Liquidity stress-tester: do Basel III and unconventional monetary policy work?

Abstract

This article presents a macro stress-testing model for liquidity risks of banks, incorporating the proposed Basel III liquidity regulation, unconventional monetary policy and credit supply effects. First and second round (feedback) effects of shocks are simulated by a Monte Carlo approach. Banks react according to the Basel III standards, endogenizing liquidity risk. The model shows how banks’ reactions interact with extended refinancing operations and asset purchases by the central bank. The results indicate that Basel III limits liquidity tail risk, in particular if it leads to a higher quality of liquid asset holdings. The flip side of increased bond holdings is that monetary policy conducted through asset purchases gets more influence on banks relative to refinancing operations.

Keywords:

• banking
• financial stability
• stress-tests
• liquidity risk

JEL Classification::

• C15
• E44
• G21
• G32

Acknowledgements

The author would like to thank colleagues from DNB for their valuable comments on earlier versions of this article.

Notes

¹ For most items the DNB and Basel Committee classifications are similar, in a limited number of cases we re-classify the exposures from the DNB report to the Basel classification on the basis of supervisory expert judgement. For some items this leads to an imperfect mapping and imprecise calculations of the LCR and NSFR. In particular, this holds for internal securitizations (classified as bond holding in the DNB report, but not classified as liquid assets in the LCR) and deposits of corporates (no clear separation between Small and Medium Enterprises (SME) and non-SME in the DNB report, while the Basel classification takes this difference into account). Furthermore, the total item values in the liquidity report do not match precisely with the item values on the balance sheet, which could distort the calculation of the NSFR, which for some items can be based on balance sheet information. In cases where the mapping leads to a serious distortion of the liquidity ratios (e.g. for banks with large securitized asset holdings) we leave those banks out of the simulations. Moreover, the scenarios are presented in terms of deviations from baseline (i.e. from LCR_t0 and NSFR_t0), which reduces the influence of the imperfect mapping.

² The table in the Appendix includes the liquidity value factors of assets; the haircut is equal to 100% minus that factor. In the model simulations we apply the weighting factors according to the decision of the overseeing body (Governors and Heads of Supervision (GHOS)) of the Basel Committee in July 2010 (BCBS, 2010); these factors are slightly different for some balance sheet items compared to the weighting factors in Appendix 1, which originate from BCBS (2009).

³ The threshold should reflect a substantial decline of the LCR; the precise threshold value could be based on supervisory intervention levels.

⁴ In the model, liquidity lost on the asset side determines the mitigating actions taken with assets and liquidity lost on the liability side determines the measures taken by liabilities. The model imposes the restriction that a bank does not expand its total balance sheet by the mitigating actions; i.e. the sum of elements in I_i ,_t2 is smaller or equal to the sum of elements in I_i ,_t0. This is applied by reducing the assets and liabilities proportionally if the balance sheet total at t2 would exceed the total at t0.

⁵ A restriction in the model is that a bank with no exposure in a certain market does not enter this market.

⁶ If ω = 1 (the downside restriction for ω), the mitigating reaction of a bank will not be counteracted by adverse reputational effects and will by definition improve a banks’ liquidity position.

⁷ The x-axis of Fig. 5(a) and (b) scale the LCR as an index with LCR _t0 = 100.

⁸ The SD of LCR _t3 is smaller than the SD of LCR _t1.

⁹ By running the model with a sample of banks in one country (i.e. Dutch banks) it is assumed that the second round effects of scenarios are caused by shocks in the local banking sector. The reactions of banks in other countries are not taken into account.

¹⁰ Note that during the peak of the crisis, in October 2008, ω was around 5, both based on corporate bond spreads and implied stock price volatility. In that respect the presented model results of the replicated credit scenario may be an underestimation of the second round effects.

¹¹ If the denominator of the LCR equals 0 because outflow = inflow, then the LCR is fixed at 100%.

¹² In the wholesale and retail run scenarios, the impact on aggregate credit supply is negligible, since the model simulations indicate that large banks do not breach the reaction threshold and hence do not adjust balance sheets.

¹³ The increase of central bank borrowing is simulated by assuming that wholesale funding is halved and that this loss of funding is made up by central bank borrowing. The exit is simulated by assuming that banks substitute the additional central bank funding again with wholesale funding, or in an alternative simulation, with retail savings.

¹⁴ This haircut (w _{sim 2}) results from the scenario simulation in the Section ‘Credit crisis scenario’; it is the average haircut on bonds included in the denominator of the LCR, weighted by the total holdings of these bonds. Compared to the weighted average haircut on bonds in the initial situation (w _LCR ≈ 9%) – which is assumed to be a three SD shock (as given in the Section ‘First round effects’) – the 26% average price fall is equal to a nine SD shock.

¹⁵ The assumptions are applied by changing the elements in vectors I_{LCR ,t0}, I_{NSFR_ST ,t0} and I_{NSFR_LT ,t0}. The first assumption implies that liquid asset holdings – the numerator in the LCR – are 25% higher and that this amount proportionally reduces other assets. The second assumption implies that high quality government bond holdings are 25% higher, while level 2 bonds are reduced by the same amount. Under the third condition, debt securities issued are 25% lower, while retail savings are increased by the same amount. The fourth assumption implies that debt securities maturing within 1 year are reduced by 25% and longer term securities increased by the same amount. The fifth condition implies that unsecured funding is reduced by 25% and secured borrowing increased by the same amount.