Economic spillovers from cloud computing: evidence from OECD countries

ABSTRACT

The objective of this paper is to estimate the impact of cloud computing on economic performance. The analysis relies on a multi-equation model in which cloud computing is complementary with broadband to maximize the economic contribution of digital technologies. The proposed model is estimated for a sample of OECD countries, depicting cloud penetration at the sector level. Our results suggest positive effects from cloud adoption on economic outcomes. However, the impact of cloud computing depends on broadband penetration and cloud adoption, which, as expected, vary by country and industry. The sector benefitting from the largest economic impact is information and communications, followed by manufacturing, while in contrast, no impact was detected for the transportation sector. This is one of the first empirical analysis estimating cloud economic spillovers on an aggregated basis, from which important public policy implications can be derived to stimulate the development and adoption of this technology.

KEYWORDS:

• Cloud computing
• broadband
• ICT
• economic performance
• productivity

Acknowlegements

A portion of the research in this article was originally developed for a study commissioned by Amazon Web Services (AWS). The authors would like to thank Nicolas Gresser, Mouchka Heller, Yazeed AlQatarneh, Ji Hyun Lee, and two anonymous referees for providing useful comments and suggestions. All of the article’s contents are sole responsibility of the authors. Javier Pérez Menéndez (Universidad Pontificia Comillas) provided duties of research assistance.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Vu et al. (2020) provided empirical evidence of the relevance of broadband as a predictor of cloud computing adoption.

2 This analysis was conducted using the country’s input-output tables provided by Statistics Indonesia (AWS, 2021).

3 Koutroumpis (2009) also includes local loop unbundling as a metric associated to regulation, although he later discarded this variable in the latest version of his model (Koutroumpis, 2019).

4 These should not be confused with the direct effects generated by cloud revenue, nor by the direct, indirect, and induced effects derived from an initial investment in cloud (the so-called ‘construction effect’).

5 For instance, year-level fixed effects proved to be critical to absorb the COVID-related shock in our model during 2020.

6 More recently, Levinsohn and Petrin (2003) proposed instead an estimator using intermediate inputs as proxy for unobservable shocks affecting productivity. Levinsohn and Petrin (2003) argue that intermediate materials can be a better proxy for unobservable productivity shocks as they are expected to react more smoothly than investment, and because some firms may decide not to invest at all in some specific years. However, we were unable to estimate the approach recommended by Levinsohn and Petrin (2003) since, unfortunately, there is no data available on intermediate materials such as electricity or fuel consumed by industry level for the sample considered. In any case, as our sample is industry-aggregated rather than firm-specific, we believe that this theoretical advantages over the Olley and Pakes (1996) approach do not longer remain relevant.

7 We thank an anonymous referee for highlighting this.

8 The 19 countries included in the sample are detailed in Table 4. Sectors included: Construction; Information and communication; manufacturing; professional, scientific, and technical activities; real estate activities; transportation and storage; accommodation, food, and administrative services.

9 Using average exchange rates as reported by the International Monetary Fund

10 The reason to use 30 Mbps broadband is threefold. First, because the original definition for broadband based on 512 Kbps speed as reported by the ITU is completely outdated nowadays. Second, because in our sample, standard broadband penetration in fixed broadband is close to 95% of firms, thus indicating homogeneity in this kind of infrastructure. Third, because speed is a crucial variable associated to the economic impact derived from broadband.

11 When a single year of cloud penetration is missing between two reported values, we imputed the correspondent figure based on the compound average growth rate across that interval.

12 Weights for firms according to size were calculated based on the most recent OECD data, and given the lack of update in this indicator, were assumed to remain constant through the period.

13 The number of broadband lines per economic industry is calculated from the broadband penetration and the number of firms by sector within each country (source: OECD).

14 According to Oxford Economics, investment by public cloud service providers is measured through investment in the hardware infrastructure necessary to supply public cloud services (servers, storage, and ethernet switches).

15 To build the sectoral dummies, information and communication, and transportation and storage sectors are grouped into a single category named transport and communications. Similarly, accommodation, food, and administrative services are grouped in a category named other services.

16 These results are aligned with some found in previous studies that analyzed the effects of ICTs in TFP, although not focusing specifically on cloud computing. To cite an example, Ahmed (2017) found a positive contribution of ICT to TFP and TFP per unit of labor for a sample of ASEAN economies.

17 We would like to thank an anonymous referee for raising up this point.

18 Complete estimations are available for those who want to request them.

Additional information

Funding

This work was supported by Amazon Web Services.