The Brave New World of Hedge Fund Indexes Desperately Seeking Pure Style Indexes Lionel Martellini Marshall School of Business University of Southern California.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The Brave New World of Hedge Fund Indexes
Desperately Seeking Pure Style Indexes
Lionel MartelliniMarshall School of Business
University of Southern California Chercheur Associé EDHEC
Motivation Problems are Amplified for Hedge Fund Indexes
• Existing indexes are not fully representative– Because of the lack of regulation on hedge fund performance disclosure, existing
data bases only cover a relatively small fraction of the hedge fund population– Probably only a little more than half of existing hedge funds choose to self-report
their performance to one of the major hedge fund databases– One of the most popular hedge fund indexes, the EACM 100, does not account for
more than a tiny percentage of all existing hedge funds– Most HF indexes are equally-weighted (all but CSFB/Tremont)
• Existing indexes are biased– Most hedge fund indexes are based upon managers' self-proclaimed styles– Given that hedge fund managers jealously protect the secret of their investment
strategies (the so-called black-box problem), relying on managers' self-proclaimed style is actually almost a necessity
– This procedure only makes sense under the following two conditions: (1) a manager follows a unique investment and (2) a manager's self-proclaimed style matches the manager's actual trading strategies
– Style drift problem (see for example Lhabitant (2001))
• Directional strategies: aimed at benefiting from market movements and trends
– Global macro– Hedge (long bias)– Long (e.g. “growth” or “value” stocks) – Short (e.g. “overvalued”, “glamour” stocks)
• Non directional strategies: – Event driven (corporate events such as takeovers, spin-offs, mergers, etc.)– Restructuring (buying or shorting securities of companies under Chapter 11
and/or undergoing some form of reorganization)– Fixed-income arbitrage (long and short via treasuries, corporate and/or asset-
backed securities)– Capital structure arbitrage (buying and selling different securities of the same
Desperately Seeking Pure Style IndexesStatistical Approach – Kalman Filter
• Simple model
– Rkt is the return on competing index k– It is the (unobservable) return on pure index– kt is the noise, measurement error resulting from presence of biases and
absence of representativeness
• Assume normally distributed pure indexes and measurement errors
• Kalman filter is used both to evaluate the likelihood function and to forecast and smooth the unobserved state variables
• Here, our motivation is to estimate (smooth) the unobserved pure index, based on the observed returns on competing indexes:
Desperately Seeking Pure Style IndexesPortfolio Approach – PCA
• Use factor analysis techniques to generate a set of pure indexes
– They can be thought of as the best possible one-dimensional summaries of information conveyed by competing indexes for a given style, in the sense of the larger fraction of the variance explained.
– Here, we are looking for the portfolio weights that make the combination of competing indexes capture the largest possible fraction of the information contained in the data from the various competing indexes
• The method– From a mathematical standpoint, it involves transforming a set of K correlated
competing indexes into a set of orthogonal variables, or implicit factors, which reproduces the original information present in the correlation structure
– Each implicit factor is defined as a linear combination of original variables– Use the first component of a PCA of competing indexes as a candidate for a
pure style index (typically captures a large proportion of cross-sectional variations because competing styles tend to be at least somewhat positively correlated)
Desperately Seeking Pure Style IndexesPortfolio Approach – Min Var Analysis
• Pure hedge fund indexes generated as the first component in a factor analysis should be as representative as possible since there is no other linear combination of competing indexes that implies a lower information loss
• Another approach consists in focusing on minimization of the bias
• Use same model as before
• Add assumptions– Noise term is zero on average– Noise term is uncorrelated with return on pure index– Homoscedastic model
Desperately Seeking Pure Style IndexesPortfolio Approach – Additional Assumption
• If one is willing to make the additional assumption of no correlation between noise terms for various competing indexes, then the variance-covariance matrix of residuals is diagonal, and one obtains the following simple solution
• May also impose positivity constraints to ensure that resulting index is a long-only portfolio of individual hedge funds
• Contribution– Document heterogeneity in competing hedge fund index providers– Attempt to provide remedies to the problem
• Extensions– We have suggested that a database of pure style indexes be
maintained at the EDHEC-MISYS Risk and Asset Management Research Center, and posted on a dedicated web site
– The index construction methodology • Step 1: Use the first 3 years of monthly returns to calibrate the model.• Step 2: Perform a PCA analysis of competing indexes for each strategy on
the data used for calibration purposes• Step 3: These portfolios are held for 3 months, their monthly returns are
recorded, and the same process is repeated
– Our results can easily be extended to traditional investment styles such as growth/value, small-cap/large-cap