Regional unemployment dynamics in Indonesia: serial persistence, spatial dependence, and common factors

Abstract

We analyze the space-time dynamics of Indonesia’s provincial unemployment by simultaneously accounting for their serial persistence, spatial dependence, and common factors. The results show that unemployment rates vary widely across provinces, but have similar patterns over time, indicating the presence of common latent factors. Using the average national unemployment rate as a proxy for common factors, the results indicate that the space-time dynamics of provincial unemployment are characterized by both significant serial persistence and spatial dependence. The results also quantify which regions are most sensitive to national unemployment shocks, providing a deeper understanding of regional unemployment heterogeneity.

Appendices

Appendix

Appendix A: Cross-sectional dynamics of unemployment

Figure 4

Fig. 4

Cross-sectional dynamics of unemployment

Appendix B: Spatial distribution of unemployment

Figure 5

Fig. 5

Spatial distribution of unemployment

Appendix C: Derivation of a spatial connectivity structure for Indonesia

Figure 6

Fig. 6

Derivation of a spatial connectivity structure (weights matrix) for Indonesia

In panel (a) of Fig. 6, the administrative boundaries of the provinces of Indonesia are not consistently continuous due to the presence of islands. As a result, commonly used contiguity criteria, which determines geographical neighbors, cannot be directly applied. To address this limitation, the original centroids of the provincial boundaries and the computational geometry approach of Brassel and Reif (1979) and Aurenhammer (1991) are employed to derive a completely connected polygon structure (panel (b)). This connected structure if formed by calculating perpendicular bisectors between every pair of neighboring points. In contemporary spatial analysis literature, this interconnected structure and its construction framework are referred to as the Thiessen polygon approach (Anselin 2020). Within this derived polygon structure, it becomes possible to identify neighboring regions based on both contiguity and distance criteria.

Appendix D: Evolution of spatial dependence in unemployment based on alternative connectivity structures (weight matrices)

Figure 7

Fig. 7

Evolution of Moran’s I based on different weight matrices

Appendix E: Evolution of spatial dependence of the residuals based on the four models of Table 1

Figure 8

Fig. 8

Evolution of spatial dependence of the residuals based on the four models of Table 1

Appendix F: Sensitivity coefficients

Table 2

Table 2 Regional sensitivity coefficients to national shocks

Note: Sensitivity parameters are calculated for the first common factor described in Eq. 4. The standard errors for these parameters are calculated using the ratio formulas for stochastic variables described by Mood et al. (1974).