[Bich Ngoc Nguyen]

Dear all,

I would like to ask about the Ramsey Reset and other postestimation to test the appropriation of models in Stata. I carried out Heckman model and PPML model as the results below:

heckman lntrade lngdpexp lngdpimp lnstariff binary_ntm LPIexp LPIimp DBIexp DBIimp TABexp
> TABimp,select(lndist lnstariff binary_ntm LPIimp DBIimp TABimp)

Heckman selection model Number of obs = 572
(regression model with sample selection) Censored obs = 12
Uncensored obs = 560

Wald chi2(10) = 499.22
Log likelihood = -1044.587 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
lntrade | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lntrade |
lngdpexp | .5063633 .080751 6.27 0.000 .3480942 .6646323
lngdpimp | .7275791 .0682668 10.66 0.000 .5937787 .8613796
lnstariff | .6063113 2.680599 0.23 0.821 -4.647566 5.860188
binary_ntm | -.0742413 .1681425 -0.44 0.659 -.4037945 .2553118
LPIexp | 2.790873 .3975619 7.02 0.000 2.011666 3.57008
LPIimp | 1.525779 .3875687 3.94 0.000 .7661587 2.2854
DBIexp | .0444199 .0073635 6.03 0.000 .0299877 .058852
DBIimp | -.017263 .0087484 -1.97 0.048 -.0344097 -.0001164
TABexp | -.1215937 .0143852 -8.45 0.000 -.1497883 -.0933992
TABimp | -.0647002 .0143956 -4.49 0.000 -.0929151 -.0364853
_cons | -25.61454 4.627234 -5.54 0.000 -34.68375 -16.54533
-------------+----------------------------------------------------------------
select |
lndist | 1.007029 .5527868 1.82 0.068 -.0764133 2.090471
lnstariff | 1.017142 5.030818 0.20 0.840 -8.843079 10.87736
binary_ntm | -.074987 .3525874 -0.21 0.832 -.7660456 .6160717
LPIimp | 2.020955 .7232403 2.79 0.005 .6034304 3.43848
DBIimp | -.0299147 .0181729 -1.65 0.100 -.0655329 .0057036
TABimp | .000585 .0310995 0.02 0.985 -.0603688 .0615389
_cons | -13.32619 9.255747 -1.44 0.150 -31.46712 4.814745
-------------+----------------------------------------------------------------
/athrho | .2664853 .2707918 0.98 0.325 -.2642568 .7972274
/lnsigma | .3678971 .0303897 12.11 0.000 .3083344 .4274599
-------------+----------------------------------------------------------------
rho | .2603513 .2524367 -.2582728 .6624839
sigma | 1.444693 .0439038 1.361156 1.533358
lambda | .3761279 .3669703 -.3431206 1.095376
------------------------------------------------------------------------------
LR test of indep. eqns. (rho = 0): chi2(1) = 0.78 Prob > chi2 = 0.3762

.xtpoisson tradevalue lngdpexp lngdpimp lndist lnstariff binary_ntm DBIexp DBIimp LPIexp LP
> Iimp TABexp TABimp

Fitting Poisson model:

Random-effects Poisson regression Number of obs = 572
Group variable: pair Number of groups = 52

Random effects u_i ~ Gamma Obs per group:
min = 11
avg = 11.0
max = 11

Wald chi2(11) = 3.02e+06
Log likelihood = -1385085.5 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
tradevalue | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lngdpexp | .9250992 .0009164 1009.55 0.000 .9233032 .9268952
lngdpimp | .1529191 .0021301 71.79 0.000 .1487442 .1570939
lndist | 5.105654 1.137417 4.49 0.000 2.876358 7.33495
lnstariff | -.8488981 .0104352 -81.35 0.000 -.8693508 -.8284454
binary_ntm | -.1983055 .0004042 -490.56 0.000 -.1990978 -.1975132
DBIexp | -.0098621 .0000587 -167.95 0.000 -.0099772 -.009747
DBIimp | -.0040748 .0000569 -71.67 0.000 -.0041862 -.0039633
LPIexp | .10246 .0013376 76.60 0.000 .0998384 .1050816
LPIimp | .105585 .0023256 45.40 0.000 .1010269 .1101431
TABexp | -.0102395 .0000545 -187.91 0.000 -.0103463 -.0101327
TABimp | -.0052576 .0000682 -77.04 0.000 -.0053914 -.0051239
_cons | -60.89114 10.26587 -5.93 0.000 -81.01187 -40.7704
-------------+----------------------------------------------------------------
/lnalpha | .6251173 .162936 .3057687 .944466
-------------+----------------------------------------------------------------
alpha | 1.868465 .3044402 1.357668 2.57144
------------------------------------------------------------------------------
LR test of alpha=0: chibar2(01) = 2.7e+07 Prob >= chibar2 = 0.000


Looking at the results, I see some problems. First, some factors do not meet the expected sign as literature (such as: distance, or tariff). Secondly, i know that the ppml have too much error robust and the database i used do not have the zero-traded value ( only 12/572 observations). thus, I wonder about which models is more appropriate.

Some papers mentioned about Ramsey Reset test. Does is use for the case?

I am a Stata beginner so that i hope to receive all supports.

Thanks

[Joao Santos Silva]

Dear Bich Ngoc Nguyen

The sample selection estimator is not appropriate in this context. Also, for xtpoisson you need fixed effect; the random effects model is not suitable.

Best wishes,

Joao

[Richa Khurana]

Dear Sir,

In an earlier you mentioned that RESET test cannot be used to compare between models with different set of regressors.
I have a similar query.
While comparing OLS and PPML with the same set of regressors, OLS passes the RESET test but PPMl does not.
The results using OLS with country specific and time fixed effects are:
reg ln_trade time_fe* exp_fe* imp_fe* ln_GDP_i ln_GDP_j ln_GDPPC_i ln_GDPPC_j ln_DISTANCE LANG_off comcol BORDER TC_ij TD_iin_jout TD_iout_
> jin ,cluster(pair_id)
note: time_fe2 omitted because of collinearity
note: exp_fe11 omitted because of collinearity
note: imp_fe11 omitted because of collinearity

Linear regression Number of obs = 39538
F(182, 3995) = .
Prob > F = .
R-squared = 0.5452
Root MSE = 2.9177

(Std. Err. adjusted for 3996 clusters in pair_id)
------------------------------------------------------------------------------
| Robust
ln_trade | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ln_GDP_i | 1.245375 .2431161 5.12 0.000 .7687323 1.722019
ln_GDP_j | 1.188227 .259224 4.58 0.000 .6800034 1.696451
ln_GDPPC_i | -1.165797 .2554174 -4.56 0.000 -1.666557 -.6650363
ln_GDPPC_j | -.6012588 .2657007 -2.26 0.024 -1.122181 -.0803371
ln_DISTANCE | -2.575251 .0718338 -35.85 0.000 -2.716086 -2.434417
LANG_off | .1444777 .1361866 1.06 0.289 -.1225241 .4114795
comcol | .7577103 .2001424 3.79 0.000 .3653195 1.150101
BORDER | .4226024 .2374454 1.78 0.075 -.042923 .8881278
TC_ij | .0651144 .2272669 0.29 0.775 -.3804556 .5106844
TD_iin_jout | .0152666 .1673667 0.09 0.927 -.3128655 .3433986
TD_iout_jin | -.0923774 .1227574 -0.75 0.452 -.3330505 .1482957
_cons | -24.14524 5.945621 -4.06 0.000 -35.80198 -12.48851
------------------------------------------------------------------------------

RESET test result:
F( 1, 3995) = 0.13
Prob > F = 0.7189

Results using PPML:
ppml trade time_fe* exp_fe* imp_fe* ln_GDP_i ln_GDP_j ln_GDPPC_i ln_GDPPC_j ln_DISTANCE LANG_off comcol BORDER TC_ij TD_iin_jout TD_iout_ji
> n ,cluster(pair_id)
Number of parameters: 184
Number of observations: 67327
Pseudo log-likelihood: -3.217e+09
R-squared: .7117156
Option strict is: off
(Std. Err. adjusted for 4013 clusters in pair_id)
------------------------------------------------------------------------------
| Robust
trade | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ln_GDP_i | .7492899 .4156265 1.80 0.071 -.065323 1.563903
ln_GDP_j | 1.881322 .4708665 4.00 0.000 .9584406 2.804203
ln_GDPPC_i | -.6207829 .4301451 -1.44 0.149 -1.463852 .2222861
ln_GDPPC_j | -1.473061 .486012 -3.03 0.002 -2.425627 -.5204947
ln_DISTANCE | -1.1692 .0574596 -20.35 0.000 -1.281818 -1.056581
LANG_off | .3033443 .1929194 1.57 0.116 -.0747707 .6814594
comcol | .1947252 .2565154 0.76 0.448 -.3080358 .6974862
BORDER | .1244633 .1567582 0.79 0.427 -.1827771 .4317037
TC_ij | -.1083682 .1010755 -1.07 0.284 -.3064726 .0897361
TD_iin_jout | -.0093952 .1330421 -0.07 0.944 -.270153 .2513626
TD_iout_jin | .1840264 .1735568 1.06 0.289 -.1561387 .5241915
_cons | -24.5791 8.71941 -2.82 0.005 -41.66883 -7.489366
------------------------------------------------------------------------------
RESET test result:
chi2( 1) = 6.98
Prob > chi2 = 0.0082

The sample has a maximum of 77 countries for 17 years. The dependent variable is exports from i to j for a particular commodity at HS 4-digit. and the variables of interest are Trade creation (TC) and Trade Diversion (TD). Also except AIFTA (which I am interested in) I haven't controlled for other FTAs here.
Would controlling for more FTAs make a difference?

Sir, please could you suggest if these estimations are correct. If not, where am I going wrong?

Thank you.

Best regards
Richa

[Joao Santos Silva]

Dear Richa Khurana,

We know that OLS is badly biased, so there is not much point in running it; not even as a robustness check. So, the question is how to improve your model.

I often see gravity models performing poorly when the sample cover only a small number of countries; can you get more data? Also, maybe you can include variables that make sense for the particular product you are considering.

Best wishes,

Joao

[Richa Khurana]

Dear Sir,

Thank you for your advice. I can try adding a few more countries, but not many as the sample already includes India's major trading partners in this particular commodity (HS code: 2710). Adding more would mostly increase the zeroes.
I will try and include more variables.
Sir will controlling for other FTAs (besides other variables) help here?

Best regards
Richa

[Richa Khurana]

Dear Sir,

Also could it be possible that controlling only for country specific fixed effects in a panel setting causing the bias?

Best regards
Richa

[Joao Santos Silva]

Including the countries with zero trade would be my first priority; they are not relevant for OLS, but matter for Poisson.

Best wishes,

Joao

[Richa Khurana]

Sir, 90 percent zeroes in dataset would also make sense?

Best regards
Richa

[Joao Santos Silva]

Yes, that's not a problem :-)

Joao

[Richa Khurana]

Great! Thank you very much for your advice and suggestions Sir!

Best regards
Richa

[Richa Khurana]

Dear Sir,

Please help me with one more thing.
The result given above was for a commodity at HS 4-digit.
At 2-digit, out of 32 commodities, the model is correctly specified for in 8 cases, but not for the rest. It is also different at aggregate level.
Why should the results vary like this with the same specification everywhere?

At aggregate level, also controlling for other FTAs gives the following result for the reset test:

ppml trade time_fe* exp_fe* imp_fe* ln_GDP_i ln_GDP_j ln_GDPPC_i ln_GDPPC_j ln_DISTANCE border lang_off comcol tc_ij td_iin_jout td_iout_ji
> n usacan eueu euzaf eumar euisr sadcsadc indlka eujor cancri gccgcc cezcez euegy paklka turtun eudza eac saftasafta turmar indbtn egytur ag
> adiragadir aseanjpn eupng turjor turchl saftaafg chlmys canjor kortur chlvnm sadcsyc turmys euciv eugha fit2, cluster(pair_id)

note: checking the existence of the estimates
WARNING: trade has very large values, consider rescaling
WARNING: ln_GDP_i has very large values, consider rescaling or recentering
WARNING: ln_GDP_j has very large values, consider rescaling or recentering
WARNING: fit2 has very large values, consider rescaling or recentering

Number of regressors excluded to ensure that the estimates exist: 4
Excluded regressors: exp_fe_52 exp_fe_68 eac sadcsyc
Number of observations excluded: 1654

note: time_fe1 omitted because of collinearity
note: exp_fe_78 omitted because of collinearity
note: imp_fe47 omitted because of collinearity

note: starting ppml estimation
note: trade has noninteger values

Iteration 1: deviance = 1.10e+11
Iteration 2: deviance = 7.10e+10
Iteration 3: deviance = 6.16e+10
Iteration 4: deviance = 5.73e+10
Iteration 5: deviance = 5.60e+10
Iteration 6: deviance = 5.58e+10
Iteration 7: deviance = 5.57e+10
Iteration 8: deviance = 5.57e+10
Iteration 9: deviance = 5.57e+10
Iteration 10: deviance = 5.57e+10
Iteration 11: deviance = 5.57e+10
Iteration 12: deviance = 5.57e+10
Iteration 13: deviance = 5.57e+10
Iteration 14: deviance = 5.57e+10
Iteration 15: deviance = 5.57e+10
Iteration 16: deviance = 5.57e+10

Number of parameters: 219
Number of observations: 105004
Pseudo log-likelihood: -2.787e+10
R-squared: .88545331
Option strict is: off
(Std. Err. adjusted for 6320 clusters in pair_id)
------------------------------------------------------------------------------
| Robust
trade | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ln_GDP_i | 1.291668 .2434751 5.31 0.000 .8144654 1.76887
ln_GDP_j | 1.204141 .1897354 6.35 0.000 .8322661 1.576015
ln_GDPPC_i | -.2112285 .2002902 -1.05 0.292 -.6037902 .1813331
ln_GDPPC_j | -.1840733 .1263164 -1.46 0.145 -.431649 .0635024
ln_DISTANCE | -.9931771 .1534526 -6.47 0.000 -1.293939 -.6924156
border | .6589599 .1267878 5.20 0.000 .4104603 .9074595
lang_off | .5744317 .1208803 4.75 0.000 .3375106 .8113528
comcol | .1915244 .1508954 1.27 0.204 -.1042252 .4872741
tc_ij | .0000781 .1065739 0.00 0.999 -.2088028 .208959
td_iin_jout | -.4089032 .080488 -5.08 0.000 -.5666568 -.2511496
td_iout_jin | -.2255814 .0631433 -3.57 0.000 -.3493401 -.1018228
usacan | .8043289 .2457584 3.27 0.001 .3226513 1.286007
eueu | .724366 .1322358 5.48 0.000 .4651886 .9835433
euzaf | 1.226822 .2210137 5.55 0.000 .7936435 1.660001
eumar | 1.631674 .3095517 5.27 0.000 1.024963 2.238384
euisr | .368643 .2242583 1.64 0.100 -.0708951 .8081811
sadcsadc | 1.989663 .432692 4.60 0.000 1.141602 2.837724
indlka | 2.530628 .6720177 3.77 0.000 1.213497 3.847758
eujor | -.475621 .2021359 -2.35 0.019 -.8718 -.0794419
fit2 | -.0193551 .0074963 -2.58 0.010 -.0340475 -.0046626
_cons | -38.39923 6.741989 -5.70 0.000 -51.61328 -25.18517
------------------------------------------------------------------------------

. do "C:\Users\Richa\AppData\Local\Temp\STD01000000.tmp "

. test fit2=0

( 1) fit2 = 0

chi2( 1) = 6.67
Prob > chi2 = 0.0098

Can I say the model is correctly specified at 5 per cent?

Is it wrong to use same dataset at all levels?

Best regards
Richa

[Manmeet Ajmani]

Dear Prof. @Joao Santos Silva,

I have couple of questions regarding ppml:

1) Clarification: Will ppml with FE dummies (country x time) in panel data yield the same results to that of xtpqml? Following commands (as example):

ppml trade(in levels) indep exporter*time_FE importer*time_FE, cluster(countrypair)

xtpqml trade(in levels) indep, fe cluster(countrypair)

Will above two commands yield the same results? Reading some of the stata threads suggests me so.

2) Under/Over trading question: Refering to 'undertrading/overtrading' point in log of gravity website, can you please provide the reference of the paper where you have used the commands? It will help me to understand how you are measuring undertrading and overtrading. I could not follow your codes from gen yhat = exp(fitted7) command:

xtpqml rtrade_w0 cu ldist lrgdp_w0 lrgdppc_w0 comlang comborder fta landl island lareap comcol curcol evercol comctry , fe i(pairid)
predict fitted7, xb
gen yhat=exp(fitted7)
egen meany=mean(rtrade_w0) if yhat !=., by(pairid)
egen meanyhat=mean(yhat), by(pairid)
gen exp_alpha=meany/meanyhat
gen error_y=rtrade_w0 -yhat*exp_alpha

su  error_y

i.e., why you have considered expotential term (exp(fitted7)), mean of trade and yhat, and error_y = rtrade_w0 - yhat*exp_alpha?

I will appreciate your guidance.

Regards,
Manmeet Ajmani
Stata 14.2

[Joao Santos Silva]

Dear Manmeet Ajmani,

1 - It is true that the results of xtpoisson can be obtained with ppml if this includes the right set of dummies. However, I do not think the commands you present will lead to the same result because there are 2 sets of FE in ppml and only one set in xtpoisson.

2 - I never used those commands on a paper; I wrote them to help someone. After taking the exponential, we then need to estimate the fixed effects and that is done based on the fact that the fixed effect sets the mean of predicted trade in each group equal to the mean of actual trade in the group.

Best wishes,

Joao

[Manmeet Ajmani]

Dear Prof. @Joao Santos Silva,

I will be grateful if you can provide your feedback.

Objective of my research

I’m interested in finding out whether South Asian countries and Southeast Asian countries are under- or over-trading.

Description of data:
My data has 229320 observations of 105 exporter and importer countries, from year 1996 to 2016.

* Example generated by -dataex-. To install: ssc install dataex
clear
input int year str18 productdescription str26(reportername partnername) str6 tradeflowname float DefTradeMill str7 countrypair
2004 "Agricultural Trade" "Ireland"              "Sudan"              "Export"         0 "IRL SUD"
2014 "Agricultural Trade" "Saudi Arabia"         "Nepal"              "Export"         0 "SAU NPL"
2015 "Agricultural Trade" "Ghana"                "Paraguay"           "Export"         0 "GHA PRY"
2010 "Agricultural Trade" "India"                "Malaysia"           "Export"  803.9243 "IND MYS"
2013 "Agricultural Trade" "Cote d'Ivoire"        "Australia"          "Export"  5.401716 "CIV AUS"
2003 "Agricultural Trade" "Israel"               "Venezuela"          "Export" .13719939 "ISR VEN"
2011 "Agricultural Trade" "Nigeria"              "Japan"              "Export"  98.63414 "NGA JPN"
2014 "Agricultural Trade" "Colombia"             "Nigeria"            "Export"  3.769131 "COL NGA"
2009 "Agricultural Trade" "Syrian Arab Republic" "Dominican Republic" "Export"   .249697 "SYR DOM"
2014 "Agricultural Trade" "Peru"                 "Algeria"            "Export"  1.764022 "PER DZA"
end

Methodology:
Using country-pair dummies in ppml command, I’m trying to measure under-and over-exporting between countries following Mishra and Roy paper.
I executed the following ppml command:

xi, prefix(_fe) noomit i.reporteriso3*i.year i.partneriso3*i.year
drop _fepartneri_* _fereporter_* _feyear_*
 
ppml DefTradeMill `cties' contig comlang_off comcol col45 ln_distw _fe*, cluster(en_ctypair)
 
where DefTradeMill = deflated trade is million USD
`cties’ = country-pair dummy
contig = Contiguity
comlang_off = Common official language
comcol = common colonizer
col45 = pair in colonial relationship
ln_distw = log of distance
_fe* = exporter * time and importer * time fe

Results:
Number of parameters: 4288
Number of observations: 186378
Pseudo log-likelihood: -3450260.3
R-squared: .86713501
Option strict is: off
                               (Std. Err. adjusted for 10,776 clusters in en_ctypair)
-------------------------------------------------------------------------------------
                    |               Robust
       DefTradeMill |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
--------------------+----------------------------------------------------------------
             contig |   .5161429   .0922634     5.59   0.000     .3353099    .6969759
        comlang_off |   .2007062   .0845267     2.37   0.018     .0350369    .3663755
             comcol |   .5511037   .1401718     3.93   0.000      .276372    .8258355
              col45 |   .8380919   .1432694     5.85   0.000     .5572891    1.118895
           ln_distw |  -1.070027   .0409086   -26.16   0.000    -1.150206   -.9898472
            BRN_KHM |  -5.321514   .5429197    -9.80   0.000    -6.385616   -4.257411
            BRN_IDN |  -1.965092   .5016265    -3.92   0.000    -2.948262   -.9819221
            BRN_LAO |  -3.147395   .5525154    -5.70   0.000    -4.230305   -2.064485
            BRN_MYS |    1.85671   .5060934     3.67   0.000      .864785    2.848635
            BRN_MMR |  -4.965856   .5699987    -8.71   0.000    -6.083033   -3.848679
            BRN_PHL |  -2.068478   .4853534    -4.26   0.000    -3.019753   -1.117202
            BRN_SGP |   .5853504   .5322933     1.10   0.271    -.4579252    1.628626
            BRN_THA |    .064716   .4767165     0.14   0.892    -.8696311    .9990632
            BRN_VNM |  -1.186524   .4697097    -2.53   0.012    -2.107138   -.2659097
            BRN_BGD |   -6.64619   .5676142   -11.71   0.000    -7.758693   -5.533686
            BRN_IND |  -6.298606   .5313226   -11.85   0.000    -7.339979   -5.257232
            BRN_NPL |  -2.871323   .5146126    -5.58   0.000    -3.879945   -1.862701
            BRN_PAK |  -6.306463   .5457102   -11.56   0.000    -7.376035   -5.236891
            BRN_CHN |  -1.187913     .49589    -2.40   0.017     -2.15984   -.2159864
            KHM_BRN |   2.220235   .5425527     4.09   0.000     1.156851    3.283619
            KHM_IDN |  -1.619641   .3956366    -4.09   0.000    -2.395075    -.844208
            KHM_LAO |   .8653939   .4947587     1.75   0.080    -.1043153    1.835103
            KHM_MYS |   .2687746    .383145     0.70   0.483    -.4821757    1.019725
            KHM_MMR |  -2.889298    .496502    -5.82   0.000    -3.862424   -1.916172
            KHM_PHL |  -1.210273   .3673196    -3.29   0.001    -1.930206   -.4903394
            KHM_SGP |  -.5198247   .4121216    -1.26   0.207    -1.327568    .2879188
            KHM_THA |  -1.927757   .3797639    -5.08   0.000     -2.67208   -1.183433
            KHM_VNM |  -2.141725   .4079412    -5.25   0.000    -2.941275   -1.342175
              _cons |   1.486915   .7780091     1.91   0.056    -.0379548    3.011785
…….*dummies removed for brevity
-------------------------------------------------------------------------------------

After this, I ran RESET test, which is as follows:

commands:
predict XB,xb
gen XB2 = XB^2
quietly ppml DefTradeMill `cties' contig comlang_off comcol col45 ln_distw XB2 _fe*, cluster(en_ctypair) keep
 
test XB2 = 0
 
results:
WARNING: XB2 has very large values, consider rescaling  or recentering
Number of regressors excluded to ensure that the estimates exist: 448
Number of observations excluded: 0
Warning:  variance matrix is nonsymmetric or highly singular
WARNING: The model appears to overfit some observations with DefTradeMill=0
 
 
 ( 1)  XB2 = 0
       Constraint 1 dropped
 
           chi2(  0) =       .
         Prob > chi2 =         .

Question: Why the test statistics is missing? Is it due to keep option?

When I run the above command without “keep” option, I got the following result:

ppml DefTradeMill `cties' contig comlang_off comcol col45 ln_distw XB2 _fe*, cluster(en_ctypair)
 
Number of parameters: 4289
Number of observations: 186378
Pseudo log-likelihood: -3432757.5
R-squared: .86834016
Option strict is: off
                               (Std. Err. adjusted for 10,776 clusters in en_ctypair)
-------------------------------------------------------------------------------------
                    |               Robust
       DefTradeMill |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
--------------------+----------------------------------------------------------------
        contig |    .681038   .1306765     5.21   0.000     .4249168    .9371593
        comlang_off |   .2224005   .0834714     2.66   0.008     .0587995    .3860015
             comcol |   .6621034   .1453955     4.55   0.000     .3771336    .9470733
              col45 |   .9477815   .1447849     6.55   0.000     .6640083    1.231555
           ln_distw |  -1.269803   .0739845   -17.16   0.000    -1.414809   -1.124796
                XB2 |  -.0149341   .0062214    -2.40   0.016    -.0271277   -.0027405
            BRN_KHM |  -3.759391    .880763    -4.27   0.000    -5.485655   -2.033127
            BRN_IDN |  -2.145936   .5367284    -4.00   0.000    -3.197905   -1.093968
            BRN_LAO |  -1.528459   .9113547    -1.68   0.094    -3.314681    .2577635
            BRN_MYS |   1.375353    .577688     2.38   0.017     .2431054    2.507601
            BRN_MMR |  -3.570065   .8520639    -4.19   0.000    -5.240079    -1.90005
            BRN_PHL |  -2.117421   .5197358    -4.07   0.000    -3.136084   -1.098757
            BRN_SGP |   .2472873   .5764133     0.43   0.668     -.882462    1.377037
            BRN_THA |  -.1081627       .516    -0.21   0.834    -1.119504    .9031788
            BRN_VNM |  -1.440556   .5119374    -2.81   0.005    -2.443935   -.4371775
            BRN_BGD |  -5.461992    .778698    -7.01   0.000    -6.988212   -3.935772
              _cons |   .9075362   .9549157     0.95   0.342    -.9640642    2.779137
-------------------------------------------------------------------------------------
 
test XB2 = 0
 
 ( 1)  XB2 = 0
 
           chi2(  1) =    5.76
         Prob > chi2 =    0.0164

So, can I say that my model specification is reliable at 1 percent?

If yes, then can you please guide me how to interpret my result, which is as follows:

1. If BRN_SGP = 0.25, then Brunei (BRN) is over-exporting to Singapore (SGP) by (e(0.25) – 1) = 0.29 million USD. Is my interpretation correct?
2. If not, then how will you interpret it?

Looking forward to your reply.

Regards,
Manmeet Ajmani

[Joao Santos Silva]

Dear Manmeet Ajmani

Please perform the RESET as

predict XB,xb
qui su XB
gen XB2 = (XB-r(mean))^2
quietly ppml DefTradeMill `cties' contig comlang_off comcol col45 ln_distw XB2 _fe*, cluster(en_ctypair)
test XB2 = 0

Best wishes,

Joao