The impact of efficiency on asset quality in banking

ABSTRACT

We investigate the impact of banks’ ability to minimise costs on asset quality, by assessing the temporal relationship between these variables in a sample of Italian banks over the period 2006–2015. We offer new insights into the channels through which bank efficiency affects non-performing loans by disentangling the short-term component of cost efficiency from its long-term component. We show that non-performing loans afflicting Italian banks can be explained by both efficiency components. A decrease in short-term cost efficiency precedes a worsening in banks’ asset quality, implying that regulators should consider adopting short-term efficiency as an early warning indicator of a deterioration in asset quality. We also present evidence of a trade-off between long-term efficiency and bank non-performing loans, which suggests that the removal of exogenous hindrances that prevent banks from allocating optimal levels of resources to the management of their loan portfolio should be a main policymakers’ objective.

KEYWORDS:

• Banking
• non-performing loans
• stochastic frontier analysis
• short-term cost efficiency
• long-term cost efficiency

JEL CLASSIFICATIONS:

• C33
• G01
• G18
• G21
• G28

Disclosure statement

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

Notes

1 In 2014, the ECB launched the Comprehensive Assessment to ensure: i) adequate levels of bank capitalisation and ii) banks’ resilience to financial shocks. The assessment comprised an Asset Quality Review, which revealed a significantly larger stock of impaired bank loans in the euro area than previously disclosed, thus triggering the ECB’s focus on resolving NPLs.

2 https://www.bankingsupervision.europa.eu/banking/priorities/html/ssm.supervisory_priorities2019.en.html

3 The Government’s response to the global financial crisis included liquidity provisions, measures for the recapitalisation of distressed banks, schemes for strengthening and supporting the banking sector, and the extension of the depositor protection scheme. However, Italy’s state was far better than that of other European counterparts. Only very modest amount of resources (0.3% of GDP) were deployed to sustained the banking system, in comparison to 55.8% in Belgium, 6.6% in France, 16.9% in Germany, 67.7% in the United Kingdom, 101.9% in Ireland, 24.5% in the Netherlands (Cosma and Gualandri, 2012). Concerning the sovereign crisis, Italy introduced in December 2011 the ‘Save Italy' decree, which included the provision of public guarantees on the liabilities of Italian banks by the Italian Ministry of Economy and Finance.

4 For instance, in 2013, bank loans represented 64.7% of Italian firms’ total financial debt, more than 20 percentage points higher than the euro area average, which stood at 42.9%. The only other country where firms were more dependent on bank credit was Greece, where bank loans accounted for approximately 70% of total firm financing. By way of comparison, bank loans represented 32.2% of total financial firm debt in France, 51.8% in Spain and 52.1% in Germany (European Commission 2015).

5 Bank of Italy (2013) defines ‘bad loans’ as ‘exposures to an insolvent counterparty (even if insolvency is not legally ascertained) or in equivalent situations, regardless of any loss estimate made by the bank and irrespective of any possible collateral or guarantee’.

6 To estimate efficiency scores, Demerjian, Lev, and McVay (2012), Demerjian et al. (2013) and Andreou et al. (2017) use Data Envelopment Analysis whereas Andreou, Philip, and Robejsek (2016) use Stochastic Frontier Analysis. In all cases, the authors use a two-step design where in the first step, efficiencies are estimated and in the second step they purge all firm specific effect from the efficiency component using tobit regressions. It is worth pointing out the two-step methodology employed by Andreou, Philip, and Robejsek (2016) has been amply criticised (see Wang and Schmidt 2002).

7 The bad management hypothesis does not explicitly discuss the voluntary accumulation of risky loans by banks. That is, a bank may choose to take on more credit risk in a specific year which in turn may lead to higher monitoring costs, forcing this bank to depart from the cost frontier. This deliberate increase in credit risk will be therefore linked to lower levels of efficiency for different reasons than the bad luck hypothesis. We thank an anonymous referee for bringing this possible interpretation to our attention.

8 A variable x is said to Granger-cause y if, given past values of y, past values of x are able to explain current values of y (Granger 1969).

9 GIPSI countries include Greece, Ireland, Portugal, Spain, Italy.

10 It is worth noting that, given the time-invariant nature of structural efficiency, it is not possible to directly test for the temporal relationship between structural cost efficiency and NPLs.

11 The use of the term ‘habits’ to identify recurring managerial behaviours as potential sources of structural inefficiency is widely accepted (see Blasch et al. 2017 and Filippini, Geissmann, and Greene 2018).

12 We confirm the model specification with BIC information criteria, as suggested by Andrews and Lu (2001).

13 The logit transformation ensures that the dependent variable spans over the interval [-∞;+∞] as opposed to the [0;1] interval and is distributed symmetrically (the untransformed distribution of BL is skewed to the right and we reject the null hypothesis of normality of the Shapiro-Francia W' test at the 1% level). Furthermore, the logit transformation prevents non-normality in the error term and accounts for non-linearities, for example if larger shocks to the explanatory variables cause a large, non-linear response in the transformed dependent variable (Ghosh 2015).

14 The advantages of cost efficiency over profit efficiency as a proxy for management quality with respect to risk have been well documented in the literature (for example, Williams 2004 and more recently Assaf et al. 2019).

15 It is worth pointing out that the variables explaining persistent inefficiency should be naturally time-invariant as we aim to capture persistent characteristics of the Italian banking industry and operating environment that could have affected the cost-minimization behaviour of banks in the long-run (see Lien, Kumbhakar, and Alem 2018).

16 Unlike Berger and DeYoung (1997), we do not test for this hypothesis only on a sub-sample of weakly capitalised banks but, in line with Fiordelisi, Marques-Ibanez, and Molyneux (2011) and Louzis, Vouldis, and Metaxas (2012), we rely on the entire sample. This approach allows us to control for managers in highly capitalised banks who could resort to a liberal credit policy under the notion that their bank is ‘too big to fail’, thus implying a positive relationship between capital and asset quality (Rajan 1994).

17 In our sample, the SIBs include UniCredit Spa, Banca Carige SpA, Veneto Banca, Unione di Banche Italiane (UBI), Intesa Sanpaolo, Mediobanca, Credito Emiliano, Banca Popolare di Vicenza, Banca Popolare di Sondrio, Banca Popolare di Milano, Banca Popolare dell’Emilia Romagna, Banco Popolare, Monte dei Paschi di Siena Spa (https://www.bankingsupervision.europa.eu/ecb/pub/pdf/list_of_supervised_entities_20160331.en.pdf).

18 In all regressions, to avoid instrument proliferation and overfitting of the endogenous variables, we restrict the number of instruments in order not to exceed the number of cross-sections in the sample (Roodman 2009). We further confirm the properties of the instruments used in the SGMM by estimating the constituent levels and differenced equations for our more inclusive models (columns 1 and 2, Table 6) and we assess the relevance of the instruments. In both cases, we reject the null of underidentification (Kleibergen and Paap 2006). It should be noted that these tests serve only as an indication of the quality of the instruments used in the SGMM, as SGMM estimates require the joint estimation of the differenced and levels equations.

19 We thank an anonymous referee for suggesting this test.

20 Schivardi, Sette, and Tabellini (2017) observe that Italian banks with low levels of capitalisation were engaging in significant zombie lending between 2008 and 2013, and conclude that ‘low capital banks may be particularly averse to absorb losses, especially during a recession, and may therefore be relatively more willing to keep lending to weak firms that otherwise would not be able to service their debt’ (Schivardi, Sette, and Tabellini 2017, 15).

21 Indeed, banks may be tempted to increase the riskiness of their loan portfolio if they are certain about government support in the event of financial troubles. At the same time, if investors recognise this implicit subsidy from the state, they tend to impose lower market discipline on the bank.

22 We cannot estimate the SGMM for the SIBs of the commercial banks as the reduced samples do not contain enough cross-sections for a SGMM (13 SIBs and 62 commercial banks).

23 We do not employ Loan Loss Provisions (LLP) or Loan Loss Reserves (LLR) as alternative proxies for risk as managers may exploit information advantages and depart from normal levels of LLP/LLR for objectives other than provisioning for NPLs. Prior research suggests that discretionary LLP behaviour (which feeds back to LLR), could be due to a number of factors, such as, income smoothing, capital management and/or signalling among others (see for example, Beatty and Liao 2014).

24 Similar steps have been undertaken by UK regulators with the introduction of the Senior Managers and Certification Regime (SM&CR), aimed at strengthening the individual accountability of firms’ management and employees. With respect to the euro area, the SSM is in charge of assessing the suitability of new and re-appointed members of management bodies of banks since November 2014. As such, the ‘fit and proper’ assessment focuses only on the highest management positions.

Additional information

Notes on contributors

Oleg Badunenko

Oleg Badunenko is senior lecturer at Brunel University, London. He is also a visiting fellow at University of Portsmouth. Before this position, Oleg Badunenko was a senior lecturer in Economics at University of Portsmouth, Assistant Professor at University of Cologne and research associate at the German Institute for Economics Research (DIW-Berlin). His primary research areas are parametric and nonparametric efficiency and productivity measurement. He is also interested in modelling behaviour of banking industry.

Aristeidis Dadoukis

Aristeidis Dadoukis is an Assistant Professor in Banking at the University of Nottingham. His research interests are in banking efficiency, finance, bank regulation and financial stability.

Giulia Fusi

Giulia Fusi is a Doctoral Candidate at the University of Nottingham. Her research interests lie in the areas of banking efficiency, bank regulation and supervision and financial stability.

Richard Simper

Richard Simper is Professor of Banking and Applied Finance at the University of Nottingham. His research interests are in two areas, ‘Banking, Finance and Banking Regulation’ and ‘Public Sector Performance Measurement’.