Attribution and Risk Analysis (From the NCREIF Academy June 2010)

Transcription

1 Attribution and Risk Analysis (From the NCREIF Academy June 2010) Page 0

2 Attribution Analysis Overview Two flavors of analysis 1. Attribution ti Analysis How much have the fund s strategic objectives affected performance? How much have specific investment decisions affected performance? 2. Risk Analysis What kinds of risks does the fund undertake? Does performance reflect the risks? Page 1

3 Performance Attribution Overview What Does Attribution Analysis Mean? A performance-evaluation evaluation tool used to analyze the abilities of portfolio or fund managers. Attribution analysis uncovers the impact of the manager s investment decisions with regard to overall investment policy, asset allocation, security selection and activity. A fund or portfolio s returns are compared to a benchmark in order to determine whether a manager is actually skilled or just lucky. Page 2 Source: Geltner/Miller/Clayton/Eichholtz

4 Performance Attribution Overview Two levels at which performance attribution is performed: Portfolio level Pertains to portfolios or investment manager (or fund) level. Property level Pertains to individual properties. Page 3 Source: Geltner/Miller/Clayton/Eichholtz

5 Performance Attribution Overview Major attributes (return components): At the PORTFOLIO LEVEL: Allocation Selection Interaction At the PROPERTY LEVEL: Initial Cash Yield Cash Flow Change Yield Change Page 4 Source: Geltner/Miller/Clayton/Eichholtz

6 Performance Attribution Overview At both levels, diagnostic purpose is facilitated by comparison between subject portfolio or mgr with an appropriate benchmark. Relative (or differential) ) performance between subject vs. benchmark is quantified, in total & within components. Portfolio Level: Portfolio Total Return Benchmark Total Return Allocation Selection Interaction Property Level: Prop.IRR Bnchmk Cohort IRR Source: Geltner/Miller/Clayton/Eichholtz Init.Yield CF Growth Yield Chge Page 5

7 Breaking down Portfolio Returns Fund Return Fund Risk Cash Debt Other Assets Direct Property Returns to Fund Returns Compared to benchmark Direct Property Total Return By Geographic area and property type relative to benchmark Components of Total Return Attribution of Relative Return Income Return Capital Return Allocation Selection Property Risk Weighted Contributions to Relative Return By Geographic area and property type relative to benchmark Held Partial Sales Purchases Development Sold Source: IPD Page 6

8 Performance Attribution Overview Basic idea is that components of the performance differential between the subject & the benchmark reflect performance of various distinct functions of investment management. For example, at the portfolio level Alloc & Selection Synergy Asset Picking & Oper.Mgt Alloc Strategy Across coss Segments Source: Geltner/Miller/Clayton/Eichholtz Allocation Selection Interaction Page 7

9 Performance Attribution Overview Basic idea is that components of the performance differential between the subject & the benchmark reflect performance of various distinct functions of investment management. For example, at the property level Mgt & Disp Selection Operational Acquisition Mgt Selection Source: Geltner/Miller/Clayton/Eichholtz Initial Yield CF Change Yield Change Page 8

10 Given Sector Attribution -- Introduction A fund invested in a mixture of sectors Returns for the fund s investments in each sector A benchmark with its own mix and returns by the same sectors How can one make sense of the above differences to Explain to investors why the fund and benchmark performances differed Suggest to managers how to improve performance Explain changes in performance Page 9

11 Sector Allocation The Concept Manager adds value in two ways: Choosing superior investments SELECTION Executing a superior strategy ALLOCATION Superior selection contributions involve: Value added by transactions team Superior asset/property management practices Executing an effective disposition policy Superior allocation contributions involve: Buying/selling in the right markets at the right times Buying/selling in the right property types at the right times Investment Choices and Strategy Both Matter! Page 10

12 Strategy Failure Historical Example Noting strong apartment returns in 2005 (21.2% vs 20.6% 06% for NPI), ),a fund dover-weights eg tsapat apartments tsin 2006 to 2008 Result: In NPI Apartment Annualized Return = 5.8% NPI Annualized Total Return = 81% 8.1% Page 11

13 Example 1 Weak Property Selection, Good Sector Allocation Strategy Note the following: Fund return = 2.75% Benchmark return = 2.50% Difference =.25% Despite under-performing the benchmark in both sectors, the fund held a higher proportion of the better-performing sector. SUCCESS from STRATEGY Total Return (%) Sector Allocation (%) Benchmark beats Fund Fund tilts to better sector 0.0 Sector A Sector B 0 Sector A Sector B Fund Benchmark Page 12

14 Example 2 Good Property Selection, Poor Sector Allocation Strategy Fund return = 2.25% Benchmark return = 2.50% Difference = -.25% Fund beat benchmark in both sectors but it allocated too much of the portfolio to the sector that performed poorly. The benchmark had less allocated to sector that performed poorly Total Return (%) Sector Allocation (%) Sector A Fund beats benchmark in both sectors Sector B 80 But fund tilts to wrong 70 sector Sector A Sector B Fund Benchmark Page 13

15 Sector Attribution the basic math For an individual sector: R f R b W f W b Fund return for sector i Benchmark return for sector i Fund proportion of sector i Benchmark proportion of sector i There are four key terms for each sector: W f R f W b R b W b R f W f R b Sector i contribution to total fund return Sector I contribution to total benchmark return Fund proportion of sector i Benchmark proportion of sector i Page 14

16 Sector Allocation the basic math (continued) Now, if we add up all of the sectors we get: R = W f R f B = W b R b S = W b R f A = W f R b Fund total return Benchmark total return Fund return if if diversified like like the the benchmark Benchmark return if if diversified like like the the fund Here is a tabular form of the above: Fund weights Benchmark weights Fund returns R = W f R f S = W b R f Benchmark returns A = W f R b B = W b R b Page 15

17 Sector Attribution the basic math (continued) R B = Total excess return Benchmark weight S B W b (R f -R b ) Selection effects applied to return difference Benchmark return +A B (W f -W b ) R b Allocation effects applied to weight difference + R S Cross product Difference in weights x (W f -W b ) (R f -R b ) A B terms difference in returns * Source: Lieblich (1995) Page 16

18 Using Previous Example 2 R B = Total excess return 2.25% % = -.25% =.60 x (4% - 3.5%) S B W b (R f -R b ) Selection effects +.40 x (1.5% - 1%) =.5% = ( ) x 3.5% + A B (W f -W b ) R b Allocation effects (.7 -.4) x 1% = -.75% = (.3 -.6) x (4% - 3.5%) + R S Cross product (W (7 A B f -W b ) (R f -R b ) + (.7 -.4) x (1.5% - 1%) terms = 0% Page 17

19 Summary for Previous Example 2 Selection effects.50% + Allocation effects -.75% + Cross Product.00% = Total excess return -.25% Page 18

20 Conclusion Attribution Analysis helps us explain why the returns for a fund differ from a benchmark Differences in performance can be due to Differences in the ability to select individual properties Differences in the allocation of the portfolio by sector (property type or location) compared to the benchmark Previous analysis assumed benchmark was same risk as portfolio Page 19

21 Multi-period Attribution Performance attribution lacks significance for any one short holding period (e.g., quarter or year). When aggregating across periods: At the portfolio level (i.e., for allocation and selection attribution), it is generally best to use the arithmetic TWRR of the attribution components At the property level (i.e., for initial yield, cash flow change,, and yield change attribution), it is probably more appropriate to use the IRR, because capital flow timing is controlled by the property (investment) manager at the property level. Page 20 Source: Geltner/Miller/Clayton/Eichholtz

22 Property-Level Performance Attribution Some insight may be obtained by analysis of property-level performance attribution. In particular: Initial yield relates to the traditional selection & acquisition functions; Cash flow change relates largely to the operational management function (but not purely or necessarily, e.g., it may reflect expiration of vintage leases); Yield change may reflect either a traditional selection (asset picking) effect or an operational management effect, or both. Page 21 Source: Geltner/Miller/Clayton/Eichholtz

23 Property-Level Performance Attribution These property-level management functions are related generally to three attributes (components) of the property- level since-acquisition IRR, essentially as indicated below. Property Selection Initial Yield (IY) Acquisition Transaction Execution Cash Flow Change Operational Management (CFC) Disposition Transaction Execution Yield Change (YC) Page 22 Source: Geltner/Miller/Clayton/Eichholtz

24 Property-Level Performance Attribution Useful IRR-Based property level performance attribution benchmarking requires the use of: Since-acquisition IRR IRR is computed since acquisition of property (or portfolio): IRR is computed for appropriate benchmark cohort, defined as universe of similar investments by competing managers, measured from same inception date (equal to property acquisition date). Page 23 Source: Geltner/Miller/Clayton/Eichholtz

25 Property-Level Performance Attribution Simple Numerical Example: Property (or static portfolio) bought at initial cash yield of 9%. Net cash flow grew at 2% per year. Property (or properties) sold (or appraised) after 10 years at a terminal yield of 10%, based on yr.11 projected cash flow (also 2% more than yr.10). IRR is 10.30%. How much of this IRR is due to 3 components: Initial Yield (IY), Cash Flow Change (CFC), and Yield Change (YC)? Page 24 Source: Geltner/Miller/Clayton/Eichholtz

26 Yr IRRs: (1) Actual Oper.CF (2) Actual Capital CF (3) Actual Total CF (=1+2) 10.30% (4) Init.Oper.CF constant (5) Capital Init.Yld.on(4) (6) Init.Yld (=4+5) 9.00% (7) Capital Init.Yld.on(1) (8) Actual Oper. Init.Yld (=1+7) 11.00% (9) Capital ActualYldon(4) ActualYld.on(4) (10) Actual Yld (=4+9) 8.32% Initial yield = 9.00%, computed from line (6) IRR. Cash flow change component = 2.00% =11%-9% 11%-9%, computed as the line (8) IRR less the line (6) IRR: = IRR with actual CF IRR with no CF growth, (with constant yield at initial rate). Yield change component = -0.68% = 8.32%-9.00%, computed as the line (10) IRR less the line (6) IRR: = IRR with actual yield change IRR with no yield change, (with constant CF at initial level). Interaction effect = -0.02% 0.02%, the difference between the line (3) overall IRR and the sum of the three other attributes [10.3% 3%-(9%+2% (9%+2% %)]. Page 25

27 Property-Level Performance Attribution Here is a graphical presentation of the IRR-Based property-level performance attribution we just performed: Subject Property: IRR & Component Breakout 14% 12% 10% 8% 6% 4% 2% 0% -2% -4% IRR InitYld CFchg YldChg Interaction IRR & Components Suppose we computed the same type of IRR component breakdown for an appropriate benchmark, that is, a NCREIF sub-index cohort spanning the same period of time Page 26 Source: Geltner/Miller/Clayton/Eichholtz

28 Property-Level Performance Attribution We could compare our subject performance with that achieved by a peer universe of managers, for similar properties (e.g., Calif. Industr. bldgs): Subject vs NCREIF Cohort Performance Comparison: IRR & Component Breakout 14% 12% 10% 8% 6% 4% 2% 0% -2% -4% IRR InitYld CFchg YldChg Interaction IRR & Components Subject NPI Cohort Page 27 Source: Geltner/Miller/Clayton/Eichholtz

29 Property-Level Performance Attribution Here is the relative performance,, the difference between our subject property and its benchmark, by attribute: Subject - NCREIF Cohort Relative Performance: IRR & Component Breakout 1.50% 1.00% 0.50% 0.00% -0.50% -1.00% -1.50% -2.00% -2.50% IRR InitYld CFchg YldChg Interaction IRR & Components The above pattern could be plausibly interpreted as tentative evidence for the following hypothesis: Subject performed relatively poorly due largely to some combination of poor selection, acquisition, and operational mgt, partially offset by some combination of good disposition execution (or optimistic terminal appraisal), future-oriented capital improvements, &/or market movements during the holding period. Page 28 Source: Geltner/Miller/Clayton/Eichholtz

30 Property-Level Performance Attribution Here is the relative performance,, the difference between our subject property and its benchmark, by attribute: Subject - NCREIF Cohort Relative Performance: IRR & Component Breakout 1.50% 1.00% 0.50% 0.00% -0.50% -1.00% -1.50% -2.00% -2.50% IRR InitYld CFchg YldChg Interaction IRR & Components Now suppose we computed these relative performance differentials across a number of different properties (or portfolios) we have invested in. Page 29 Source: Geltner/Miller/Clayton/Eichholtz

31 Property-Level Performance Attribution We might gain some insights about our property-level investment and management performance: Three Properties Comparison: Subject - NCREIF Cohort Relative Performance 3.00% 2.00% 1.00% 0.00% -1.00% -2.00% -3.00% -4.00% -5.00% IRR InitYld CFchg YldChg Interaction Subject 1 Subject 2 Subject 3 In this case Subject Properties #1 & 3 have similarly poor performance (rel to benchmk), due to poor initial yield & poor CF change, suggesting poor acquisition & poor operational mgt. Property #2 did better, with good InitYld & CFchg, but poor YldChg (suggesting good acquisition, but poor disposition or mgt actions that hurt future outlook (e.g., inadequate Cap.Improvement). Mkt movements can also affect these results (less so the longer the holding period). Page 30 Source: Geltner/Miller/Clayton/Eichholtz

32

33 okey Concepts Portfolio performance, in part, should consist of compensation for incurring risk Total performance decomposable into risk-adjusted and risk premium segments oapproach Compute metrics for fund performance and fund risk. Page 32

34 Measure Sharpe Ratio Beta Fund Risk Measures Definition iti ( Avg. fund return risk free rate ) / std. dev. of fund returns (Covariance between fund returns and benchmark returns) / (Variance of benchmark returns)* Treynor Ratio ( T i ) Tracking Error (Avg. fund return risk free rate) / beta Std. dev. of difference in returns for fund versus benchmark Information Ratio Avg. of difference in returns for fund vs. benchmark / tracking error *Also calculated by the coefficient of a linear regression of the fund returns against the benchmark returns. Page 33

35 Fund Risk Measures contu d Measure Jensen s Alpha Definition Difference between actual fund return and fund return expected from CAPM* *Capital Asset Pricing Model (CAPM): Expected Return = Risk Free Rate + Beta x (Market return Risk Free Rate) Page 34

36 Example Property Benchmark Year Return Return % 8.00% % 10.00% Beta for fund is % 12.00% % 13.00% % 14.00% % 14.00% Returns for Fund vs Benchmark % 12.00% % 11.00% 25% % 10.00% 00% 20% y = x % 9.00% 15% % 8.00% % 5.00% 10% % 3.00% % 800% 100% 1.00% 5% % 2.00% 0% % 3.00% 0% 2% 4% 6% 8% 10% 12% 14% 16% -5% % 5.00% % 8.00% -10% % 10.00% Benchmark returns % 12.00% Fund returns Property Return = -5.83% x Benchmark Return Page 35

37 Year Fund Return Benchmark Return Fund - Benchmark Risk-free rate FUnd Exces return Benchmark Excess return % 8.00% 6.00% 3.00% 11.00% 5.00% % 10.00% -1.00% 3.00% 6.00% 7.00% % 12.00% 4.00% 3.00% 13.00% 9.00% % 13.00% -1.00% 3.00% 9.00% 10.00% % 14.00% 2.00% 3.00% 13.00% 11.00% % 14.00% 6.00% 3.00% 17.00% 11.00% % 12.00% 2.00% 3.00% 11.00% 9.00% % 11.00% 4.00% 3.00% 12.00% 8.00% % 10.00% 2.00% 3.00% 9.00% 7.00% % 9.00% 4.00% 3.00% 10.00% 6.00% % 8.00% -2.00% 3.00% 3.00% 5.00% % 5.00% 3.00% 3.00% 5.00% 2.00% % 3.00% -9.00% 3.00% -9.00% 0.00% 00% % 1.00% -9.00% 3.00% % -2.00% % 2.00% -8.00% 3.00% -9.00% -1.00% % 3.00% 3.00% 3.00% 3.00% 0.00% % 5.00% -3.00% 3.00% -1.00% 2.00% % 8.00% 0.00% 3.00% 5.00% 5.00% % 10.00% 8.00% 3.00% 15.00% 7.00% % 12.00% 0.00% 3.00% 9.00% 9.00% Variance 0.62% 0.16% 0.22% 0.62% 0.16% Stdev 7.86% 3.97% 4.74% 7.86% 3.97% Mean 9.05% 8.50% 0.55% 6.05% 5.50% 50% Sharpe Ratio: ( Avg. fund return risk free rate ) / std. dev. of fund returns = (9.05% - 3%) / 7.86% =.77 Tracking error: 4.74% (see above) Page 36

38 Risk Attribution Fund Contributions: Basic View Fund return = Risk free rate + Risk contribution Risk contribution = Fund risk x Return per unit of risk r F Fund Return F Note that the risk contribution tib ti is equal to the fund standard deviation multiplied by the fund Sharpe ratio Axis rreturn r 0 Risk Free Return Risk Contribution S F F Risk Free Rate Contribution Risk Axis F Page 37

39 Risk Attribution Fund Contribution vs. Benchmark Fund return = Risk free rate + Market risk contribution + Fund risk premium + Manager value added r F Fund Return F Note: Manager value added d does not include the fund risk premium the investor bears the risk and receives the market compensation for it. Advantages Risk contributions computed using a market metric rather than the fund metric. Return Axis r B r 0 Excess Return r F r B Benchmark Return Risk Free Return Manager Value Added Return Enhancement Alpha (Jensen) B B 1 B F F Fund Risk Premium S B F Market Risk Contribution S B B Risk Free Rate Contribution Risk Axis B Page 38

40 Conclusion Attribution Analysis helps explain why a portfolio and individual properties differed in performance from a benchmark Differences from benchmark can be broken down into a number of different components Allocation o vs. Selectionec Acquisitions, dispositions, development, held Initial yield, cash flow growth, terminal cap rate Sharpe ratio and other risk measures