Banking diversity, financial complexity and resilience to financial shocks: evidence from Italian provinces

ABSTRACT

In this paper, we investigate the influence of banking and financial diversity on stability. We compute an index of banking diversity for Italian provinces and, drawing on network theory, we propose a measure of the diversity and development of the overall provincial financial sector. Our results show that diversity in the banking and financial markets promotes greater stability. Such beneficial effects are particularly evident during periods of financial distress. We ascribe our findings to the better diversification achieved by more diverse financial systems, as documented by lower loan concentration and higher loan diversification in terms of economic destination and borrower category.

KEYWORDS:

• Financial diversity
• financial stability
• non-performing loans
• financial complexity
• financial crises
• banking diversity

JEL CODES:

• G01
• G20
• P34

Acknowledgments

I am grateful to Pasquale Scaramozzino, the participants to the “Giancarlo Marini Young Economists Session” of the XXXII Villa Mondragone International Economic Seminar (Rome, July 2021) and to the Tor Vergata Lunch Seminar (Rome, November 2021) for the helpful comments and suggestions. A previous version of this paper has been awarded the 9^th Suerf/Unicredit Foundation Research Prize.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. The topic has gained attention not only from academics but also from policy makers. See, for instance, the European Parliament resolution of 5 June 2008 on Competition: Sector inquiry on retail banking (2007/2201(INI)), the Great Britain Treasury (2010) note “A new approach to financial regulation: judgement, focus and stability” and the European Commission Liikanen (2012) report “High-level Expert Group on reforming the structure of the EU banking sector”.

2. The paper is closely related to our work, as both share the aim to investigate whether more diverse territories enjoy a sort of shield during financial turmoil. On the other hand, we differentiate from it along several dimensions. First, we focus on sub-national territories and extend the sample to include the recent Covid-19 pandemic. Second, (Baum et al. 2020) measure stability at bank level, while we consider financial stability at provincial level. Third, the authors compute their indexes of diversity by only focusing on two dimensions (ownership and competition), while we employ a broader concept that also encompasses the geographic spread of financial intermediaries within each province and their balance sheet diversity. Finally, we also propose a measure of diversity and development for the overall financial system.

3. While the Great Financial Crisis and the Covid-19 pandemic exogenously hit the Italian economy, the nature of the sovereign debt crisis might be debatable. Indeed, as it is well-known, the crisis emerged as the result of a loop dynamic at the European level. From one point of view, sovereign debts deteriorated because of the public rescuing huge financial institutions. At the same time, the crisis was exacerbated by the holding of large amount of public debt by domestic banks. Endogeneity in the form of reverse causality would arise if the instability caused by the crisis affected the diversity of local financial systems, and not vice versa. However, government support to the banking sector during the crisis was very limited in Italy, so that it should not have affected the diversity of local systems. Finally, banking diversity and financial complexity measures do not change much within-province over time (see Appendix D, so that we tend to exclude that crisis episodes affected diversity in our sample.

4. This is the period under scrutiny as for the main econometric analysis. However, we are able to extend further our sample in ancillary exercises and consider the period 1995–2020, thanks to greater data availability. We prefer to exploit such additional information rather than ignore it, so that we do not reduce the sample (starting from 2006) when implementing such additional models.

5. The province of Fermo was created by dividing in two the province of Ascoli Piceno. The same happened for the province of Monza e della Brianza, from the territory of Milan. As for Barletta-Andria-Trani, four out of ten current municipalities previously belonged to the province of Foggia, while the rest was part of the province of Bari. However, the three main cities of the new entity belonged to the province of Bari, and as of December 2017, 82% of the population of the new province lived in municipalities that were part of the province of Bari http://demo.istat.it/bilmens2017gen/index02.html. Hence, we aggregate the territory of Barletta-Andria-Trani to the latter. Sardinian new provinces were not established by splitting previous entities. On the contrary, the creation of the new provinces in 2001 was followed by several waves of changes of single municipalities across different provinces. Such a process ended in 2016, when a single additional province was created in order to incorporate most of the territory that belonged to the newly-2001-established entities. As a result of the different reforms, we would recur to single municipalities to reconcile data at NUTS3 level for Sardinia. Since financial data at municipality-level are not generally available or unreliable, we prefer to consider Sardinia as a single unit.

6. A value of 1 is only theoretical, since it would be attributed to a province that has no branch in its territory. Since we consider four corporate categories, the maximum score would be 0.75, reflecting a completely diverse financial structure (each institutional category accounting for 25% of the shares in provincial branches).

7. Bankfocus data are available since 2005 for Italian banks. On the other hand, banking data at provincial level coming from the Bank of Italy are available since 1995. Hence, aggregate indexes that considers the balance sheet diversity of provincial banks are available only since 2005. However, such dimension is rather stable over time. Hence, to extend our analysis backwards, we compute the average DIVERSITY_F by province from 2005 to 2020. Then, we consider such average and add it to the other dimensions of diversity to obtain our aggregate indexes in the period before 2005.

8. Provincial main city is the provincial capoluogo, the official provincial capital.

9. Hence, branches that are located in the main city are attributed a distance of 0 km.

10. See, for instance, Belloc et al. (2016) and Landini et al. (2020).

11. When applied to export data, the authors suggest eliminating nodes and territories that do not satisfy a certain threshold. In this context, we cannot apply such threshold, but the introduction of provinces in which no intermediary is located introduces a severe bias in the algorithm. To avoid it, we eliminate such provinces and attribute to them the minimum financial complexity as observed in the overall sample. However, such procedure is prone to criticism. On the other hand, when computing complexity from the Aida dataset, we do not incur in such problem. Indeed, Aida not only reports information on firms that have their legal headquarter in the province, but also on subsidiaries (for example Aida reports information on insurance agencies, not only on the insurance company they belong to). This in turn implies that no province in our sample has zero firm imputed.

12. See Bank of Italy (2020) for details. Among the others, the non-performing loan rate (on margin) at country/region or bank level is used as a proxy for financial risk and instability by Sundararajan et al. (2001), Barth et al. (2004), Gonzalez (2005), Podpiera (2006), Agoraki et al. (2011), Cubillas and Gonzalez (2014), Ghosh (2015), Lee and Lu (2015), Chau et al. (2020).

13. The Great Financial Crisis originated in 2007, but Italy was hit later on, after the collapse of Lehman Brothers. The GFC is usually dated between 2008 and 2009 for Italy, see for instance the influential database on banking crises by Laeven and Valencia, (2018). The European sovereign debt crisis’ first event is usually attributed to October/November 2009, when the newly elected Greek government revisited its budget deficit forecast. It started to propagate to other countries (Ireland, Portugal, Spain and Italy) in 2010, even if Italy was affected by major problems only since May 2011 (see Lane, 2012). We impute to 2013 the last year of crisis for Italy since in 2014, contrary to the previous two years, the Italian GDP did not experience a severe drop and remained stable (see World Bank data available at https://data.worldbank.org/indicator/NY.GDP.MKTP.KD?locations=IT). Such dates seem reliable and offer a consistent picture of the evolution of non-performing loans rate for Italy (Figures 3 and 4).

14. For details, see the IMF monitoring of policy responses to the pandemic (https://www.imf.org/en/Topics/imf-and-covid19/Policy-Responses-to-COVID-19#I) and the monitoring by the joint Task force between Ministero dell’Economia e delle Finanze, Ministero dello Sviluppo Economico, Bank of Italy, l’Associazione Bancaria Italiana (ABI), Mediocredito Centrale (MCC) and Sace, available at https://www.bancaditalia.it/focus/covid-19/task-force/index.html.

15. Ambiguous or inconclusive results on the relationship between social capital measures and resilience in Italy appears also in works that consider a longer time span (see for instance Sabatino, 2019).

16. Results are confirmed when considering the time span of section 4.1 (2006–2020).

17. A one standard deviation increase in DIVERSITY_OCG translates into a 2.70% decrease in loan concentration, while a one standard deviation increase in DIVERSITY_ORIGINAL predicts a 2.10% decrease in the dependent variable.

18. For brevity, we do not report regressions that consider single elements of the loan diversity index as dependent. However, higher COMPLEXITY_AIDA is associated to higher levels of less traditional loans (e.g. financial investment, public infrastructure, purchases of real estate not as consumer households’ dwellings, durables by firms, etc.) than DIVERSITY_ORIGINAL. The latter translates mainly into higher levels of traditional loans, such as mortgages for the purchase of dwellings by consumer households. This supports the explanation provided in text. Results can be imputed to the broader plethora of intermediaries considered by our complexity measure. These, in turn, are better able to finance a broader array of investment. On the contrary, banking diversity measures are too narrow to capture such effect.

19. First loan moratoria were introduced with the “Cure Italy” decree adopted on March 17^th. However, such measures needed time to become fully operational and the resort to the moratoria was limited until April.

20. This is confirmed by the monitoring of the Bank of Italy on the participation in debt moratoria. End of December data are higher than those coming from the end of June for both moratoria by government impulse (‘Cure Italy’ and ‘Liquidity’ decree laws for firms, access to the ‘Gasparrini’ Fund for households’ mortgages) and financial sector initiatives. The same is true for requests for financing backed by the Central Guarantee Fund by SMEs (under Article 13 of the ‘Liquidity’ decree law). See data on the monitoring available at https://www.bancaditalia.it/focus/covid-19/tabelle-moratorie.pdf. See also Ciocchetta et al. (2021) and data from the joint Task force between Ministero dell’Economia e delle Finanze, Ministero dello Sviluppo Economico, Bank of Italy, l’Associazione Bancaria Italiana (ABI), Mediocredito Centrale (MCC) and Sace, available at https://www.bancaditalia.it/focus/covid-19/task-force/index.html.

21. These studies use population diversity (in terms of age and gender) as instruments for board diversity of resident firms and banks, respectively. Other studies have used measures of diversity of the external environment as instruments for other financial phenomena (Liu et al. 2014; Shim, 2019).

22. We also use the foreign diversity index and foreign population as instruments in this setting. Results on financial complexity are confirmed. However, the Hansen test points to endogeneity problems affecting our instruments. Hence, we modify the instruments set used in these specifications. Moreover, results form Table 18. are confirmed used cars and car accidents as instruments, not reported for brevity.

23. Conversely, the PCI is defined as $PCI=\frac{\vec{Q}-\vec{Q}}{std\left(\vec{Q}\right)}$ , Where $\vec{Q}$ ,is the eigenvector of the matrix ${{\tilde{M}}_{xx}}^{\mathrm{\prime }}$ (product level counterpart of ${\tilde{M}}_{i{i}^{\prime }}$) associated to the second largest eigenvalue.

24. The correlation between the index in 1996 and 2020 is about 76%, while that between the index in 1996 and the index in the period of the GDC and debt crisis is about 88%. The correlation improves when considering the average value of diversity in 1996–2006 and the value observed in 2020.