Abstract
In recent years, different alternatives have been suggested to specify and estimate gravity models for bilateral trade. Presently, the so-called Poisson pseudomaximum likelihood (PPML) with log-linear index is probably the most commonly used method. A method is proposed for panel data that targets to reconcile the pros and cons of fixed and random effects models, respectively. It applies equally to two- and three-way panel models and those with country-specific time-varying effects. It allows to filter out potential correlation between observed and unobserved heterogeneity and to identify the effects of time-invariant factors. It can also be used when panels are short in time, and to other specifications than the PPML-like gamma PML, zero-inflated, or Tobit-like models. We introduce and illustrate the proposed estimator with a study of bilateral trade flows across the European Union before the recent economic crisis.
Similar content being viewed by others
Notes
Note that like for the log-linear OLS, the consistency of PPML depends on the assumptions made about the error structure.
Except if the time-invariant covariate exhibits variation in a different dimension than the fixed effects, for example when we have covariates \(X_{ij}\) and FEs \((\eta _{i},\eta _{j})\).
Exceptions are, for example, Logit or Poisson-distributed responses where conditioning on their mean eliminates the country-pair fixed effects.
They actually do it in the context of small-area statistics.
When including temporal means of covariates, one has to be careful for the coefficients’ comparison: for example, including \(\beta \ln x_{ijt} \) and \(\xi \cdot \frac{1}{n}_t \sum _t \ln x_{ijt}\), the total impact of \(\ln x_{ijt}\) is \(\beta +\frac{1}{n_t}\xi \). One can therefore find both, studies looking only at \(\hat{\beta }\) and those looking at \(\hat{\beta }+\frac{1}{n_t}\hat{\xi }\).
The notation is not unique in the literature, while many use it as a synonym for penalized splines in general, others refer exclusively to its implementation with B-spline basis.
Santos Silva and Tenreyro (2006) use the notation of pseudo referring to McCullagh and Nelder (1989) who speak of quasi- likelihood. Nelder (2000) differentiates between them. Along his classification, one might be more interested in the quasi-maximum likelihood methods, but in abuse of notation, we keep the abbreviation PPML.
See Racine and Parmeter (2014) for a method of model selection based on in-sample prediction that accommodates non- and semiparametric estimations and panel data.
This selection is arbitrary and only to simplify the simulation study; it is neither a selection based on the findings from the previous sections nor a recommendation for empirical studies.
Note that we conducted many more simulation studies, available from the authors, which were even more in favor of the SEM approach.
We thank Enrique Martínez-Galán and Eliano Marques for their help in obtaining these data.
References
Anderson JE (1979) A theoretical foundation for the gravity equation. Am Econ Rev 69:106–116
Anderson JE, Van Wincoop E (2003) Gravity with gravitas: a solution to the border puzzle. Am Econ Rev 93:170–192
Anderson JE, Yotov YV (2009) The changing incidence of geography. Am Econ Rev 100:2157–2186
Baltagi HB, Egger P, Pfaffermayr M (2003) A generalized design for bilateral trade flow models. Econ Lett 80:391–397
Bergstrand JH (1985) The gravity equation in international trade: some micro-economic foundations and empirical evidence. Rev Econ Stat 67:474–481
Blackburn M, Neumark D (1992) Unobserved ability, efficiency wages, and interindustry wage differentials. Q J Econ 107:1421–1435
Breslow NE, Clayton DG (1993) Approximate inference in generalized linear mixed models. J Am Stat Assoc 88:9–25
Chamberlain G (1982) Multivariate regression models for panel data. J Econ 18:5–46
Deardorff AV (1998) Determinants of bilateral trade: does gravity work in a neoclassical world? In: Frankel JA (ed) The regionalization of the world economy. University of Chicago Press, Chicago, pp 7–22
Egger P, Pfaffermayr M (2003) The proper panel econometric specification of the gravity equation: a three-way model with bilateral interaction effects. Empir Econ 28:571–580
Gomez-Herrera E (2013) Comparing alternative methods to estimate gravity models of bilateral trade. Empir Econ 44:1087–1111
Green PJ, Silverman BW (1994) Nonparametric regression and generalized linear models. Chapman & Hall, London
Gurmu S, Rilstoneb P, Sternc S (1999) Semiparametric estimation of count regression models. J Econ 88:123–150
Henderson DJ, Millimet DL (2008) Is gravity linear? J Appl Econ 23:137–172
Kauermann G (2005) A note on smoothing parameter selection for penalized spline smoothing. J Stat Plann Inference 127:53–69
Köhler M, Schindler S, Sperlich S (2014) A review and comparison of bandwidth selection methods for Kernel regression. Int Stat Rev 82:243–274
Krivobokova T, Kneib T, Claeskens G (2010) Simultaneous confidence bands for penalized spline estimators. J Am Stat Assoc 105:852–863
Lin X, Carroll RJ (2006) Semiparametric estimation in general repeated measures problems. J Royal Stat Soc B 68:69–88
Lin X, Zhang D (1999) Inference in generalized additive mixed models by using smoothing splines. J R Stat Soc B 55:381–400
Lombardía MJ, Sperlich S (2008) Semiparametric inference in generalized mixed effects models. J R Stat Soc B 70:913–930
Lombardía MJ, Sperlich S (2012) A new class of semi-mixed models and its application in small area estimation. Comput Stat Data Anal 56:2903–2917
Martinez-Zarzoso I (2013) The log of gravity revisited. Appl Econ 45:311–327
Mátyás L (1997) Proper econometric specification of the gravity model. World Econ 20:363–368
McCullagh P, Nelder JA (1989) Generalized linear models, 2nd edn. Chapman and Hall, London
Mundlak Y (1978) On the pooling of time series and cross section data. Econometrica 46:69–85
Nelder JA (2000) Quasi-likelihood and pseudo-likelihood are not the same thing. J Appl Stat 27:1007–1011
Nie Z, Racine J (2012) The crs package: nonparametric regression splines for continuous and categorical predictors. R J 4:48–56
Niedercorn JH, Bechdolt BV (1969) An economic derivation of the ‘gravity law’ of spatial interaction. J Reg Sci 9:273–282
Nuroğlu E, Kunst RM (2013) Competing specifications of the gravity equation: a three-way model, bilateral interaction effects, or a dynamic gravity model with time-varying country effects? Empir Econ. doi:10.1007/s00181-013-0696-3
Proença I, Fontoura MP, Martínez-Galán E (2008) Trade in the enlarged European Union: a new approach on trade potential. Port Econ J 7:205–224
Racine JS, Parmeter C (2014) Data-driven model evaluation: a test for revealed performance. In: Ullah A, Racine JS, Su L (eds) Handbook of applied nonparametric and semiparametric econometrics and statistics. Oxford University Press, Oxford, pp 308–345
Santos Silva J, Tenreyro S (2006) The log of gravity. Rev Econ Stat 88:641–658
Serlenga L, Shin Y (2007) Gravity models of intra-EU trade: application of the CCEP-HT estimation in heterogeneous panels with unobserved common time-specific factors. J Appl Econ 22:361–381
Sperlich S, Lombardia MJ (2010) Local polynomial inference for small area statistics: estimation, validation and prediction. J Nonparametr Stat 22:633–648
Sperlich S, Tjøstheim D, Yang L (2002) Nonparametric estimation and testing of interaction in additive models. Econ Theory 18:197–251
Venables WN, Ripley BD (2002) Modern applied statistics with S, 4th edn. Springer, Berlin
Wellner JA, Zhang Y (2007) Two likelihood-based semiparametric estimation methods for panel count data with covariates. Ann Stat 35:2106–2142
Westerlund J, Wilhelmsson F (2011) Estimating the gravity model without gravity using panel data. Appl Econ 43:641–649
Wood SN (2006) Generalized additive models: an introduction with R. Chapman and Hall/CRC Press, London
Wood SN (2008) Fast stable direct fitting and smoothness selection for generalized additive models. J R Stat Soc B 70:495–518
Wood SN (2011) Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models. J R Stat Soc B 73:3–36
Wooldridge JM (2005) Fixed-effects and related estimators for correlated random-coefficient and treatment-effect panel data models. Rev Econ Stat 87:385–390
Zheng X (2008) Semiparametric Bayesian estimation of mixed count regression models. Econ Lett 100:435–438
Acknowledgments
Isabel Proença is grateful for the financial support received from FCT (Fundação para a Ciência e Tecnologia) through the PEst-OE/EGE/UI0491/2011 program and Stefan Sperlich for funding from the Swiss National Science Foundation 100018-140295. We thank helpful discussions with Jaya Krishnakumar, Inmaculada Martínez-Zarzoso, Marcelo Olarreaga, and Walter Zucchini, of participants of ESEM/EEA-2012, the seminars at ISEG and GSEM, an anonymous referee, and in particular the AE Christopher Parmeter who essentially contributed to the stepwise improvement of this article.
Author information
Authors and Affiliations
Corresponding author
Appendix: Addition information on used data
Appendix: Addition information on used data
Countries included in the data set: Austria (AU), Belgium (BE), Bulgaria (BU), the Czech Republic (CZ), Denmark (DK), Estonia (EE), Finland (FI), France (FR), Germany (DE), Greece (GR), Hungary (HU), Ireland (IR), Italy (IT), Latvia (LV), Lithuania (LH), Luxembourg (LU), the Netherlands (NE), Poland (PL), Portugal (PT), Romania (RO), Slovakia (SK), Slovenia (SV), Spain (SP), Sweden (SW), and the United Kingdom (UK).
Details about the used variables:Footnote 13
-
T: nominal import (cif) flows in \(10^{3}\) euros.
-
MGDP/XGDP: importer/exporter country’s nominal GDP at market prices in millions of euro, from Eurostat’s New Cronos Database.
-
MPOP/XPOP: importer/exporter country’s population, expressed in thousands of people at the end of the period, from Eurostat’s New Cronos Database.
-
DIST: absolute distance expressed in kilometers, the geodesic distance between capitals (in the case of the Netherlands, Amsterdam substitutes Den Haag), measured as the surface distance between two points of latitude and longitude (great circle distance) obtained from www.wcrl.ars.usda.gov/cec/java/lat-long.htm.
-
NEIGH: neighboring dummy variable is equal to one if two trading partners share a land or sea border, zero otherwise. From CIA’s World Factbook 2003 as published on www.cia.gov/cia/publications/factbook/index.html.
-
EU15: dummy equal to 1 if the exporting country belongs to the European 15.
-
MLOCK/XLOCK: landlockedness dummy for importer/exporter country; equals one if country has no direct connection to sea.
-
COMLANG: common language dummy variable is equal to one if two trading partners share the same official language, zero otherwise. From CIA’s Factbook 2003 on www.cia.gov/cia/publications/factbook/index.html.
-
ETHNIC: ethnic dummy, equal to one if there is an ethnic minority of the exporter country in the importer country that represents more than 5 % of total population of the latter. From CIA’s The World Factbook 2003.
-
MAREA/XAREA: importer/exporter country’s area.
-
PAT: the number of patents of importer country recorded as EPO (European patent office) patent applications (Direct EPO filings + EURO-PCT in regional phase); source OECD (Table 9).
Rights and permissions
About this article
Cite this article
Proença, I., Sperlich, S. & Savaşcı, D. Semi-mixed effects gravity models for bilateral trade. Empir Econ 48, 361–387 (2015). https://doi.org/10.1007/s00181-014-0891-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00181-014-0891-x