-
Login or Register
- Log in with
- You are not logged in. You can browse but not post. Login or Register by clicking 'Login or Register' at the top-right of this page. For more information on Statalist, see the FAQ.
-
Sebastian Kripfganz thanks for the suggestion. It works fine now. -
xtseqreg does not support xtabond2 for the first stage, because xtabond2 does not have the necessary functionality to compute parameter-level scores. You can use xtdpdgmm instead of xtabond2. For simple GMM estimations, you can also simply estimate the first stage directly with xtseqreg itself. For examples, please see further up in this Statalist topic (in particular note the instruction in post #81) or the replication files to my 2019 Journal of Applied Econometric article.
Time dummies are usually included in the first stage (i.e., inside the brackets).Leave a comment:
-
Hi All,
I estimate the following model:
Then, I use:Code:xtabond2 investment l.investment mtb time*, gmm (investment mtb, collapse lag (2 4)) iv(time*) ar(4) twos r nol
Stata keeps saying:Code:xtseqreg investment (l.investment mtb time*) efi, vce(robust) iv(time*) first(gmm10)
I cannot figure out why I got that error message. Have you any suggestion about?Code:. xtseqreg investment (l.investment mtb time*) efi, vce(robust) iv(time*) first( > gmm10) option first() incorrectly specified -- variable names do not match
On the side, should I include the set of period dummies, time*, in the brackets or should I put them close to efi, the time invariant regressor?
Thanks in advance.
DarioLeave a comment:
-
That's actually spot on. EV_EBITDA is not country-specific. I did not realize this would cause trouble. Thank you!Leave a comment:
-
There appears to be perfect collinearity in your data set between one of the regressors and the full set of time dummies.. The only explanation I have is that one of your regressors might only vary over time but not across groups.Leave a comment:
-
With xtdpdgmm 2021 is omitted, while xtseqreg omits the constant.
Code:. xtdpdgmm investments l.investments fundraising loans realrate EV_EBITDA GDPGrowth, gmm(l.investments fundraising loans realrate EV_EBITDA GDPGrowth, lagrange > (1 1) collapse model(diff)) teffects vce(robust) note: 2021.Year omitted because of collinearity Generalized method of moments estimation Fitting full model: Step 1 f(b) = 4.713e-24 Group variable: Country Number of obs = 322 Time variable: Year Number of groups = 23 Moment conditions: linear = 19 Obs per group: min = 14 nonlinear = 0 avg = 14 total = 19 max = 14 (Std. Err. adjusted for 23 clusters in Country) ------------------------------------------------------------------------------ | Robust investments | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- investments | L1. | -.2415121 .3279978 -0.74 0.462 -.884376 .4013518 | fundraising | -.1464715 .1552177 -0.94 0.345 -.4506925 .1577496 loans | -.0327191 .0366048 -0.89 0.371 -.1044631 .0390249 realrate | .0224572 .1848913 0.12 0.903 -.3399231 .3848375 EV_EBITDA | .1772153 .1587355 1.12 0.264 -.1339006 .4883312 GDPGrowth | -.1380841 .1200476 -1.15 0.250 -.3733731 .0972049 | Year | 2009 | -.9359642 1.229798 -0.76 0.447 -3.346323 1.474395 2010 | -.9254255 .8985977 -1.03 0.303 -2.686645 .8357937 2011 | -.410679 1.003757 -0.41 0.682 -2.378006 1.556648 2012 | -.8875602 .9375887 -0.95 0.344 -2.7252 .9500798 2013 | -1.618549 1.080531 -1.50 0.134 -3.73635 .4992529 2014 | -1.60488 1.295792 -1.24 0.216 -4.144586 .9348267 2015 | -.6483632 .8649522 -0.75 0.453 -2.343638 1.046912 2016 | -.5990037 .6912807 -0.87 0.386 -1.953889 .7558815 2017 | -.0421427 .3971043 -0.11 0.915 -.8204528 .7361674 2018 | -.6543601 .9897612 -0.66 0.509 -2.594256 1.285536 2019 | -.6679732 .6109213 -1.09 0.274 -1.865357 .5294105 2020 | -1.680386 .7663415 -2.19 0.028 -3.182388 -.1783846 2021 | 0 (empty) | _cons | 14.97025 6.142948 2.44 0.015 2.930292 27.01021 ------------------------------------------------------------------------------ Instruments corresponding to the linear moment conditions: 1, model(diff): L1.L.investments L1.fundraising L1.loans L1.realrate L1.EV_EBITDA L1.GDPGrowth 2, model(level): 2009bn.Year 2010.Year 2011.Year 2012.Year 2013.Year 2014.Year 2015.Year 2016.Year 2017.Year 2018.Year 2019.Year 2020.Year 3, model(level): _consLeave a comment:
-
There are effectively only 19 instruments for 20 coefficients. From the output, I cannot tell why one of the instruments was omitted. Can you run the same specification with my other command xtdpdgmm and check if there is any of the instruments omitted in the list of instruments below the regression output?
The time effects capture the trends in all variables, not just the dependent variable.Code:xtdpdgmm investments l.investments fundraising loans realrate EV_EBITDA GDPGrowth, gmm(l.investments fundraising loans realrate EV_EBITDA GDPGrowth, lagrange(1 1) collapse model(diff)) teffects vce(robust)
Leave a comment:
-
If I specify it correctly, it seems that the constant is always omitted? What is the reason for this?
Another question I cannot seem to understand: The investment series is a growing series, but in most specifications the time effects are negative? Why is that?
Code:xtseqreg investments l.investments fundraising loans realrate EV_EBITDA GDPGrowth, gmm(l.investments fundraising > loans realrate EV_EBITDA GDPGrowth, lagrange(1 1) collapse model(diff)) teffects vce(robust) Group variable: Country Number of obs = 322 Time variable: Year Number of groups = 23 Obs per group: min = 14 avg = 14 max = 14 Number of instruments = 19 (Std. Err. adjusted for 23 clusters in Country) ------------------------------------------------------------------------------ | Robust investments | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- investments | L1. | -.2415121 .3279978 -0.74 0.462 -.884376 .4013518 | fundraising | -.1464715 .1552177 -0.94 0.345 -.4506925 .1577496 loans | -.0327191 .0366048 -0.89 0.371 -.1044631 .0390249 realrate | .0224572 .1848913 0.12 0.903 -.3399231 .3848375 EV_EBITDA | 2.054362 .7227027 2.84 0.004 .6378913 3.470834 GDPGrowth | -.1380841 .1200476 -1.15 0.250 -.3733731 .0972049 | Year | 2009 | 2.912187 .8515469 3.42 0.001 1.243186 4.581189 2010 | -.7377108 .8516696 -0.87 0.386 -2.406952 .9315309 2011 | -1.020752 1.136603 -0.90 0.369 -3.248452 1.206948 2012 | -1.216061 .9925282 -1.23 0.220 -3.161381 .7292586 2013 | -2.744837 1.390924 -1.97 0.048 -5.470998 -.0186759 2014 | -2.825025 1.701869 -1.66 0.097 -6.160627 .5105765 2015 | -3.792585 1.853554 -2.05 0.041 -7.425483 -.1596861 2016 | -3.696297 1.428735 -2.59 0.010 -6.496566 -.896027 2017 | -5.063511 2.241953 -2.26 0.024 -9.457659 -.6693639 2018 | -3.235437 .9713064 -3.33 0.001 -5.139163 -1.331712 2019 | -6.440201 2.549344 -2.53 0.012 -11.43682 -1.443578 2020 | -6.138611 2.101067 -2.92 0.003 -10.25663 -2.020595 2021 | -7.367803 3.023332 -2.44 0.015 -13.29342 -1.442181 | _cons | 0 (omitted) ------------------------------------------------------------------------------Leave a comment:
-
Actually, now your model is underidentified. You have 1 instrument too few, which is why one of the time dummies is omitted. If you include 2 lags of the dependent variable as regressors, you need to also include at least two lags as instruments. As a minimum, the following would give you a just-identified model:
Code:xtseqreg L(0/2).investments fundraising loans realrate EV_EBITDA GDPGrowth, gmm(L(1/2).investments fundraising loans realrate EV_EBITDA GDPGrowth, lagrange(1 1) collapse model(diff)) teffects vce(robust)Leave a comment:
-
Great insights - I am sorry for my ignorance, but I have a final specification. As you suggested above I replaced the instrument with the lagged dependent variable, removed time effects in the second-stage regression, and added two lags for the dependent variable. For the GMM instruments I reduced the lag order in addition to collapsing the instruments. Now estimates make sense, the number of instruments is reasonable and the instruments are differenced. Does the specification make sense? Sorry to bother you, but I think these comments might be valuable for other beginners within the field as well!
Code:xtseqreg L(0/2).investments fundraising loans realrate EV_EBITDA GDPGrowth, gmm(L.investments fundraising lo > ans realrate EV_EBITDA GDPGrowth, lagrange(1 1) collapse model(diff)) teffects vce(robust) Group variable: Country Number of obs = 299 Time variable: Year Number of groups = 23 Obs per group: min = 13 avg = 13 max = 13 Number of instruments = 18 (Std. Err. adjusted for 23 clusters in Country) ------------------------------------------------------------------------------ | Robust investments | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- investments | L1. | .3526592 .1260307 2.80 0.005 .1056435 .5996749 L2. | .4248817 .1087693 3.91 0.000 .2116978 .6380657 | fundraising | -.0670592 .059737 -1.12 0.262 -.1841416 .0500232 loans | -.0138039 .0342988 -0.40 0.687 -.0810283 .0534205 realrate | .0078039 .1847154 0.04 0.966 -.3542315 .3698393 EV_EBITDA | .3774473 .1448065 2.61 0.009 .0936318 .6612628 GDPGrowth | .014976 .1742886 0.09 0.932 -.3266235 .3565754 | Year | 2010 | -.0048678 1.011757 -0.00 0.996 -1.987874 1.978139 2011 | .7605277 1.020011 0.75 0.456 -1.238657 2.759712 2012 | -.1814969 .6433709 -0.28 0.778 -1.442481 1.079487 2013 | -.968242 .6885015 -1.41 0.160 -2.31768 .3811961 2014 | -.4326073 .6633004 -0.65 0.514 -1.732652 .8674375 2015 | .3612557 .7168759 0.50 0.614 -1.043795 1.766307 2016 | -.330125 .7287244 -0.45 0.651 -1.758399 1.098149 2017 | -.0637389 .5153792 -0.12 0.902 -1.073864 .9463857 2018 | -.7606723 1.134643 -0.67 0.503 -2.984532 1.463187 2019 | -.6009757 .7758376 -0.77 0.439 -2.121589 .9196381 2020 | -1.339748 .6713034 -2.00 0.046 -2.655478 -.0240173 2021 | 0 (omitted) | _cons | 0 (omitted) ------------------------------------------------------------------------------ . xtseqreg investments (L(1/2).investments fundraising loans realrate EV_EBITDA GDPGrowth) qe_D, iv(qe_D, mode > l(level)) vce(robust) Group variable: Country Number of obs = 299 Time variable: Year Number of groups = 23 ------------------------------------------------------------------------------ Equation _first Equation _second Number of obs = 299 Number of obs = 299 Number of groups = 23 Number of groups = 23 Obs per group: min = 13 Obs per group: min = 13 avg = 13 avg = 13 max = 13 max = 13 Number of instruments = 18 Number of instruments = 2 (Std. Err. adjusted for clustering on Country) ------------------------------------------------------------------------------ | Robust investments | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- _first | investments | L1. | .3526592 .1260307 2.80 0.005 .1056435 .5996749 L2. | .4248817 .1087693 3.91 0.000 .2116978 .6380657 | fundraising | -.0670592 .059737 -1.12 0.262 -.1841416 .0500232 loans | -.0138039 .0342988 -0.40 0.687 -.0810283 .0534205 realrate | .0078039 .1847154 0.04 0.966 -.3542315 .3698393 EV_EBITDA | .3774473 .1448065 2.61 0.009 .0936318 .6612628 GDPGrowth | .014976 .1742886 0.09 0.932 -.3266235 .3565754 _cons | 0 (omitted) -------------+---------------------------------------------------------------- _second | qe_D | .5846122 .6130905 0.95 0.340 -.617023 1.786247 _cons | -.5007053 .5002986 -1.00 0.317 -1.481273 .4798619 ------------------------------------------------------------------------------ . estat overid Hansen's J-test for equation _first chi2(0) = 0.0000 note: coefficients are exactly identified Prob > chi2 = . Hansen's J-test for equation _second chi2(0) = 0.0000 note: coefficients are exactly identified Prob > chi2 = .Leave a comment:
-
The first lag of the dependent variable, investments, is not a valid instrument for the first-differenced model because it is correlated by construction with the first-differenced error term. You would need to start with lag 2 or replace investments by L.investments in your gmm() option.Leave a comment:
-
If I transform the instruments as you suggest, the results does not make any sense. However, I know for a fact that there indeed are unobserved group-specific effects. Is there something else which can be done?
Code:xtseqreg investments l.investments fundraising loans realrate EV_EBITDA GDPGrowth, gmm(investments fundraising loans realrate EV_EBITDA GDPGrowth, model(diff) lagrange(1 2) collapse) teffects vce(robust) Group variable: Country Number of obs = 322 Time variable: Year Number of groups = 23 Obs per group: min = 14 avg = 14 max = 14 Number of instruments = 24 (Std. Err. adjusted for 23 clusters in Country) ------------------------------------------------------------------------------ | Robust investments | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- investments | L1. | -.7970249 .1026553 -7.76 0.000 -.9982257 -.5958241 | fundraising | -.4651334 .1440001 -3.23 0.001 -.7473684 -.1828984 loans | -.1435853 .0646596 -2.22 0.026 -.2703158 -.0168547 realrate | .2401136 .1424713 1.69 0.092 -.039125 .5193523 EV_EBITDA | 3.351303 .2631574 12.73 0.000 2.835524 3.867082 GDPGrowth | -.2026799 .1112768 -1.82 0.069 -.4207784 .0154185 | Year | 2009 | 4.538746 .8283898 5.48 0.000 2.915132 6.16236 2010 | -1.653224 .8996291 -1.84 0.066 -3.416464 .1100171 2011 | -2.570672 1.353209 -1.90 0.057 -5.222913 .0815686 2012 | -2.214624 1.331253 -1.66 0.096 -4.823832 .394584 2013 | -4.821718 1.443958 -3.34 0.001 -7.651824 -1.991612 2014 | -4.759001 1.480642 -3.21 0.001 -7.661005 -1.856996 2015 | -6.212976 1.41284 -4.40 0.000 -8.982093 -3.44386 2016 | -4.813099 1.603158 -3.00 0.003 -7.95523 -1.670967 2017 | -7.415403 1.589979 -4.66 0.000 -10.5317 -4.299102 2018 | -3.058209 1.407552 -2.17 0.030 -5.81696 -.2994591 2019 | -9.845316 1.816203 -5.42 0.000 -13.40501 -6.285623 2020 | -7.98553 1.785062 -4.47 0.000 -11.48419 -4.486872 2021 | -10.98603 1.805061 -6.09 0.000 -14.52389 -7.448179 | _cons | 0 (omitted) ------------------------------------------------------------------------------Leave a comment:
-
Now you have specified only untransformed instruments for the level model (the default). This would only be valid if there were no unobserved group-specific effects. Not surprisingly, the Hansen test rejects. Also, time effects should normally be included in the first stage, not the second stage.Leave a comment:
-
Thank you, Sebastian. I know understand the difference between one-step and two-step significantly better. Having GMM-instrumented the variables with the collapse function provides for a reasonable estimate given my hypotheses, along with a reasonable number of instruments. However the J-test for instrumental validity is rejected. Is there anything I can do about this? Do you have any advice on the specification?
Code:xtseqreg investments l.investments fundraising loans realrate EV_EBITDA GDPGrowth, gmm(investments fundraising loans realrate EV_EBITDA GDPGrowth, lagrange(1 2) collapse) teffects vce(robust) Group variable: Country Number of obs = 322 Time variable: Year Number of groups = 23 Obs per group: min = 14 avg = 14 max = 14 Number of instruments = 24 (Std. Err. adjusted for 23 clusters in Country) ------------------------------------------------------------------------------ | Robust investments | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- investments | L1. | .343524 .0742916 4.62 0.000 .1979151 .4891328 | fundraising | .4023131 .0872731 4.61 0.000 .2312609 .5733653 loans | .0858774 .063126 1.36 0.174 -.0378473 .2096021 realrate | -.0598164 .1445509 -0.41 0.679 -.3431309 .2234982 EV_EBITDA | .4172634 .1554146 2.68 0.007 .1126565 .7218704 GDPGrowth | -.2297269 .2305901 -1.00 0.319 -.6816753 .2222214 | Year | 2009 | .6828628 .8297635 0.82 0.411 -.9434437 2.309169 2010 | .7206616 .7358084 0.98 0.327 -.7214964 2.16282 2011 | 1.530941 1.077079 1.42 0.155 -.5800947 3.641978 2012 | .4080854 .6441439 0.63 0.526 -.8544135 1.670584 2013 | .194311 .7907806 0.25 0.806 -1.355591 1.744213 2014 | -.1823429 .8501904 -0.21 0.830 -1.848686 1.484 2015 | -.3632815 1.126338 -0.32 0.747 -2.570863 1.8443 2016 | -1.887032 1.074414 -1.76 0.079 -3.992845 .2187815 2017 | -1.318689 2.079065 -0.63 0.526 -5.393582 2.756204 2018 | -2.829009 1.367854 -2.07 0.039 -5.509954 -.1480636 2019 | -1.583258 1.552625 -1.02 0.308 -4.626348 1.459831 2020 | -3.601026 1.462555 -2.46 0.014 -6.467581 -.7344706 2021 | -2.024731 2.20412 -0.92 0.358 -6.344728 2.295265 | _cons | 0 (omitted) ------------------------------------------------------------------------------I am grateful for your help so far!Code:xtseqreg investments (l.investments fundraising loans realrate EV_EBITDA GDPGrowth) qe_D, iv(qe_D, model(level)) teffects vce(robust) Group variable: Country Number of obs = 322 Time variable: Year Number of groups = 23 ------------------------------------------------------------------------------ Equation _first Equation _second Number of obs = 322 Number of obs = 322 Number of groups = 23 Number of groups = 23 Obs per group: min = 14 Obs per group: min = 14 avg = 14 avg = 14 max = 14 max = 14 Number of instruments = 24 Number of instruments = 15 (Std. Err. adjusted for clustering on Country) ------------------------------------------------------------------------------ | Robust investments | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- _first | investments | L1. | .343524 .0742916 4.62 0.000 .1979151 .4891328 | fundraising | .4023131 .0872731 4.61 0.000 .2312609 .5733653 loans | .0858774 .063126 1.36 0.174 -.0378473 .2096021 realrate | -.0598164 .1445509 -0.41 0.679 -.3431309 .2234982 EV_EBITDA | .4172634 .1554146 2.68 0.007 .1126565 .7218704 GDPGrowth | -.2297269 .2305901 -1.00 0.319 -.6816753 .2222214 _cons | 0 (omitted) -------------+---------------------------------------------------------------- _second | qe_D | -.4532447 .3270456 -1.39 0.166 -1.094242 .1877528 | Year | 2008 | 1.709431 2.193838 0.78 0.436 -2.590412 6.009273 2009 | 2.412 2.519466 0.96 0.338 -2.526063 7.350063 2010 | 2.449798 2.392151 1.02 0.306 -2.238731 7.138328 2011 | 3.240372 2.062877 1.57 0.116 -.8027918 7.283536 2012 | 2.117516 2.220177 0.95 0.340 -2.233951 6.468983 2013 | 1.903742 2.004936 0.95 0.342 -2.025861 5.833344 2014 | 1.546794 1.895133 0.82 0.414 -2.167598 5.261186 2015 | 1.622037 1.574558 1.03 0.303 -1.464039 4.708114 2016 | .0785806 1.263686 0.06 0.950 -2.398198 2.55536 2017 | .6469234 .6114013 1.06 0.290 -.551401 1.845248 2018 | -.8436903 1.324827 -0.64 0.524 -3.440303 1.752923 2019 | .4020603 .9941911 0.40 0.686 -1.546518 2.350639 2020 | -1.576295 1.46051 -1.08 0.280 -4.438842 1.286252 2021 | 0 (omitted) | _cons | -1.689724 2.192586 -0.77 0.441 -5.987114 2.607666 ------------------------------------------------------------------------------ . estat overid Hansen's J-test for equation _first chi2(5) = 23.0000 H0: overidentifying restrictions are valid Prob > chi2 = 0.0003 Hansen's J-test for equation _second chi2(0) = 0.0000 note: coefficients are exactly identified Prob > chi2 = .Leave a comment:
Leave a comment: