Specialisation in mortgage risk under Basel II

Transcription

1 Specialisation in mortgage risk under Basel II Matteo Benetton 1, Peter Eckley 2, Nicola Garbarino 2, Liam Kirwin 2, Georgia Latsi 3 1 London School of Economics 2 Bank of England 3 4-most EBA Research Workshop, November 2016 The views presented here are those of the authors and do not necessarily reflect the views of the Bank of England, the Monetary Policy Committee, the Financial Policy Committee, or the PRA.

2 Motivation Residential mortgage market Epicentre of financial crisis (Mian and Sufi, 2015) Large share of total bank lending (Jordà et al, 2016)

3 Motivation Residential mortgage market Epicentre of financial crisis (Mian and Sufi, 2015) Large share of total bank lending (Jordà et al, 2016) Methodology-driven heterogeneity in capital requirements BCBS (2016)

4 Motivation Residential mortgage market Epicentre of financial crisis (Mian and Sufi, 2015) Large share of total bank lending (Jordà et al, 2016) Methodology-driven heterogeneity in capital requirements BCBS (2016) Specialization distribution of risk Current debate on reforms of Basel II-III

5 Risk weight (as on 31 Dec 2015) Heterogeneity in risk weights - UK mortgages 60% IRB distribution IRB average SA 40% 20% 0% Loan-to-value (LTV) band K min = RWA KReq Two approaches: models (IRB) and standardised (SA)

6 Do regulatory risk models affect market outcomes? Mechanism: Similar risk, different methodologies capital requirements specialisation Theory: Repullo & Suarez (2004) Empirics: Behn et al (2016a & 2016b) for corporate lending in Germany

7 This paper Identification challenge: isolating effect of methodology one borrower, many lenders (Khwaja Mian, 2008) mortgages: one borrower, one lender?

8 This paper Identification challenge: isolating effect of methodology one borrower, many lenders (Khwaja Mian, 2008) mortgages: one borrower, one lender? Micro-data on 7 million UK mortgages ( ) Two identification strategies based on: 1. Quasi-experimental variation from switch to Basel II 2. New LTV-level risk weight data for post-basel II

9 Outline Identification Results Policy

10 Outline Identification Results Policy

11 Switch to Basel II as a quasi-experiment Switch to Basel II as an exogenous supply-side shock Selection into IRB group approx. exogenous w.r.t. risk High costs of IRB adoption (CMA, 2015) Mainly driven by firm size (economies of scale)

12 Average risk weight Risk weights variation Basel I 50% for ALL loans Basel II-III (2008 onwards) Firms choose SA or IRB 50% LTV 40% 30% 20% IRB-SA gap was larger at lower LTV High Low High SA IRB 10% Low SA IRB LTV>75 LTV 75

13 Average interest rate Mortgage price variation Basel I 50% for ALL loans Basel II-III (2008 onwards) Firms choose SA or IRB 6% Mortgage rates fall for both SA and IRB firms 4% 2% IRB-SA gap is larger at lower LTV LTV High High Low Low SA IRB SA IRB LTV>75 LTV<=75 Benchmark rate

14 Diff. between IRB & SA avg. rates Mortgage price variation (IRB-SA price difference) Basel I 50% for ALL loans Basel II-III (2008 onwards) Firms choose SA or IRB 0.0 price IRB > price SA price IRB < price SA Price gap opens up at low LTV High LTV Low LTV LTV>75 LTV 75

15 Triple difference model ( ) Rate ibst =

16 Triple difference model ( ) Rate ibst = Common impact δ 1 BaselII t +

17 Triple difference model ( ) Rate ibst = Common impact δ 1 BaselII t + Differential impact δ 12 BaselII t IRB }{{ b + δ } 13 BaselII t LowLTV s + }{{} for IRB firms low LTV

18 Triple difference model ( ) Rate ibst = Common impact δ 1 BaselII t + Differential impact δ 12 BaselII t IRB }{{ b + δ } 13 BaselII t LowLTV s + }{{} for IRB firms low LTV Structural differences + δ 3 LowLTV s + δ 2 IRB }{{ b } for IRB } {{ } for low LTV + δ 23 IRB b LowLTV }{{} s for IRB firms at low LTV

19 Triple difference model ( ) Rate ibst = Common impact δ 1 BaselII t + Differential impact δ 12 BaselII t IRB }{{ b + δ } 13 BaselII t LowLTV s + }{{} for IRB firms low LTV Structural differences + δ 3 LowLTV s + δ 2 IRB }{{ b } for IRB } {{ } for low LTV + δ 23 IRB b LowLTV }{{} s for IRB firms at low LTV DDD: Differential impact for IRB firms at low LTV δ 123 BaselII t IRB b LowLTV s

20 Triple difference model ( ) Rate ibst = Common impact δ 1 BaselII t + Differential impact δ 12 BaselII t IRB }{{ b + δ } 13 BaselII t LowLTV s + }{{} for IRB firms low LTV Structural differences + δ 3 LowLTV s + δ 2 IRB }{{ b } for IRB } {{ } for low LTV + δ 23 IRB b LowLTV }{{} s for IRB firms at low LTV DDD: Differential impact for IRB firms at low LTV δ 123 BaselII t IRB b LowLTV s +αcontrols ibst + ɛ ibst

21 Triple difference model ( ) Rate ibst = Common impact δ 1 BaselII t + Differential impact δ 12 BaselII t IRB }{{ b + δ } 13 BaselII t LowLTV s + }{{} for IRB firms low LTV Structural differences + δ 3 LowLTV s + δ 2 IRB }{{ b } for IRB } {{ } for low LTV + δ 23 IRB b LowLTV }{{} s for IRB firms at low LTV Hypotheses: DDD: Differential impact for IRB firms at low LTV δ 123 BaselII t IRB b LowLTV s +αcontrols ibst + ɛ ibst 1. Interest rates: δ 123 < 0 2. Portfolio shares: δ 123 > 0

22 Risk weights pass-through model ( ) Rate ibst =

23 Risk weights pass-through model ( ) Rate ibst = pairwise-interacted { fixed effects }} { γ bt }{{} bank-time + γ bs }{{} bank-ltv + γ st }{{} LTV-time

24 Risk weights pass-through model ( ) Rate ibst = pairwise-interacted { fixed effects }} { γ bt }{{} bank-time + γ bs }{{} bank-ltv + γ st }{{} LTV-time + risk weights by bank, LTV and time βrw bst

25 Risk weights pass-through model ( ) Rate ibst = pairwise-interacted { fixed effects }} { γ bt }{{} bank-time + γ bs }{{} bank-ltv + γ st }{{} LTV-time + risk weights by bank, LTV and time βrw bst + αcontrols ibst + ε ibst

26 Risk weights pass-through model ( ) Rate ibst = pairwise-interacted { fixed effects }} { γ bt }{{} bank-time + γ bs }{{} bank-ltv + γ st }{{} LTV-time + risk weights by bank, LTV and time βrw bst + αcontrols ibst + ε ibst Hypothesis: β > 0

27 Risk weights pass-through model ( ) Rate ibst = pairwise-interacted { fixed effects }} { γ bt }{{} bank-time + γ bs }{{} bank-ltv + γ st }{{} LTV-time + risk weights by bank, LTV and time βrw bst + αcontrols ibst + ε ibst Hypothesis: β > 0 Also with RW bst CapReq bt

28 Outline Identification Results Policy

29 Triple difference model Results ( ) Benchmark LTV threshold Panel A: interest ibst DDD bst (0.088) (0.083) (0.090) Adjust R Observations Panel B: portfolio share bst DDD bst (0.008) (0.008) (0.009) Adjust R Observations IRB prices fall by an additional 32bp at low LTV (vs. high) IRB portfolio share of low LTV increases by 12pp

30 Risk weights model Results ( ) Dependent variable: interest ibst (1) (2) RW bst RW bst Cap req bt (0.003) (0.018) Fixed effects: Lender-quarter Yes Yes Lender-segment Yes Yes Segment-quarter Yes Yes Individual controls Yes Yes Adjusted R Observations pp RW 1bp Rates LTV 50: 30pp RW 30bp Rates

31 Outline Identification Results Policy

32 Main results: impact of Basel II internal models Basel II: specialisation of smaller firms (SA) in high LTV Lower systemic importance But less sophisticated risk management Within Basel II: 1pp RW 1bp Rates Below 75% LTV, implies 20-30bp price advantage Jump from 10th to 1st in best buy tables (at 75% LTV)

33 Risk weight Risk weight Basel: reduction in variability of models and in IRB-SA gap 0.6 IRB SA 0.6 IRB SA (1) (2) Loan-to-value (LTV) band Loan-to-value (LTV) band Options: (1) more risk sensitive SA, (2) floors on IRB

34 Appendix

35 Alternative channels Triple difference model Basel II t Low LTV b Dependent variable: interest ilbt (1) (2) (3) (4) IRB l (0.088) (0.086) Low buffer l (0.090) (0.092) Funding shock l (0.118) Adjusted R Observations 6,931,773 6,931,773 6,931,773 5,032,264 Exposure to the crisis (low capital buffer) Effect of the crisis (high funding cost)

36 Heterogeneous effects Risk weights model Dependent variable: interest ibst Capital buffer LTV High Low High Low (1) (2) (3) (4) RW bst Fixed effects: (0.003) (0.004) (0.005) (0.003) Lender-quarter Yes Yes Yes Yes Lender-segment Yes Yes Yes Yes Segment-quarter Yes Yes Yes Yes Individual controls Yes Yes Yes Yes Adjusted R Observations Pass-through driven by lenders with low buffers Similar at high and low LTV

37 Data Product Sales Database: UK residential mortgages Rates, product characteristics, property and loan values, borrower characteristics At origination c. 14 million loans CMA/PRA survey Risk weights by loan-to-value band 17 solo entities on IRB

38 Two complementary identification strategies Triple difference RW pass-through Period Risk weight data No Yes Variation only IRB v SA also IRB v IRB Focus Regime change IRB models

39 Portfolio shares 100% Portfolio share at low LTV ( 75%) SA IRB 75% 50% 25% 0% Pre-2008 Post-2008 Pre-2008 Post-2008

40 Market shares 100% Market shares in each segment High LTV (>75%) Low LTV (<=75%) 90% 80% 70% 60% Pre-2008 Post-2008 Pre-2008 Post-2008 SA IRB