8 minute read

By Tom Arend (tom.arend@oecd.org), Annabelle Mourougane (annabelle.mourougane@oecd.org), OECD Statistics and Data Directorate and Julia Schmidt (julia.schmidt@oecd.org), Directorate for Science, Technology and Innovation

Key findings

• The share of domestic value added in exports is estimated to have increased marginally in 2023 in most OECD and in key emerging-market economies (China, India, Brazil, Indonesia and South Africa).
• Most countries experienced a rise in the share of domestic value added in exports in manufacturing, while the picture was mixed in services.
• Domestic services shares of exports are estimated to have increased slightly in 2023 in most countries covered, especially in manufacturing. The foreign services share is estimated to have been broadly stable on average across countries.

Main trends in 2023

Getting timely information on global value chains is of key interest to policymaking, given the renewed interest in their resilience since the COVID-19 crisis (Schwellnus, Haramboure and Samek, 2023; Jaax, Miroudot and van Lieshout, 2023). This article updates nowcasts of selected TiVA indicators for 2021-23 for 41 countries (36 OECD countries and Brazil, China, India, Indonesia, South Africa) and 24 industries, building on the OECD TiVA database, most recent data on balance of payments, national accounts and short-term indicators of the business cycles. It relies on the methodology set up in Mourougane et al. (2023). The focus of this article is on the most recent year, 2023.

A fragmented regulatory environment, high – although receding – inflation, heightened uncertainty, and geo-political tensions, all contributed to subdued economic growth and international trade in 2023 (OECD, 2024a; OECD, 2024b).

The share of domestic value added in export flows is estimated to have increased marginally in 2023 on average across the 41 economies covered and across the OECD (Figure 1). This is consistent with the foreseen stabilisation in global trade reported in OECD (2024a). This was preceded by a fall in the share in 2022.

Most countries experienced a rise in the share of domestic value added in export flows in 2023, albeit to a varying degree across industries (Figure 2). The growth was particularly marked in manufacturing, where most countries experienced an increase. There was no clear pattern in services, where about half of the countries saw a small increase and half a decrease or a stabilisation.

The increase in 2023 was more pronounced in South and Latin American countries than in the other regions (Figure 3). Ireland stands out, with a marked decline in the share of domestic value added in exports in most industries (Figure 2). Norway, Luxembourg and South Africa also showed a decline in a few industries. All these countries, particularly Ireland, display above-average confidence intervals, pointing to large uncertainties around 2023 estimates (see How have the nowcasts been computed?). Domestic services share of exports is also estimated to have slightly increased in 2023 on average across the 41 countries covered, returning to its 2020 level. As for the share of domestic value added in exports, the small rise was broad-based in manufacturing but not in services (Figure 4). By contrast the share of foreign services share of exports is estimated to have been stable, on average across countries (Figure 5).

Definition of TiVA indicators
Martins Guilhoto, Webb and Yamano (2022) provide the following definition of the TiVA indicators nowcasted in this article.

Domestic value added share (or content) of gross exports by industry i to partner region p, represents the exported value added that has been generated anywhere in the domestic economy. This is an ‘intensity measure’ and reflects how much value added, generated anywhere in the domestic economy, is embodied in total gross exports by industry.

Foreign value added share (or content) of gross exports captures the value of imported intermediate goods and services that are embodied in a domestic industry’s exports. The value added can come from any foreign industry upstream in the production chain. This is an ‘intensity measure’, often referred to as ‘import content of exports’ and considered as a measure of ‘backward linkages’ in analyses of GVCs. It reflects how much value added, generated abroad, is embodied in total gross exports by industry.

Domestic services share (or content) of gross exports includes intermediate inputs coming from upstream domestic services industries and exports of final services. This indicator is often used to measure services content embodied in manufacturing exports, to capture the rising importance of services integration in manufacturing production and exports.

Figure 1. Trends in key TiVA indicators, average of 41 countries and across the OECD
Per cent

Note: Average over the domestic value-added shares of exports weighted with exports, not corrected for intra-region trade. This differs from estimates reported in the OECD TiVA database (Guilhoto, Webb and Yamano, 2022[1]). The OECD average does not include Chile and Costa Rica. The OECD aggregate in TiVA database treats the OECD as a single economy and thus value added flows between OECD countries are regarded as domestic flows, while exports are to non-OECD members only.
Source: Authors’ calculations.

Figure 2. Changes in the domestic value-added share of exports by sector
Annual change, percentage points

Note: Differences between developments at the economy-wide level on the one hand and manufacturing and services at the other hand may be explained by developments in sectors which are not shown. Manufacturing refers to ISIC Rev.4 Divisions 10 to 33, and Services to ISIC Rev.4 Divisions 45 to 98.
Source: Authors’ calculations.

Figure 3. Share of domestic value-added in exports at the economy-wide level by region
Per cent

Note: Regional averages over the corresponding country’s domestic value-added shares of exports weighted with exports, not corrected for intra-region trade. This differs from estimates reported in the OECD TiVA database (Guilhoto, Webb and Yamano, 2022[1]). Error bands correspond to the nowcasted value -/+ the corresponding Root Mean Squared Error (RMSE).
Source: Authors’ calculations.

Figure 4. Changes in the domestic services value-added share of exports by sector
Annual changes, percentage points

Note: Differences between developments at the economy-wide level and manufacturing and services may be explained by developments in sectors which are not shown. Manufacturing refers to ISIC Rev.4 Divisions 10 to 33, and Services to ISIC Rev.4 Divisions 45 to 98.
Source: Authors’ calculations.

Figure 5. Changes in the foreign services value-added share of exports by sector
Annual change, percentage points

Differences between developments at the economy-wide level and manufacturing and services may be explained by developments in sectors which are not shown. Manufacturing refers to ISIC Rev.4 Divisions 10 to 33, and Services to ISIC Rev.4 Divisions 45 to 98.
Source: Authors’ calculations.

How have the nowcasts been computed?
The quality of the nowcasts is best understood by considering the empirical strategy deployed (Figure 6). A range of standard econometric models were combined with machine learning across the 36 OECD countries (all but Chile and Costa Rica) and China, Brazil, India, Indonesia and South Africa to nowcast several TiVA indicators (See Definitions on p. 2). The methodology follows Mourougane et al. (2023) although for the purpose at hand, models were entirely re-estimated and extended to nowcast three-year ahead.
A panel data approach was used to increase the robustness of results (Woloszko, 2020 Fosten and Nandi, 2023). In addition, models are estimated in quasi real time taking publication lags into account.

Figure 6. Main steps of the empirical strategy

Source: OECD illustration.

The first step in the nowcasting approach is to collect and process the input data that are used to predict the target variable. Predictors include national accounts data, labour market indicators, trade and business statistics as well as measurements of geopolitical risks and uncertainty (see Table 1 for a selection). The data are transformed in case they were not stationary. Indicators whose predictive accuracy is expected to be large, but which are not sufficiently timely, have been prolonged with “bridges” using the same methods and model selection criteria as for nowcasting selected TiVA indicators.

Table 1: Overview of explanatory variables

Note: All the data are official estimates, except the global economic policy uncertainty index (Policyuncertainty.com).
Source: OECD compilation.

The second step is to select the best model to nowcast the target variable. The methods include penalised regressions (Lasso, Ridge) as well as tree-based approaches such gradient boosted trees. In addition, a consensus model, which takes the average of all machine learning models, and an autoregressive model of order 1, the standard benchmark model in the nowcasting literature, is calculated. A cross-validation process, common in machine learning approaches is implemented to prevent that the model performs well in-sample but fails in out-of-sample predictions (Hastie, Tibshirani and Friedman, 2009). The best models are selected based on the root mean squared errors (RMSEs) for one-year ahead predictions. Those models continue to perform well in nowcasting three-year ahead.

The last step is to use the best models to nowcast trade in value added for 2023. Note that for each country-sector instance, we select the model that performs best, i.e., one best model per country and sector, rather than using the model that would perform best on average across all the countries.

Model performance appears to be unequal across countries (Figure 7). Large errors translate into more uncertain nowcasts. On average across countries and sectors, errors are found to be lower than in Mourougane et al. (2023) (1.25 percentage points as compared to 1.8 percentage points in Mourougane et al. (2023)). Ireland and Luxembourg, and to a lesser extent Lithuania, Latvia, Island, Greece and Slovakia, display above average RMSE. Confidence bands are computed using the in-sample on-year ahead RMSE.

Figure 7. Absolute RMSEs across countries
Average RMSE, percentage points

Note: Country-average RMSEs were estimated for across all years of the in-sample training period. The horizontal dashed line corresponds to the overall average absolute RMSE (1.25 percentage points). Countries highlighted in dark blue perform worse than the overall average RMSE.
Source: Authors’ calculations.

Figure 8. Performance across sectors in the United States
Average RMSE, percentage points

Note: The United States is presented as RMSEs were estimated for across all years of the in-sample training period.
Source: Authors’ calculations.

References

Fosten, J., and S. Nandi (2023), “Nowcasting from cross-sectionally dependent panels”, Journal of Applied Econometrics, https://doi.org/10.1002/jae.2980.
Hastie, T., R. Tibshirani and J. Friedman (2009), The Elements of Statistical Learning, Data Mining, Inference, and Prediction, Second Edition.
Jaax, A., S. Miroudot and E. van Lieshout (2023), “Deglobalisation? The Reorganisation of Global Value Chains in a Changing World”, OECD Trade Policy Papers, No. 272, OECD Publishing, Paris, https://doi.org/10.1787/b15b74fe-en.
Martins Guilhoto, J., C. Webb and N. Yamano (2022), “Guide to OECD TiVA Indicators, 2021 edition”, OECD Science, Technology and Industry Working Papers, No. 2022/02, OECD Publishing, Paris, https://doi.org/10.1787/58aa22b1-en.
Mourougane, A., P. Knutsson, R. Pazos, J. Schmidt and F. Palermo (2023), “Nowcasting trade in value added indicators”, OECD Statistics Working Papers, No. 2023/03, OECD Publishing, Paris, https://doi.org/10.1787/00f8aff7-en.
Schwellnus, C., A. Haramboure and L. Samek (2023), “Policies to Strengthen the Resilience of Global Value Chains: Empirical Evidence from the COVID-19 Shock”, OECD Science, Technology and Industry Policy Papers, No. 141, OECD Publishing, Paris, https://doi.org/10.1787/fd82abd4-en.
OECD (2024a), Interim Economic Outlook, https://www.oecd.org/economic-outlook/february-2024/.
OECD (2024b), OECD Services Trade Restrictiveness Index, policy trends up to 2024, https://www.oecd.org/publications/oecd-services-trade-restrictiveness-index-b9e5c870-en.htm.
Woloszko, N. (2020), “Tracking activity in real time with Google Trends”, OECD Economics Department Working Papers, No. 1634, OECD Publishing, Paris, https://doi.org/10.1787/6b9c7518-en