On the endogeneity of telecommunications and economic growth: evidence from Asia

Abstract

The impact of mobile and fixed telephones on economic growth has been the subject of increasing scrutiny in the literature on economic development. It is even of interest to theoretical macroeconomists, as it provides a useful test of the positive network externalities that should be present if endogenous growth theory is correct. We study the relationship between teledensity and growth in Asia, as the countries there have experienced wildly different levels of telephone penetration per capita, and of rates of growth of GDP per capita. We estimate several econometric models, one which explicitly treats telecom as strictly exogenous, and others which treat it as endogenous. Our conclusions are robust to the econometric specification. We find that the impact of teledensity on growth is positive, and increases with the level of telephone penetration. This provides support for endogenous growth theory.

Keywords:

• telecommunications
• development
• economic growth

Jel Codes:

• O1
• O3
• O4
• E1
• E2

Acknowledgments

The authors gratefully acknowledge the useful advice of Professors Jonathan Hamilton, David Roodman, and David Drukker. Valuable research assistance was provided by Emily Hildore, Elia El-Hajj, and Fernanda Cuadra.

Notes

Mina Baliamoune-Lutz is the accepting Associate Editor for this article.

See, for example, Barro (1991) and Mankiw, Romer, and Weil (1992).

For example, Knack and Keefer (1995), Keefer and Knack (1997), Knack (1996).

Examples include Gwartney, Holcombe, and Lawson (2004) and (2006).

The convergence debate is typified by the exchange in Mankiw, Phelps, and Romer (1995).

See Cameron and Trivedi (2009) and Roodman (2009) for an introduction to the Arellano–Bond estimator, as well as its implementation in Stata.

See Cameron and Trivedi (2009).

The discussion on the validity of lagged differences as an instrument is heavily drawn from Roodman (2009).

The null hypothesis of the Arellano–Bond test for AR(1) is that there is no autocorrelation in residuals in first differences.

See Baum (2006).