Abstract
Due to the scarcity of subnational interregional input–output (IRIO), various approaches to their estimation are actively under investigation in the literature. This paper focuses on the application of spatial econometric method. It determines intra- and interregional coefficients through a joint procedure which successfully avoids the direct recycling of estimates for other geographies and granularities. Instead, the use of Bayesian methods is proposed, which formally integrate limited evidence from existing regional tables (Finland) with a set of sectoral data on value added for 73 NUTS-3 regions in Poland, the latter being dominant. An empirical test of replicating the Korean survey-based IRIO table demonstrates that the accuracy of this approach slightly outperforms an alternative IRIOLQ procedure. The incorporation of time-based distance measurement has only modest effects on empirical fit, and the use of big geolocation dataset to account for commuting relocates 18.9% of the induced effect from a city to its periphery.
Acknowledgements
The support of the National Science Centre in Poland is gratefully acknowledged (research project 2018/31/D/HS4/00316). The author is grateful for helpful comments to the Editor and anonymous Referees, the participants of 47th Macromodels International Conference 2021 in Wieliczka (especially Jacek Osiewalski and Justyna Wróblewska), 26th International Conference on Macroeconomic Analysis and International Finance 2022 in Rethymno (especially Joan Paredes as a discussant), 16th World Conference of Spatial Econometrics Association 2022 in Warsaw (especialy Philipp Otto) and 10th Summer Workshop on Macroeconomics and Finance 2022 in Warsaw, as well as to Piotr Pȩkała for invaluable support in the data acquisition process. The usual disclaimer applies.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 See Figure 17 in Online Appendix F for a visualization on the map of Finnish regions.
2 Further evidence available in Figure 16 in Online Appendix F. It illustrates the driving times from Warsaw to the given NUTS-3 region's ‘capital’ (i.e. the center of the most populated city) and reveals that some destinations are available more quickly than others, in spite of the apparently equal distance from Warsaw.
3 Two features of the simulating process should be noted. First, given the specific shape of the domain (see Figure ), I prefer to work with and
rather than
and
in a procedure where the step is drawn with a constant variance and the responsiveness of distance profiles to the level of shape and scale is (graphically) very different in various parts of the domain. Second, as a candidate generating density, I use multivariate normal density with zero covariances and parameter-specific variances:
for
and
for
,
for
,
for error variances
and
for cross-sectoral error correlations
(or, more precisely, for their transformations
such that
).
5 https://www.bok.or.kr/eng/bbs/E0000634/view.do?nttId=10059403&menuNo=400069, accessed 29.06.2023.
6 Estimation results for Korea can be found in Online Appendix C.
7 https://www.meti.go.jp/english/statistics/tyo/tiikiio/index.html, accessed 29.06.2023.