMM weight matrix:     Robust
                                                   No. of obs      =      4910
                                                   No. of panels   =      1132
                                                   Ave. no. of T   =     4.337


------------------------------------------------------------------------------
             | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
FP           |
          FP |
         L1. |  -.0933012   .0424248    -2.20   0.028    -.1764522   -.0101502
         L2. |   -.141975   .0393213    -3.61   0.000    -.2190434   -.0649066
             |
      stdCSR |
         L1. |  -.0190224   .0196139    -0.97   0.332    -.0574649    .0194201
         L2. |   .0009352    .019469     0.05   0.962    -.0372234    .0390939
             |
      stdCSI |
         L1. |   .0102805   .0188394     0.55   0.585    -.0266441    .0472051
         L2. |  -.0336241   .0199325    -1.69   0.092    -.0726911     .005443
             |
          FS |   .6865474   .5089899     1.35   0.177    -.3110544    1.684149
          FL |   .2476658   .6587071     0.38   0.707    -1.043376    1.538708
         ROA |  -.1102748   .1488491    -0.74   0.459    -.4020138    .1814641
          AI |  -2.436316   1.814436    -1.34   0.179    -5.992546    1.119913
         RDI |    .187271   .1303987     1.44   0.151    -.0683058    .4428478
-------------+----------------------------------------------------------------
stdCSR       |
          FP |
         L1. |   .0377219   .0266121     1.42   0.156    -.0144369    .0898807
         L2. |   .0031176   .0277156     0.11   0.910    -.0512041    .0574392
             |
      stdCSR |
         L1. |  -.2342126   .0494584    -4.74   0.000    -.3311493   -.1372759
         L2. |  -.1269942   .0386467    -3.29   0.001    -.2027403   -.0512481
             |
      stdCSI |
         L1. |  -.1221039    .031395    -3.89   0.000    -.1836371   -.0605707
         L2. |   -.002635   .0339494    -0.08   0.938    -.0691745    .0639046
             |
          FS |   1.622537   .6768078     2.40   0.017     .2960181    2.949056
          FL |  -1.192995   1.203124    -0.99   0.321    -3.551075    1.165085
         ROA |  -.1504952   .4971552    -0.30   0.762    -1.124901    .8239111
          AI |  -1.045704   .6942656    -1.51   0.132    -2.406439    .3150316
         RDI |   .0723352   .0503954     1.44   0.151     -.026438    .1711083
-------------+----------------------------------------------------------------
stdCSI       |
          FP |
         L1. |   .1024386   .0334146     3.07   0.002     .0369471      .16793
         L2. |   .0026724   .0297487     0.09   0.928     -.055634    .0609789
             |
      stdCSR |
         L1. |   .0434939   .0326719     1.33   0.183    -.0205418    .1075296
         L2. |    .000669   .0299564     0.02   0.982    -.0580445    .0593826
             |
      stdCSI |
         L1. |  -.4073032   .0385334   -10.57   0.000    -.4828273    -.331779
         L2. |  -.0608807   .0406853    -1.50   0.135    -.1406225    .0188611
             |
          FS |   2.708939   .7824248     3.46   0.001     1.175414    4.242463
          FL |  -1.378027   .9219878    -1.49   0.135     -3.18509     .429036
         ROA |  -.2814539   .3006427    -0.94   0.349    -.8707027     .307795
          AI |  -4.168512    2.84898    -1.46   0.143    -9.752409    1.415386
         RDI |   .2794474   .2065948     1.35   0.176     -.125471    .6843657
------------------------------------------------------------------------------
Instruments : l(1/3).(FP stdCSR stdCSI) FS FL ROA AI RDI

. pvarstable

   Eigenvalue stability condition

  +----------------------------------+
  |      Eigenvalue      |           |
  |   Real     Imaginary |  Modulus  |
  |----------------------+-----------|
  | -.0398048   .3819295 |  .3839981 |
  | -.0398048  -.3819295 |  .3839981 |
  | -.1254591   .3410381 |  .3633826 |
  | -.1254591  -.3410381 |  .3633826 |
  | -.2021446  -.1269117 |  .2386818 |
  | -.2021446   .1269117 |  .2386818 |
  +----------------------------------+

   All the eigenvalues lie inside the unit circle.
   pVAR satisfies stability condition.

. pvargranger

  panel VAR-Granger causality Wald test
    Ho: Excluded variable does not Granger-cause Equation variable
    Ha: Excluded variable Granger-causes Equation variable

  +------------------------------------------------------+
  |  Equation \ Excluded |    chi2     df   Prob > chi2  |
  |----------------------+-------------------------------|
  |FP                    |                               |
  |               stdCSR |      0.967    2        0.617  |
  |               stdCSI |      4.433    2        0.109  |
  |                  ALL |      5.119    4        0.275  |
  |----------------------+-------------------------------|
  |stdCSR                |                               |
  |                   FP |      2.037    2        0.361  |
  |               stdCSI |     15.852    2        0.000  |
  |                  ALL |     17.488    4        0.002  |
  |----------------------+-------------------------------|
  |stdCSI                |                               |
  |                   FP |     11.179    2        0.004  |
  |               stdCSR |      1.921    2        0.383  |
  |                  ALL |     11.786    4        0.019  |
  +------------------------------------------------------+

.
. pvar FP stdCSR stdCSI, lags(4) exog(FS FL ROA AI RDI) instlags(1/5) fd

Panel vector autoregresssion



GMM Estimation

Final GMM Criterion Q(b) =     .0152
Initial weight matrix: Identity
GMM weight matrix:     Robust
                                                   No. of obs      =      2960
                                                   No. of panels   =       851
                                                   Ave. no. of T   =     3.478


------------------------------------------------------------------------------
             | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
FP           |
          FP |
         L1. |  -.2010331   .0543192    -3.70   0.000    -.3074968   -.0945693
         L2. |  -.1058913   .0583113    -1.82   0.069    -.2201794    .0083968
         L3. |  -.0158555   .0415418    -0.38   0.703     -.097276    .0655649
         L4. |    .013332   .0348559     0.38   0.702    -.0549843    .0816482
             |
      stdCSR |
         L1. |  -.0841953   .0391285    -2.15   0.031    -.1608857   -.0075049
         L2. |  -.0236957   .0260386    -0.91   0.363    -.0747303     .027339
         L3. |  -.0295019   .0305056    -0.97   0.333    -.0892918     .030288
         L4. |   -.049729   .0267544    -1.86   0.063    -.1021665    .0027086
             |
      stdCSI |
         L1. |    .030289   .0230159     1.32   0.188    -.0148213    .0753993
         L2. |  -.0167317   .0256692    -0.65   0.515    -.0670424    .0335789
         L3. |  -.0500466   .0225097    -2.22   0.026    -.0941649   -.0059284
         L4. |  -.0543191   .0243584    -2.23   0.026    -.1020607   -.0065776
             |
          FS |  -.9081441   .6896272    -1.32   0.188    -2.259789    .4435003
          FL |    .575455   .6935945     0.83   0.407    -.7839653    1.934875
         ROA |   1.765481   1.019416     1.73   0.083    -.2325384      3.7635
          AI |    .419563   3.652459     0.11   0.909    -6.739124     7.57825
         RDI |  -.0092439   .0213871    -0.43   0.666    -.0511618    .0326739
-------------+----------------------------------------------------------------
stdCSR       |
          FP |
         L1. |   .0681826   .1258785     0.54   0.588    -.1785348       .3149
         L2. |  -.2263459   .2002171    -1.13   0.258    -.6187642    .1660725
         L3. |   .0307803   .0947676     0.32   0.745    -.1549609    .2165214
         L4. |  -.0345759   .1003497    -0.34   0.730    -.2312578     .162106
             |
      stdCSR |
         L1. |  -.0460077   .1291093    -0.36   0.722    -.2990574    .2070419
         L2. |  -.2091202   .0710807    -2.94   0.003    -.3484357   -.0698046
         L3. |  -.1800161   .0902909    -1.99   0.046    -.3569831   -.0030491
         L4. |  -.0206146   .0701438    -0.29   0.769    -.1580939    .1168646
             |
      stdCSI |
         L1. |  -.1471257   .0718198    -2.05   0.041    -.2878898   -.0063615
         L2. |     .01573   .0808954     0.19   0.846    -.1428221    .1742822
         L3. |  -.0410234   .0720094    -0.57   0.569    -.1821593    .1001125
         L4. |   .0869333   .0710501     1.22   0.221    -.0523224     .226189
             |
          FS |   1.840911   1.912244     0.96   0.336    -1.907019    5.588841
          FL |  -1.199202   2.542412    -0.47   0.637    -6.182237    3.783834
         ROA |  -4.718612   4.394699    -1.07   0.283    -13.33206     3.89484
          AI |   11.04245   11.34444     0.97   0.330    -11.19225    33.27715
         RDI |   .0170672   .0604607     0.28   0.778    -.1014337     .135568
-------------+----------------------------------------------------------------
stdCSI       |
          FP |
         L1. |   .0298645     .02432     1.23   0.219    -.0178019    .0775308
         L2. |  -.0006999   .0276879    -0.03   0.980    -.0549673    .0535674
         L3. |   .0053581   .0216091     0.25   0.804    -.0369949     .047711
         L4. |  -.0242667   .0205151    -1.18   0.237    -.0644757    .0159422
             |
      stdCSR |
         L1. |   .0370716   .0351849     1.05   0.292    -.0318896    .1060328
         L2. |    .046217   .0329528     1.40   0.161    -.0183693    .1108033
         L3. |  -.0338227   .0364327    -0.93   0.353    -.1052296    .0375841
         L4. |  -.0164244   .0309369    -0.53   0.595    -.0770596    .0442109
             |
      stdCSI |
         L1. |   -.261747   .0488698    -5.36   0.000      -.35753   -.1659641
         L2. |  -.1102297   .0487178    -2.26   0.024    -.2057149   -.0147445
         L3. |  -.1219805   .0397168    -3.07   0.002     -.199824   -.0441369
         L4. |  -.0043247   .0364676    -0.12   0.906       -.0758    .0671505
             |
          FS |   .3989989   .5256405     0.76   0.448    -.6312375    1.429235
          FL |   .0507916   .4944811     0.10   0.918    -.9183736    1.019957
         ROA |   -1.58433   .7924316    -2.00   0.046    -3.137467   -.0311927
          AI |   3.058156   1.506054     2.03   0.042     .1063449    6.009968
         RDI |   .0290919   .0146451     1.99   0.047      .000388    .0577958
------------------------------------------------------------------------------
Instruments : l(1/5).(FP stdCSR stdCSI) FS FL ROA AI RDI

. pvarstable

   Eigenvalue stability condition

  +----------------------------------+
  |      Eigenvalue      |           |
  |   Real     Imaginary |  Modulus  |
  |----------------------+-----------|
  |  .2407073  -.6426825 |  .6862804 |
  |  .2407073   .6426825 |  .6862804 |
  |  -.615926          0 |   .615926 |
  |  -.021556  -.4870446 |  .4875214 |
  |  -.021556   .4870446 |  .4875214 |
  | -.2534015   .3706492 |  .4489913 |
  | -.2534015  -.3706492 |  .4489913 |
  |  .2397114  -.3665646 |  .4379853 |
  |  .2397114   .3665646 |  .4379853 |
  |  .3753265          0 |  .3753265 |
  | -.3395553  -.1161905 |  .3588844 |
  | -.3395553   .1161905 |  .3588844 |
  +----------------------------------+

   All the eigenvalues lie inside the unit circle.
   pVAR satisfies stability condition.

. pvargranger

  panel VAR-Granger causality Wald test
    Ho: Excluded variable does not Granger-cause Equation variable
    Ha: Excluded variable Granger-causes Equation variable

  +------------------------------------------------------+
  |  Equation \ Excluded |    chi2     df   Prob > chi2  |
  |----------------------+-------------------------------|
  |FP                    |                               |
  |               stdCSR |      6.091    4        0.192  |
  |               stdCSI |      8.932    4        0.063  |
  |                  ALL |     13.076    8        0.109  |
  |----------------------+-------------------------------|
  |stdCSR                |                               |
  |                   FP |      2.218    4        0.696  |
  |               stdCSI |      7.469    4        0.113  |
  |                  ALL |      8.330    8        0.402  |
  |----------------------+-------------------------------|
  |stdCSI                |                               |
  |                   FP |      2.974    4        0.562  |
  |               stdCSR |      4.064    4        0.397  |
  |                  ALL |      6.154    8        0.630  |
  +------------------------------------------------------+

Both are stable. My real question is I guess, can you elaborate on how you find the optimal lag and instlag, because for me it feels like lag 2 is optimal, but then I worry whether I pick the right instlags. For the code I provide above, do you think I should specify instlags or leave them out at the actual pvar command?

[George Ford]

The third lag does nothing in this model. Keep it to 2. If you use fod, then use inst lags 1/2. With fd, then skip the first lag and use 2/3. fod will save a year.

[Jesse Nooijen]

The third lag does nothing in this model. Keep it to 2. If you use fod, then use inst lags 1/2. With fd, then skip the first lag and use 2/3. fod will save a year.

Thanks for your reply once again. I will be running fd to accomodate the first-differencing as Kang et al. applied that in their model. Regarding the decision to use instlags 2/3 with the fd option compared to fod, could you perhaps explain why you would choose for instrumental lags 2 and 3, instead of e.g. 1/3 or not specifying them? I am trying to understand the logic but I do not grasp it yet, as my knowledge in this area could be improved. I really appreciate the advice though, I have decided to move forward with 2 lags.

My latest trial was as follows:

pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(2) pvaropts(instlags(1/3)) // overidentifies
pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(2) pvaropts(instlags(2/3)) //just identifies it
pvar FP stdCSR stdCSI, lags(2) exog(FS FL ROA AI RDI) instlags(1/3) fd //applying 3 instrumental lags with FD
pvarstable //test for stability
pvargranger //test for granger causality
pvar FP stdCSR stdCSI, lags(2) exog(FS FL ROA AI RDI) instlags(2/3) fd //applying 2 instrumental lags, disregarding lag 1 with FD
pvarstable //test for stability
pvargranger //test for granger causality
pvar FP stdCSR stdCSI, lags(2) exog(FS FL ROA AI RDI) fd //applying standard instrumental lags with FD
pvarstable //test for stability
pvargranger //test for granger causality

 //Testing models
. pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(2) pvaropts(instlags(1/3)) // overidentifies
Running panel VAR lag order selection on estimation sample
..

 Selection order criteria
 Sample:  2013 - 2018                              No. of obs      =      3898
                                                   No. of panels   =       997
                                                   Ave. no. of T   =     3.910

  +--------------------------------------------------------------------------+
  |   lag |    CD          J      J pvalue     MBIC       MAIC       MQIC    |
  |-------+------------------------------------------------------------------|
  |     1 |  .9693167   52.56524   .0000306   -96.2627   16.56524  -23.48185 |
  |     2 |   .989256   20.19164   .0167659  -54.22233   2.191642   -17.8319 |
  +--------------------------------------------------------------------------+

. pvarsoc FP stdCSR stdCSI, exog(FS FL ROA AI RDI) maxlag(2) pvaropts(instlags(2/3)) //just identifies it
Running panel VAR lag order selection on estimation sample
..

 Selection order criteria
 Sample:  2013 - 2018                              No. of obs      =      3898
                                                   No. of panels   =       997
                                                   Ave. no. of T   =     3.910

  +--------------------------------------------------------------------------+
  |   lag |    CD          J      J pvalue     MBIC       MAIC       MQIC    |
  |-------+------------------------------------------------------------------|
  |     1 |  .9575771   19.94225   .0182721  -54.47172   1.942254  -18.08129 |
  |     2 |  .9452751          .          .          .          .          . |
  +--------------------------------------------------------------------------+

. pvar FP stdCSR stdCSI, lags(2) exog(FS FL ROA AI RDI) instlags(1/3) fd //applying 3 instrumental lags with FD

Panel vector autoregresssion



GMM Estimation

Final GMM Criterion Q(b) =     .0215
Initial weight matrix: Identity
GMM weight matrix:     Robust
                                                   No. of obs      =      4910
                                                   No. of panels   =      1132
                                                   Ave. no. of T   =     4.337


------------------------------------------------------------------------------
             | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
FP           |
          FP |
         L1. |  -.0933012   .0424248    -2.20   0.028    -.1764522   -.0101502
         L2. |   -.141975   .0393213    -3.61   0.000    -.2190434   -.0649066
             |
      stdCSR |
         L1. |  -.0190224   .0196139    -0.97   0.332    -.0574649    .0194201
         L2. |   .0009352    .019469     0.05   0.962    -.0372234    .0390939
             |
      stdCSI |
         L1. |   .0102805   .0188394     0.55   0.585    -.0266441    .0472051
         L2. |  -.0336241   .0199325    -1.69   0.092    -.0726911     .005443
             |
          FS |   .6865474   .5089899     1.35   0.177    -.3110544    1.684149
          FL |   .2476658   .6587071     0.38   0.707    -1.043376    1.538708
         ROA |  -.1102748   .1488491    -0.74   0.459    -.4020138    .1814641
          AI |  -2.436316   1.814436    -1.34   0.179    -5.992546    1.119913
         RDI |    .187271   .1303987     1.44   0.151    -.0683058    .4428478
-------------+----------------------------------------------------------------
stdCSR       |
          FP |
         L1. |   .0377219   .0266121     1.42   0.156    -.0144369    .0898807
         L2. |   .0031176   .0277156     0.11   0.910    -.0512041    .0574392
             |
      stdCSR |
         L1. |  -.2342126   .0494584    -4.74   0.000    -.3311493   -.1372759
         L2. |  -.1269942   .0386467    -3.29   0.001    -.2027403   -.0512481
             |
      stdCSI |
         L1. |  -.1221039    .031395    -3.89   0.000    -.1836371   -.0605707
         L2. |   -.002635   .0339494    -0.08   0.938    -.0691745    .0639046
             |
          FS |   1.622537   .6768078     2.40   0.017     .2960181    2.949056
          FL |  -1.192995   1.203124    -0.99   0.321    -3.551075    1.165085
         ROA |  -.1504952   .4971552    -0.30   0.762    -1.124901    .8239111
          AI |  -1.045704   .6942656    -1.51   0.132    -2.406439    .3150316
         RDI |   .0723352   .0503954     1.44   0.151     -.026438    .1711083
-------------+----------------------------------------------------------------
stdCSI       |
          FP |
         L1. |   .1024386   .0334146     3.07   0.002     .0369471      .16793
         L2. |   .0026724   .0297487     0.09   0.928     -.055634    .0609789
             |
      stdCSR |
         L1. |   .0434939   .0326719     1.33   0.183    -.0205418    .1075296
         L2. |    .000669   .0299564     0.02   0.982    -.0580445    .0593826
             |
      stdCSI |
         L1. |  -.4073032   .0385334   -10.57   0.000    -.4828273    -.331779
         L2. |  -.0608807   .0406853    -1.50   0.135    -.1406225    .0188611
             |
          FS |   2.708939   .7824248     3.46   0.001     1.175414    4.242463
          FL |  -1.378027   .9219878    -1.49   0.135     -3.18509     .429036
         ROA |  -.2814539   .3006427    -0.94   0.349    -.8707027     .307795
          AI |  -4.168512    2.84898    -1.46   0.143    -9.752409    1.415386
         RDI |   .2794474   .2065948     1.35   0.176     -.125471    .6843657
------------------------------------------------------------------------------
Instruments : l(1/3).(FP stdCSR stdCSI) FS FL ROA AI RDI

. pvarstable //test for stability

   Eigenvalue stability condition

  +----------------------------------+
  |      Eigenvalue      |           |
  |   Real     Imaginary |  Modulus  |
  |----------------------+-----------|
  | -.0398048   .3819295 |  .3839981 |
  | -.0398048  -.3819295 |  .3839981 |
  | -.1254591   .3410381 |  .3633826 |
  | -.1254591  -.3410381 |  .3633826 |
  | -.2021446  -.1269117 |  .2386818 |
  | -.2021446   .1269117 |  .2386818 |
  +----------------------------------+

   All the eigenvalues lie inside the unit circle.
   pVAR satisfies stability condition.

. pvargranger //test for granger causality

  panel VAR-Granger causality Wald test
    Ho: Excluded variable does not Granger-cause Equation variable
    Ha: Excluded variable Granger-causes Equation variable

  +------------------------------------------------------+
  |  Equation \ Excluded |    chi2     df   Prob > chi2  |
  |----------------------+-------------------------------|
  |FP                    |                               |
  |               stdCSR |      0.967    2        0.617  |
  |               stdCSI |      4.433    2        0.109  |
  |                  ALL |      5.119    4        0.275  |
  |----------------------+-------------------------------|
  |stdCSR                |                               |
  |                   FP |      2.037    2        0.361  |
  |               stdCSI |     15.852    2        0.000  |
  |                  ALL |     17.488    4        0.002  |
  |----------------------+-------------------------------|
  |stdCSI                |                               |
  |                   FP |     11.179    2        0.004  |
  |               stdCSR |      1.921    2        0.383  |
  |                  ALL |     11.786    4        0.019  |
  +------------------------------------------------------+

. pvar FP stdCSR stdCSI, lags(2) exog(FS FL ROA AI RDI) instlags(2/3) fd //applying 2 instrumental lags, disregarding lag 1 with FD

Panel vector autoregresssion



GMM Estimation

Final GMM Criterion Q(b) =  2.36e-33
Initial weight matrix: Identity
GMM weight matrix:     Robust
                                                   No. of obs      =      4910
                                                   No. of panels   =      1132
                                                   Ave. no. of T   =     4.337


------------------------------------------------------------------------------
             | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
FP           |
          FP |
         L1. |   .2491734   .1724856     1.44   0.149    -.0888922    .5872391
         L2. |  -.0762901   .0382813    -1.99   0.046      -.15132   -.0012602
             |
      stdCSR |
         L1. |  -.0958029   .0567649    -1.69   0.091    -.2070601    .0154542
         L2. |  -.0352636   .0218013    -1.62   0.106    -.0779932    .0074661
             |
      stdCSI |
         L1. |    .067259    .067732     0.99   0.321    -.0654932    .2000112
         L2. |   .0216374   .0281628     0.77   0.442    -.0335606    .0768355
             |
          FS |   .0902883    .431242     0.21   0.834    -.7549304     .935507
          FL |    .426314   .4118176     1.04   0.301    -.3808336    1.233462
         ROA |  -.0684304   .0824373    -0.83   0.406    -.2300045    .0931437
          AI |  -.1103108   .3348383    -0.33   0.742    -.7665817    .5459602
         RDI |   .0480963   .0231589     2.08   0.038     .0027056     .093487
-------------+----------------------------------------------------------------
stdCSR       |
          FP |
         L1. |    .051045   .1611787     0.32   0.751    -.2648593    .3669494
         L2. |   .0136529   .0228602     0.60   0.550    -.0311524    .0584581
             |
      stdCSR |
         L1. |   .4673169    .109549     4.27   0.000     .2526048     .682029
         L2. |    .066263   .0471775     1.40   0.160    -.0262032    .1587292
             |
      stdCSI |
         L1. |   .0916666    .115033     0.80   0.426     -.133794    .3171272
         L2. |   .1064008   .0478362     2.22   0.026     .0126436     .200158
             |
          FS |  -.1535738   .7540356    -0.20   0.839    -1.631456    1.324309
          FL |  -.2865209   .2720853    -1.05   0.292    -.8197982    .2467564
         ROA |  -.0279548    .057604    -0.49   0.627    -.1408565    .0849469
          AI |   .0177948   .0790051     0.23   0.822    -.1370524     .172642
         RDI |  -.0008553   .0095108    -0.09   0.928     -.019496    .0177855
-------------+----------------------------------------------------------------
stdCSI       |
          FP |
         L1. |   -.059051   .2350901    -0.25   0.802    -.5198192    .4017171
         L2. |   .0083337   .0333732     0.25   0.803    -.0570766    .0737441
             |
      stdCSR |
         L1. |   .1193336   .1103728     1.08   0.280    -.0969932    .3356604
         L2. |   .0049958    .046079     0.11   0.914    -.0853174     .095309
             |
      stdCSI |
         L1. |   .5624861   .2844747     1.98   0.048      .004926    1.120046
         L2. |   .2906869   .1034732     2.81   0.005     .0878832    .4934907
             |
          FS |  -.5715328   1.089281    -0.52   0.600    -2.706484    1.563418
          FL |  -.0401006   .4632002    -0.09   0.931    -.9479563    .8677552
         ROA |    .026394   .0929802     0.28   0.777    -.1558438    .2086318
          AI |  -.0565756   .1413096    -0.40   0.689    -.3335374    .2203862
         RDI |  -.0002746   .0148472    -0.02   0.985    -.0293746    .0288254
------------------------------------------------------------------------------
Instruments : l(2/3).(FP stdCSR stdCSI) FS FL ROA AI RDI

. pvarstable //test for stability

   Eigenvalue stability condition

  +----------------------------------+
  |      Eigenvalue      |           |
  |   Real     Imaginary |  Modulus  |
  |----------------------+-----------|
  |   .936376          0 |   .936376 |
  |    .49361          0 |    .49361 |
  | -.3150579          0 |  .3150579 |
  |   .141987  -.2367258 |  .2760424 |
  |   .141987   .2367258 |  .2760424 |
  | -.1199254          0 |  .1199254 |
  +----------------------------------+

   All the eigenvalues lie inside the unit circle.
   pVAR satisfies stability condition.

. pvargranger //test for granger causality

  panel VAR-Granger causality Wald test
    Ho: Excluded variable does not Granger-cause Equation variable
    Ha: Excluded variable Granger-causes Equation variable

  +------------------------------------------------------+
  |  Equation \ Excluded |    chi2     df   Prob > chi2  |
  |----------------------+-------------------------------|
  |FP                    |                               |
  |               stdCSR |      3.377    2        0.185  |
  |               stdCSI |      1.002    2        0.606  |
  |                  ALL |      3.482    4        0.481  |
  |----------------------+-------------------------------|
  |stdCSR                |                               |
  |                   FP |      0.629    2        0.730  |
  |               stdCSI |      7.349    2        0.025  |
  |                  ALL |      8.563    4        0.073  |
  |----------------------+-------------------------------|
  |stdCSI                |                               |
  |                   FP |      1.296    2        0.523  |
  |               stdCSR |      1.991    2        0.370  |
  |                  ALL |      3.208    4        0.524  |
  +------------------------------------------------------+

. pvar FP stdCSR stdCSI, lags(2) exog(FS FL ROA AI RDI) fd //applying standard instrumental lags with FD

Panel vector autoregresssion



GMM Estimation

Final GMM Criterion Q(b) =  1.86e-33
Initial weight matrix: Identity
GMM weight matrix:     Robust
                                                   No. of obs      =      3878
                                                   No. of panels   =       993
                                                   Ave. no. of T   =     3.905


------------------------------------------------------------------------------
             | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
FP           |
          FP |
         L1. |  -.1773475   2.230095    -0.08   0.937    -4.548253    4.193558
         L2. |   .5496897   .5208577     1.06   0.291    -.4711726    1.570552
             |
      stdCSR |
         L1. |   .1139305   1.180423     0.10   0.923    -2.199657    2.427518
         L2. |  -.1166222   .2435565    -0.48   0.632    -.5939841    .3607398
             |
      stdCSI |
         L1. |   -.059024   .8078697    -0.07   0.942     -1.64242    1.524372
         L2. |   .0623657   .2055265     0.30   0.762    -.3404588    .4651901
             |
          FS |  -.9912795   2.470709    -0.40   0.688    -5.833781    3.851222
          FL |  -.4432589   .8380447    -0.53   0.597    -2.085796    1.199279
         ROA |    1.17778   1.980043     0.59   0.552    -2.703034    5.058593
          AI |   .1216162   .4053416     0.30   0.764    -.6728387    .9160712
         RDI |  -.0961469   .6452293    -0.15   0.882    -1.360773    1.168479
-------------+----------------------------------------------------------------
stdCSR       |
          FP |
         L1. |  -.1458248   1.198782    -0.12   0.903    -2.495395    2.203745
         L2. |  -.0765576   .4267877    -0.18   0.858    -.9130462     .759931
             |
      stdCSR |
         L1. |  -1.250254    .930874    -1.34   0.179    -3.074734     .574225
         L2. |    .741407   .3258828     2.28   0.023     .1026884    1.380126
             |
      stdCSI |
         L1. |   .2594347     .49997     0.52   0.604    -.7204885    1.239358
         L2. |  -.0291951   .1428511    -0.20   0.838    -.3091781     .250788
             |
          FS |    -.97222   1.690359    -0.58   0.565    -4.285264    2.340824
          FL |  -.1012981   .7332424    -0.14   0.890    -1.538427    1.335831
         ROA |   .0295783   1.024799     0.03   0.977     -1.97899    2.038147
          AI |   .0046329   .2156369     0.02   0.983    -.4180077    .4272735
         RDI |   .0482099   .3334451     0.14   0.885    -.6053305    .7017503
-------------+----------------------------------------------------------------
stdCSI       |
          FP |
         L1. |  -.6960819   4.384238    -0.16   0.874    -9.289031    7.896867
         L2. |  -.0076298   .9798175    -0.01   0.994    -1.928037    1.912777
             |
      stdCSR |
         L1. |  -.7885475   2.401032    -0.33   0.743    -5.494484    3.917389
         L2. |   .1961766   .5107851     0.38   0.701    -.8049439    1.197297
             |
      stdCSI |
         L1. |   .2865997   1.625896     0.18   0.860    -2.900097    3.473296
         L2. |   .2312547   .4253815     0.54   0.587    -.6024778    1.064987
             |
          FS |  -.0583344   4.981025    -0.01   0.991    -9.820964    9.704295
          FL |   .1031141   1.322031     0.08   0.938     -2.48802    2.694248
         ROA |   -.413345    3.91423    -0.11   0.916    -8.085095    7.258405
          AI |  -.1126825   .7221135    -0.16   0.876    -1.527999    1.302634
         RDI |   .1946427   1.301599     0.15   0.881    -2.356445     2.74573
------------------------------------------------------------------------------
Instruments : l(3/4).(FP stdCSR stdCSI) FS FL ROA AI RDI

. pvarstable //test for stability

   Eigenvalue stability condition

  +----------------------------------+
  |      Eigenvalue      |           |
  |   Real     Imaginary |  Modulus  |
  |----------------------+-----------|
  | -1.574969          0 |  1.574969 |
  |  -.881089          0 |   .881089 |
  |  .6360584          0 |  .6360584 |
  |  .5293539  -.0490251 |  .5316192 |
  |  .5293539   .0490251 |  .5316192 |
  | -.3797101          0 |  .3797101 |
  +----------------------------------+
   At least one eigenvalue lie outside the unit circle.
   pVAR does not satisfy stability condition.

. pvargranger //test for granger causality

  panel VAR-Granger causality Wald test
    Ho: Excluded variable does not Granger-cause Equation variable
    Ha: Excluded variable Granger-causes Equation variable

  +------------------------------------------------------+
  |  Equation \ Excluded |    chi2     df   Prob > chi2  |
  |----------------------+-------------------------------|
  |FP                    |                               |
  |               stdCSR |      2.097    2        0.350  |
  |               stdCSI |      0.990    2        0.610  |
  |                  ALL |      3.721    4        0.445  |
  |----------------------+-------------------------------|
  |stdCSR                |                               |
  |                   FP |      0.039    2        0.981  |
  |               stdCSI |      0.343    2        0.842  |
  |                  ALL |      0.475    4        0.976  |
  |----------------------+-------------------------------|
  |stdCSI                |                               |
  |                   FP |      0.088    2        0.957  |
  |               stdCSR |      0.171    2        0.918  |
  |                  ALL |      0.870    4        0.929  |
  +------------------------------------------------------+

.
end of do-file

Given these results if I am not mistaken, I do conclude that the model with 2 lags and instlags(2/3) is best given it has excellent GMM criterion, stable eigenvalues and limited granger causality. The model with lags 2 instlags(1/3) has a good fit, stable eigenvalues and significant granger causality for some variables. Given the better fit for instlags(2/3) and your advice I will proceed with that model. If possible if you could enlighten me on how you determine the instlags to use given fd and FOD that would be highly appreciated.


Added later:
I found the command to display the J-value after pvar, for the model pvar FP stdCSR stdCSI, lags(2) exog(FS FL ROA AI RDI) instlags(2/3) fd overid it resulted in no value being displayed for J?

Test of overidentifying restriction: 
  Hansen's J chi2(0) = 1.158e-29 (p =     .)

[George Ford]

As for the 2/3 lag, it is my understanding that with fd you need the instrument to be free of the variables they are instrumenting for and that lag1 would not be so.

[Jesse Nooijen]

As for the 2/3 lag, it is my understanding that with fd you need the instrument to be free of the variables they are instrumenting for and that lag1 would not be so.

Thanks for clarifying, I see that now. If not mistaken this means it is because you want the instrument lag to be uncorrelated with the error terms in the first equation, therefore rendering lag1 invalid as instrument.

I am still rather unsure regarding which model I should definitively pick, since analyzing the hansen J-statistic on the actual pvar model (rather than pvarsoc trials) presented me with no results for the model specifying 2 lags with instlags(2/3) fd. I tried running some other pvar models to see if I could get acceptable P-values (>0.01), but I did not get any acceptable values. Do you have any advice regarding this? The pvarsoc command determined that the lag of 2 was appropriate, yet the actual pvar model no longer confirms it. Any thougts are appreciated.

[George Ford]

As you've realized, there's a bit of art to this, and pvar involves a lot of art.

A few recommendations.

(1) what is your intuition about how the outcomes are related over time? don't rely simply on what Stata puts out.

(2) Is var even the right approach?

(3) Replicate/reproduce the Kang (2016) results so you fully understand what they did and whether you can.

(4) take the time to become intimate with your data. autocorrelation structure? patterns, oddities outliers? plot the individual series and stare at them.

(5) start as simple as possible. 1 lag in either fd or fod (try both, you'll have more data with fod), and try one instlag.

(6) keep in mind, like all datasets, there may be nothing there. you aren't fishing for significance--that's p-hacking. you want a model that best represents the data generating process, and sometimes that requires a very careful study of the data, especially when you're using multiple lags.

(7) what is your purpose? Are you trying to publish, or are you trying to write a class paper? There are different standards for each.

[Jesse Nooijen]

As you've realized, there's a bit of art to this, and pvar involves a lot of art.

A few recommendations.

(1) what is your intuition about how the outcomes are related over time? don't rely simply on what Stata puts out.

(2) Is var even the right approach?

(3) Replicate/reproduce the Kang (2016) results so you fully understand what they did and whether you can.

(4) take the time to become intimate with your data. autocorrelation structure? patterns, oddities outliers? plot the individual series and stare at them.

(5) start as simple as possible. 1 lag in either fd or fod (try both, you'll have more data with fod), and try one instlag.

(6) keep in mind, like all datasets, there may be nothing there. you aren't fishing for significance--that's p-hacking. you want a model that best represents the data generating process, and sometimes that requires a very careful study of the data, especially when you're using multiple lags.

(7) what is your purpose? Are you trying to publish, or are you trying to write a class paper? There are different standards for each.

Thanks for your extensive reply once again. Given the discussions with my supervisor, and the approach by Kang et al. (which I slightly deviate from to improve it) I believe pvar to be the right approach. I ran various tests with different lags, and after some further discussion, my supervisor agreed that a lag of 2 with 3 instrumental lags is most likely the right way to go. We also concluded to use FOD compared to FD as this improves the estimation (does not have the flaws of FD).

I agree with your remark on p-hacking, I was not looking for singificant variables, I am merely interested in a non-significant p-value for the Hansen J-stastic, to confirm the validity of the instruments.

Regarding the model specifications, I do have one further question. I wanted to include Year Fixed effects as control, but pvar does not directly allow this. Therefore, as advised, I tried running code with this result:

. xi: pvar FP stdCSR stdCSI, lags(2) exog(FS FL ROA AI RDI i.Year) instlags(1/3) overid
i.Year            _IYear_2010-2019    (naturally coded; _IYear_2010 omitted)

But this did not produce any results. Therefore I tried to manually create dummies using

tabulate Year, generate(Year_dummy)
global YearDummy Year_dummy1 Year_dummy2 Year_dummy3 Year_dummy4 Year_dummy5 Year_dummy6 Year_dummy7 Year_dummy8 Year_dummy9 Year_dummy10

I read online that I should exclude 1 year dummy to prevent multicollinearity, do you have any advice for which dummy you would recommend to exclude (first or last?) I tried finding past sources particularly for pvar with dummy years but I merely found random/fixed effect references. Furthermore, I also investigated teh manual by Abrigo & Love (2016) some more, and realized that perhaps I should specify GMM as estimator, given I am using alternative instrumental lags, should I not move away from the regular estimation process to GMM? Any thoughts on this?

Regarding #7, I am writing a master thesis, not looking to publish, but my professor challenged me to take a more difficult topic than I originally had planned and that is how I ended up here. I am looking to make it as good and robust as possible, hence all these questions.

Thanks once again!

[George Ford]

Not sure why you need a year fixed effect when working with differenced data. Time fixed effects are very much like differencing.

If you use them, it normally makes no difference which you exclude (the coef on the FE change, but they are generally of no interest, and of no interest here). I'm not sure with pvar (or the resulting irf). Try the first and lastto see if the results change. I'd have to think about what the FE are doing in that model (maybe the same thing they normally do, but not sure).

[Jesse Nooijen]

Not sure why you need a year fixed effect when working with differenced data. Time fixed effects are very much like differencing.

If you use them, it normally makes no difference which you exclude (the coef on the FE change, but they are generally of no interest, and of no interest here). I'm not sure with pvar (or the resulting irf). Try the first and lastto see if the results change. I'd have to think about what the FE are doing in that model (maybe the same thing they normally do, but not sure).

Interesting, I was not aware of that but it makes sense, thanks for clarifying that.

Regarding specifying these effects, initially I left out year fixed effects since I assumed the specification of FD/FOD would indeed handle the time fixed effects. However, since my professor specifically told me to include them and presented me with a command to try, I assumed they are necessary.

In Kang et al., they report the following: "Furthermore, recognizing the panel nature of our data, we include firm- and time-specific fixed effects and adapt the Blundell and Bond (1998) estimator that uses lagged values and change in lagged values for endogenous variables as instruments to identify the causal effects." and they specifically report Year Fixed Effects in their regression table with a note specifiying they use first differencing, which makes me again doubt whether I have to specifically include these effects through dummies as well. I will discuss with my supervisor whether this is something we overlooked.

I did also see that you can specifiy gmmstyle as option to alter the estimation method. Given that I am using instrumental lags that deviate, should I not include this specification? The regular estimation method is less reliablle when using different instrumental lags if not mistaken?