Dear Statlist members,
Good afternoon.
I am estimating a translog cost function intending to measure the efficiency of the Syrian banking sector, my sample includes 17 banks (out of a total of 20 banks) for 12 years starting from 2005 till 2016 (yielding an unbalanced panel data). To check the variability of the my sample, my supervisor proposed to perform a regression between on 3 stages: the first stage would be to run a regression (according to the Hausman test, the results were in favor of using the fixed effects model) to check for the coefficients of my regressors, then the second step would be to run the same regression but adding a time dummy (represents a crisis dummy that takes the value of zero in the pre-war period 2005-2011, and takes the value of 1 in the war period 2012-2016), and finally the third step is to interact the crisis dummy with all my regressors and check the coefficients from this step with the value of the coefficients from the first step.
I chose my dependent variable as the ln(total cost/one of the input prices) and my regressors are three outputs (in logarithms), 2 price inputs (taken in logarithm form and divided by the third input price as we did with the total costs), 2 control variables (in logarithm form) and the interactions between the outputs, the input prices, and the control variables, this would yield 30 regressors in the initial model. In the second model, I added the crisis dummy, in the third model I added the interactions between the crisis dummy with all the regressors in the first model so we would have 61 regressors. The values of the coefficients in the first model and second model were almost the same but to my surprise, the coefficients of the regressors in the third model were so huge (and almost all of the coefficients are significant) compared to to the same regressors in the first and second models.
Many thanks in advance
Good afternoon.
I am estimating a translog cost function intending to measure the efficiency of the Syrian banking sector, my sample includes 17 banks (out of a total of 20 banks) for 12 years starting from 2005 till 2016 (yielding an unbalanced panel data). To check the variability of the my sample, my supervisor proposed to perform a regression between on 3 stages: the first stage would be to run a regression (according to the Hausman test, the results were in favor of using the fixed effects model) to check for the coefficients of my regressors, then the second step would be to run the same regression but adding a time dummy (represents a crisis dummy that takes the value of zero in the pre-war period 2005-2011, and takes the value of 1 in the war period 2012-2016), and finally the third step is to interact the crisis dummy with all my regressors and check the coefficients from this step with the value of the coefficients from the first step.
I chose my dependent variable as the ln(total cost/one of the input prices) and my regressors are three outputs (in logarithms), 2 price inputs (taken in logarithm form and divided by the third input price as we did with the total costs), 2 control variables (in logarithm form) and the interactions between the outputs, the input prices, and the control variables, this would yield 30 regressors in the initial model. In the second model, I added the crisis dummy, in the third model I added the interactions between the crisis dummy with all the regressors in the first model so we would have 61 regressors. The values of the coefficients in the first model and second model were almost the same but to my surprise, the coefficients of the regressors in the third model were so huge (and almost all of the coefficients are significant) compared to to the same regressors in the first and second models.
- To define my regressors in a macro:
Code:
global xvar lntl lnoea lnnii lnw2D lnw3D lntl2 lnoea2 lnnii2 lnw2D2 lnw3D2 lneq lnllp lneq2 lnllp2 iact1 iact2 iact3 iact4 iact5 iact6 iact7 iact8 iact9 iact10 iact11 iact12 iact13 iact14 iact15 iact16
-
Code:
xtset id year
-
Code:
xtreg lntcD $xvar, fe
- The results of the first model:
-
Code:
Fixed-effects (within) regression Number of obs = 111 Group variable: id Number of groups = 17 R-sq: Obs per group: within = 0.9795 min = 2 between = 0.5923 avg = 6.5 overall = 0.8121 max = 9 F(30,64) = 101.82 corr(u_i, Xb) = -0.1038 Prob > F = 0.0000 ------------------------------------------------------------------------------ lntcD | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------- lntl | -.8543398 1.502416 -0.57 0.572 -3.85576 2.147081 lnoea | -2.668691 1.575779 -1.69 0.095 -5.816671 .4792895 lnnii | .2089025 .2709448 0.77 0.444 -.3323719 .750177 lnw2D | 1.004794 .700106 1.44 0.156 -.3938285 2.403417 lnw3D | .0003386 1.399373 0.00 1.000 -2.79523 2.795908 lntl2 | .0758403 .0854099 0.89 0.378 -.0947857 .2464662 lnoea2 | -.1743747 .1219659 -1.43 0.158 -.4180296 .0692803 lnnii2 | .010878 .0099781 1.09 0.280 -.0090555 .0308116 lnw2D2 | .0005951 .0409005 0.01 0.988 -.081113 .0823032 lnw3D2 | -.0235098 .0699826 -0.34 0.738 -.1633161 .1162966 lneq | .108945 1.304998 0.08 0.934 -2.498087 2.715977 lnllp | 1.133728 .3713425 3.05 0.003 .3918864 1.87557 lneq2 | -.2203478 .104942 -2.10 0.040 -.4299935 -.0107021 lnllp2 | .0146593 .010815 1.36 0.180 -.0069461 .0362647 iact1 | -.0136558 .0438129 -0.31 0.756 -.1011822 .0738706 iact2 | .1070544 .0616944 1.74 0.088 -.0161942 .230303 iact3 | .1039922 .0739666 1.41 0.165 -.0437731 .2517575 iact4 | .0393443 .0991637 0.40 0.693 -.158758 .2374465 iact5 | .0147782 .0188256 0.79 0.435 -.0228301 .0523866 iact6 | .0110684 .0205931 0.54 0.593 -.0300711 .0522078 iact7 | -.1555763 .0583796 -2.66 0.010 -.272203 -.0389496 iact8 | -.0425453 .0895008 -0.48 0.636 -.2213437 .136253 iact9 | .0114676 .0189465 0.61 0.547 -.0263824 .0493175 iact10 | -.0923999 .0198445 -4.66 0.000 -.1320438 -.0527559 iact11 | -.0366808 .0548468 -0.67 0.506 -.1462499 .0728882 iact12 | .2972163 .099944 2.97 0.004 .0975553 .4968774 iact13 | -.0242802 .0185953 -1.31 0.196 -.0614285 .0128681 iact14 | -.0499301 .0195805 -2.55 0.013 -.0890467 -.0108136 iact15 | .0304576 .0207391 1.47 0.147 -.0109737 .0718888 iact16 | -.0072393 .0116585 -0.62 0.537 -.0305299 .0160513 _cons | 31.09566 17.05464 1.82 0.073 -2.974905 65.16622 -------------+---------------------------------------------------------------- sigma_u | .60556741 sigma_e | .16679111 rho | .92948776 (fraction of variance due to u_i) ------------------------------------------------------------------------------ F test that all u_i=0: F(16, 64) = 6.21 Prob > F = 0.0000
- To add the crisis dummy:
-
Code:
global xvar1 lntl lnoea lnnii lnw2D lnw3D lntl2 lnoea2 lnnii2 lnw2D2 lnw3D2 lneq lnllp lneq2 lnllp2 iact1 iact2 iact3 iact4 iact5 iact6 iact7 iact8 iact9 iact10 iact11 iact12 iact13 iact14 iact15 iact16 cd
-
Code:
xtreg lntcD $xvar1, fe
- The results of the second model:
-
Code:
Fixed-effects (within) regression Number of obs = 111 Group variable: id Number of groups = 17 R-sq: Obs per group: within = 0.9795 min = 2 between = 0.5999 avg = 6.5 overall = 0.8157 max = 9 F(31,63) = 97.03 corr(u_i, Xb) = -0.0964 Prob > F = 0.0000 ------------------------------------------------------------------------------ lntcD | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------- lntl | -.859437 1.514366 -0.57 0.572 -3.885655 2.166781 lnoea | -2.6956 1.597465 -1.69 0.096 -5.887879 .4966785 lnnii | .2134203 .2745974 0.78 0.440 -.3353188 .7621594 lnw2D | 1.005569 .7055249 1.43 0.159 -.4043095 2.415448 lnw3D | -.0188541 1.415633 -0.01 0.989 -2.847771 2.810063 lntl2 | .0773606 .0866299 0.89 0.375 -.0957555 .2504767 lnoea2 | -.1715172 .1242915 -1.38 0.172 -.419894 .0768596 lnnii2 | .010869 .0100553 1.08 0.284 -.0092249 .0309628 lnw2D2 | .0017487 .0418873 0.04 0.967 -.0819563 .0854537 lnw3D2 | -.0235178 .0705225 -0.33 0.740 -.1644458 .1174102 lneq | .132109 1.32359 0.10 0.921 -2.512874 2.777092 lnllp | 1.125977 .3775578 2.98 0.004 .3714874 1.880466 lneq2 | -.2189145 .106158 -2.06 0.043 -.4310543 -.0067746 lnllp2 | .0143052 .0111369 1.28 0.204 -.0079501 .0365606 iact1 | -.0139907 .0442042 -0.32 0.753 -.1023256 .0743443 iact2 | .10676 .0621995 1.72 0.091 -.0175358 .2310558 iact3 | .1041202 .0745418 1.40 0.167 -.0448397 .2530802 iact4 | .0403293 .100132 0.40 0.688 -.1597686 .2404272 iact5 | .0144918 .0190612 0.76 0.450 -.023599 .0525826 iact6 | .0105911 .0209808 0.50 0.615 -.0313356 .0525178 iact7 | -.1551023 .05891 -2.63 0.011 -.2728245 -.0373801 iact8 | -.0417318 .0903449 -0.46 0.646 -.2222717 .1388081 iact9 | .0110464 .0192864 0.57 0.569 -.0274943 .0495872 iact10 | -.0925893 .0200352 -4.62 0.000 -.1326263 -.0525522 iact11 | -.0370671 .0553265 -0.67 0.505 -.1476282 .073494 iact12 | .2943027 .1024665 2.87 0.006 .0895397 .4990656 iact13 | -.0244373 .0187663 -1.30 0.198 -.0619388 .0130641 iact14 | -.0502809 .0198618 -2.53 0.014 -.0899715 -.0105903 iact15 | .0314099 .0217898 1.44 0.154 -.0121335 .0749534 iact16 | -.0069394 .0119078 -0.58 0.562 -.0307352 .0168564 cd | .0153238 .0992154 0.15 0.878 -.1829423 .2135899 _cons | 31.30776 17.24099 1.82 0.074 -3.145606 65.76113 -------------+---------------------------------------------------------------- sigma_u | .59855999 sigma_e | .16807782 rho | .92691233 (fraction of variance due to u_i) ------------------------------------------------------------------------------ F test that all u_i=0: F(16, 63) = 5.64 Prob > F = 0.0000
- To interact the crisis dummy with the regressors
-
Code:
foreach x of varlist lntl lnoea lnnii lnw2D lnw3D lntl2 lnoea2 lnnii2 lnw2D2 lnw3D2 lneq lnllp lneq2 lnllp2 iact1 iact2 iact3 iact4 iact5 iact6 iact7 iact8 iact9 iact10 iact11 iact12 iact13 iact14 iact15 iact16{ -
Code:
gen double cd`x' = cd * (`x')
-
Code:
}
- Include all the regressors and the interaction with the regressors in the third model:
-
Code:
global xvar2 lntl lnoea lnnii lnw2D lnw3D lntl2 lnoea2 lnnii2 lnw2D2 lnw3D2 lneq lnllp lneq2 lnllp2 iact1 iact2 iact3 iact4 iact5 iact6 iact7 iact8 iact9 iact10 iact11 iact12 iact13 iact14 iact15 iact16 cd cdlntl cdlnoea cdlnnii cdlnw2D cdlnw3D cdlntl2 cdlnoea2 cdlnnii2 cdlnw2D2 cdlnw3D2 cdlneq cdlnllp cdlneq2 cdlnllp2 cdiact1 cdiact2 cdiact3 cdiact4 cdiact5 cdiact6 cdiact7 cdiact8 cdiact9 cdiact10 cdiact11 cdiact12 cdiact13 cdiact14 cdiact15 cdiact16
-
Code:
xtreg lntcD $xvar2, fe
- The resultsof the third model:
-
Code:
Fixed-effects (within) regression Number of obs = 111 Group variable: id Number of groups = 17 R-sq: Obs per group: within = 0.9947 min = 2 between = 0.2097 avg = 6.5 overall = 0.5612 max = 9 F(61,33) = 101.97 corr(u_i, Xb) = -0.2020 Prob > F = 0.0000 ------------------------------------------------------------------------------ lntcD | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------- lntl | 47.09133 14.91185 3.16 0.003 16.75293 77.42973 lnoea | -46.40368 14.0999 -3.29 0.002 -75.09013 -17.71722 lnnii | 4.37507 1.88242 2.32 0.026 .5452585 8.204882 lnw2D | 1.100498 3.348964 0.33 0.745 -5.71302 7.914015 lnw3D | 19.17671 6.098829 3.14 0.004 6.768545 31.58487 lntl2 | -1.799829 .4239165 -4.25 0.000 -2.662294 -.9373648 lnoea2 | .8813601 .4463555 1.97 0.057 -.0267571 1.789477 lnnii2 | .0544818 .0397587 1.37 0.180 -.026408 .1353715 lnw2D2 | -1.032923 .2472421 -4.18 0.000 -1.535941 -.529905 lnw3D2 | -.2459903 .4326006 -0.57 0.573 -1.126123 .6341423 lneq | -29.03087 9.461581 -3.07 0.004 -48.2806 -9.781139 lnllp | -3.008367 1.966399 -1.53 0.136 -7.009036 .9923016 lneq2 | .3430149 .5905285 0.58 0.565 -.8584244 1.544454 lnllp2 | -.0981912 .0384162 -2.56 0.015 -.1763495 -.0200328 iact1 | 1.841016 .5717113 3.22 0.003 .6778605 3.004171 iact2 | -1.801202 .6411307 -2.81 0.008 -3.105592 -.4968115 iact3 | -.7978385 .4025068 -1.98 0.056 -1.616745 .0210676 iact4 | .6622181 .3090564 2.14 0.040 .0334381 1.290998 iact5 | .1290717 .1018969 1.27 0.214 -.0782392 .3363825 iact6 | -.2824996 .1258163 -2.25 0.032 -.5384747 -.0265244 iact7 | -1.362191 .4328597 -3.15 0.003 -2.24285 -.481531 iact8 | .8699046 .4723241 1.84 0.075 -.0910461 1.830855 iact9 | .3260643 .091666 3.56 0.001 .1395684 .5125601 iact10 | -.5697394 .0937982 -6.07 0.000 -.7605734 -.3789055 iact11 | -.765502 .484363 -1.58 0.124 -1.750946 .2199418 iact12 | 1.807147 .4855092 3.72 0.001 .8193708 2.794923 iact13 | .0254151 .1274403 0.20 0.843 -.2338641 .2846944 iact14 | .7476658 .1699423 4.40 0.000 .4019157 1.093416 iact15 | -.0806342 .1316575 -0.61 0.544 -.3484933 .187225 iact16 | -.201224 .0706172 -2.85 0.007 -.3448958 -.0575523 cd | -146.8053 65.18711 -2.25 0.031 -279.4295 -14.18115 cdlntl | -44.43249 14.84393 -2.99 0.005 -74.63269 -14.2323 cdlnoea | 49.45785 14.24765 3.47 0.001 20.47078 78.44492 cdlnnii | -4.480562 1.991074 -2.25 0.031 -8.531434 -.429691 cdlnw2D | -.0143426 3.363131 -0.00 0.997 -6.856683 6.827998 cdlnw3D | -25.36333 6.401982 -3.96 0.000 -38.38826 -12.3384 cdlntl2 | 1.727536 .4276136 4.04 0.000 .8575497 2.597522 cdlnoea2 | -1.39611 .4913676 -2.84 0.008 -2.395805 -.3964154 cdlnnii2 | -.0619074 .0396564 -1.56 0.128 -.142589 .0187741 cdlnw2D2 | 1.217328 .268301 4.54 0.000 .6714655 1.763191 cdlnw3D2 | .0988369 .4576092 0.22 0.830 -.832176 1.02985 cdlneq | 31.12527 9.460709 3.29 0.002 11.87731 50.37323 cdlnllp | 3.286925 2.004197 1.64 0.111 -.7906452 7.364495 cdlneq2 | -.5015864 .5895298 -0.85 0.401 -1.700994 .697821 cdlnllp2 | .168012 .0649718 2.59 0.014 .0358258 .3001981 cdiact1 | -1.816611 .5699775 -3.19 0.003 -2.976239 -.6569834 cdiact2 | 1.881834 .648641 2.90 0.007 .5621645 3.201504 cdiact3 | .7273724 .3893951 1.87 0.071 -.0648578 1.519603 cdiact4 | -.2111172 .3302419 -0.64 0.527 -.8829994 .4607651 cdiact5 | -.1428134 .1035163 -1.38 0.177 -.3534189 .067792 cdiact6 | .2599114 .1259827 2.06 0.047 .0035976 .5162252 cdiact7 | 1.379313 .4379577 3.15 0.003 .4882818 2.270345 cdiact8 | -.9126804 .4920562 -1.85 0.073 -1.913776 .0884154 cdiact9 | -.3629481 .0998283 -3.64 0.001 -.5660503 -.1598459 cdiact10 | .5378179 .1133859 4.74 0.000 .3071326 .7685032 cdiact11 | .6821611 .4743148 1.44 0.160 -.2828396 1.647162 cdiact12 | -1.682702 .5045611 -3.33 0.002 -2.709239 -.6561646 cdiact13 | -.0402936 .1287717 -0.31 0.756 -.3022816 .2216945 cdiact14 | -.8163593 .1742462 -4.69 0.000 -1.170866 -.4618527 cdiact15 | .0707761 .1403209 0.50 0.617 -.2147089 .3562611 cdiact16 | .2462194 .0743865 3.31 0.002 .0948788 .39756 _cons | 125.5604 68.69779 1.83 0.077 -14.20634 265.3271 -------------+---------------------------------------------------------------- sigma_u | .89100955 sigma_e | .11779067 rho | .98282358 (fraction of variance due to u_i) ------------------------------------------------------------------------------ F test that all u_i=0: F(16, 33) = 4.69 Prob > F = 0.0001
Many thanks in advance
Comment