Estimating digital product trade through corporate revenue data

The explosive growth of digital trade presents significant challenges in measurement and understanding, particularly concerning the definition and classification of digital products versus goods and services. While organizations like the OECD, WTO, UNCTAD, and IMF define digital trade broadly, discrepancies remain, particularly around the classification of "digital products," which encompass digital goods (e.g., video games), productized services (e.g., digital advertising), and digital intermediation fees (e.g., hotel booking fees). These discrepancies are crucial as they impact policy decisions such as tariff application, significantly affecting economic statistics. This study focuses on estimating trade in these "digital products," aiming to address the gap in accurate and granular data for a better understanding of digital trade's impact on global trade balances, sustainable development, and economic complexity.

Existing methods for estimating digital trade often rely on surveys, resulting in probabilistic estimates lacking the granularity and provenance of physical trade data. These methods typically employ broad categories from balance of payment data, hindering a detailed analysis of specific digital sectors. The authors highlight the importance of studying digital product trade due to its impact on trade balances (e.g., offsetting physical goods deficits), its potential connection to the decoupling of economic growth from greenhouse gas emissions (through the "twin transition" hypothesis), and its role in influencing international estimates of economic complexity. The lack of detailed data on digital product trade impedes a complete picture of these economic phenomena.

The researchers developed a novel method to estimate bilateral digital product trade using corporate revenue data from over 2500 large online firms (with revenues exceeding USD 1 billion) and consumption data from a market intelligence company (AppMagic). The methodology involves several steps: 1. **Data Collection:** Revenue data were primarily obtained from the Orbis database, supplemented by Statista and other public sources. The researchers manually identified subsidiaries and allocated revenue across 31 digital product sectors based on Statista's classifications and firm financial statements. 2. **Consumption Data Integration:** AppMagic provided consumption data for mobile games and applications in 60 countries, serving as a ground truth for model training and extrapolation. 3. **Machine Learning:** A gradient-boosted regression tree model was employed to extrapolate consumption data to 189 countries and 29 additional digital sectors. The model incorporates features inspired by gravity models of trade, including parent company revenues, total world consumption for each sector, shared language, geographic distance, GDP, and ICT capacity. 4. **Optimal Transport:** An optimal transport procedure was used to allocate the estimated consumption to firm revenues based on geographical proximity. This procedure prioritized the allocation of revenues to domestic consumption to ensure conservative estimations. 5. **Confidence Intervals:** 95% confidence intervals were calculated using a linear regression model to provide upper and lower bounds for the trade estimates. The resulting dataset includes bilateral digital trade estimates for 15,515 firms, 189 countries, and 31 sectors for the years 2016-2021. The authors acknowledge limitations in data availability and the assumptions made in the model, particularly concerning revenue allocation and the extrapolation of consumption patterns across sectors.

The study yielded several significant findings: 1. **Scale and Growth of Digital Product Trade:** The total value of digital product trade in 2021 was estimated at almost USD 1 trillion, exceeding the GDPs of several countries. This trade grew at an annualized rate of 24.5% between 2016 and 2021, surpassing the growth rates of both services and physical goods trade. 2. **Geographical Concentration:** Digital product exports were found to be more spatially concentrated than other forms of trade, originating primarily in a few countries (including the US, Ireland, Luxembourg, and the Cayman Islands depending on revenue assignment criteria). Imports, however, were more evenly distributed, suggesting demand-driven factors play a significant role. 3. **Impact on Trade Balances:** The study demonstrates that digital product trade can significantly affect trade balances, particularly for net exporters (e.g., the US) and importers of digital products, offsetting deficits in physical goods. 4. **Decoupling and Digital Trade:** High-income economies that decoupled economic growth from emissions tended to exhibit larger digital product export sectors, suggesting a potential link between digitalization and sustainable development. However, the authors note that further research is needed to explore the causal mechanisms. 5. **Economic Complexity:** Incorporating digital product exports into economic complexity calculations (ECI and PCI) revised the complexity estimates of several economies, highlighting the significance of digital sectors in economic sophistication. Digital sectors, on average, exhibited higher complexity scores than physical goods sectors.

The findings significantly contribute to the understanding of digital product trade's multifaceted impact on the global economy. The method presented offers a novel approach to estimating bilateral trade flows with unprecedented sectoral granularity, providing insights into trade balances, the link between digitalization and sustainable development, and the dynamics of economic complexity. The observed higher concentration of digital product exports suggests the influence of knowledge agglomeration and potential supply-side constraints. The relationship between decoupling and digital exports highlights the potential role of digitalization in achieving sustainable economic growth. The inclusion of digital product trade data in economic complexity metrics provides a more comprehensive representation of an economy's sophistication.

This paper offers a novel methodology and dataset for estimating digital product trade, revealing significant insights into its scale, geography, and impact on various economic indicators. The findings highlight the importance of considering digital trade in understanding global trade balances, the relationship between economic growth and emissions, and economic complexity. Future research should focus on expanding data coverage, refining the methodology to address limitations, and exploring the causal mechanisms linking digitalization and sustainable development.

The study's limitations include the reliance on data from a limited universe of large firms, potentially overestimating growth and concentration. The extrapolation of consumption patterns from app and game data to other sectors might introduce biases. The assumptions about revenue allocation and optimal transport might not fully capture the complexities of digital trade. The relatively short time series (2016-2021) limits the analysis of long-term trends. Furthermore, while the study suggests a correlation between decoupling and digital exports, it does not establish causality and requires further investigation.