Hedge Fund Indexes and Strategy Classification

509 views

Published on

Invited presentation at AIMA Research Day 2003 conference: a study of hedge fund index biases, data quality and cleaning methods. Review of five proposals for hedge fund strategy classifications by leading experts.

Published in: Economy & Finance, Business
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
509
On SlideShare
0
From Embeds
0
Number of Embeds
4
Actions
Shares
0
Downloads
3
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Hedge Fund Indexes and Strategy Classification

  1. 1. Hedge Fund Strategy Classification: AIMA Survey and Analysis of Commercial Classifications Drago Indjic Fauchier Partners AIMA Research Day, 20 October 2003, Paris
  2. 2. Overview • AIMA initiative (April 2003) • AIMA Classification practice survey (June, published in Sept 2003) • Analysis of commercial databases classification (Aug-Sep 2003) • Classification methodology proposals • Acknowledgements: Alexander Ineichen, Francois-Serge L’Habitant, Lionel Martellini, Narayan Naik, Aasmund Heen • Standard disclaimer
  3. 3. 1 Introduction • Early 2003: An index family for every commercial data source: too many indices but a lack of definitions – Implications: legal, performance attribution etc. • Ad-hoc committee under the auspices of AIMA called for “Expressions of interest” in April 2003 – 72 members (Aug 2003) • ‘Non-commercial’, coordinated long-term research effort leading to the development of a set of definition “guidelines”
  4. 4. 2 Survey: Classification Source and Limits No Reply No classification 3% Using outside (external) classification system 47% Is s u e s Using own (internal) classification system 50% Other issues Verification difficult Strategy classification too narrow Strategy classification too broad 0 5 10 15 Frequency AIMA HF Strategy Classification Survey: Sample of 36 out of 73 institutions, June 2003. Source: AIMA Journal, Sep 2003 20
  5. 5. 3 Others 14% Hedgefund.net 9% “External” Classification Sources – Commercial Databases CSFB/Tremont 27% MSCI 23% HFR 27% Source: AIMA Journal, Sep 2003
  6. 6. Classification Source by User Category 4 100% 75% 50% 25% 0% Bank (1) Fund of funds (5) Own classification system Doesn't classify HF manager (16) Investor (3) Service Provider (10) Total (35) External classification provider Most HF managers and investors rely on commercial classifications. Source: AIMA Journal, Sep 2003
  7. 7. AIMA Committee Member’s Involvement Ba nk HF M fu nd s of In ve st or Fu nd ce m Se rv i An on y pr ov id er s 18 16 14 12 10 8 6 4 2 0 ou s Number of instances 5 Passive Active Service providers may dominate “active” membership – commercial pressure. Source: AIMA Journal, Sep 2003
  8. 8. 6 Survey Findings • Fact: almost 50% of professionals rely on commercial sources – Some reply on more than one source • Demand for more specific, verifiable classifications – True meaning of hedge fund indices, investment guidelines, RFP, performance attribution … • What classifications are commercially available? – No “best” index - unequal risk of different indices for the same strategy
  9. 9. 7 Expectation Management • 100% classification accuracy is not feasible – Limited by transparency (IAFE IRC recommendations – even valuation is problematic) and consistency of manager’s behaviour – Limited coverage of risk platforms and exchanges - is transparency welcome? Are new funds investor-friendlier? • Who should be providing classifications? – Fund administrators (or risk measurers)? – How often vendors re-classify strategies? • Pricing accurate classifications?
  10. 10. 8 Strategy Emerging Markets Foreign Exchange Global Emer. Global Macro Macro Managed Futures Market Timing Sector Short Selling Total Directional Arbitrage Eq Market Neutral Fixed Income Arbitrage Market Neutral Merger Arbitrage Relative Value Arb Total Relative Valu HFR TASS 129 28 CISDM Strategy 108 99 53 108 89 163 47 137 16 446 89 153 141 20 399 130 147 121 25 298 90 393 67 73 523 367 Direct Count of Hedge Fund Strategies 393 Note: Data as at September 1st, 2003 Arbitrage Eq Market Neutral Fixed Income Arbitrage Market Neutral Merger Arbitrage Relative Value Arb Total Relative Value Global Est. Equity Hedge Equity Non-Hedge Global Intl Long Only Long/Short Equity Total Security Selec Securities Event Driven Total Multi Process HFR 89 153 141 TASS CISDM 130 147 90 393 67 73 523 367 393 325 551 85 46 16 636 65 231 296 836 836 387 104 104 153 153
  11. 11. 9 Strategy Hedge Fund Index Median Other Unclassified Composite Total Fund of Funds HFR Direct Count of Hedge Fund Strategies (2) TASS CISDMHedge 2479 2433 1676 102 210 22 20 82 447 101 2682 2725 2165 564 524 445 5928 5974 4775 Event driven and short selling are the only strategy descriptions common to all three data providers. Note: Data as at September 1st, 2003
  12. 12. 10 Classification Purity Martellini (2003) Classification methodologies – concern over purity Index Provider N ° of Indices Classification Methodology EACM 18 Classified by EACM HFR 37 CSFB 14 Zurich 5 Classified by the manager and then checked by the Index Committee Classified by Zurich Van Hedge 16 Classified by Hennessee 24 Manager self proclaimed style Classified by the ma Van Hedge nager and then checked by the Index Committee HF Net 37 LJH 16 CISDM 19 Manager self proclaimed style Altvest 14 MSCI over 160 Manager self proclaimed style Classified by the manager and then checked by Committee S&P 10 Classified by S&P Feri 16 Classified by Feri Blue X 1 MondoHedge 7 EurekaHedge 3 HFIntelligence 9 InvestHedge + 12 EuroHedg e + 7 AsiaHedge Bernheim 1 TalentHedge 3 Manager self proclaimed style Classified by Classified by LJH BlueX Classified by the manager and then checked by the Index Committee Not reported Not reported Not reported Classified by TalentHedge the Index
  13. 13. 11 Commercial Strategy Classifications • How are funds are classified by commercial databases? – Get a “baseline” classification estimates using HFR, Tass and CISDM hedge fund databases – How consistent are the classifications of the same fund? – Related study: Meriot Jones (Pertrac), Apr 2003, unpublished • “Noisy” database fund identifiers and strategy classification fields
  14. 14. 12 Hedge fund database Classification analysis • Fauchier Partners research project – 3 man-months (G. Thompson, A. Heen, A. Lahiri) • Not taxonomical analysis of strategy descriptions but collecting evidence • Pertrac data format - database cleaning, name matching and counting • Not database market research: – Not a comparison of data vendors
  15. 15. 13 Approach • “Top-Down” Strategy Classification Approach – Map the “narrow” vendor strategies to “broad” strategies (by convention) – “Count” classifications and “vote” – Estimate overall consistency of the broad strategy classifications and identify conflicts • Identify “unique” funds in different databases – Problem: No ISIN, no sector classification – LP/Ltd, USD/EUR share classes etc causes funds to be identified as the same when they are not
  16. 16. 14 “Top-Down” Strategy Grouping: A Strategy Mapping Convention (1) Directional Multi Process Security Selection Relative Value Emerging Markets (H*,T*) Distressed Securities (H*) Global Est. (C*) Convertible Arbitrage (H*,T*) Foreign Exchange (H) Event Driven (H*,T*,C*) Equity Hedge (H*) Eq Market Neutral (H*,T*) Global Emer. (C*) Equity Non-Hedge (H*) Fixed Income (H*) Global Macro (T*,C*) Global Intl (C*) Fixed Income Arbitrage (T*) Macro (H*) Long Only (C) Market Neutral (C*) Managed Futures (T*) Long/Short Equity (T*) Merger Arbitrage (H*) Market Timing (H*) Relative Value Arb (H*) Sector (H*,C*) Short Selling (H*,T*,C*) Notes: H = HFR98, T = Tass, C = CISDMHedge. * = index exists. FOF excluded • • Subject to discussion: convention based on compilation of several sources. Note: Altvest classifies non-exclusively (“tick all that apply”)
  17. 17. 15 “Top-Down” Strategy Grouping: A Strategy Mapping Convention (2) Directional Event Driven Security Selection Relative Value Emerging Markets (H*,T*) Distressed Securities (H*) Global Est. (C*) Convertible Arbitrage (H*,T*) Foreign Exchange (H) Event Driven (H*,T*,C*) Equity Hedge (H*) Eq Market Neutral (H*,T*) Global Emer. (C*) Merger Arbitrage (H*) Equity Non-Hedge (H*) Fixed Income Arbitrage (T*) Global Macro (T*,C*) Global Intl (C*) Market Neutral (C*) Macro (H*) Long/Short Equity (T*) Relative Value Arb (H*) Managed Futures (T*) Market Timing (H*) Sector (H*,C*) Short Selling (H*,T*,C*) Fixed Income (H*) Long Only (C) Notes: H = HFR98, T = Tass, C = CISDMHedge. * = index exists. FOF excluded • Following to Naik and Ineichen; large multi-strategy funds should be in separate group (Inechien)
  18. 18. 16 Fund Matching Heuristics • Descriptive + numerical criteria : match of fund name (substrings) and fund return (±%tollerance) on two specific dates ID MatchID Name Source Return Return Previous 1422 2042 Pioneer Global Macro (PGM) USD ALTVEST -1.28% -0.32% 1414 2042 Pioneer Global Macro (USD) Tass -1.28% -0.32% 4600 2042 Pioneer Global Macro PGM (USD) HFR98 -1.28% -0.32% • Runtime: merged database cleaning for 15,000 funds takes ~1 hour on PC
  19. 19. Automatic HF Universe Count 17 Tass Tremont (54%) 28% 12% 4% 10% 25% HFR (52%) 5% 16% CISDMHedge (35 %) Total of 6363 funds in 3 major databases (table for >3 databases available), after filtering duplicate records 4589 “unique” funds (28% less). Includes dead and alive funds for classification analysis purpose. Source: Fauchier (August 2003)
  20. 20. 18 Strategy Classification “Matching” • Following to identification of an unique fund present in 1 or more databases: • Cases of classification multiplicity: – Only 1: trivial, fund present in only one database, no 2nd opinion on its classification – 2: fund present in two databases – 3: fund present three databases – >3: fund present in three databases • Algorithm: count modified Pertrac “des” database field descriptors where “narrow” vendor classification are replaced by “broad” classifications
  21. 21. Case of Two Available Classifications 19 Fund Broad Strategy Broad Strategy YYY Relative Value YYY Fund Directional XXX Relative XXX Relative 2 “name” matches Directional Multi-Process Relative Value Security Selection Fund of funds #Agreement 312 132 236 476 468 Nonagreement 156 94 240 272 32 468 226 476 748 500 % Directional Multi-Process Relative Value Security Selection Fund of funds 1 strategy 19% 8% 15% 29% 29% 2 strategies 20% 12% 30% 34% 4% Out of 794 funds classified into different broad strategies there are 156 instances where one of the “broad” strategies is “Directional”. “Non-agreements” counts instances, while “agreement” counts instances of unique fund pairs (thus equals 2 x the number of funds).
  22. 22. Case of Three Available Classifications 20 Broad Strategy Fund Broad Strategy Fund Broad Strategy Fund ZZZ Relative Val. YYY Relative Val. XXX Relative Val. ZZZ Directional YYY Sec. Select XXX Relative Val. ZZZ Sec. Select YYY Relative Val. XXX Relative Val. 3 “name” Matches Directional Multi-Process Relative Value Security Selection Fund of Funds # Agreement 166 106 253 331 259 2 to 1 198 116 318 438 46 46 6 22 46 27 410 228 593 815 332 Security Selection Fund of Funds Non-agreement Percentages Directional Multi-Process Relative Value 1 strategy 15% 10% 23% 30% 23% 2 strategies 18% 10% 28% 39% 4% 3 strategies 31% 4% 15% 31% 18% Note: some funds are classified in 3 different “broad” strategies.
  23. 23. 21 Further Database Classification Research • Estimate size of universe and attrition rates – quarterly trend analysis of strategy growth • Marginal utility of additional databases – how many? • What is behind inconsistencies? – Identify classification trouble spots – Estimate misclassification rate and bias – Induce vendor’s classification rule • Verify HF index compositions
  24. 24. 22 Part 2: Methodology Requirements • Threshold transparency level (non-transparent funds cannot be classified) • 1: performance estimates (NAV) • 2: consolidated exposure (sensitivities) • 3: position level (daily copy of portfolio statement) • 4: trade level (intra-daily - ideal) • Accuracy, precision, confidence … • Econometrics: data (history) requirements, “drift” detection discriminate styles within strategy, adapt to evolving strategies
  25. 25. 23 Current Classification Methodology Proposals • Initiate discussion • Several proposals made by ad-hoc committee: – Statistical: clustering, PCA – Structural: risk factors, syntactical • Further proposals are welcome – Explanation facility
  26. 26. 24 F. –S. L’Habitant (2003) • Cluster Analysis: the best way to classify hedge funds without bias – Suggested algorithm: partition around metroids (PAM) • Center of each style = first principal component of all indices publicly available for a style (e.g. EDHEC indices) • Leverage effects should be normalized
  27. 27. 25 Related Research • Brown and Goetzmann (2001) style analysis using clustering – Does not distinguish between (equally correlated) share classes with varying leverage • Gyger and Gibson (2001) – “Hard” vs “Soft” (fuzzy/probabilistic) classification, robust distance measures – Normalise leverage by average strategy variance (or by “gross” balance sheet exposure?) • Produces peer-relative measure (“tracking error”)
  28. 28. 26 Naik (2003) • “Asset based style” factor analysis, Fung and Hsieh (2001) – Linear and nonlinear (option) payoffs • Standardise taxonomy of strategies – Managers should self-declare %risk exposure to strategies • Mutual fund industry – re-classification lessons – Some 700 managers asking to be reclassified by Morningstar exhibited better performance under new benchmarks (Goetzmann)
  29. 29. 27 Martellini (2003) • Two problems: right categories + classification method • Using a manager’s self-proclaimed style is not a good option because of style biases and style drifts. – William Sharpe’s insight: “If it acts like a duck, I’ll consider it’s a duck” • Perform a rolling-window regression analysis of the fund performance on a set of indices, and look for patterns – One should use pure indices perfectly representative of a given pure strategy • Many index providers exist but none is entirely reliable – EDHEC Indices: Portfolio of indices derived using PCA
  30. 30. 28 Indjic (2003) • Verification and validation problem Issuer X X X Y Z Type Equity CB CDS Equity Equity Sector A A A B B Position Short Long Long Long Short • What does managers’ portfolio holdings say about strategy? – Strategy reasoning system
  31. 31. 29 AIMA Conference Feedback • Why not classifying strategies on the basis of VaR? • Can discretionary traders be ever classified using systematic factors?
  32. 32. 30 Future: Methodology • Guidelines/ endorsement (for investors, FoF, performance attribution) – Standard definitions – “Blind classification” competition – Are you prepared to “override” your classifications? • Classification “clearing house” / web server – Consensus building – Data fusion (statistics, factor analysis)
  33. 33. 31 Future: Committee • Open Forum – Public dissemination – classification workshop in 2004? – Consensus is slowly moving: how to facilitate the process? • Format for constructive dialogue with vendors – Publish names of inconsistently classified funds and resolve conflicts? – Implication for index “products” and benchmarking • For-profit or not? – “Open” academic “standard” – Independency guarantee vs (charitable) funding

×