Effects of economic conditions and policy interventions on OECD broadband adoption

Abstract

The positive effects of broadband networks and services on productivity and economic growth are well established. Looking at broadband as an engine of economic prosperity, the OECD and its member states are seeking to foster its widespread adoption. However, which public policies best promote the adoption of broadband remains controversial. This article contributes in two ways to this discussion. It offers a comprehensive discussion of the factors that influence broadband adoption and uses an econometric approach that is well-suited to overcome the challenges of modelling broadband adoption. This framework allows drawing more robust and nuanced policy recommendations.

Keywords:

• broadband
• OECD
• unbundling
• industrial policy

JEL Classification:

• L96
• L86
• L51

Acknowledgement

Johannes Bauer gratefully acknowledges the hospitality of the Institute of Mass Communication and Media Research (IPMZ) at the University of Zurich, Switzerland, which has facilitated completing this article. The authors are responsible for all errors.

Notes

¹ For example, across digital subscriber line (DSL), cable and wireless broadband.

² Similar practice was found in Biddle (1991) who tested whether a bandwagon effect exists for the demand for personalized licence plates. In particular, the bandwagon effect (‘network effect’ in this article) is tested by including the market demand for the product in the previous year as an explanatory variable in modelling the current market demand.

³ Additional information on data sources and expected signs of the relations is provided in the following section

⁴ The difference estimator is asymptotically consistent but has low asymptotic precision and is biased in small samples (Blundell and Bond, 1998). Blundell and Bond, using Monte Carlo simulations, demonstrate the efficiency of the difference GMM estimator is increased by combining it with a levels estimator. In particular, the system GMM estimator combines into a system additional moment conditions for equations in levels with the moment conditions for equations in first-differences. Including the level equation enables use of cross-sectional variation, increasing estimate precision and enabling inclusion of time-invariant arguments. GMM models are well suited to panels with persistent series, short time dimensions and large cross-sectional dimensions (Blundell and Bond, 1998).

⁵ Data for years when countries have zero broadband penetration are omitted from estimation. Omitted data points are Czech Republic (2001, 2002, 2003), Greece (2001, 2002), Ireland (2001), Mexico (2001), Poland (2001), Slovak Republic (2001, 2002). Some of the independent variables used in our model become available with considerable delay so that the most recent year that could be used was 2009.

⁶ Because of data unavailability, 39 COMPUTER data points were extrapolated using the Exponential Smoothing option in EViews v. 4. Extrapolated data amount to 1.3% of the total database used in estimation.

⁷ Correctly specifying the number of instruments in system-GMM estimation is important for identification. Roodman’s (2007) rule of thumb is that the number of instruments should not exceed the number of cross-sectional units. The two-step estimator incorporates the one-step estimator disturbance covariance matrix (Edwards and Garland, 2009). Blundell and Bond (1998) argue the one-step estimator is robust to time and cross-sectional sources of heteroscedasticity.