An analysis of market access

Abstract

Market access matters. This article creates, for the first time, a quantitative measure of market access, the Market Access Index, which enables the creation of a rank order, or league table, of market access for a large sample of world economies. The article uses Structural Equation Modelling and a data set from the World Economic Forum and provides a broader framework for the analysis of market access arguing that market access is driven by the regulatory environment, public institutions and network industries. The article finds that network industries are the most important contributors to market access, followed by public institutions. The regulatory environment, covering trade policy is the least important contributor. These findings have important implications for the role of trade policy in influencing greater market access.

Acknowledgement

We thank the WEF for providing these data in readily useable electronic form.

Notes

¹Initial tests included seven variables representing the PISI, five representing the RESI and five representing the NISI. The variables were reduced to four in each sub-index after a battery of correlation tests revealed perfect correlation across some of the variables. This step is crucial as an SEM model cannot be estimated when multicollinearity exists.

²Because of missing data for East Timor, Egypt, Kazakhstan and Tajikistan, the sample size is reduced to 113 countries.

³Because of the scaling used by the WEF, no observations can be sufficiently large or influential to result in biased or inefficient estimates. This claim is supported by the robustness tests undertaken by Squalli and Wilson (2008).

⁴This represents a model where all the directional linear relationships between the latent variables are removed and replaced with co-varying relationships between the latent sub-indexes.

⁵Variables are well behaved when normally distributed, when no data are missing or in the absence of outliers.

⁶A saturated model generally has an equal number of parameters to nonredundant elements.

⁷See Raykov and Marcoulides (2000) for details.

⁸The independence model represents a model in which all the correlations/covariances are equal to zero.

⁹No modification indices or standardized residual covariances were significant.