ABSTRACT
This paper examines cyclical behaviour of banks’ systemic risk contribution and exposure. Using a panel of 787 banks from country members of the Organization for Economic Co-operation and Development and the European Union covering the period 2000–2017, we document that both systemic risk contribution and exposure are positively related to business cycle. Systemic risk starts to accumulate in the financial sector during periods of boom when the output gap is positive. Furthermore, during periods of robust economic growth, the level of credit tends to increase dramatically, going hand in hand with asset and property prices developments. We also find that contribution and exposure to system-wide distress move procyclically during credit and house cycles, meaning that during upturns in credit and house cycles bank interconnectedness increases, but tend to fall during the downturns. However, individual risk of the banks evolves countercyclically during business and financial cycles.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 Berger, Molyneux, and Wilson (Citation2020) note two potential links between finance and the real economy: one in which financial institutions and markets have a strong negative impact on the real economy due to financial imbalances that build-up during crises, and one in which banks, other financial institutions and markets have strong positive effects on the real economy because the reduction in banks and other financial services causes economic deterioration. However, the general conclusion is that the latter effect prevails.
2 Throughout this paper, business and economic cycle terms will be used interchangeably.
3 For a recent survey on banks and the real economy link, see Berger, Molyneux, and Wilson (Citation2020).
4 As pointed-out by Berger, Molyneux, and Wilson (Citation2020), one of the issues in the analysis of the financial sector-real economy nexus are causal relationships in which changes in the financial sector are driven by the changes in the real economy. One way of addressing this problem is to employ quasi-natural experiments with exogenous shocks to bank output.
5 See Section 2.2 for details.
6 Bisias et al. (Citation2012) and Silva, Kimura, and Sobreiro (Citation2017) provide two extensive surveys of systemic risk measures. Also, Giglio, Kelly, and Pruitt (Citation2016) study 19 systemic risk measures and their impact on real economic activity, and their ability to predict quantiles of future macroeconomic shocks.
7 For recent application of ∆CoVaR see, among others, Andrieş, Nistor, and Sprincean (Citation2020), Bakkar, Rugemintwari, and Tarazi (Citation2020), and Ben Ameur et al. (Citation2020).
8 Following Adrian and Brunnermeier (Citation2016), all our systemic risk indicators are estimated for a 5% quantile.
9 For recent application of MES see, among others, De Sola Perea et al. (Citation2019), Oordt and Van, Zhou (Citation2019), and Bakkar, Rugemintwari, and Tarazi (Citation2020).
10 Consider the following example. Let’s say that one has a time-series that starts in 2000:Q1 and wants to extract the cycle for the period from 2005:Q1 to 2017:Q4. Applying the one-sided HP filter, the cycle at 2005:Q1 is first calculated employing the data from 2000:Q1 to 2005:Q1, the cycle at 2005:Q2 is calculated from the data from 2000:Q1 to 2005:Q2, and so on. In this case, when new data is released for 2018:Q1, the cycle and trend calculations are not altered as long as the past data is not revised. With the two-sided HP filter, however, the cycle is calculated using the entire time-series, and when new data is available, the past trend and cycles values have to be replaced with the new ones because of the change in the length of the time-series. Therefore, the one-sided HP filter is more robust comparing to its two-sided analogue.
11 While the Kalman filter estimates give the one-sided HP filter, the Kalman smoother estimates give the two-sided HP filter (see Hamilton Citation2017).
12 Other studies (e.g., Basel Committee on Banking Supervision Citation2010; Drehmann, Borio, and Tsatsaronis Citation2012) use a smoothing parameter of 400,000 in order to extract financial cycles. Our results remain robust for both systemic risk contribution and exposure after employing the one-sided HP filter with lambda set at 400,000.
13 Where data is not already seasonally adjusted, we use the Census X-13 procedure.
14 Because some variables are already computed as indexes, we re-scale them to 2010:Q4 = 100.
15 We apply the PCA to demeaned data and pick only the first principal component, which is our synthetic index.
16 According to World Bank classification criteria, all OECD and EU members are developed economies, classified as either upper-middle-income or high-income.
17 Drehman et al. (2012) point out that the average length of financial cycle for seven developed countries since 1960 is 16 years. Schüler et al. (2015) find that the average financial cycle length for 13 European countries is 7.2 years over 1970–2013 period. Hence, given the structure of our dataset (18 years), there is enough information to perform a study on financial variables and their cyclical behaviour.