When gambling for resurrection is too risky

Transcription

1 When gambling for resurrection is too risky Divya Kirti IMF MFM Winter Meeting March 2017 The views expressed herein are my own and should not be attributed to the IMF, its Executive Board, or its management. 1/ 35

2 What drives risk taking by financial institutions? Workhorse corporate finance model: risk shifting Particularly applied to finance, regulation Jensen Meckling (1976), Stiglitz Weiss (1981), Rajan (2006), Acharya Viswanathan (2011), Helmann Murdock Stiglitz (2000), Rochet (2008), Plantin Rochet (2007) What happens when really in trouble? Risk shifting framework makes clear prediction: gamble for resurrection 2/ 35

3 Can gambling for resurrection be too risky? Study risk taking by US financial institutions during the crisis Compare life insurers hit hard by crisis to those hit less hard Instead of doubling down, insurers hit hard pulled back Reduced credit risk Reduced interest-rate risk Maybe franchise value can make gambling for resurrection too risky 3/ 35

4 What about regulation? Not just effectively tighter capital requirements: risk reduction within assets with identical regulatory treatment State level US insurance regulation: coordinated moral suasion unlikely Focus on insurance helps sharpen results and interpretation Results are broader than insurance: same approach yields similar results for banks 4/ 35

5 Literature Risk shifting In general and in finance: Jensen Meckling (1976), Stiglitz Weiss (1981), Rajan (2006), Acharya Viswanathan (2011), Helmann Murdock Stiglitz (2000), Rochet (2008), Plantin Rochet (2007) Empirical literature: Becker Ivashina (2015), IMF (2016), Foley-Fisher Narajabad Verani (2016), Dell Ariccia Laeven Suarez (2017), Plosser Santos (2014), Hong (2017) Insurance Risks, strategy: Domanski Shin Sushko (2015), Koijen ogo (2016), Chodorow-Reich Ghent Haddad (2016) Regulation, response: Becker Ivashina (2015), Koijen ogo (2016), Ellul Jotikasthira Lundblad Wang (2015), Becker Opp (2014), Merrill Nadauld Strahan (2014), Merrill Nadauld Stulz Sherland (2014), Chodorow-Reich (2014) Context: Berends McMenamin Plestis Rosen (2013), IMF (2016), S&P (2014), Poterba (1997), Foley-Fisher Narajabad Verani (2015), Briys de Varenne (1997), Paulson Plestis Rosen McMenamin Mohey-Deen (2014), NAIC (2013) Other connections: derivatives, crisis frameworks, credit supply 5/ 35

6 Approach and results 6/ 35

7 Data Use data for 2005Q1-2014Q4 Rich data on life insurers assets Quarterly position-level data on all bond holdings Daily position-level data on all bond transactions Contract-level data on all interest-rate swap positions/transactions Caveats Detail is for general accounts (backing term/life rather than VA) Less detail on liabilities, use simple duration assumption Look at pro-forma consolidated entities 7/ 35

8 Approach Focus on large insurers Top two asset deciles by quarter Large insurers account for 90% by general account bonds Measures of risk taking on two dimensions Interest-rate risk: gap between liability DV01 and asset DV01 (bonds and derivatives), as a fraction of liability DV01, in percentage points Credit risk: average TM of all bonds purchased in quarter, in basis points Divide sample into three periods Pre-crisis: 2005Q1-2007Q2 (cutoff following Becker Ivashina 2015) Crisis: 2007Q3-2010Q4 Post-crisis: 2011Q1-2014Q4 8/ 35

9 Identify insurers hit hard at parent level Measure crisis experience and level of financial health during crisis at parent level Look at dividends, equity/assets and equity issuance and construct insurer-level flag Flag as hit hard if dividend or equity growth below 10th percentile in , or issued equity in This flags 24 out of 50 large insurers with matched parents 9/ 35

10 Insurers hit hard pulled back Risk taking based on crisis experience for large insurers with matched parent Interest-rate risk Credit Risk Net DV01 Gap Net DV01 Gap TM TM Crisis Hit Flag 6.85 (1.06) (1.15) Crisis Hit Flag 2007Q3-2010Q4 Crisis Hit Flag 2011Q1-2014Q (-2.63) (-1.26) (-2.45) (-2.40) (-2.74) (-2.20) (-1.08) (-0.89) Log(Assets) 0.29 (0.12) (-0.05) (-2.85) (-0.36) Quarter FE Insurer FE SE clustered by R 2 Insurer-Quarters Insurers N I,Q , I,Q , N I,Q , I,Q , Crisis hit flag: insurer-level dummy for severe dividend cuts, reduction in equity/assets ratio or equity issuance during crisis ( ). SE double clustered, t-stats in parentheses. 10/ 35

11 Insurers hit hard reduced net interest-rate exposure Risk taking based on crisis experience for large insurers with matched parent 11/ 35

12 Insurers hit hard took less credit risk Risk taking based on crisis experience for large insurers with matched parent 12/ 35

13 Franchise value, not risk shifting? Results suggest a type of anti-jensen Meckling effect Risk shifting framework predicts that insurers hit hard should have increased risk taking Insurers hit hard pulled back, relative to insurers hit less hard Consistent with value from ensuring survival Financial institutions depend on trust, which is unlikely to survive bankruptcy Chodorow-Reich Ghent Haddad (2016): maybe franchise value relates to existing assets Other versions: private benefits from management/employment 13/ 35

14 What about capital requirements? Perhaps capital requirements became effectively tighter for insurers hit hard Key advantage of insurance data: can examine risk taking within regulatory buckets Insurers hit hard reduced risk within assets with identical regulatory treatment 14/ 35

15 Risk reduction within regulatory buckets Approach: follow Becker Ivashina (2015) within large insurers Share of newly issued bonds bought by insurers hit hard: NAIC 1 (AAA-A) corporate bonds H1 2007H TM Hit hard fraction Hit hard fraction Hit hard fraction 6.79 (0.79) (-2.54) (-3.34) Duration (-0.77) 2.42 (2.86) 2.20 (2.86) Tot insurer purchases 1.44 (0.29) 5.75 (2.49) 7.36 (5.87) Month FE Issuer FE SE clustered by R 2 Issues Issuers Iss,M Iss,M Iss,M , / 35

16 What about moral suasion? US insurers are regulated at the state level Coordination on moral suasion is unlikely Top 10 states only cover three quarters of assets Working Group for nationally significant insurers can apply peer pressure on lead regulator (NAIC 2013) Literature suggests moral suasion is unlikely Koijen ogo (2016) document regulatory arbitrage across states Ellul et al (2015) show differences in regulatory implementation across states Capital requirements were substantially lowered for MBS (Becker Opp 2014, Merrill Nadauld Stulz Sherland 2014) 16/ 35

17 Additional results Insurers hit hard took more risk ex-ante Details Bond level analysis shows risk reduction within: NAIC 2 (BBB) corporate bonds Details High yield corporate bonds Details All privately issued investment grade bonds Details Role of interest-rate derivatives Details Credit risk vs. duration Details All large insurers Details 17/ 35

18 Results are broader than insurance Same approach yields similar results for banks Look at dividends, equity/assets and construct bank-level flag Large banks hit hard pulled back relative to banks hit less hard Lower credit growth Less interest rate risk post-crisis Regressions and figures Helpful to compare insurers and banks Different exposure to interest rates: really about risk Maybe insurers bailout probability not high enough? Franchise value about new business or existing assets? 18/ 35

19 Discussion 19/ 35

20 Franchise value can make gambling for resurrection risky Insurers hit hard by crisis pulled back Reduced net interest-rate exposure Took less credit risk Insurance setting addresses concerns about regulation Risk reduction within regulatory categories State level regulation makes moral suasion unlikely Results are broader, apply to banks as well 20/ 35

21 What do we learn about risk shifting? Literature shows more risk taking in response to some shocks e.g. Becker Ivashina (2015), IMF (2016), Foley-Fisher Narajabad Verani (2016), Dell Ariccia Laeven Suarez (2017), Plosser Santos (2014), Hong (2017) How should this be reconciled with an anti-jensen Meckling effect? Two possibilities Workhorse corporate finance model fails locally (in the neighborhood of what happened in crisis) Something else explains increased risk documented by literature (e.g. fixed return targets, earnings and dividend smoothing) 21/ 35

22 Implications for macro-prudential policy Macro-prudential perspective: might want to loosen capital requirements in a crisis What if capital requirements aren t the binding constraint? If franchise value matters, want to make survival clear: stress tests, sufficient recapitalization 22/ 35

23 Appendix 23/ 35

24 Insurers hit hard bought more MBS in Net MBS purchases as a fraction of total net purchases from for insurers hit hard Notes: Net MBS purchases adjusted to exclude prepayments. Back to main presentation 24/ 35

25 ... than insurers not hit hard Net MBS purchases as a fraction of total net purchases from for insurers not hit hard Notes: Net MBS purchases adjusted to exclude prepayments. Back to main presentation 25/ 35

26 MBS bought in 2005 were subsequently downgraded Corporate bonds bought in 2005 were much less likely to be downgraded Private structured bonds bought by insurers hit hard in 2005 by rating Notes: NAIC 1 is omitted category. Restricted to private structured bonds bought in 2005, weighted by purchases in Back to main presentation 26/ 35

27 Bond level analysis: NAIC 2 corporate bonds Share of newly issued bonds bought by insurers hit hard: NAIC 2 (BBB) corporate bonds TM H1 2007H Hit hard fraction Hit hard fraction Hit hard fraction 3.53 (0.46) 2.66 (1.14) (-2.66) Duration 0.11 (0.10) 0.85 (1.21) 1.46 (2.41) Tot insurer purchases (-0.09) 3.96 (2.41) 6.88 (5.46) Month FE Issuer FE SE clustered by R 2 Issues Issuers Iss,M Iss,M Iss,M , Back to main presentation 27/ 35

28 Bond level analysis: H corporate bonds Share of newly issued bonds bought by insurers hit hard: High yield (not NAIC 1 or 2) corporate bonds H1 2007H TM Hit hard fraction Hit hard fraction Hit hard fraction (-0.21) (-0.37) (-4.00) Duration (-0.54) 1.19 (2.02) 1.55 (3.39) Tot insurer purchases 5.50 (1.15) 4.55 (3.77) 6.77 (7.49) Month FE Issuer FE SE clustered by R 2 Issues Issuers Iss,M Iss,M , Iss,M , Back to main presentation 28/ 35

29 Bond level analysis: all private IG bonds Share of newly issued bonds bought by insurers hit hard: all privately issued investment grade bonds H1 2007H TM Hit hard fraction Hit hard fraction Hit hard fraction 0.37 (0.08) (-0.59) (-2.65) NAIC (0.79) 3.88 (0.68) 2.94 (0.80) Structured (-0.16) (-0.85) (-1.30) Duration (-0.91) 1.10 (2.36) 0.59 (1.59) Tot insurer purchases 5.53 (4.16) 4.70 (5.15) 7.13 (8.70) Month FE Issuer FE SE clustered by R 2 Issues Issuers Iss,M , Iss,M ,686 1,055 Iss,M ,359 1,484 Back to main presentation 29/ 35

30 Insurance: role for derivatives in managing risk Risk taking based on crisis experience for all large insurers Including derivatives Excluding derivatives DV01 Gap DV01 Gap DV01 Gap DV01 Gap Crisis Hit Flag Crisis Hit Flag 2007Q3-2010Q4 Crisis Hit Flag 2011Q1-2014Q4 Log(Assets) 6.85 (1.06) (-2.63) (-2.74) 0.29 (0.12) (-1.26) (-2.20) (-0.05) 6.22 (0.96) (-2.39) (-1.41) 0.98 (0.39) (-0.26) (-0.95) (-0.04) Quarter FE Insurer FE SE clustered by R 2 Insurer-Quarters Insurers N I,Q , I,Q , N I,Q , I,Q , Crisis hit flag: insurer-level dummy for severe dividend cuts, reduction in equity/assets ratio or equity issuance during crisis ( ). SE double clustered, t-stats in parentheses. Back to main presentation 30/ 35

31 Insurance: credit risk or duration? Insurers hit hard bought lower yielding bonds, not shorter-term bonds Risk taking based on crisis experience for all large insurers Credit risk Duration of purchases TM TM Duration Duration Crisis Hit Flag Crisis Hit Flag 2007Q3-2010Q4 Crisis Hit Flag 2011Q1-2014Q4 Log(Assets) (1.15) (-2.45) (-1.08) (-2.85) (-2.40) (-0.89) (-0.36) 0.52 (1.21) 0.69 (1.90) (-0.30) 0.02 (0.15) 0.60 (1.59) (-0.27) 0.81 (1.27) Quarter FE Insurer FE SE clustered by R 2 Insurer-Quarters Insurers N I,Q , I,Q , N I,Q , I,Q , Crisis hit flag: insurer-level dummy for severe dividend cuts, reduction in equity/assets ratio or equity issuance during crisis ( ). SE double clustered, t-stats in parentheses. Back to main presentation 31/ 35

32 Insurance: all large insurers Risk taking based on crisis experience for all large insurers Net DV01 Gap Interest-rate risk Net DV01 Gap Crisis Hit Flag 1.47 Crisis Hit Flag 2007Q3-2010Q4 (0.30) Crisis Hit Flag 2011Q1-2014Q4 Log(Assets) TM 9.92 (1.01) Credit risk TM (-1.82) (-0.86) (-1.81) (-1.76) (-2.64) (-2.28) (-1.26) (-1.12) 2.61 (1.29) 1.78 (0.31) (-2.17) (-0.16) Quarter FE Insurer FE SE clustered by R 2 Insurer-Quarters Insurers N I,Q , I,Q , N I,Q , I,Q , Crisis hit flag: insurer-level dummy for severe dividend cuts, reduction in equity/assets ratio or equity issuance during crisis ( ). SE double clustered, t-stats in parentheses. Back to main presentation 32/ 35

33 Banks hit hard pulled back Credit growth and risk taking based on crisis experience for large banks Credit risk Interest rate risk log 100 log 100 Loans >1r Loans >1r Crisis Hit Flag Crisis Hit Flag 2007Q3-2010Q4 Crisis Hit Flag 2011Q1-2014Q4 Log(Assets) 0.31 (0.87) (-2.47) (-1.56) (-3.45) (-2.41) (-1.24) 1.17 (2.11) 2.23 (0.57) (-0.17) (-1.93) (-2.23) 0.37 (0.17) (-1.69) (-2.38) Quarter FE BHC FE SE clustered by R 2 BHC-Quarters BHCs N B,Q , B,Q , N B,Q , B,Q , Crisis hit flag: bank-level dummy for severe dividend cuts, reduction in equity/assets ratio during crisis ( ). SE double clustered, t-stats in parentheses. Back to main presentation 33/ 35

34 Banks hit hard reduced credit provision Credit growth (RE loans) based on crisis experience for large banks Back to main presentation 34/ 35

35 Banks hit hard took on less interest-rate risk Asset maturities (Call reports) based on crisis experience for large banks Back to main presentation 35/ 35