Spatial dependence in the non-performing loans of small Italian cooperative banks

ABSTRACT

This paper investigates the presence of spatial dependence in the non-performing loans (NPLs) ratio of small Italian cooperative banks. As these financial intermediaries operate in a delimited area, the strategies they adopt to recover NPLs can produce spatial spillover effects on the ability of neighbouring banks to recover credit. Furthermore, small banks’ reliance on the relationship lending technique to select and monitor borrowers suggests a similar ability to identify underperforming borrowers, which may have a negative impact on the community in which those banks operate. Our empirical estimations provide strong evidence for both spatial and spatial–temporal variables driving impaired loans in local banks. The results indicate different effects of the spatial terms, showing a direct positive effect of the contemporaneous spatial lag variable and a negative effect of the space–time autoregressive coefficient. Whereas the former effects can be ascribed to changes in the macroeconomic cycle, the latter confirms the insight that the recovery capacities of local banks can be harmed by neighbouring banks’ credit recovery policies.

KEYWORDS:

• non-performing loans
• local banks
• spatial dynamic data panel model
• spatial dependence
• bad loan recovery
• Lerner index

JEL:

• C31
• D43
• G21

ACKNOWLEDGEMENTS

The authors thank the editor and two anonymous referees for their critical comments and suggestions. We also thank the participants at the 9th International Conference on Risk Analysis and at the 62nd Annual Conference of the Italian Economic Association. The usual disclaimer applies. The authors contributed equally to each section of the article.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Notes

1. NPLs are usually defined as loans for which interest and principal payments have been past/due for at least 90 days (Lee et al., 2019).

2. The credit recovery procedures undertaken by Italian banks are foreclosures, bankruptcies, arrangements with creditors, restructuring agreements, recovery plans and out-of-court agreements, but also straight portfolio sales or securitization (Carpinelli et al., 2017).

3. Subsection 2.2 reviews the literature analysing the geographical effects.

4. Our model’s estimates should be robust given that the specification controls for common shocks affecting provincial financial markets.

5. CCBs have undergone reform in recent years that has confirmed their characteristics of mutualism and localism but has also prescribed limits on management independence due to high levels of risk. The reform requires that each CCB joins and enters into an agreement with a holding group (Cooperative Banking Group) authorized by the Bank of Italy. CCBs have retained ownership of their assets and generally retained their managerial independence, but this can be limited in high-risk circumstances.

6. To treat co-movements among cross-sectional units located in different geographical areas, methods have been devised in the spatial econometrics literature to incorporate spatial autocorrelation and spatial heterogeneity within linear regression models (Elhorst, 2014). Such a relationship is termed spatial autocorrelation, or weak spatial dependence.

7. $\mathit{W}$ is a non-negative, time-invariant and deterministic matrix of order n containing the spatial weights $\left(w_{ij}\right)$ representing the weights assigned to the spatial connection. Because no unit is spatially correlated with itself, $w_{i,j}=0$ when $i=j$ (e.g., Kelejian & Prucha, 2010), while the other elements define the spatial connection according an assigned weight $w_{i,j}\ne 0$.

8. Based on the intermediaries listed on the Bank of Italy’s website (https://infostat.bancaditalia.it/GIAVAInquiry-public/ng/), there were 268 CCBs as of 31 December 2018.

9. To build a balanced dataset, the missing values were filled in from the balance sheets published on the banks’ websites.

10. Given that the elasticity ($ϵ$) of TC with respect to Q is: ${ϵ}_{TC,Q}={\displaystyle \frac{dTC/TC}{dQ/Q}=MC\times {\displaystyle \frac{1}{AC}\to MC={ϵ}_{TC,Q}\times AC,}}$where AC is the average cost $\left({\displaystyle \frac{TC}{Q}}\right)$ and $MC={\displaystyle \frac{dTC}{dQ}}$.

11. We owe the inclusion of Int rate diff to the suggestion of an anonymous referee. The suggestion is appreciated as we found that including a variable comparing CCB credit costs with those of other banks improved our estimates. It is essential to note, however, that as provincial-level data on the interest rate of CCBs are unavailable, we calculate it from our sample, which represents a limitation of our analysis.

12. To stress test our results for stability, we also considered additional macroeconomic variables, such as population density and the amount of provincial NPLs net of those associated with CCBs. The estimated results of our model’s main variables did not change but the multicollinearity issues associated with the estimated coefficients led us to exclude these additional variables from our models. Overall, our specification does not suffer from multicollinearity bias, with a mean variance inflation factor (VIF) of 1.57 and no value exceeding 2.

13. The Bank of Italy provides data on NPLs and loans for the Italian provinces but does not specify the type of banks to which they refer. Therefore, we calculated the subtrahend of the difference (NPL for CCBs) from our sample of bank-level data.

14. The results of the LM tests shown in Table A1 in Appendix A in the supplemental data online are based on the minimum bandwidth, but the results have also been confirmed when the model uses the other two bandwidths.

15. These test results for the TSR models, reported in Table A2 in Appendix A in the supplemental data online, also support the need for spatial specification.

16. Z Score is the result of the sum of return on assets (ROA) and the equity-to-assets ratio over the standard deviation of ROA, and is calculated with a five-year rolling timeframe.