Skip to main content
Log in

The dynamic effects of non-performing loans on banks’ cost of capital and lending supply in the Eurozone

  • Original Paper
  • Published:
Empirica Aims and scope Submit manuscript


This paper analyses the transmission channel from non-performing loans to the cost of capital, credit provision and liquidity creation in the banks of the Eurozone. The empirical results suggest that holdings of non-performing loans increase both the long- and short-term cost of capital for banks. Moreover, the less capitalized the bank, the greater the reduction in credit provision and liquidity creation due to the increased cost of capital. This phenomenon is found to be more economically significant for European periphery country banks than for core country banks. The identification of the transmission channel is robust to the Granger predictability test.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Subscribe and save

Springer+ Basic
EUR 32.99 /Month
  • Get 10 units per month
  • Download Article/Chapter or Ebook
  • 1 Unit = 1 Article or 1 Chapter
  • Cancel anytime
Subscribe now

Buy Now

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1

Source: Authors’ estimation based on Bankscope and Thomson Reuters Datastream. The solid line represents the periphery countries, and the dashed line, the periphery countries. The vertical line splits the sample into the pre- and post crisis subperiods (t = 2007Q2). a displays the evolution of the non-performing loans ratio (NPLit) with the sample broken down first by sub-periods, and then by core and periphery countries. b displays the evolution of the cost of capital (rit) by sample sub-periods, and by core and periphery countries. c displays the distribution of NPLit by sub-periods, and by core and periphery countries. d displays the distribution of the Beta CAPM (βit) by sub-periods, and by core and periphery countries. The whiskers represent the maximum and the minimum of the distribution. The box is divided into two parts by the median, i.e., the 50th percentile of the distribution. The upper (lower) box represents the 25% of the sample above (below) the median, i.e., the upper (lower) quartile

Similar content being viewed by others


  1. Recent authors demonstrate that high levels of NPLs might deteriorate creditworthiness and reduce the demand for credit (Accornero et al. 2017; Balgova et al. 2016; Bending et al. 2014; Cucinelli 2015).

  2. The Eurozone Members included in our sample are Austria, Belgium, Cyprus, Estonia, Finland, France, Germany, Greece, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Portugal, Slovakia, Slovenia, and Spain.

  3. See King (2009) for a similar approach.

  4. The assumption of constant betas for 5-year periods would be justified if betas were changing as close enough to a snail’s pace as is the case in diversified portfolios. Since new information is incorporated following the banking and sovereign debt crises, betas for individual stocks may change rapidly and the assumption of a 5-year window may not be applicable.

  5. See Carbó-Valverde et al. (2017) for a similar approach.

  6. Note that the ADL model is equivalent to error-correction by substituting \(y_{it} = y_{i,t - 1} + \Delta y_{it}\) and \(NPL_{it} = NPL_{i,t - 1} + \Delta NPL_{it}\). In the error correction mechanism, the adjustment of y towards its equilibrium is defined by the deviations of both variables lagged by one period: \(y_{i,t - 1} - \frac{{\alpha_{2} + \alpha_{3} }}{{1 - \alpha_{1} }}NPL_{i,t - 1}\).

  7. Marginal costs are calculated following the transcendental logarithmic costs function which includes operating (labour, capital and deposits) and financial costs, and a trend to control for technological changes over time (e.g., Fernandez de Guevara et al. 2005; Carbó-Valverde et al. 2009; Cruz-García et al. 2017;Mansilla-Fernández 2020).

  8. Two further solutions are possible. The unstable solution or hysteresis (\(\alpha = 1\)) means that the solution contains a linear trend and that the initial condition exerts full influence on yit. The explosive solution (\(\left| \alpha \right| > 1\)) is the opposite of the ‘stable’ solution, i.e., the effect of the regressor is divergent on yit.

  9. This test is conducted under the null \(\gamma^{*} = \left( {\sum\nolimits_{i = 1}^{p} {\gamma_{i} } } \right) - 1 = 0\).

  10. Anastasiou (2017) and Anastasiou et al. (2019) use cointegration and causality techniques, respectively, to test for persistent macroeconomic and business cycle effects on non-performing loans. This article goes further by extending the analysis to the long-term repercussions of accumulating non-performing loans on capital financing and bank functioning.

  11. Two-, three-, and four-period-period lagged instrumental variables are used to control for possible endogeneity issues deriving from correlations of errors over time. The Sargan test and serial autocorrelation tests of second (AR(2)) and third order (AR(3)) are performed to test for orthogonality of the instruments.

  12. Recall that periphery country banks accumulated relatively larger volumes of impaired loans on their balance sheets (Barba Navaretti et al. 2017; Mansilla-Fernández 2017; Climent-Serrano 2019). Consequently, according to the transmission channel under investigation in this study, equity investors perceive these institutions as relatively riskier than core country banks (see, Chiesa and Mansilla-Fernández 2018).

  13. Following Vides et al. (2018) who perform a thorough analysis of the integration of European stock markets after the sovereign-debt crisis, we include the CDS of sovereign debt at the country level to avoid confounding effects.

  14. Results upon request.

  15. The definition of causality, as defined by Granger (1969) and Sims (1972), states that lagged values of \(y_{it}\) should not have explanatory power on movements of \(x_{it}\) beyond that provided by lagged values of \(x_{it}\); more formally \(f\left( {x_{t} |x_{t - 1} ,y_{t - 1} } \right) = f(x_{t} |x_{t - 1} )\). Importantly, the variable \(x_{it}\) is weakly exogenous if it has no explanatory power on any other variable in the regression. Finally, if \(x_{it}\) is weakly exogenous and if \(y_{t - 1}\) is non-significant, then \(x_{it}\) is strongly exogenous (see also Greene 2012, p. 358).

  16. We follow the methodology used by Holtz-Eakin et al. (1988) for panel data with individual fixed effects.


Download references


We would like to express our gratitude to Fritz Breuss (Editor-in-Chief) and an anonymous referee for his/her thorough review, comments and suggestions, which significantly contributed to improving the quality of this article. We are indebted to Isabel Abinzano, Giorgio Barba Navaretti, Massimiliano Barbi, Giacomo Calzolari, Santiago Carbó-Valverde, Pilar Corredor, Luigi Filippini, Paolo Manasse, Luis Muga, Alberto Franco Pozzolo, Francisco Rodríguez-Fernández, Massimo Spisni, and seminar participants at the Department of Economics and the Department of Management of the University of Bologna, and the Business Administration Department of the Public University of Navarre for their helpful feedback and suggestions. José Manuel Mansilla-Fernández gratefully acknowledges Financial Support from ECO2016-77631-R (Ministerio de Economía y Competitividad). All remaining errors are our own.

Author information

Authors and Affiliations


Corresponding author

Correspondence to José Manuel Mansilla-Fernández.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.


Appendix 1

See Table 8.

Table 8 Liquidity classification of bank activities

Appendix 2: Granger causality test

We use the Granger causality test to assess the direction of the causality between NPLs and our variables of study: the cost of capital, CAPM beta, ROE and the gap between the cost of capital and ROE. We employ four lags (l) of the variables in order to capture the long-term effects of NPLs on the target variables. Since we are using panel data, we follow the Holtz-Eaking et al.’s (1988) methodology with individual fixed effects (fi). The statistical significance of the test is measured by using an F-test.

In order to test whether NPLs predict our variables of study, two conditions should be meet:

  1. 1.

    The NPLs ratio (NPLit) should be statistically significant to the cost of capital (rit):

    $$\begin{array}{*{20}c} {r_{it} = \varphi_{0} + \mathop \sum \limits_{l = 1}^{L} \varphi_{l} r_{i,t - l} + \mathop \sum \limits_{l = 1}^{L} \psi_{l} NPL_{i,t - l} + \nu_{t} + f_{i} + \varepsilon_{it} } \\ \end{array}$$
  2. 2.

    The cost of capital (rit) should not be significant in explaining NPLs (NPLit):

    $$\begin{array}{*{20}c} {NPL_{it} = \varphi_{0}^{{\prime }} + \mathop \sum \limits_{l = 1}^{L} \varphi_{l}^{{\prime }} r_{i,t - l} + \mathop \sum \limits_{l = 1}^{L} \psi_{l}^{{\prime }} NPL_{i,t - l} + \nu_{t} + f_{i} + \varepsilon_{it}^{{\prime }} } \\ \end{array}$$

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Chiesa, G., Mansilla-Fernández, J.M. The dynamic effects of non-performing loans on banks’ cost of capital and lending supply in the Eurozone. Empirica 48, 397–427 (2021).

Download citation

  • Published:

  • Issue Date:

  • DOI:


JEL Classification