Looking at the determinants of efficiency in banking: evidence from Italian mutual-cooperatives

Abtract

Italy has experienced a restructuring and consolidation process in the banking industry since the 1990s that is expected to foster efficiency and competition. Despite the reforms, a peculiarity of the industry is the persistence of small mutual-cooperative banks (Banche di Credito Cooperativo, BCCs) active in narrow markets. The scope of this paper is to analyze the determinants of BCCs’ efficiency in the 2006–2011 period. In the first step of the study, a stochastic cost frontier is used to yield bank efficiency. Then the cost efficiency becomes the dependent variable of fixed and random effect models. The reference market of BCCs is the province (NUTS3). We find that BCC cost efficiency is positively affected by market concentration and demand density and inversely related to branching. Importantly, these results are robust to any sample restriction anchored to the distribution of efficiency. While the evidence regarding the credit quality is inconclusive for all BCCs, the sensitivity analysis shows that the risk in local markets is a source of BCC cost inefficiency.

Keywords:

• Mutual-cooperative banks
• local markets
• efficiency determinants
• stochastic frontier
• costs

JEL Classifications:

• G21
• C13
• D22
• O16
• P13

Acknowledgements

The authors would like to thank Sergio Destefanis, Maurizio La Rocca, Fabio Piluso, Damiano Silipo and two anonymous referees for very useful comments on an earlier version of this paper. Editorial assistance by John Richard Broughton and Alessia Via is also acknowledged. The usual disclaimer applies.

Notes

1. The process of institutional reforms has been regulated by several norms, such as, for instance, the 2002 budget law, the 262/2005 law and the 353/2006 Legislative Decree. Details on these reforms are in Giannola (2009), Messori, Tamburini, and Zazzaro (2003) and Silipo (2009).

2. This result is found in Ayadi et al. (2009), Battaglia et al. (2010), Giannola et al. (1997), Giannola and Scarfiglieri (1998), Girardone, Molyneux, and Gardener (2004), Giordano and Lopes (2006, 2012), Fontani and Vitali (2007), Dongili, Rossi, and Zago (2008) and Turati (2004).

3. Another recent paper focusing on BCC performance is Fiordelisi and Mare (2013), which defers from ours because they analyze how efficiency affects the probability of default of cooperatives instead of analyzing the determinants of individual efficiency as we do. After controlling for regional environmental variables meant to be good predictors of default, Fiordelisi and Mare (2013) prove that, over the 1997–2009 period, the probability of BCC to survive increases with efficiency.

4. As Bos and Kool (2006) argue, studies that do not take into account differences between bank-type yield inappropriate conclusions about bank performance. On the contrary, using a wide sample of banks allows net efficiency measures to predict how BCCs are ranked under the assumption that banks operate in an equivalent environment.

5. The relationship between individual efficiency and external determinants might be evaluated at branch level, whatever the bank-type. However, data at branch level are not available in Italy – as well as in many other countries – because they are classified as sensitive-statistics.

6. Two different data-aggregations are needed for addressing the issues we pose. The first concerns data at bank level, while the second regards the geographical aggregation we refer to. Data on individual banks are from the Italian Banking Association (ABI). When considering the provincial level (NUTS3) we use different data sources (Bank of Italy, Italian Institute of Statistics, Istituto Tagliacarne). The period under scrutiny covers the years 2006–2011. This is why the implementation of International Accounting Standards (IAS) occurred in 2005 and banks’ balance sheets before-and-after IAS are not comparable.

7. Data needed to calculate HH2 is the value of total assets by the ith bank in every province j (TA_ij). Because this information is not freely available in Italy, as well as in many other countries, we proceed through this calculation: TA_ij=TA_i*b_ij, where TA_i is the balance-sheet amount of Total Asset (TA) of the ith bank and b_ij is the proportion of branches of bank i in province j (b_ij=BB_ij/BB_j). This procedure is proposed by Carbò Valverde et al. (2003).

8. Alessandria, Aosta, Como, Imperia, Mantova, Milan, Novara, Pavia, Torino, Belluno, Arezzo, Grosseto, Massa, Siena, Lecce, Agrigento, Caltanissetta, Enna, Messina, Ragusa, Siracusa, Trapani (data at level of single province are available upon request).

9. Following Battese and Coelli (1995) allows us to address the issues brought up by Lensink and Meesters (2014) and Wang and Schmidt (2002). Phrased differently, we use a variant of the SFA traditional two-step approach, as in the first step we basically exploit all the advantages provided by the stochastic frontiers specification proposed by Battese and Coelli (1995), while the common use of two-step procedure refers to Battese and Coelli (1992).

10. Using a translog, linear homogeneity also requires standard symmetry (β_js = β_sj and ω_nq = ω_qn) and linear restrictions of the cost function ( and ).

11. As in many other recent papers in the banking efficiency literature (see, for example, Battaglia et al. 2010; Giordano and Lopes 2008; Lensink and Mester 2012) the assumptions on v_it and u_it are those originally proposed by Battese and Coelli (1995), also because modeling other ‘possible correlated structures of the technical inefficiency effects and the random errors in the frontier’ (Battese and Coelli 1995, 327) goes beyond the scope of this work.

12. We implement an LR test to verify the correctness of the Cobb-Douglas versus the Translog. Under H₀ there is the more parsimonious model, which is rejected at 1%.

13. The balance-sheet ratios are (a) the income diversification defined as [Income Commissions/(Income Commissions+Net Interests Income)]; the loans diversification expressed as (1–Loans/Total Assets); the Loans/Deposits ratio and the Equity/Total Assets ratio.

14. Here it is important to provide some model diagnostics. To this end we consider two tests. The Hausman test is conducted to assess the appropriateness of random or fixed effects models. Failure to reject Ho indicates that the random specification is valid. Results are in favor of fixed effect specification. Furthermore, the Hausman-Taylor specification is compared with the fixed effects model. In the Hausman-Taylor specifications, all variables at bank level are treated as endogenous, while environmental variables are assumed to be exogenous. Even in this case, the test supports the fixed effect model, given that the difference in estimated coefficients is not statistically different. This implies that bank-level variables may be treated as exogenous.

15. It is important to say that the size-efficiency nexus may not be the same whatever the size, because nonlinear effects can arise (Andries 2011; Berger and Mester 1997). To this end, we have augmented the basic equation with SIZE² and, alternatively, with the logarithm of SIZE. In both cases, estimations are not significant, implying that there is no-linearity (results are available upon request).