Does time difference between countries reduce bilateral trade? An application of the correlated random effects method using panel data

ABSTRACT

This article implements the correlated random effects (CRE) panel data technique in a gravity framework to analyse the effect of time difference between countries on bilateral trade. One major advantage of the CRE approach over the fixed-effects approach is that it is able to estimate the effect of variables that remain unchanged within panel clusters (e.g. time difference between countries), while these variables get dropped from regressions that use fixed-effects methods. Regression results based on the CRE Poisson pseudo-maximum likelihood estimator indicate statistically significant negative effect of time difference between countries on bilateral trade. An additional hour of time difference between countries is found to reduce bilateral merchandise exports by approximately 8%, even after controlling for the effect of distance in the regressions.

KEYWORDS:

• Gravity model
• panel data
• correlated random effects model
• fixed-effects model
• Poisson model
• time difference

JEL CLASSIFICATION:

• F10
• F14
• C23
• F10

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ For cross-country time-series data on bilateral trade, when a country-pair is defined as the panel unit, time difference between countries will remain unchanged within panel over time, and will vary between panels only.

² Time-varying exporter- and importer-fixed-effects Poisson regression for the entire time period does not converge.

³ Poisson FE estimator is implemented by the Stata module -XTPQML- developed by Simcoe (2008) and the Poisson CRE estimator is implemented by -XTPOISSON-, both of which are maximum likelihood estimators. The Poisson pseudo-maximum likelihood (PPML) estimator is implemented by the Stata module -PPML- developed by Santos Silva and Tenreyro (2011). This estimator addresses potential convergence problems in the Poisson regression by identifying and dropping variables that cause nonconvergence or result in spurious estimates.