A simple solution to the distance puzzle: balanced data and Poisson estimation

Abstract

We propose the use of a balanced panel data set and Poisson regression as a solution to the distance puzzle. Employing annual data for the period 1972–2010, we confirm the existence of the puzzle by applying OLS regression to both an unbalanced panel data set and a narrowly defined balanced panel. We find that Poisson regression remedies the distance puzzle, producing a constant trend for the distance coefficient when the unbalanced panel is examined and a positive trend for the balanced data. The findings confirm the common intuition that the influence of transport costs on trade flows has decreased over time.

Keywords:

• distance puzzle
• gravity model
• imports
• OLS
• Poisson regression

JEL Classification:

• F14
• F60

Notes

¹ See Disdier and Head (2008) and Carrère et al. (2013) for more extensive analyses of the topic. The puzzle is also called the ‘missing globalization puzzle’ by Coe et al. (2007) and Siliverstovs and Schumacher (2008).

² Another vein of literature focuses on the average distance of trade and the intensive and extensive margins of trade. See, for example, Carrère and Schiff (2005) and Berthelon and Freund (2008).

³ The unbalanced panel data set includes 210 countries, while the balanced panel data set includes 65 countries. Using 1972 as the initial year in our reference period produces the largest number of observations in our balanced data set. Use of any earlier year results in fewer observations as the number of countries for which data are available decreases. Beginning the reference period more recently increases the number of country pairs; however, the number of observations in the data set decreases due to the inclusion of fewer annual observations.

⁴ The dependent variable is not log-transformed when the Poisson estimation technique is employed.

⁵ Internal distance, when k = j, is calculated as (Head and Mayer, 2000).