Lenders on the storm of wholesale funding shocks: saved by the central bank?

ABSTRACT

We provide empirical evidence on banks’ responses to shocks in the wholesale funding market, using data of 181 euro area banks over the period from August 2007 to June 2013. Responses to funding liquidity shocks for both banks’ lending volumes and loan rates, to households and corporates, are analysed in a panel VAR framework. We thereby distinguish banks by country, extent of Eurosystem borrowing, bank size and capitalization. The results show that shocks in the securities and interbank markets have significant effects on loan rates and credit supply, particularly of banks in stressed countries of the periphery. The results also suggest that central bank liquidity has mitigated this effect on lending volumes. Lending to nonfinancial corporations is more sensitive to wholesale funding shocks than lending to households. Lending volumes of large banks that are typically more dependent on wholesale funding and banks with large exposure to sovereign bonds show stronger responses to wholesale funding shocks.

KEYWORDS:

• Banking
• financial intermediation
• financial crisis

JEL CLASSIFICATION:

• G21
• G32

Acknowlegment

We thank two anonymous referees, Hans Degryse, Jakob de Haan, Vasileios Georgakopoulos, Wim Goes, Cornelia Holthausen, Marcus Pramor, Peter van Els, seminar participants at DNB and participants of the ECB/MPC research workshop of the IBSI taskforce (Paris, 2014), the New York and Boston Federal Reserve Bank conference on Risks of Wholesale Funding (New York, 2014), and the World Finance Conference (Buenos Aires, 2015) for valuable comments and advice. Views expressed are those of the authors and do not necessarily reflect official positions of De Nederlandsche Bank or the European Central Bank.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ We experimented with different lag lengths and it turned out that the number of lags did not make a significant difference in the impulse responses.

² For more details, we refer to Love and Zicchino (2006), whose Stata code we gratefully used for the estimation.

³ The use of real GDP growth as a control variable for loan demand is common in, e.g., the extensive empirical literature on the credit channel (e.g. Kashyap and Stein 2000; De Haan 2003, where bank loan supply effects are examined).

⁴ However, for both indicators, it can be argued that they are crude in that they reflect demand conditions in a particular country, so that biases may arise if banks lend to different firms or households within that country.

⁵ According to Levin–Lin–Chu unit-root test (excluding panel means and time trends), unit roots for all series can be rejected. Results are available on request.

⁶ Access to funding may depend on banks’ risk management strategies as well, but most likely with a lag.

⁷ An extreme change is defined as the 0.1% tail of the distribution of changes for all banks. Defined this way, a major change is a month-on-month increase or decrease of the total balance sheet larger than 75%.

⁸ These countries are Austria, Belgium, Cyprus, Germany, Estonia, Spain, Finland, France, Greece, Ireland, Italy, Luxembourg, Malta, The Netherlands, Portugal, Slovenia and Slovakia.

⁹ Summary statistics for the balance sheet items of the subsamples are presented in Appendix 2.

¹⁰ Data on borrowing from the Eurosystem, taken from another internal ECB data source, is used to exclude the Eurosystem borrowing from total MFI borrowing.

¹¹ The data set contains banks’ main assets, not all assets.

¹² Summary statistics for the model variables of the subsamples are presented in Appendix 3, Table A2.

¹³ The third sample split allows us to analyse the effect of central bank borrowing, which could not be included as a variable in the p-VAR models, because many banks do not borrow from the central bank (complicating log transformation and model estimation as inclusion of central bank borrowing would require dropping another variable from the model to preserve a minimum number of degrees of freedom).

¹⁴ The statistical significance is tested by comparing the confidence bands of the impulse responses of the two subsamples (see Figure 5). If the confidence bands overlap for all lags, the impulse responses are not significantly different.

¹⁵ Of the banks in stressed countries, 83% is a high central bank borrower, compared to 33% of all banks in nonstressed countries, implying that the subsamples of banks in stressed versus nonstressed countries and high versus low central bank borrowers are related but not identical.

¹⁶ The impulse–response functions of a parsimonious panel-VAR model including MFI, S and CB as endogenous variables confirm that Central Bank borrowing reacts significantly and negatively to a shock in wholesale funding, implying that a fall in either MFI or S increases the demand for central bank funding.