Three strikes and you re out: a simple econometric model of systemic banking crises David Aikman, Oliver Bush, Julia Giese, Rodrigo Guimarães and Hanno Stremmel Bank of England CEMLA/World Bank/Banca d Italia Conference on Macroprudential Policies, 20 th June 2013 The views given in this presentation are those of the authors and not necessarily those of the Bank of England or any other institution. 1
Outline Background: the UK s macroprudential framework Potential role for indicators Synthesising via statistical techniques A simple econometric model: three strikes and you re out (preliminary analysis) Policy 2
Role of the Financial Policy Committee (FPC) FPC set up to take a top-down macroprudential view Mandate to remove or reduce systemic risks with a view to enhancing and protecting the resilience of the UK financial system Secondary objective to support the economic policy of the Government, including its objectives for growth and employment Composition: 10 members, including four externals 3
FPC s powers General recommendations eg to HM Treasury over regulatory perimeter Directions over specific macroprudential tools Countercyclical capital buffer (CCB) Sectoral capital requirements for mortgages and intra-financial system exposures In future: leverage ratio; liquidity tool; margining requirements; LTV / LTIs? Comply-or-explain recommendations to PRA and FCA Structural risks 4
Core indicators Serve two broad purposes Internally: starting point for analysis, consistency of decisionmaking Externally: transparency, accountability, predictability But not meant as a substitute for judgment: limited knowledge about regime; trade-off between rules and discretion Which indicators? Basel III: Credit-GDP gap Complements to the credit-to-gdp gap 5
6
Synthesising the indicators Percentage of Variance Explained by Principal Components PC1 PC2 PC3 PC4 PC5 PC6 Individual 44 22 12 7 4 3 Cumulative 44 66 78 84 88 91 8 PC1 PC2 PC3 6 4 2 0-2 -4-6 -8 1969 1979 1989 1999 2009 7
Literature Kaminsky and Reinhart (1999), Demirgüç-Kunt and Detragiache (1998, 2000) Borio and Lowe (2002), Borio and White (2004), Schularick and Taylor (2012), Drehmann et al (2011) Barrell et al (2010), Kato et al (2010) Minsky (1972), Kindleberger (1987), Adrian and Boyarchenko (2012), Brunnermeier and Sannikov (2012), He and Krishnamurthy (2012) Giese et al (2013) 8
Data 25 countries (mixture of AEs and EMEs), annual data from 1980 to 2010 Crisis indicators (L&V and R&R) adjusted for post crisis bias identified by Bussière and Fratzscher (2006) Only 213 observations, with 9 crises (L&V)/11 crises (R&R) Credit data from BIS; GDP data from IMF/OECD; leverage data from Worldscope/Datastream; VIX proxy constructed from stock market indices from Datastream; corporate tax rate data from IMF/OECD 9
Credit-to-GDP gap Median 25th percentile Ratio 0.8 75th percentile Sample median 0.6 0.4 0.2 0-0.2-0.4-0.6-4 -3-2 -1 Crisis 1 2 3 4 Years 10
Accounting leverage Median 25th percentile % 6 75th percentile Sample median 5 4 3 2 1 0-4 -3-2 -1 Crisis 1 2 3 4 Years 11
Equity market volatility Median 25th percentile Index 25 75th percentile Sample median 20 15 10 5 0-4 -3-2 -1 Crisis 1 2 3 4 Years 12
Three strikes and you re out Country Signal year Crisis? Credit 1 Leverage 1 VIX 1 Country Signal year Crisis? Credit 1 Leverage 1 VIX 1 Three full strikes Other crises Denmark 2004-05 2008 92 16 15 Australia 1986 1989 76 39 37 Sweden 1989 1991 83 22 24 Belgium 2005 2008 18 20 4 Canada 1980 1983 44 9 50 Two full strikes and one half strike Greece 2005 2008 96 52 27 Australia 2006 n/a 85 49 25 Switzerland 2005 2008 32 5 4 Canada 1997 n/a 59 24 20 UK 2004 2007 36 46 9 Denmark 1990 n/a 81 49 19 1 Percentiles of the sample distribution France 2004-05 2008 59 12 16 Ireland 1992 n/a 90 42 19 Ireland 1995 n/a 90 48 5 Italy 2005 2008 78 48 8 Portugal 2000 n/a 99 17 47 Portugal 2005 2008 89 36 2 Spain 1999 n/a 81 25 48 Spain 2005 2008 89 36 2 Sweden 2005 2008 63 25 17 1 Percentiles of the sample distribution 13
Benchmark model Estimation method Probit coef/se Credit gap 2.43** (1.07) Change in credit gap 7.31** (3.59) Leverage -0.64** (0.30) Change in leverage -0.89* (0.54) VIX -0.57*** (0.18) Change in VIX -0.23 (0.14) Constant 3.86** (1.65) Number of observations 213 Log-Likelihood -16.36 Notes: all variables are third lags; *** p<0.01, ** p<0.05, * p<0.1 14
Signal ratio 0.00 0.25 0.50 0.75 1.00 ROC 1: credit-to-gdp gap model 0.00 0.25 0.50 0.75 1.00 Noise ratio Noise ratio at signal-maximising cutoff 0.71 Signal ratio at noise-minimising cutoff 0.22 AUROC 0.71 15
Signal ratio 0.00 0.25 0.50 0.75 1.00 ROC 2: credit gap and benchmark models 0.00 0.25 0.50 0.75 1.00 Noise ratio Benchmark Credit-to-GDP gap Model CG BM Noise ratio at signal-maximising cutoff 0.71 0.06 Signal ratio at noise-minimising cutoff 0.22 0.11 AUROC 0.71 0.98 16
Signal ratio 0.00 0.25 0.50 0.75 1.00 ROC 3: pre-2006 model 0.00 0.25 0.50 0.75 1.00 Noise ratio Noise ratio at signal-maximising cutoff 0.35 Signal ratio at noise-minimising cutoff 0.00 AUROC 0.82 17
Signal ratio 0.00 0.25 0.50 0.75 1.00 ROC 4: out of sample fit of pre-2006 model 0.00 0.25 0.50 0.75 1.00 Noise ratio Noise ratio at signal-maximising cutoff 0.71 Signal ratio at noise-minimising cutoff 0.57 AUROC 0.83 18
Causation? Estimation method IV coef/se Credit gap 0.99*** (0.37) Leverage -0.53*** (0.06) Change in VIX -0.09*** (0.03) Constant 1.71*** (0.41) Null: exogeneity 2.74*** (0.31) Number of observations 142 Log-Likelihood -269.42 Notes: all variables are third lags; *** p<0.01, ** p<0.05, * p<0.1; the instrument is corporation tax 19
Straw man policy rule for the UK* Actual leverage ratio Leverage ratio, p=0.01 Leverage ratio, p=0.001 9 8 7 6 5 4 3 2 1920 1940 1960 1980 2000 * Don t take this too seriously (especially the level calibration) 20
Near term: Improve coverage Remove misspecification Use foreign assets Outstanding questions Medium term: More historical analysis (like Schularick and Taylor) Tailor for EMEs (FX angle) VAR, including impact on GDP Wish list: A better empirical understanding of the channels through which the quantity and price of lending affect financial stability A better theoretical understanding 21
Summary UK s macroprudential framework Potential role for indicators Three danger signs for macroprudential policymakers A potentially useful econometric model A role for indicators in calibrating macroprudential policy? 22