Banks’ riskiness over the business cycle: a panel analysis on Italian intermediaries

Abstract

A comprehensive investigation is provided on the issue of the possible cyclical nature of banks’ behaviour using a large panel of Italian intermediaries over the period 1985 to 2002. Estimating both static and dynamic models, the article investigates whether loan loss provisions and non-performing loans show a cyclical pattern. The econometric results confirm that business cycle affects banks’ loan loss provisions and new bad debts. The impact of recessionary conditions is significant and long-lasting. Moreover, the empirical evidence provides some support for the income-smoothing hypothesis. The estimated relations may be employed to carry out stress tests to assess the effects of macroeconomic shocks on banks' balance sheets.

Acknowledgements

I am grateful to S. Laviola for continuous encouragement and useful suggestions; S. Grassi and A. Vezzulli for challenging discussions on panel data; M. Agostino, F. Cannata, J. Marcucci, A. Sironi, an anonymous referee and the participants at the CCBS Forum of Financial Stability Experts at the Bank of England and the seminar at the University of York for their comments. My special thanks go to P. N. Smith for his patient and thoughtful guidance and to A. Ozkan and P. Spencer for their valuable advice.

Notes

¹ Outlier banks are excluded by eliminating the observations in which the values of the bank-specific variables (except SIZE) are above and below the last and the first percentile respectively.

² During the 1990s the Italian banking system experienced an intense process of mergers and acquisitions. To deal with the impact of these operations on the sample it is assumed that they took place at the beginning of the sample period, consolidating the balance-sheet items of the banks involved.

³ The sample includes banks with total assets equal to at least 20 billion euros; it represents more than 65% of Italian banks’ consolidated total assets.

⁴ For simplicity only the unit root tests for the dependent variables are presented; tests are, however, carried out for all the regressors as well. For the microeconomic explanatory variables, except RISKST, the tests generally do not find significant evidence of the presence of a unit root. Interestingly, the standard Augmented Dickey–Fuller tests (ADF) performed on the aggregate time series fail to reject nonstationarity, thus confirming the advantage in terms of power of also exploiting cross-sectional information. Finally, it is worth noting that most of the macroeconomic series, even the first-differenced ones, seem to be nonstationary according to the ADF tests. This result is affected, however, by the low power of the test, especially in small samples and for near unit root processes (Enders, 1995). More powerful tests, such as Kwiatkowski et al. (1992) unit root tests for the null hypothesis of trend stationarity, fail to reject stationarity at the 5% level.

⁵ It is beyond the scope of this paper to set up a complete structural model, even though a system of simultaneous equations might be an appealing tool to describe the co-movements of the variables.

⁶ ECB (2001), Cavallo and Majnoni (2002), Valckx (2003) use a static model only; Salas and Saurina (2002) prefer the dynamic equation, while Pain (2003) estimates both the static and the dynamic specifications.

⁷ In Italy three-quarters of medium and long-term loans are secured; nearly 60% of such loans are granted against real collaterals.

⁸ The use of different measures of the spread (namely, the difference between loan and several inter-bank rates) leaves the results virtually unchanged.

⁹ Similar results were obtained using alternative interest rates such as those on the Treasury bills or the inter-bank rates.

¹⁰ In particular, no dependent variable Granger causes GDP growth at any conventional significance level.

¹¹ The Sargan test from the one-step estimator is not heteroscedasticity-consistent (see Arellano and Bond, 1991).

¹² Adjusted bad loans are those outstanding when a borrower is reported to the Central Credit Register: (a) as a bad debt by the only bank that disbursed credit; (b) as a bad debt by one bank and as having an overshoot by the only other bank exposed; (c) as a bad debt by one bank and the amount of the bad debt is at least 70% of its exposure towards the banking system or as having overshoots equal to or more than 10% of its total loans outstanding; (d) as a bad debt by at least two banks for amounts equal to or more than 10% of its total loans outstanding.

¹³ For a review, see Gambacorta and Mistrulli (2003). Data on other indicators of banks’ capitalization, such as the capital buffer (i.e. the capital above the regulatory minimum), are not available for the whole period under examination.

¹⁴ As for the LLP equation, a test was carried out for the stability of the coefficients. It failed to reject the null hypothesis of parameter constancy.