Dear Stata users, Joao Santos Silva, and fellow gravity modelers,
I have encountered some problems in my empirical research related to carbon leakage and trade.
Context
My dataset comprises bilateral trade data at the sectoral level, with the primary dependent variable being emissions embodied in imports. Specifically, I estimate the following equation:
2. Symmetrical Fixed Effects: As an alternative, I have estimated the model with symmetrical importer-time and exporter-time fixed effects (i.e., γit is constrained to be equal to δjt). In this setup, the interaction term is identifiable. However, I am wondering to which extent these symmetrical fixed effects account for fluctuations in the economy of the importing country. What are your thoughts on this?
3. Trade Cost Variable: Unrelated to the above issues, I'd like to check whether it is correct to include the time-varying trade cost variable Dij in logs, since it is specified in ad-volorem form?
Snippet of my dataset:
Thank you all in advance for helping me out on this.
Best wishes,
Yannick
Sources:
Heid, B., Larch, M., & Yotov, Y. V. (2021). Estimating the effects of non‐discriminatory trade policies within structural gravity models. Canadian Journal of Economics/Revue canadienne d'économique, 54(1), 376-409.
I have encountered some problems in my empirical research related to carbon leakage and trade.
Context
My dataset comprises bilateral trade data at the sectoral level, with the primary dependent variable being emissions embodied in imports. Specifically, I estimate the following equation:
Cijs,t = exp[α2 d(EPS) × INTLij + γit + δjt + vst + μijs + μbt + α1 log(1+ Dijt)] + ϵijs,t (1)
- Cijs,t denotes the CO2 content of imports of country i from country j in sector s, including both international and intra-national (i.e. domestic) trade (in CO2).
- d(EPS) × INTLij is the interaction between the difference in environmental policy stringency (d(EPS) = EPSit - EPSjt) and a dummy variable INTLij that equals 1 for international trade. I expect α2 to be positive.
- γit and δjt represent exporter-time and importer-time fixed effects, respectively. vst captures sector-time fixed effects. μijs denotes country-pair-sector fixed effects. μbt accounts for border-year fixed effects, which control for changes in international trade costs relative to intra-national trade costs.
- Dij is a time-varying bilateral trade cost variable from the ESCAP-WB Trade Cost Database, expressed in ad valorem equivalent form.
- Eq. 1 is estimated using the PPML estimator, i.e. the ppmlhdfe command in Stata.
- Identification Issue: According to Heid et al. (2021), including intra-national trade allows for the identification of any non-discriminatory trade policies as well as non-trade policies even when exporter- and importer-time fixed effects are included. Despite this, STATA drops the interaction term d(EPS) × INTLij when estimating eq. 1. Could anyone explain why this might be happening and suggest potential solutions? See the construction of the interaction below, as well as the regression output.
Code:
gen intl = 1 if importer!=exporter (13,566 missing values generated) . replace intl=0 if importer==exporter (13,566 real changes made) . gen intl_dif_ij_EPS = intl * (EPS21_i - EPS21_j) (266,266 missing values generated) ppmlhdfe co2_content intl_dif_ij_EPS log_trade_cost, absorb(j#i#s s#year i#year j#year inter#year) cluster(i#j#s) (dropped 1796 observations that are either singletons or separated by a fixed effect) warning: dependent variable takes very low values after standardizing (7.9168e-12) note: 1 variable omitted because of collinearity: inter_dif_ij HDFE PPML regression No. of obs = 366,454 Absorbing 5 HDFE groups Residual df = 21,966 Statistics robust to heteroskedasticity Wald chi2(1) = 86.73 Deviance = 1.34680e+10 Prob > chi2 = 0.0000 Log pseudolikelihood = -6735979874 Pseudo R2 = 0.9965 Number of clusters (i#j#s) = 21,967 (Std. err. adjusted for 21,967 clusters in i#j#s) -------------------------------------------------------------------------------- | Robust co2_content | Coefficient std. err. z P>|z| [95% conf. interval] ---------------+---------------------------------------------------------------- intl_dif_ij_EPS | 0 (omitted) log_trade_cost | -.6371896 .0684216 -9.31 0.000 -.7712935 -.5030856 _cons | 19.199 .0476588 402.84 0.000 19.10559 19.2924 --------------------------------------------------------------------------------
3. Trade Cost Variable: Unrelated to the above issues, I'd like to check whether it is correct to include the time-varying trade cost variable Dij in logs, since it is specified in ad-volorem form?
Snippet of my dataset:
Code:
* Example generated by -dataex-. For more info, type help dataex clear input str9(importer exporter) int year str6 IND float(co2_content intl_dif_ij_EPS log_trade_cost) "ARG" "ARG" 2000 "D10T12" 6844510 0 0 "ARG" "ARG" 2001 "D10T12" 6636712 0 0 "ARG" "ARG" 2002 "D10T12" 7670751 0 0 "ARG" "ARG" 2003 "D10T12" . 0 0 "ARG" "ARG" 2004 "D10T12" 14759698 0 0 "ARG" "ARG" 2005 "D10T12" 14213412 0 0 "ARG" "ARG" 2006 "D10T12" 14847242 0 0 "ARG" "ARG" 2007 "D10T12" 14501948 0 0 "ARG" "ARG" 2008 "D10T12" 15278275 0 0 "ARG" "ARG" 2009 "D10T12" 16977000 0 0 "ARG" "ARG" 2010 "D10T12" 16744057 0 0 "ARG" "ARG" 2011 "D10T12" 18253434 0 0 "ARG" "ARG" 2012 "D10T12" 19692390 0 0 "ARG" "ARG" 2013 "D10T12" 20047822 0 0 "ARG" "ARG" 2014 "D10T12" 19017780 0 0 "ARG" "ARG" 2015 "D10T12" 22300582 0 0 "ARG" "ARG" 2016 "D10T12" 20025090 0 0 "ARG" "ARG" 2017 "D10T12" 20574622 0 0 "ARG" "ARG" 2018 "D10T12" 18297452 0 0 "ARG" "ARG" 2000 "D13T15" 2203199.8 0 0 "ARG" "ARG" 2001 "D13T15" 2000815.5 0 0 "ARG" "ARG" 2002 "D13T15" 1695733.8 0 0 "ARG" "ARG" 2003 "D13T15" . 0 0 "ARG" "ARG" 2004 "D13T15" 5339110 0 0 "ARG" "ARG" 2005 "D13T15" 5409009 0 0 "ARG" "ARG" 2006 "D13T15" 5589320 0 0 "ARG" "ARG" 2007 "D13T15" 5600412 0 0 "ARG" "ARG" 2008 "D13T15" 5364749 0 0 "ARG" "ARG" 2009 "D13T15" 5094218 0 0 "ARG" "ARG" 2010 "D13T15" 5585709 0 0 "ARG" "ARG" 2011 "D13T15" 6007067 0 0 "ARG" "ARG" 2012 "D13T15" 5912168 0 0 "ARG" "ARG" 2013 "D13T15" 5945553 0 0 "ARG" "ARG" 2014 "D13T15" 5762494 0 0 "ARG" "ARG" 2015 "D13T15" 5971323 0 0 "ARG" "ARG" 2016 "D13T15" 5253716 0 0 "ARG" "ARG" 2017 "D13T15" 4744749 0 0 "ARG" "ARG" 2018 "D13T15" 4328012 0 0 "ARG" "ARG" 2000 "D16" 154349.52 0 0 "ARG" "ARG" 2001 "D16" 122990.27 0 0 "ARG" "ARG" 2002 "D16" 127674.02 0 0 "ARG" "ARG" 2003 "D16" . 0 0 "ARG" "ARG" 2004 "D16" 923668.8 0 0 "ARG" "ARG" 2005 "D16" 807155.3 0 0 "ARG" "ARG" 2006 "D16" 885259.3 0 0 "ARG" "ARG" 2007 "D16" 910833.1 0 0 "ARG" "ARG" 2008 "D16" 991047 0 0 "ARG" "ARG" 2009 "D16" 820032.3 0 0 "ARG" "ARG" 2010 "D16" 855373.7 0 0 "ARG" "ARG" 2011 "D16" 927323.4 0 0 "ARG" "ARG" 2012 "D16" 917962 0 0 "ARG" "ARG" 2013 "D16" 927639.4 0 0 "ARG" "ARG" 2014 "D16" 819333.5 0 0 "ARG" "ARG" 2015 "D16" 862390.1 0 0 "ARG" "ARG" 2016 "D16" 737336.1 0 0 "ARG" "ARG" 2017 "D16" 704515.9 0 0 "ARG" "ARG" 2018 "D16" 630378.1 0 0 "ARG" "ARG" 2000 "D17T18" 1695272 0 0 "ARG" "ARG" 2001 "D17T18" 1516552 0 0 "ARG" "ARG" 2002 "D17T18" 1433771 0 0 "ARG" "ARG" 2003 "D17T18" . 0 0 "ARG" "ARG" 2004 "D17T18" 2964082 0 0 "ARG" "ARG" 2005 "D17T18" 2804171 0 0 "ARG" "ARG" 2006 "D17T18" 3010510 0 0 "ARG" "ARG" 2007 "D17T18" 3075446 0 0 "ARG" "ARG" 2008 "D17T18" 3185869 0 0 "ARG" "ARG" 2009 "D17T18" 2949060 0 0 "ARG" "ARG" 2010 "D17T18" 3004854 0 0 "ARG" "ARG" 2011 "D17T18" 3290745 0 0 "ARG" "ARG" 2012 "D17T18" 3235521.5 0 0 "ARG" "ARG" 2013 "D17T18" 3329973.5 0 0 "ARG" "ARG" 2014 "D17T18" 3192178 0 0 "ARG" "ARG" 2015 "D17T18" 3385247 0 0 "ARG" "ARG" 2016 "D17T18" 3151508 0 0 "ARG" "ARG" 2017 "D17T18" 3081584 0 0 "ARG" "ARG" 2018 "D17T18" 2902027 0 0 "ARG" "ARG" 2000 "D19" 5685618 0 0 "ARG" "ARG" 2001 "D19" 5809087 0 0 "ARG" "ARG" 2002 "D19" 8098276 0 0 "ARG" "ARG" 2003 "D19" . 0 0 "ARG" "ARG" 2004 "D19" 8438462 0 0 "ARG" "ARG" 2005 "D19" 6761695 0 0 "ARG" "ARG" 2006 "D19" 5338450 0 0 "ARG" "ARG" 2007 "D19" 5825263 0 0 "ARG" "ARG" 2008 "D19" 7250279 0 0 "ARG" "ARG" 2009 "D19" 9712234 0 0 "ARG" "ARG" 2010 "D19" 10521705 0 0 "ARG" "ARG" 2011 "D19" 11448484 0 0 "ARG" "ARG" 2012 "D19" 13411457 0 0 "ARG" "ARG" 2013 "D19" 13968090 0 0 "ARG" "ARG" 2014 "D19" 14092115 0 0 "ARG" "ARG" 2015 "D19" 14753895 0 0 "ARG" "ARG" 2016 "D19" 14871400 0 0 "ARG" "ARG" 2017 "D19" 14944739 0 0 "ARG" "ARG" 2018 "D19" 14026114 0 0 "ARG" "ARG" 2000 "D20T21" 3231112 0 0 "ARG" "ARG" 2001 "D20T21" 3504779.5 0 0 "ARG" "ARG" 2002 "D20T21" 3108923.5 0 0 "ARG" "ARG" 2003 "D20T21" . 0 0 "ARG" "ARG" 2004 "D20T21" 6660840 0 0 end
Best wishes,
Yannick
Sources:
Heid, B., Larch, M., & Yotov, Y. V. (2021). Estimating the effects of non‐discriminatory trade policies within structural gravity models. Canadian Journal of Economics/Revue canadienne d'économique, 54(1), 376-409.
Comment