Financial Intermediaries and the Cross-Section of Asset Returns by Adrian, Etula, Muir Discussion Pietro Veronesi The University of Chicago Booth School of Business 1
What does this paper do? 1. From Broker-Dealer Flow of Funds data, define Leverage = Total Financial Assets Total Financial Assets Total Liabilities 2
What does this paper do? 1. From Broker-Dealer Flow of Funds data, define Leverage = Total Financial Assets Total Financial Assets Total Liabilities 2. Compute Shocks LevFac =Δln(Leverage) 2
What does this paper do? 1. From Broker-Dealer Flow of Funds data, define Total Financial Assets Leverage = Total Financial Assets Total Liabilities 2. Compute Shocks LevFac =Δln(Leverage) 3. Finds: 2
My Discussion 1. Why is this interesting? 2. Fishing expedition? 3. Interpretation and measurement 4. Other Comments 3
The factor is disarmingly simple. Whyisthisinteresting? It follows immediately from assets and liabilities of broker-dealers 4
Whyisthisinteresting? The factor is disarmingly simple. It follows immediately from assets and liabilities of broker-dealers Note all interesting spikes are not in recessions. = Financial channel 4
Whyisthisinteresting? There are some theories behind the use of intermediary-based SDF: E.g. Brunnermeier Pedersen (2009, RFS): Unconstrained risk neutral speculators subject to potential future funding constraints generate a SDF SDF = φ 1 E 0 [φ 1 ] where φ 1 = shadow cost of capital φ 1 =1+max j { max ( v j 1 pj 1 m j+ 1 )}, (vj 1 pj 1 ) m j 1 v j 1 = fundamental value of asset; pj 1 = price of asset; m j+ 1,mj 1 = endogenous margin requirements on long/short positions. 5
Whyisthisinteresting? There are some theories behind the use of intermediary-based SDF: E.g. Brunnermeier Pedersen (2009, RFS): Unconstrained risk neutral speculators subject to potential future funding constraints generate a SDF SDF = φ 1 E 0 [φ 1 ] where φ 1 = shadow cost of capital φ 1 =1+max j { max ( v j 1 pj 1 m j+ 1 )}, (vj 1 pj 1 ) m j 1 v j 1 = fundamental value of asset; pj 1 = price of asset; m j+ 1,mj 1 = endogenous margin requirements on long/short positions. Financiers impose margins depending on volatility / liquidity = E[R j ]= Cov(φ 1,R j ) E 0 [φ 1 ] 5
Whyisthisinteresting? There are some theories behind the use of intermediary-based SDF: E.g. Brunnermeier Pedersen (2009, RFS): Unconstrained risk neutral speculators subject to potential future funding constraints generate a SDF SDF = φ 1 E 0 [φ 1 ] where φ 1 = shadow cost of capital φ 1 =1+max j { max ( v j 1 pj 1 m j+ 1 )}, (vj 1 pj 1 ) m j 1 v j 1 = fundamental value of asset; pj 1 = price of asset; m j+ 1,mj 1 = endogenous margin requirements on long/short positions. Financiers impose margins depending on volatility / liquidity = E[R j ]= Cov(φ 1,R j ) E 0 [φ 1 ] Here, φ 1 = a b ln(leverage): Lower leverage = tighter funding constraint Bad times = high effective risk aversion of financiers = Lower leverage & tighter funding constraints 5
Whyisthisinteresting? There are some theories behind the use of intermediary-based SDF: E.g. He and Krishnamurthy (2009, RFS): Only specialists/intermediaries can invest in the stock market = SDF depends on intermediary consumption process Households unwilling to provide unlimited funds to intermediary: Time varying constraint of available funds depends on specialists/intermediary wealth SDF dependson constraints: whenspecialists wealth declines, the capital constraint is tighter = higher risk premia Here, Broker-Dealer Leverage proxies for capital constraint: Low leverage = capital constraint. 6
Whyisthisinteresting? It compares well with other factors, even with the tough FF25 ( Value / Size) 7
Whyisthisinteresting? Using all the portfoios (FF25, 10 momentum, 6 T-Bonds): 8
Fishing expedition? The brocker-dealer leverage factor prices a lot of portfolios The empirical analysis is reasonably well done Performed all the possible tests that seem now required to dispell the fishing expedition critique Use large number of portfolios Check significance of first-pass betas Do formal tests and use simulations to compute confidence intervals for cross-sectional R2s Interpret the magnitudes of the coefficients (whenever possible) For instance, Lewellen, Nagel, and Shanken (JFE, 2010) show that many proposed factors would not survive further scrutiny. 9
Model / assets Variables OLS R 2 GLS R 2 T 2 q LL (2001) const. cay c cay c FF25 3.33-0.81 0.25 0.00 0.58 0.05 33.9 0.44 (2.28) (-1.25) (0.84) (0.42) [0.30, 1.00] [0.00, 0.50] [p=0.08] [0.00, 0.72] FF25 + 30 ind. 2.50-0.48 0.09-0.00 0.00 0.01 193.8 1.31 (3.29) (-1.23) (0.53) (-1.10) [0.00, 0.35] [0.00, 0.20] [p=0.00] [0.18, 1.08] LVN (2004) const. my c my c FF25 3.58 4.23 0.02 0.10 0.57 0.02 20.8 0.45 (2.22) (0.76) (0.04) (1.57) [0.35, 1.00] [0.00, 0.35] [p=0.57] [0.00, 0.48] FF25 + 30 ind. 2.78 0.37-0.02 0.03 0.09 0.00 157.1 1.32 (3.51) (0.13) (-0.09) (1.40) [0.00, 1.00] [0.00]* [p=0.04] [0.00, 0.96] SV (2006) const. R M s w R M FF25 3.07-0.95-0.21 0.27 0.02 26.0 0.46 (1.96) (-0.58) (-2.06) [0.00, 1.00] [0.00, 0.40] [p=0.63] [0.00, 0.30] FF25 + 30 ind. 2.57-0.49-0.09 0.08 0.02 160.8 1.31 (2.77) (-0.44) (-1.99) [0.00, 1.00] [0.00, 0.40] [p=0.07] [0.00, 0.72] LVX (2006) const. I HH I Corp Ncorp FF25 2.47-0.80-2.65-8.59 0.80 0.26 25.2 0.34 (2.13) (-0.39) (-1.03) (-1.96) [0.75, 1.00] [0.05, 1.00] [p=0.29] [0.00, 0.48] FF25 + 30 ind. 2.22 0.20-0.93-5.11 0.42 0.04 141.2 1.27 (3.14) (0.19) (-0.58) (-2.32) [0.20, 1.00] [0.00, 0.55] [p=0.11] [0.00, 0.84] Yogo (2006) const. c Ndur c Dur R M FF25 1.98 0.28 0.67 0.48 0.18 0.04 24.9 0.46 (1.36) (1.00) (2.33) (0.29) [0.00, 1.00] [0.00, 0.55] [p=0.69] [0.00, 0.30] FF25 + 30 ind. 1.95 0.18 0.19 0.12 0.02 0.05 159.3 1.24 (2.27) (1.01) (1.52) (0.11) [0.00, 0.60] [0.00, 1.00] [p=0.06] [0.00, 0.78] CAPM const. R M FF25 2.90-0.44-0.03 0.01 77.5 0.46 (3.18) (-0.39) [0.00, 0.55] [0.00, 0.25] [p=0.00] [0.12, 0.48] FF25 + 30 ind. 2.03 0.10-0.02 0.00 225.2 1.34 (2.57) (0.09) [0.00, 0.35] [0.00, 0.05] [p=0.00] [0.18, 0.96] Cons. CAPM const. c FF25 1.70 0.24 0.05 0.01 60.6 0.46 (2.47) (0.89) [0.00, 1.00] [0.00, 0.50] [p=0.01] [0.06, 0.66] FF25 + 30 ind. 2.07 0.03-0.02 0.00 224.5 1.34 (3.51) (0.15) [0.00, 0.65] [0.00]* [p=0.00] [0.18, 1.02] Fama French const. R M SMB HML FF25 2.99-1.42 0.80 1.44 0.78 0.19 56.1 0.37 (2.33) (-0.98) (1.70) (3.11) [0.60, 1.00] [0.05, 0.65] [p=0.00] [0.06, 0.42] FF25 + 30 ind. 2.21-0.49 0.60 0.87 0.31 0.06 200.4 1.24 (2.14) (-0.41) (1.24) (1.80) [0.00, 0.90] [0.05, 0.35] [p=0.00] [0.12, 0.90] 10
Model / assets Variables OLS R 2 GLS R 2 T 2 q LL (2001) const. cay c cay c FF25 3.33-0.81 0.25 0.00 0.58 0.05 33.9 0.44 (2.28) (-1.25) (0.84) (0.42) [0.30, 1.00] [0.00, 0.50] [p=0.08] [0.00, 0.72] FF25 + 30 ind. 2.50-0.48 0.09-0.00 0.00 0.01 193.8 1.31 (3.29) (-1.23) (0.53) (-1.10) [0.00, 0.35] [0.00, 0.20] [p=0.00] [0.18, 1.08] LVN (2004) const. my c my c FF25 3.58 4.23 0.02 0.10 0.57 0.02 20.8 0.45 (2.22) (0.76) (0.04) (1.57) [0.35, 1.00] [0.00, 0.35] [p=0.57] [0.00, 0.48] FF25 + 30 ind. 2.78 0.37-0.02 0.03 0.09 0.00 157.1 1.32 (3.51) (0.13) (-0.09) (1.40) [0.00, 1.00] [0.00]* [p=0.04] [0.00, 0.96] SV (2006) const. R M s w R M FF25 3.07-0.95-0.21 0.27 0.02 26.0 0.46 (1.96) (-0.58) (-2.06) [0.00, 1.00] [0.00, 0.40] [p=0.63] [0.00, 0.30] FF25 + 30 ind. 2.57-0.49-0.09 0.08 0.02 160.8 1.31 (2.77) (-0.44) (-1.99) [0.00, 1.00] [0.00, 0.40] [p=0.07] [0.00, 0.72] LVX (2006) const. I HH I Corp Ncorp FF25 2.47-0.80-2.65-8.59 0.80 0.26 25.2 0.34 (2.13) (-0.39) (-1.03) (-1.96) [0.75, 1.00] [0.05, 1.00] [p=0.29] [0.00, 0.48] FF25 + 30 ind. 2.22 0.20-0.93-5.11 0.42 0.04 141.2 1.27 (3.14) (0.19) (-0.58) (-2.32) [0.20, 1.00] [0.00, 0.55] [p=0.11] [0.00, 0.84] Yogo (2006) const. c Ndur c Dur R M FF25 1.98 0.28 0.67 0.48 0.18 0.04 24.9 0.46 (1.36) (1.00) (2.33) (0.29) [0.00, 1.00] [0.00, 0.55] [p=0.69] [0.00, 0.30] FF25 + 30 ind. 1.95 0.18 0.19 0.12 0.02 0.05 159.3 1.24 (2.27) (1.01) (1.52) (0.11) [0.00, 0.60] [0.00, 1.00] [p=0.06] [0.00, 0.78] CAPM const. R M FF25 2.90-0.44-0.03 0.01 77.5 0.46 (3.18) (-0.39) [0.00, 0.55] [0.00, 0.25] [p=0.00] [0.12, 0.48] FF25 + 30 ind. 2.03 0.10-0.02 0.00 225.2 1.34 (2.57) (0.09) [0.00, 0.35] [0.00, 0.05] [p=0.00] [0.18, 0.96] Cons. CAPM const. c FF25 1.70 0.24 0.05 0.01 60.6 0.46 (2.47) (0.89) [0.00, 1.00] [0.00, 0.50] [p=0.01] [0.06, 0.66] FF25 + 30 ind. 2.07 0.03-0.02 0.00 224.5 1.34 (3.51) (0.15) [0.00, 0.65] [0.00]* [p=0.00] [0.18, 1.02] Fama French const. R M SMB HML FF25 2.99-1.42 0.80 1.44 0.78 0.19 56.1 0.37 (2.33) (-0.98) (1.70) (3.11) [0.60, 1.00] [0.05, 0.65] [p=0.00] [0.06, 0.42] FF25 + 30 ind. 2.21-0.49 0.60 0.87 0.31 0.06 200.4 1.24 (2.14) (-0.41) (1.24) (1.80) [0.00, 0.90] [0.05, 0.35] [p=0.00] [0.12, 0.90] 10
Model / assets Variables OLS R 2 GLS R 2 T 2 q LL (2001) const. cay c cay c FF25 3.33-0.81 0.25 0.00 0.58 0.05 33.9 0.44 (2.28) (-1.25) (0.84) (0.42) [0.30, 1.00] [0.00, 0.50] [p=0.08] [0.00, 0.72] FF25 + 30 ind. 2.50-0.48 0.09-0.00 0.00 0.01 193.8 1.31 (3.29) (-1.23) (0.53) (-1.10) [0.00, 0.35] [0.00, 0.20] [p=0.00] [0.18, 1.08] LVN (2004) const. my c my c FF25 3.58 4.23 0.02 0.10 0.57 0.02 20.8 0.45 (2.22) (0.76) (0.04) (1.57) [0.35, 1.00] [0.00, 0.35] [p=0.57] [0.00, 0.48] FF25 + 30 ind. 2.78 0.37-0.02 0.03 0.09 0.00 157.1 1.32 (3.51) (0.13) (-0.09) (1.40) [0.00, 1.00] [0.00]* [p=0.04] [0.00, 0.96] SV (2006) const. R M s w R M FF25 3.07-0.95-0.21 0.27 0.02 26.0 0.46 (1.96) (-0.58) (-2.06) [0.00, 1.00] [0.00, 0.40] [p=0.63] [0.00, 0.30] FF25 + 30 ind. 2.57-0.49-0.09 0.08 0.02 160.8 1.31 (2.77) (-0.44) (-1.99) [0.00, 1.00] [0.00, 0.40] [p=0.07] [0.00, 0.72] LVX (2006) const. I HH I Corp Ncorp FF25 2.47-0.80-2.65-8.59 0.80 0.26 25.2 0.34 (2.13) (-0.39) (-1.03) (-1.96) [0.75, 1.00] [0.05, 1.00] [p=0.29] [0.00, 0.48] FF25 + 30 ind. 2.22 0.20-0.93-5.11 0.42 0.04 141.2 1.27 (3.14) (0.19) (-0.58) (-2.32) [0.20, 1.00] [0.00, 0.55] [p=0.11] [0.00, 0.84] Yogo (2006) const. c Ndur c Dur R M FF25 1.98 0.28 0.67 0.48 0.18 0.04 24.9 0.46 (1.36) (1.00) (2.33) (0.29) [0.00, 1.00] [0.00, 0.55] [p=0.69] [0.00, 0.30] FF25 + 30 ind. 1.95 0.18 0.19 0.12 0.02 0.05 159.3 1.24 (2.27) (1.01) (1.52) (0.11) [0.00, 0.60] [0.00, 1.00] [p=0.06] [0.00, 0.78] CAPM const. R M FF25 2.90-0.44-0.03 0.01 77.5 0.46 (3.18) (-0.39) [0.00, 0.55] [0.00, 0.25] [p=0.00] [0.12, 0.48] FF25 + 30 ind. 2.03 0.10-0.02 0.00 225.2 1.34 (2.57) (0.09) [0.00, 0.35] [0.00, 0.05] [p=0.00] [0.18, 0.96] Cons. CAPM const. c FF25 1.70 0.24 0.05 0.01 60.6 0.46 (2.47) (0.89) [0.00, 1.00] [0.00, 0.50] [p=0.01] [0.06, 0.66] FF25 + 30 ind. 2.07 0.03-0.02 0.00 224.5 1.34 (3.51) (0.15) [0.00, 0.65] [0.00]* [p=0.00] [0.18, 1.02] Fama French const. R M SMB HML FF25 2.99-1.42 0.80 1.44 0.78 0.19 56.1 0.37 (2.33) (-0.98) (1.70) (3.11) [0.60, 1.00] [0.05, 0.65] [p=0.00] [0.06, 0.42] FF25 + 30 ind. 2.21-0.49 0.60 0.87 0.31 0.06 200.4 1.24 (2.14) (-0.41) (1.24) (1.80) [0.00, 0.90] [0.05, 0.35] [p=0.00] [0.12, 0.90] 10
How does the Leverage Factor do with FF25 + 30 Ind.? 11
Lev Growth Lev Growth Interpretation How do we interpret the leverage factor? 5 Household 3 BrokerDealer 4 2 3 2 1 1 0 0-1 -1-2 -2-3 -3-4 -6-4 -2 0 2 4 Asset Growth -4-4 -2 0 2 4 Asset Growth 12
Interpretation How do we interpret the leverage factor? 1.1 Consider Leverage2 = Total Liabilities Total Financial Assets 1.05 1 0.95 0.9 0.85 0.8 0.75 1952 1957 1962 1967 1972 1977 1982 1987 1992 1997 2002 2007 2012 13
Interpretation How do we interpret the leverage factor? 1.1 Consider Leverage2 = Total Liabilities Total Financial Assets 1.05 1 0.95 0.9 0.85 0.8 0.75 1952 1957 1962 1967 1972 1977 1982 1987 1992 1997 2002 2007 2012 13
Interpretation It would be important to figure out the source of variation In the paper: Active deleveraging 14
Interpretation It would be important to figure out the source of variation In the paper: Active deleveraging Alternatives: 1. Higly levered institutions maybe get kicked out from sample in times of crisis? (and viceversa?) 14
Interpretation It would be important to figure out the source of variation In the paper: Active deleveraging Alternatives: 1. Higly levered institutions maybe get kicked out from sample in times of crisis? (and viceversa?) 2. Asymmetric measurement of assets and liabilities? What if assets are at book value but liabilities are marked-to-market? 14
Interpretation It would be important to figure out the source of variation In the paper: Active deleveraging Alternatives: 1. Higly levered institutions maybe get kicked out from sample in times of crisis? (and viceversa?) 2. Asymmetric measurement of assets and liabilities? What if assets are at book value but liabilities are marked-to-market? Can we look at the components of assets and liabilities and figure out what is the main source of variation? 14
Active Deleveraging: Figure 3.2 in Adrian and Shin (2010) Total Assets and Leverage Lehman Brothers Merrill Lynch Morgan Stanley Total Asset Growth -.2 -.1 0.1.2 1998-4 2008-1 2007-3 1998-3 2007-4 Total Asset Growth -.2 -.1 0.1.2 1998-4 2008-1 1998-3 2007-3 2007-4 Total Asset Growth -.2 -.1 0.1.2 1998-4 2008-1 1998-3 2007-4 2007-3 -.2 -.1 0.1.2 Leverage Growth -.2 -.1 0.1.2 Leverage Growth -.2 -.1 0.1.2 Leverage Growth Bear Stearns Goldman Sachs Citigroup Markets 98-04 Total Asset Growth -.1 0.1.2 1998-4 2007-3 2008-1 1998-3 2007-4 Total Asset Growth -.05 0.05.1.15 2007-4 2007-3 2008-1 Total Asset Growth -.3 -.2 -.1 0.1 1998-3 1998-4 -.2 -.1 0.1.2 Leverage Growth -.2 -.1 0.1 Leverage Growth -.2 -.1 0.1.2 Leverage Growth 15
Interpretation Funding constraints? Ted Spread = LIBOR - 3m T-Bill 1 Ted Spread Leverage2 4 0.9 2 0.8 1985 1990 1995 2000 2005 2010 2015 0 16
Interpretation Funding constraints? Shocks to Ted Spread 2.5 2 LevFac TedSpreadFac 1.5 1 0.5 0 0.5 1 1.5 1985 1990 1995 2000 2005 2010 2015 Correlation =37.23% (mainlydrivenby2datapoints) 17
Interpretation Liquidity constraints? Market liquidity measure from Pastor and Stambaugh (2003, JPE) 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1 liquidity factor leverage factor 1.2 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 Correlation =4% 18
Interpretation Proxy for returns on financial sector? Returns on Bank Industry Portfolio 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1 Banks Leverage Factor 1.2 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 Correlation = 22% 19
Conclusions The proposed intermediary-based SDF prices a lot of different portfolios that are typically hard to price. This is quite interesting, but this findings really spurs a lot of other questions What does this findings tells us about the underlying economy? For instance, structurally, why should value stocks be riskier and growth stocks safer when SDF is related to leverage of broker-dealers? If it is really funding constraints, does the margin constraint bind more for value stocks, or for momentum stocks? Is the channel through a tightening of financing constraints for some type of firms, such as distressed value firms? Or is it an omitted variable problem? The risk of the economy increase, which lead broker-dealers to delever, and value stocks drop by more than growth stocks? More information of the sources of variation in SDF and its links to the real economy would be quite important to better understand the economics behind the empirical findings. 20