Quantitative Investment Strategies: The Unintended Consequences Vadim Zlotnikov Chief Investment Strategist and Director of Quantitative Research Sanford C. Bernstein & Co. LLC. January 8, 2008 osure Appendix of this report for important disclosures and analyst certi
37
Embed
Quantitative Investment Strategies: The Unintended Consequences
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.
Returns to Changes in Current Accruals are Non-linear
1979–June YTDRelative Annual
Returns (%)
Source: Bernstein Analysis.
Bernstein Research14
Constructing Models – Critical Considerations
Bernstein Research15
Critical Aspects of Stock-Selection Model Construction
Investment Management’s Issues
Universe/style
Time horizon/turnover/liquidity
Hit rates, persistence, IR
Fundamental analysis/transparency
Analytical Issues
Negatively correlated factors
Simplicity (“good enough”)
Avoidance of data mining/overfitting
Updating of factors/weights
Bernstein Research16
Analysis of Quant Models: Factors Are Similar, but...
Source: Bernstein survey of 25 buy-side quantitative models, where rankings for S&P 500 stocks were provided; January, 2007
Survey ResultsFactor Exposure of Quantitative Models
0.0 0.5 1.0 1.5 2.0 2.5
Price-to-Normalized Earnings Ratio
LBO Value
Total Cash-to-Market Capitalization
P/E-to-Long-Term Growth Forecast
Average 9-Month Price Trend
Total Yield
Enterprise Value-to-EBIT
Price-to-Gross Cash Flow
Deviation from Average
TraditionalFactors
LessTraditional
Factors
Bernstein Research17
…Completely Different Buy/Sell Recommendations - Model Construction Is Key
Survey ResultsDegree of Overlap in the Rankings of S&P 500 Stocks
First Quintile and Fifth Quintile Based on Quantitative Models
Degree of Overlap in Rankings
# of Stocks Ranked
369
108
23
368
102
30
0
100
200
300
400
500
Low <20% Medium 21-40% High >40%
Q1 Q5
Source: Bernstein survey of 25 buy-side quantitative models, where rankings for S&P 500 stocks were provided; January, 2007
Bernstein Research18
“Controversy” Stocks Illustrate the Problem
"Controversy" Stocks: Simultaneously Rankedas Q1 and Q5 by More than 25% of Models
Ranked as of 6/30/06
DHI D R Horton Inc. 38.1% 38.1%KBH KB Home 33.3 33.3WPI Watson Pharmaceuticals Inc. 23.8 23.8XTO Xto Energy Inc. 23.8 23.8MWV Meadwestvaco Corp. 23.8 23.8WHR Whirlpool Corp. 33.3 23.8JBL Jabil Circuit Inc. 23.8 23.8AIG American International Group 23.8 23.8DELL Dell Inc. 28.6 23.8CZN Citizens Communications Co. 28.6 23.8AAPL Apple Computer Inc. 23.8 28.6PFG Principal Financial Group Inc. 33.3 23.8NVLS Novellus Systems Inc. 33.3 23.8PHM Pulte Homes Inc. 23.8 33.3Average 28.0% 27.1%
Ticker Company NameQ1 by %of Models
Q5 by %of Models
Source: Bernstein analysis June 30, 2006The companies discussed are for illustrative purposes only. Any fund managed by AllianceBernstein L.P. and distributed through its subsidiaries securities or investment interests in these companies at any given time.
Bernstein Research19
Limitations and Challenges of Quantitative Approaches
Battling the Efficient Market Hypothesis
Data mining and spurious correlations
Risk factor vs. source of excess return
Rational agents with constraints vs. behavioralists
"Knowing" When a Strategy Failed, Is Failing or Will Fail
Can't always wait for statistical significance
Sometimes don't know why it worked
Easier if you have robust expectations
Underwriting volatility (or risk) for short-term profit
Bernstein Research20
Issues in Quantitative Research
What are the new factors?
When should they work?
Where are they most effective?
Methods for integrating signals and constructing portfolios
Bernstein Research21
What Are the New Factors?Moving Beyond the Compustat/FactSet
Nature of investor ownership; attention
Internet as source of fundamental data, e.g., Webcrawlers
Alternative asset classes as signals, e.g., options, futures, swaps…
Third-party market share, patent and other data
Bernstein Research22
When They Should Work? Dynamic Factor Timing
Changes in macro-economic setting; risk regimes
Seasonality/cyclicality
Technical: serial correlation vs. mean reversion
Bayesian updating
Presence of an opportunity (e.g., dispersion)
Bernstein Research23
Where Are They Most Likely to be Effective?Universe and Factor Conditioning
Static vs. dynamic universe definition
Level of granularity: style, sector, industry, stock
Tails of the returns distribution; shorts vs. longs
Bernstein Research24
Integration and Updating of Factor Weights
Numerous approaches for determining initial factor weights:
In-sample regressions
Principal component analysis
Optimizer of factor weights
Likewise, several approaches for updating the factor weights:
Bayesian updating
Rebalancing of the conditional universes
Desirable to match target portfolio turnover and factor efficacy duration
Integration of investment and trading alphas – explicit trading costs
Bernstein Research25
0
10
20
30
40
Signals Basedon AlternativeAsset Classes
StochasticFactor Weights:
LearningModels
Varying FactorWeights:
DeterministicModels
PortfolioConstruction
Other
This Year Last Year
Portfolio Construction and Factor Timing Are Primary Areas for Future Research
Most Promising Research Area to Deliver Future Outperformance in U.S. Equity Market Is:
(%)
Source: Bernstein Survey of Analysts and Portfolio Managers 2007
Bernstein Research26
Models Are Being Actively Modified to Incorporate Findings
During the Next 12 Months, the Most Significant, Revolutionary Change to Our Quantitative Models Will Include:
0
10
20
30
No MajorChangesPlanned
Risk Models,Portfolio
Construction
Dynamic FactorWeight
Allocation
More GranularModels
Other
This Year
Last Year
(%)
Source: Bernstein Survey of Analysts and Portfolio Managers 2007
Bernstein Research27
Summary
Growth in deployment of quantitative tools is likely to persist
"Commoditization" of factors means a shift in the nature of value-added
Integration of quantitative and fundamental research is still suboptimum
Bernstein Research28
Quantitative Approaches: What Went Wrong?
Bernstein Research29
Margin Sustainability is Key to Investment Outlook
S&P 500: Price-to-Sales v s. FCF Yield Minus 10-Year Treasury1965 Through Early-November 2007
Source: Bernstein Analysis.
Bernstein Research30
Collective Extrapolation of Historically Lowest Volatility Drove Turmoil and Failure of Quantitative Strategies
S&P 500: Market Volatility*1874 Through End-October 2007
Source: Bloomberg, Ibbotson, Robert Shiller, Bernstein Analysis.
* Standard deviation of trailing-six-months of S&P500 monthly total returns; data smoothed over trailing-12-months.
Bernstein Research31
Volatility Shock Drove “Anti-value” Market
Large-Cap Core Universe Discriminate Analysis of Top/Bottom 10% of Stocks,
Past 3 Months
Source: Bernstein Analysis.
Past Three MonthsDiscriminant Q1-Q5
Function Partial Avg. MonthlyFactor Coefficient* R-Square SpreadBook-to-Price - 17.3% -4.2%Market Cap (Log) + 7.6% 3.1%EBIT-to-Enterprise Value - 5.9% -3.9%Monthly Earnings Revisions + 2.4% 2.7%Five-Year Sales Stability + 2.6% 3.0%YoY Sales Growth + 1.5% 3.5%ROE Volatility** + 1.4% -0.5%* "+" sign represent that factor correlates positively with performance; "-" sign represents that factor correlates negative with performance** top quintile is lowest roe volatility
Bernstein Research32
Only Modest Misvaluations Emerged Among Large Caps
Large-Cap UniverseDispersion of Book-to-Price vs. Free Cash Flow Yield*
Through Late-October 2007
Source: Bernstein Analysis.
* Data smoothed over three-months.
Bernstein Research33
However, Illiquidity Premium Up Sharply
Dispersion Across Stocks in Book/Price and Free Cash Flow Yield vs. Past 7 Years
Small-/Mid-Cap Universe 1968-Mid-October 2007
Source: Bernstein Analysis.
Bernstein Research34
Low Multiple Stocks with High Turnover Underperformed – Capitulation and De-leveraging are the Culprits
Annualized Monthly ReturnsCrosstabs of Abnormal Turnover
December 2006 through October 2007
Source: Bernstein Analysis.
Enterprise ValueEnterprise Value-to-EBIT -to-EBIT as of
Dispersion of Book-to-PriceFinancials Vs Consumer Cyclicals vs. Technology
Through Early-November 2007
Source: Bernstein Analysis.
Bernstein Research36
Factors worth monitoring to determine persistence of stability
Key Risks to the Persistence of Stability
Risk Current Situation Variables to Monitor Potential Path to Instability1. Contagion from subprime Delinquencies rising Housing inventory rise Drop in housing wealth leads
Credit tightening Contagion to prime to accelerating defaults andAdverse impact on housing credit; state-level data lower consumer spending
2. Slower foreign inflows into US/Euro rate narrowed and Strengthening in Japanese Higher long-term ratesUS fixed income instruments contributed to declining economy and yields exacerbate housing problemsas dollar weakens dollar on long-bond and reduce Fed's flexibility
3. Wage inflation in Asia and Chinese wages rising 15-20% Size of margin pressureRaises doubt as to sustainability
weaker dollar Strength in Euro, Rupee on key outsourcers and of historically peak margins;Rise in prices on goods importers reignites inflation concerns
4. Dysfunctional policy for Every Democratic Signs of protectionist Market sell-off as investorsincome disparity candidate favors higher sentiment revival; attempt to lock-in capital gains
income, capital gains and Congressional support for at lower rates; negative fordividend tax rate tax rate increases high dividend payers
Bernstein Research37
What is Next? Adaptive Systems at Work
Re-emergence of exploitable illiquidity premium
Greater emphasis on:
Earnings quality, stability
Relative growth
Absolute, as opposed to relative, value
Avoidance of recent mistakes, pursuit of ones from log ago