DEVELOPING ASIA’S FISCAL LANDSCAPE AND CHALLENGES

This paper analyzes how substantial fiscal policy stimulus in response to COVID-19, combined with the impact of the downturn on revenues, has severely weakened public finances in many developing Asian economies. Analysis drawing on newly compiled data shows that while tax revenues in developing Asia steadily rose in the 2 decades before the COVID–19 pandemic, they continued to lag well behind high-income economies and some developing peers. The region relies on indirect taxes, creating a relatively efficient but less progressive tax structure, while government expenditures on education and health were comparatively modest.

In contrast, the IMF-GFS covers significantly more economies: 181. The IMF-GFS has data for 41 developing Asian economies of ADB, compared with only 20 for OECD data. The GFS database is updated on a weekly basis and, at the time of writing in early 2022, included some data for 2020, valuable for an initial assessment of the effects of COVID-19 on fiscal revenues. However, its temporal coverage is patchier, and the presentation of tax subcomponents are less harmonized across economies. For example, Singapore's value-added tax is missing in the IMF-GFS and subsumed under taxes on goods and services. By comparison, the two components are reported separately for Thailand. Finally, data in the IMF-GFS is a mix of general government (36%) and central government (64%) reporting across economies.
By combining data sources, we create a dataset that maximizes the desirable features of the IMF-GFS and OECD revenue statistics. Table A1 provides a summary of economy, temporal, and general government coverage comparisons across the three datasets. Our core dataset includes data for 192 economies from 1995 to 2020, including the main tax aggregates of personal and corporate income, value-added tax and other goods and services, and international trade. The gains are most evident in terms of coverage of economies. Data for 66% of economies is presented on a general government basis. The lower temporal and general government coverage are simply artefacts of the dataset combination rather than poorer coverage. Note: (i) Green-shaded columns represent superior country coverage.
(ii) Green-highlighted numbers represent superior temporal coverage measured as the share of non-missing entries to total potential entries from 1995 to 2019. It must be noted that the practice of combining databases is common for fiscal data analyses across economies. For example, the IMF World Revenue Longitudinal Data (WoRLD) was compiled using data from the OECD, the IMF-GFS, World Economic Outlook, and IMF staff estimates. The UN-WIDER Government Revenue Dataset similarly draws from multiple sources.
For our purposes, the process of supplementing databases with each other was subject to careful comparisons across the databases, ensuring that (i) data remains comparable within an economy over time; (ii) data across economies are as comparable as possible; and (iii) data remains internally consistent, i.e., the subcomponents, in principle, add up to the aggregate components. The process is described in the next section.

B. Data Compilation Notes
The OECD revenue database served as the base dataset. Data from 75 economies not covered in the OECD were imported from the IMF-GFS. Data for four economies (Brunei Darussalam; Taipei,China; Turkmenistan; and Tuvalu), which are neither in OECD nor in IMF-GFS were sourced from the ADB Key Indicators Database (KIDB). In total, our core database covers 192 economies. Table A2 presents the matching of variables across the three databases. The ultimate data source for each economy was decided using the following criteria: (i) The database with the most complete and unbroken series from 1995 to 2019.
For example, the data for the PRC in the OECD database only begins from 2009 to 2019, whereas the data in the IMF-GFS is from 1995 to 2019. In this case, we opted to keep the data from the latter source. The final source for each economy is documented in the source column of the database.
(ii) Preference for general over central government reporting: Whereas the OECD presents data for general government, the IMF-GFS presents both central and general government data for some economies. There is a preference for general government, which ideally represents the totality of an economy's revenues across government levels as described in equation (1).
general government = central + state + local + social security (1) When possible, general government series were derived by adding up subcomponents of equation (1 Notwithstanding the derivations, less than a third of the economies in the IMF-GFS dataset have good temporal and subcomponent coverage on general government entries. The choice between general or central government was determined based on the series that provided the maximum temporal coverage. Our core dataset is, therefore, a mix of general government (65%) and central government reporting (35%).
In general, we refrain from mixing central and government entries within an economy. However, there are a few cases where an economy shifted from central to general government reporting without any overlapping years when both are reported: Albania (2004), Armenia (2004), the Kyrgyz Republic (2014), and Turkmenistan (2011). We joined the two series after verifying that doing so (a) did not introduce large deviations from overall time trends, and (b) resulting revenue trends are comparable to the total revenue reported in the World Economic Outlook. These cases are indicated by the variable break, which is equal to one for the joining year for each of these economies. Changes in fiscal and calendar year reporting are likewise marked by the break variable. Annotations on the nature of the break can be found in breaknotes.
There are cases where general government is the same as central government (i.e., Hong Kong, China; and Singapore), and this is most common for smaller economies such as Fiji, the Lao People's Democratic Republic, Maldives, Papua New Guinea, and Vanuatu. In these cases, we consider the series as general government.
(iii) Within economy mixing of sources and internal consistency: Both the OECD and GFS databases exhibit data gaps for some economies. An example is Kenya where OECD only reports data from 2001-2018 for general government, whereas GFS has data from 1995 to 2019 with central government reporting. In such cases, we supplement the OECD data with the GFS data if it can be verified from overlapping observations that the two sources are close in magnitude. Specifically, we supplement the OECD data (missing observations) with the GFS data (non-missing observations) provided that ≤ . for ± , where t represents the year where data is missing from OECD dataset. This discrepancy check is carried out variable-by-variable to preserve the internal consistency of our database.
For developing Asia, ADB KIDB provides a valuable resource for extending temporal coverage. For example, in the GFS, data for Armenia (not covered in OECD), begins in 2003, whereas in ADB KIDB, it starts in 2000. We augmented Armenia's data using ADB KIDB provided the 5% discrepancy rule of thumb noted above is satisfied. The Asian Development Outlook Update database also proved a useful source albeit only for the aggregate tax revenue variable.
One area where mixing of data sources is common surrounds the years 2019 and 2020 because data coverage for OECD ends in 2018 and 2019 for most economies, whereas ADB KIDB has 2020 data for 19 developing Asian economies, and the IMF-GFS has 2020 data entries for 64 economies, nine of which are from developing Asia (with coverage expected to grow gradually over its weekly updates. The data in this background reflect GFS data as of 22 January 2022). Data augmentations for 2019 and 2020 generally follow the 5% discrepancy guideline.