The main data sources for the current implementation are the United Nations Analysis of Main Aggregates (UN AMA) database, and the OECD BIMTS and BATIS databases. The UN AMA data provide timely national accounts information for virtually all economies covered in the ICIO, including value added by broad industry group, total imports and exports, and the components of final demand. OECD BIMTS and BATIS provide data on bilateral trade in goods and services, respectively.
Because of structural and methodological differences between datasets, the data obtained from these sources cannot be treated simply as drop-in replacements for the corresponding out-of-date ICIO components. The UN AMA data, for example, are available at a coarser industry aggregation than the ICIO. Conversely, BIMTS and BATIS use different accounting principles from the ICIO to measure international trade, including with respect to the role of change of ownership versus border crossing, the treatment of trade and transport margins, and related issues.
To address these differences, the timely exogenous data are used as proxy variables in bridge regressions. Series by series, an econometric relationship is estimated between an ICIO variable of interest and the corresponding proxy over the period 1995‑2022. The estimated model is then used, together with the observed proxy, to predict the ICIO component for 2023 and 2024. In the case of value added, for example, each ICIO country-industry series is modelled on the basis of the corresponding UN AMA aggregate. Gross output is projected using the same approach, subject to constraints that prevent implausible shifts in the ratio of value added to gross output. International trade flows are also modelled using bridge regressions, with bilateral trade in goods and services in the ICIO proxied by the corresponding series from BIMTS and BATIS, respectively. In all cases, the disturbance term is modelled as an ARIMA process, with orders selected automatically through information-criterion minimisation. By contrast, aggregate final-use series are extended forward using growth rates calculated from UN AMA, as preliminary testing suggested little gain from the additional complexity of bridge regressions.
Before entering the final balancing step, the bilateral trade estimates are scaled using RAS to match each country’s national-accounts-consistent export and import totals. This ensures that the bilateral and aggregate constraints imposed on the GRAS procedure are mutually consistent and that the balancing problem is feasible.
The methodology has been validated both retrospectively — for example, by projecting year t from data available through year t‑1 and comparing the results with the published tables — and against independent estimates from Eurostat FIGARO and the Asian Development Bank, with encouraging results. The projected tables for 2023 and 2024 should nonetheless be treated as preliminary, particularly at the full level of country-industry disaggregation, where projections are naturally subject to greater uncertainty.
Work is underway to extend the projection framework to make use of a broader set of data inputs. One line of development focuses on incorporating more granular national accounts data from OECD SNA databases and national statistical offices, so as to progressively replace estimates derived from UN AMA aggregates with more authoritative national data where available. A second focuses on integrating more detailed trade data, covering both merchandise trade at the product level and services trade broken down by detailed balance-of-payments categories. This will strengthen the alignment of the projected tables with observed bilateral trade patterns.
