3/12/2015 Phd Forum - Harsh Parikh

Transcription

1 3/12/2015 Phd Forum - Harsh Parikh 1

2 Active fixed income in rising rates environment Harsh Parikh Advisor: Lionel Martellini, PhD London, 24th March /12/2015 Phd Forum - Harsh Parikh 2

3 Agenda Stylized facts about manager performance conditional on market environment Timing active and passive strategy Active and Passive policy allocation Evaluate fundamental law of active management in fixed income using a structural VAR framework Conduct style analysis using exogenous variables Determine catalysts for enhanced alpha using endogenous variables 3/12/2015 Phd Forum - Harsh Parikh 3

4 Contributions Enhance the active management literature to demonstrate not only time varying opportunities but also time varying effectiveness of skills in different market environments particularly in fixed income asset class Besides recommending mixed active and passive allocation, we demonstrate benefit of dynamic active risk budgeting for an investor by showing a timing strategy between active vs. passive and duration vs. spreads Lastly evaluate fundamental law of active management and conduct style analysis in fixed income asset class where we propose couple of additional risk factors 3/12/2015 Phd Forum - Harsh Parikh 4

5 Executive Summary Rising yield environment tend to be associated with increased uncertainty and higher implied volatility in bond yields Higher implied volatility is associated with increased dispersion of returns across fixed income sectors thereby presenting larger than normal opportunities for active managers Not only do the active fixed income managers benefit from increased opportunity but also from relatively shorter duration and longer spread duration exposure, during rising yield environment Active managers generate economically significant alpha commensurate with active risk budget spent, in rising yield periods Investors can enhance fee productivity by controlling the active risk budget and allocating between active and passive strategy commensurate with the availability of the opportunity 3/12/2015 Phd Forum - Harsh Parikh 5

6 Rising rates trend Investors have experienced a trend of falling yields for more than two decades Source: Datastream, Analysis Period: 12/31/1988 9/30/2013, Please refer to Appendix for additional details 3/12/2015 Phd Forum - Harsh Parikh 6

7 Rising rates periods Since Dec-1988 we ve recorded 10 episodes of yields rising by 1% or more, with a median length of 11 months and 124 bps rise in yield Regime Period Yield Diff Months Begin End Begin End BPS 9 7/31/1989 4/30/ /30/ /30/ /29/1995 8/30/ /30/1998 1/31/ /31/2001 3/29/ /30/2003 7/30/ /30/2005 6/30/ /31/ /31/ /31/2010 3/31/ /31/2012 8/20/ Source: Datastream, Analysis Period: 12/31/1988 9/30/2013, Please refer to Appendix for additional details 3/12/2015 Phd Forum - Harsh Parikh 7

8 12/1/1988 3/1/1990 6/1/1991 9/1/ /1/1993 3/1/1995 6/1/1996 9/1/ /1/1998 3/1/2000 6/1/2001 9/1/ /1/2003 3/1/2005 6/1/2006 9/1/ /1/2008 3/1/2010 6/1/2011 9/1/2012 Manager vs benchmark performance For purpose of research we use evestment US core plus universe median and 25 th percentile manager performance We evaluate active performance against equal weighted sector performance and for illustration here show against US aggregate as well As evident managers do have significant systematic style biases 10 $1 growth (log scale) 12/88-9/13 Returns Risk Dur Sprd Dur Median 7.8% 3.9% th Percentile 10.9% 3.9% Avg Sector 7.5% 4.7% US Aggregate 6.9% 3.8% Avg. Sector US Agg. Median 25th Ptile Source: Datastream, evestment Alliance, Factset, Analysis Period: 12/31/1988 9/30/2013, Please refer to Appendix for additional details 3/12/2015 Phd Forum - Harsh Parikh 8

9 Manager alpha in different yield environment For core plus universe both median and 25 th percentile manager, ranked by performance, have added significantly more alpha in rising yield environment Similar to most academic findings we confirm median active managers delivered insignificant alpha on the whole (Wermers and Fischer, Dec 2012, Barras et al., 2010 and Busse et al., 2010) Periods Change in Yield (BPS) Median Manager Alpha% 25 th Percentile Manager Alpha% Monthly Average Rising Yield Remainder Total Rising Yield - Remainder Rising Yield - Remainder (T-Stat) Rising Yield - Total Rising Yield - Total (T- Stat) Source: Datastream, evestment Alliance, Factset, Analysis Period: 12/31/1988 9/30/2013, Please refer to Appendix for additional details 3/12/2015 Phd Forum - Harsh Parikh 9

10 Retail manager alpha (Lipper) in different yield environment We observe significant t-stats for difference in means of rising yield and remainder periods average outperformance relative to equal weighted sector benchmark across the major fixed income Lipper categories as well Not surprisingly this is also true for other spread sensitive categories Hence, our observation with Evestment core-plus universe is not an artifact of the choice of database and the universe Monthly Outperformance 12/31/1992-9/30/2013 General Bond Multi- Sector Income Lipper Categories - Primary Share Class Corporate Debt A Corporate Debt BBB Intermedia te Investment Grade High Yield US Mortgage Rising Yield Remainder Total Rising Yield - Remainder Rising Yield Remainder (T-Stat) Rising Yield - Total Rising Yield - Total (T- Stat) Source: Factset, Lipper Net performance 3/12/2015 Phd Forum - Harsh Parikh 10

11 Activeness and rising yield environment Manager skills do seem to be time varying and clearly rising yield environment accentuates active bets with tail wind from increased breadth Refutes Gorman, 2010 conclusion that managers information ratios are expected to remain unchanged as realized tracking error is expected to be proportional to manager s active returns. Similar to Kacperczyk, 2013 JoF we show effectiveness of skills are time varying and market environment dependent Similar to Anna von Reibnitz, 2013 we show outperformance of most active vs. least active managers is concentrated in high dispersion environment. Quartile by Activeness Evestment Core Plus (Median Outperformance) Annualized M_25 (Most Active) M_50 M_75 M_100 (Least Active) Rising Yield Avg. Alpha Remainder Avg. Alpha Total Avg. Alpha Rising Yield Tracking Error Remainder Tracking Error Total Tracking Error Rising Yield IR Remainder IR Total IR Source: Datastream, evestment Alliance, Factset, Analysis Period: 12/31/1988 9/30/2013, Please refer to Appendix for additional details 3/12/2015 Phd Forum - Harsh Parikh 11

12 Active manager characteristics Using snapshot data as of 9/30/2013 we find that The median AUM is higher for most active quartile vs. least active quartile managers While most active managers mostly manage institutional strategies least active managers have significantly higher proportion of non institutional strategies Most active managers indeed have significantly higher turnover and larger portfolio management team Quartile by Activeness Total AUM Total Accounts Total Institutional Institutional Accounts AUM Portfolio Mgrs/Dual Role PMs Annual Turnover LTM Most Active M_25 1, , nd 2, , rd 2, , Least Active M_ Source: evestment Alliance, Analysis Period: 12/31/1988 9/30/2013 3/12/2015 Phd Forum - Harsh Parikh 12

13 Alpha catalysts Iwanowski JPM Summer 1996, conducted US Fixed Income sector allocation research to find systematic sector risk premia Simulating active management performance in fixed income have been addressed in literature for several regions including US, Japan and Euroland The 2 key drivers essentially are rate anticipation spread trades and intermarket spread trades (Boyd et al JFI Spring 2010, Fabozzi 2004) This work inspired us to focus on uncertainty as measured by MOVE and fixed income sector dispersion as a measure of cross sectional variation (CSV)/opportunity in fixed income MOVE Bond implied volatility estimate CSV Cross Sectional Variation in fixed income sector returns 3/12/2015 Phd Forum - Harsh Parikh 13

14 12/1/1988 7/1/1989 2/1/1990 9/1/1990 4/1/ /1/1991 6/1/1992 1/1/1993 8/1/1993 3/1/ /1/1994 5/1/ /1/1995 7/1/1996 2/1/1997 9/1/1997 4/1/ /1/1998 6/1/1999 1/1/2000 8/1/2000 3/1/ /1/2001 5/1/ /1/2002 7/1/2003 2/1/2004 9/1/2004 4/1/ /1/2005 6/1/2006 1/1/2007 8/1/2007 3/1/ /1/2008 5/1/ /1/2009 7/1/2010 2/1/2011 9/1/2011 4/1/ /1/2012 6/1/ /1/1988 7/1/1989 2/1/1990 9/1/1990 4/1/ /1/1991 6/1/1992 1/1/1993 8/1/1993 3/1/ /1/1994 5/1/ /1/1995 7/1/1996 2/1/1997 9/1/1997 4/1/ /1/1998 6/1/1999 1/1/2000 8/1/2000 3/1/ /1/2001 5/1/ /1/2002 7/1/2003 2/1/2004 9/1/2004 4/1/ /1/2005 6/1/2006 1/1/2007 8/1/2007 3/1/ /1/2008 5/1/ /1/2009 7/1/2010 2/1/2011 9/1/2011 4/1/ /1/2012 6/1/ /1/1988 6/1/1989 1/1/1990 8/1/1990 3/1/ /1/1991 5/1/ /1/1992 7/1/1993 2/1/1994 9/1/1994 4/1/ /1/1995 6/1/1996 1/1/1997 8/1/1997 3/1/ /1/1998 5/1/ /1/1999 7/1/2000 2/1/2001 9/1/2001 4/1/ /1/2002 6/1/2003 1/1/2004 8/1/2004 3/1/ /1/2005 5/1/ /1/2006 7/1/2007 2/1/2008 9/1/2008 4/1/ /1/2009 6/1/2010 1/1/2011 8/1/2011 3/1/ /1/2012 5/1/2013 History of MOVE, CSV and manager alpha MOVE Index (MOVE) BPS Cross-sectional Sector Volatility % 25th Percentile Manager Alpha% Median Manager Alpha% Monthly Manager Alpha% Source: Bloomberg, Datastream, evestment Alliance, Factset, Analysis Period: 12/31/1988 9/30/2013, Please refer to Appendix for additional details 3/12/2015 Phd Forum - Harsh Parikh 14

15 MOVE, CSV and manager alpha Overall median manager Information Ratio IR is not significantly different from zero, however it is positive and significant in rising MOVE periods The fact that median manager IR is negative in falling MOVE periods also implies time varying skills Improvement in IR in rising MOVE periods is combined with increased dispersion Contrary to Gorman 2010 that mentions IR should stay constant if dispersion rises, we show that IR can improve from effectiveness of skills Raubenheimer 2011, argue that active risk should be commensurate with crosssectional dispersion. While we acknowledge that, we argue that active risk should also be commensurate with market environments when skills are effective. Count Annualized MOVE CSV Avg % Std. Dev IR Average Average Median Manager 25th %ile Median Manager 25th %ile Median Manager 25th %ile All Periods Falling MOVE Rising MOVE Source: Datastream, evestment Alliance, Factset, Analysis Period: 12/31/1988 9/30/2013, Please refer to Appendix for additional details 3/12/2015 Phd Forum - Harsh Parikh 15

16 1/1/ /1/1999 7/1/2000 4/1/2001 1/1/ /1/2002 7/1/2003 4/1/2004 1/1/ /1/2005 7/1/2006 4/1/2007 1/1/ /1/2008 7/1/2009 4/1/2010 1/1/ /1/2011 7/1/2012 4/1/2013 Timing Active versus Passive Strategy For portfolio formation at time t allocate to active strategy (proxy with median or median of 25 th percentile by activeness outperformance) for three months when t 3 t 1 3 CSV ( t n t 1 n CSV * σ CSV t n t 1 ) Allocate to passive strategy in remainder instances Reibnitz 2013, deciding when to invest in active funds, therefore, can be as important as deciding which funds to invest in. Annualized Outperformance Median Most Active Quartile Median 1/99-9/30 Strategy Strategy Alpha Tracking Error IR Year Alpha Tracking Error IR Source: Datastream, evestment Alliance, Factset, Analysis Period: 12/31/1998 9/30/2013, Please refer to Appendix for additional details Active Regime Classification (Out of Sample Strategy) 3/12/2015 Phd Forum - Harsh Parikh 16

17 1/1/ /1/1999 9/1/2000 7/1/2001 5/1/2002 3/1/2003 1/1/ /1/2004 9/1/2005 7/1/2006 5/1/2007 3/1/2008 1/1/ /1/2009 9/1/2010 7/1/2011 5/1/2012 3/1/2013 Timing Active versus Passive Strategy We earlier showed the strategy using cross-sectional variation across all fixed income sectors, however using cross sectional variation in spread sectors improved the timing strategy significantly Understandably managers deviate using spread sector bets IR improved from 0.41 to 0.65 for the timing between passive strategy and the most active median manager strategy During the evaluation period the strategy is 40% times in active median manager strategy and 60% times in passive strategy Annualized Outperformance Median Most Active Quartile Median 1.2 Active Regime Classification (Out of Sample Strategy) 1/99-9/30 Strategy Strategy Alpha Tracking Error IR Year Alpha Tracking Error IR Source: Datastream, evestment Alliance, Factset, Analysis Period: 12/31/1998 9/30/2013, Please refer to Appendix for additional details 3/12/2015 Phd Forum - Harsh Parikh 17

18 12/1/1998 8/1/1999 4/1/ /1/2000 8/1/2001 4/1/ /1/2002 8/1/2003 4/1/ /1/2004 8/1/2005 4/1/ /1/2006 8/1/2007 4/1/ /1/2008 8/1/2009 4/1/ /1/2010 8/1/2011 4/1/ /1/2012 8/1/2013 Timing Duration vs Spread If the cross-sectional variation is any indication of timing it should also effectively time between long duration treasury and high yield sectors, having known ex-post that the prior sector outperformed all other sectors when yields fell and the later when yields rose The strategy employs same timing signal and thus is 40% of times in high yield sector and 60% times in long duration treasury sector Annualized Performance Timing HYLD vs LD TSY 40% HYLD 60% LD TSY Alpha 1/99-9/30 Returns Risk Returns/Risk (IR) Year Returns Risk Returns/Risk (IR) Source: Datastream, evestment Alliance, Factset, Analysis Period: 12/31/1998 9/30/2013, Please refer to Appendix for additional details Cumm. Alpha 3/12/2015 Phd Forum - Harsh Parikh 18

19 Active and Passive policy allocation decision Specific allocations will be dependent on a variety of factors including investment policies/guidelines, risk tolerance of the decision-makers, manager research program and transition management services Allocation to active and passive depend on the segment efficiency (Flood et al, JPM Fall 2000) For more efficient segment they recommend 50% - 65% Active and 50% - 35% Passive Investor can also employ tactical bands roughly +/-15% for tactically shifting based on cyclical changes in performance of active versus passive It also depends on information ratio, active risk and investment horizon assumptions (Crook, JII Summer 2014) Allocation to active strategies is increasing in information ratio and investment horizon Reasonable IR assumption of 0.25 would imply allocation of 55% in active in case of equities However presence of active risk creates uncertainty and merits both active and passive allocation Often ignored but extremely necessary condition in considering allocation to active and passive is outperformance opportunity also known as the breadth Information ratio is a function of skill and breadth as measured by cross sectional variation (Gorman et al, JIMFI 2010) 3/12/2015 Phd Forum - Harsh Parikh 19

20 1998M M M M M M M M M M M M M M M M M M12 Active and Passive allocation in fixed income Not so much discussed in the active passive allocation literature is the time varying breadth and its impact on policy allocation to active (Anna von Reibnitz, 2013 illustrate timing strategy using dispersion) We find not only time varying dispersion but also persistence in short term cross sectional variation (3 month average) in excess of long term average cross sectional variation Our Markov regime switching work shows 33% average probability in higher cross sectional variation regime for last 15 years In absence of views investor can use probability weighted active passive allocation as starting point Variable Coefficient Std. Error z-statistic Prob. 1.2 Cross-sectional Variation Regime probabilities Regime 1 C LOG(SIGMA) Regime 2 C LOG(SIGMA) Constant transition probabilities: P(i, k) = P(s(t) = k s(t-1) = i) (row = i / column = j) High Low Constant expected durations: Source: Datastream, Factset, Please refer to Appendix for additional details 3/12/2015 Phd Forum - Harsh Parikh 20

21 Validating FLAM in the data Fundamental Law of Active Management (Gorman et al 2010, Grinold 89, 94, Clarke et al 06) Relative expected returns of a strategy can be expressed as: E(r) = IC * CSV * z Where: E(r) is the expected value added IC is the information coefficient or measure of manager s skill CSV is the cross-sectional variation in returns of securities z is the manager s opinion on relative attractiveness of the securities For a given level of skill the manager s expected return will rise as cross sectional variation (CSV) or activeness (z) increases 3/12/2015 Phd Forum - Harsh Parikh 21

22 Validate using a structural VAR approach Understand interdependencies of 3 endogenous variables namely bond implied volatility (MOVE Index), fixed income sector returns dispersion (cross-sectional variation CSV) and 25 th percentile manager alpha (TOPA) Use changes in yield (D(Yield)) as an exogenous variable (style analysis) Findings Results imply that both implied volatility and manager alpha rise with rise in bond yields Cross-sectional variation increases with increase in implied volatility 25 th percentile manager alpha increases with increase in cross-sectional variation over and above what is explained by changes in yield Coefficients MOVE CSV TOPA MOVE(-1) T-Stats CSV(-1) T-Stats TOPA(-1) T-Stats C T-Stats D(YIELD) T-Stats Source: Bloomberg, Datastream, evestment Alliance, Eviews, Factset, Analysis Period: 12/31/1988 9/30/2013, Refer to appendix for additional details VAR stands for Vector Auto Regression, an econometric regression technique 3/12/2015 Phd Forum - Harsh Parikh 22

23 Test endogeneity All the 3 variables namely MOVE, CSV and TOPA do not Granger Cause changes in yield there by implying these don t precede rising yield and hence we regard D(YIELD) as exogenous variable Both MOVE and CSV Granger Cause each other and hence are endogenous. Both CSV and TOPA Granger Cause each other and thus are endogenous TOPA and MOVE don t Granger Cause each other and thus implying the real transmission mechanism is via CSV from MOVE to TOPA Pairwise Granger Causality* Tests Date: 11/25/13 Time: 14:06 Sample: 1988M M09 Lags: 1 Null Hypothesis: Number of Observations F-Statistic Probability MOVE does not Granger Cause D(YIELD) D(YIELD) does not Granger Cause MOVE CSV does not Granger Cause D(YIELD) D(YIELD) does not Granger Cause CSV TOPA does not Granger Cause D(YIELD) D(YIELD) does not Granger Cause TOPA CSV does not Granger Cause MOVE MOVE does not Granger Cause CSV TOPA does not Granger Cause MOVE MOVE does not Granger Cause TOPA TOPA does not Granger Cause CSV E-05 CSV does not Granger Cause TOPA *Granger Causality leans heavily on the idea that the cause occurs before the effect. 3/12/2015 Phd Forum - Harsh Parikh 23

24 Alpha response to cross-sectional variation Over and above benefitting from changes in yield our analysis as below shows that there is persistent alpha added from increase in implied volatility and increase in cross sectional variation Response of cross-sectional variation in fixed income sector returns (CSV) to 1 standard deviation shock to MOVE Index accumulates to 81 bps over 12 months. To contrast mean CSV is 98 bps in the sample i.e. CSV would increase by almost 80% with the shock in MOVE. Response of 25 th percentile manager alpha (TOPA) to 1 standard deviation shock to cross-sectional variation in fixed income sector returns (CSV) accumulates to 21 bps over 12 months. To contrast mean TOPA is 26 bps in the sample i.e. TOPA would increase by almost 80% with the shock in CSV Response of CSV to 1 Std. Dev. Shock in MOVE (In Percentage) Response of TOPA to 1 Std. Dev. Shock in CSV (In Percentage) Source: Bloomberg, Datastream, Evestment, Eviews, Factset, Analysis Period: 12/31/1988 9/30/2013, Please refer to Appendix for additional details 24 3/12/2015 Phd Forum - Harsh Parikh 24

25 Adding more variables in the system, list of endogenous variables VAR analysis for each of 7 strategies (Alpha in excess of equally weighted fixed income sectors) using principal component approach and without Median alpha 25 th percentile alpha Most Active quartile median alpha High Yield Alpha (To verify how systematic style bias would stand) TSY Long Duration Alpha (To verify how systematic bias would stand) Timing Most Active Quartile median alpha and Passive (Using CSV timing signal) Timing High Yield and TSY Long Duration (Using CSV timing signal) Other endogenous variables included 10Yr CMT Yield Moody s BAA Yield 10Yr CMT Yield (BAA spread) Cross-sectional variation in spread sectors MOVE 3/12/2015 Phd Forum - Harsh Parikh 25

26 Simultaneous style analysis, list of exogenous variables And exogenous variables include (Fung Hsieh 97, Martellini et al Wiley 2003) Location Factors (buy-and-hold policy): Sharpe s traditional factors The bond market factor: The monthly change in the 10-year treasury constant maturity yield D(H15Yield) The credit spread factor: The monthly change in the Moody's Baa yield less 10- year treasury constant maturity yield D(H15BAAYield) Trading Strategy Factors (option like payoff) Return of Primitive Trend Following Strategy (PTFS ): Bond lookback straddle (FHBONDTREND) Return of PTFS: Short Term Interest Rate Lookback Straddle Additionally we propose using Monthly change in cross-sectional variation in spread sectors Monthly change in MOVE 3/12/2015 Phd Forum - Harsh Parikh 26

27 Without using PCA, Results Looking at exogenous variables We note that median, 25 th percentile performance, 25 th percentile most active and timing strategy have shorter duration and have longer spread duration We note 25 th percentile performance and the timing strategies have exposure to monthly change in cross sectional variation thereby validating fundament law of active management Both median and 25 th percentile performance also have significant exposure to monthly change in MOVE Reviewing endogenous lagged variables The timing strategies have significant exposure to lagged cross sectional variation suggesting moment in CSV and strategy being able to capture it The 25 th percentile performance has significant exposure to lagged spreads suggesting price reversals in subsequent months adding to alpha MEDIAN 3/12/2015 Phd Forum - Harsh Parikh 27 TOPA MOST ACTIVE HYLD ALPHA TSYLD ALPHA TIMING MOST ACTIVE TIMING SPREAD GOVT H15YIELD(-1) T-Stat H15BAA_YIELD(- 1) T-Stat CSVSPRD(-1) T-Stat MOVE(-1) T-Stat DEPENDENT(-1) T-Stat C T-Stat EXOGENOUS D(H15YIELD) T-Stat D(H15BAA_YIELD) T-Stat FHBONDTREND T-Stat FHSIRTREND T-Stat D(CSVSPRD) T-Stat D(MOVE) T-Stat R-squared Adj. R-squared Source: Datastream, evestment Alliance, Factset, Analysis Period: 12/31/1998 9/30/2013, Please refer to Appendix for additional details

28 Using PCA, Results Looking at exogenous variable related components We note that median, 25 th percentile performance, 25 th percentile most active and timing strategy have shorter duration and have longer spread duration We note that median, 25 th percentile performance, 25 th percentile most active and timing strategy have exposure to monthly change in cross sectional variation thereby validating fundament law of active management Both median and 25 th percentile performance also have significant exposure to trading strategy related component We validate significant alpha for 25 th percentile performance and timing strategies Reviewing endogenous lagged variables The timing strategies have significant exposure to lagged cross sectional variation suggesting moment in CSV and strategy being able to capture it The 25 th percentile performance has significant exposure to lagged spreads suggesting price reversals in subsequent months adding to alpha MEDIAN Source: Datastream, evestment Alliance, Factset, Analysis Period: 12/31/1998 9/30/2013, Please refer to Appendix for additional details 3/12/2015 Phd Forum - Harsh Parikh 28 TOPA MOST ACTIVE HYLD ALPHA TSYLD ALPHA TIMING MOST ACTIVE TIMING SPREAD GOVT PCOMP101(-1) T-Stat PCOMP201(-1) T-Stat PCOMP301(-1) T-Stat PCOMP401(-1) T-Stat DEPENDENT(-1) T-Stat C T-Stat EXOGENOUS PCOMP1s T-Stat PCOMP2s T-Stat PCOMP3s T-Stat PCOMP4s T-Stat PCOMP5s T-Stat PCOMP6s T-Stat R-squared Adj. R-squared

29 Correlation of exogenous variables and their principal components First principal component explains 35% of variance and is related to spreads The second principal component explains close to 20% and is related to 10 Yr TSY rates The third principal component explains close to 16% and is related to dispersion The fourth and fifth principal component explains close to 22% and are related to trading strategy factors D(H15 BAA_Y IELD) FHBO NDTRE ND PCOM P1S PCOMP 2S PCOM P3S PCOM P4S PCOM P5S PCOM P6S D(H15 YIELD) FHSIRT REND D(MOV E) PCOMP1S PCOMP2S PCOMP3S PCOMP4S PCOMP5S PCOMP6S D(H15YIELD) D(H15BAA_YIELD) FHBONDTREND FHSIRTREND D(MOVE) D(CSVSPRD) Source: Factset, Lipper Net performance 3/12/2015 Phd Forum - Harsh Parikh 29

30 Correlation of endogenous variables and their principal components First principal component explains 56% and is related to spreads The second principal component explains close to 29% and is related to 10 Yr TSY rates The third principal component explains close to 12% and is related to dispersion H15YIEL PCOMP101 PCOMP201 PCOMP301 PCOMP401 D H15BAA_YIEL D MOVE CSVSPRD PCOMP PCOMP PCOMP PCOMP H15YIELD H15BAA_YIEL D MOVE CSVSPRD Source: Factset, Lipper Net performance 3/12/2015 Phd Forum - Harsh Parikh 30

31 Appendix Terminology Description Start End Bond Yield US Benchmark 10 Year Datastream Government Index Yield 12/31/88 9/30/13 Cross-Sectional Variation (CSV) Monthly cross-sectional standard deviation in returns of Fixed Income Sectors* 25 th Percentile Manager Performance Calculated using 25 th percentile performance for universe comprising of core plus fixed income managers in evestment Alliance database 25 th Percentile Manager Alpha (TOPA) Difference in 25 th Percentile Manager Performance and average return of Fixed Income Sectors* Bond Implied Volatility Merrill Option Volatility Estimate MOVE Index - This is a yield curve weighted index of the normalized implied (MOVE) volatility on 1-month Treasury options. 12/31/88 9/30/13 *Fixed Income Sectors BARCLAYS US CORP HIGH YIELD Index tracks high yield corporate securities performance 12/31/88 9/30/13 BARCLAYS US AGG CORP AA INTERMEDIATE Index tracks performance of AA corporate securities with intermediate maturity that are in US Aggregate Index 12/31/88 9/30/13 BARCLAYS US AGG CORP AA LONG Index tracks performance of AA corporate securities with long maturity that are in US Aggregate Index 12/31/88 9/30/13 BARCLAYS US CORP : INVESTMENT GRADE Index tracks investment grade corporate securities performance 12/31/88 9/30/13 BARCLAYS MUNICIPAL BOND Index tracks municipal securities performance 12/31/88 9/30/13 BARCLAYS US AGENCY INTERMEDIATE Index tracks performance of agency securities with intermediate maturity 12/31/88 9/30/13 BARCLAYS US AGENCY LONG Index tracks performance of US agency securities with long maturity 12/31/88 9/30/13 BARCLAYS US AGG SUPRANATIONAL Index tracks performance of supranational securities that are in US Aggregate Index 12/31/88 9/30/13 BARCLAYS US AGG SPNAT. LONG Index tracks performance of supranational securities with long maturity that are in US Aggregate Index 12/31/88 9/30/13 BARCLAYS US SOVEIGN: LONG Index tracks performance of US sovereign securities with long maturity 12/31/88 9/30/13 BARCLAYS US AGG SOVERIGN Index tracks performance of sovereign securities that are in US Aggregate Index 12/31/88 9/30/13 BARCLAYS US TREASURY LONG Index tracks performance of US treasury securities with long maturity 12/31/88 9/30/13 BARCLAYS US TREASURY INTERMEDIATE Index tracks performance of US treasury securities with intermediate maturity 12/31/88 9/30/13 BARCLAYS FHLMC 15Y Index tracks performance of FHLMC (Freddie Mac) securities with 15Y maturity 12/31/88 9/30/13 BARCLAYS GOVT NAT MTGE ASSN (GNMA) Index tracks performance of GNMA securities 12/31/88 9/30/13 BARCLAYS FNMA 15Y Index tracks performance of FNMA (Fannie Mae) securities with 15Y maturity 12/31/88 9/30/13 BARCLAYS US MORTGAGE BACKED SECS - Index tracks performance of US mortgage backed securities 12/31/88 9/30/13 Source: Factset, Lipper Net performance 3/12/2015 Phd Forum - Harsh Parikh 31

32 Appendix evestment Alliance Database: We used data from evestment Alliance to address these research questions. evestment Alliance collects monthly data about performance and portfolio characteristics on active money managers who self-report the data. It classifies managers as core-plus and it relies both on self-reporting and its own analysis. The database was launched in mid-2000 but data is available as far back as 1980s. The database is widely used in manager search and performance measurement by consultants and institutional investors. There is no survivorship bias in the database after 2000 but there is a back-filling bias (managers might back-fill historical data once they start reporting) and a self-reporting bias. However there is no reason to believe that these biases impact the conditional analysis and the relative performance in different market environment is likely to be bias free. As of 9/30/2013 there were 149 managers. Source: Factset, Lipper Net performance 3/12/2015 Phd Forum - Harsh Parikh 32