Quants: Beyond multi-factor models
Nomura Global Quantitative Equity Conference
May 21st 2010
Generating alpha from quant models…
It is like the old story of the man who was going to fight a duel the next day.
His second asked him, "Are you a good shot?“
"Well," said the duelist, "I can snap the stem of a wine glass at twenty paces"
…and he looked modest.
"That's all very well," said the unimpressed second.
"But can you snap the stem of the wineglass while the wineglass is pointing a loaded
pistol straight at your heart?"
(Jesse Livermore Quote)
The story so far… or how we arrived here
Quant has been a long-term value creator
US Quantitative managers versus the S&P 500. Average relative
return and Range around it (1995 through first-half 2009)
%
25
20
15 Average
10
(5)
(10)
(15)
(20)
(25)
95 96 97 98 99 00 01 02 03 04 05 06 07 08 09
Source: eVestment Alliance, Empirical Research Partners Analysis.
1 Fifteen large core funds including products managed by Acadian Asset Management, Analytic Investors, Aronson + Johnson + Ortiz, Barclays Global Investors (select poducts),
Battertmarch Financial Management, Chicago Equity Partners, INTECH Investment Management, Jacobs Levy Equity Management, LSV Asset Management, Nicholas-Applegate
Capital Management, Numeric Investors, PanAgora Asset Management, Quantitative Management Associates and State Street Global Advisors.
Performance since 2006 has not been stellar, but positive nonetheless.
2009 deserves an explanation
US Quantitative managers: Share of Actively-Managed Equities Held
in Defined-Benefit Plans (LH) and average performance (RS)
20 % 6.0
18 5.0
16 4.0
14
3.0
12
Average = 2.9% 2.0
10
1.0
8
0.0
6
4 (1.0)
Average = 0%
2 (2.0)
0 (3.0)
00 01 02 03 04 05 06 07 08 09
Source: Federal Reserve Board, Pensions & Investments, Greenwich Associates, eVestment Alliance, Empirical Research Partners Analysis
A tragedy in four acts: how quant moved from value-based models
to momentum and „dynamic‟ factor allocation
Hedge Fund Market Neutral Index (LHS)
Correlation with the performance of momentum factors (RHS)
4
Momentum breaks
down
1150 0.80
Average = 60%
1100 2 0.60
Correlation to
momentum
increases
1050 0.40
1000 0.20
3
950 Aug 07: models -
Average = 0.0 become more
‘dynamic’
900 1 (0.20)
Value starts to fail
850 (0.40)
Mar- Sep- Mar- Sep- Mar- Sep- Mar- Sep- Mar- Sep- Mar- Sep- Mar- Sep- Mar-
03 03 04 04 05 05 06 06 07 07 08 08 09 09 10
Source: HFRX Equity Market Neutral Index; Aviva Investors
Our story so far…
● Multi-factor models were initially value driven models
● Gradually, as value became out of favour, models became more momentum driven
● Increasing quant research and new entrants caused them to converge
● As we all realised this in August 2007, we enhanced the models
● Part of this enhancement meant more ‘dynamic’ models: which lead to more momentum
● These dynamic approaches broke down in May 2009
Evolution… and the fallacy of optimisation
Multi-factor models evolved in different ways
Evolution of multi-factor models
1
● Models based on either momentum or dispersion of factors
● Momentum based rotation creates significant exposure to reversals
Dynamic models
● Dispersion based rotation can lead to periods of poor performance
2
● Efforts to identify new alpha factors lead to a lot of ‘data mining’
● No compelling evidence that new factors were identified after Aug 07
New alpha models
● Better and more diversified models were a tangible output
3
● Do not fall into the ‘momentum’ and ‘dispersion’ traps
● Replicable by passive factor strategies
Static models
● Consultants and other buyers believe it lacks innovation / research
Source: Aviva Investors
Multi-factor models are created from a combination of market
anomalies
Building up multi-factor models
In building multi-factor
models, we aim to…
Relative 12%
Return
(Q1-Q5) 11%
Capital Allocation ● Maximise our
Quant information ratio (i.e.
Momentum
Model
10%
9%
return per unit of
volatility)
8%
● Maximise our hit-rate
(i.e. number of months
Value we outperform the
7%
Growth market)
6%
● Minimise our turnover
5%
4% Earnings Quality ● Minimise our
drawdowns
3%
2%
5% 7% 9% 11% 13% 15%
Tracking error
Source: Aviva Investors
The fallacy of optimisation
● Our desire for monthly/quarterly leveraged out-performance lead us to all of this
● We forgot that these inefficiencies exist because they do not work all the time
● In optimising our models for higher hit-rates, we were undermining the alpha of the
underlying strategies
● Value investors did not stop being value investors in 1998-1999, but we stopped
believing in what we were doing because it did not work in 2007-2008
● …But the alpha from these strategies is still there
Breaking the paradigm: decomposing our alpha
A number of sub-strategies have emerged
Quant sub-strategies
1
Enhanced Multi-Factor
Models
Static and dynamic models 5
2
Quant +
!
Alpha Decomposition
Fundamental
Quant
Selling Implied alpha in segregated strategies
Quant model + analyst input Universe
3 4
High Frequency trading Mathematical
Algorithmic trading Relative volatility and correlation
Source: Aviva Investors
Quants can do more by acting at SAA level
Strategic Asset Allocation Process
1 2 3
Constraints Strategic Asset Quant
Manager Selection manager
Established Allocation
● ALM Study ● Long term economic drivers ● Philosophy
● Competitor’s Review ● Sustainable asset returns ● Process
● Broader Constraints ● Portfolio optimisation ● People
● Performance
‘Equities’ as
Performanc
cap-
e versus
weighted
benchmark
benchmark
Source: Aviva Investors
Asset allocators see “equities” as a single asset class represented
by a cap-weighted index
Mean-variance comparison: S&P500 versus non-cap weighted alternatives
10%
9%
EM equities
EM Inflation-Linked Bonds
8% UK equities AP equities
Global equities
7% EU equities
Global Converts NA equities
UK Property
Sustainable return
6% UK IG Credits
JP equities
EM USD Bonds
5% Global IG Credits
UK Gilts
EU Bonds
4% Global Developed Bonds
US Bonds
UK index-linked
3%
2% JP Bonds
1%
UK Cash
0%
0% 5% 10% 15% 20% 25% 30% 35% 40%
Expected Volatility
Source: Aviva Investors
„Equities‟ should not be synonymous with to cap-weighted indexation
Mean-variance comparison: S&P500 versus non-cap weighted alternatives
Realised 18%
Return
Intellidex
DJ Select Dividend
17%
16%
Mergent VTL
15%
GWA
14% RAFI 1000
13%
WisdomTree
Dividend
12%
S&P 500
11%
10%
10% 11% 12% 13% 14% 15% 16% 17% 18%
EDHEC Risk and Asset Management Research Centre
Alpha decomposition: Illustrative example (1)
Fundamental strategies, as a proxy for the value premium
Fundamental Strategy (Value Driven): UK Market 1990-2010
160
150
140
130
120
110
Potential for significant periods of
underperformance
100
90
80
90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10
Source: Aviva Investors
Alpha decomposition: Illustrative example (1)
There are ways to measure the potential from this strategy
United Kingdom
Composite valuation dispersion
1988-2010 United Kingdom
Composite Valuation Dispersion
As Of 04/23/10
4.0
3.0
Very High
2.0
1.0
High
0.0 Normal
Low
-1.0
Very Low
-2.0
88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10
Source: Bernstein Quant Research
Alpha decomposition: Illustrative example (2)
Low vol strategies present much higher risk-adjusted returns
Global portfolio: relative returns and Sharpe ratios of low vs high volatility strategies
5.0 0.60
Alpha (LHS) %
4.0
Sharpe Ratio (RHS) %
3.0 0.50
2.0
0.40
1.0
0.0
0.30
-1.0
-2.0
0.20
-3.0
-4.0 0.10
-5.0
-6.0 0.00
D1 D2 D3 D4 D5 D6 D7 D8 D9 D10
Source: Aviva Investors
Alpha decomposition: Illustrative example (2)
There are also ways to measure the potential from this strategy
Volatility Strategies: Hit Rates versus direction of the market
90.0
80.0
Up
Down
70.0
60.0
Hit rate (%)
50.0
40.0
30.0
20.0
10.0
0.0
D1 D2 D3 D4 D5 D6 D7 D8 D9 D10
Decile
Source: Aviva Investors
Alpha decomposition: Illustrative example (3)
Index funds continue to grow… and so arbitrage opportunities
US: Index Fund and ETF Assets as a share of all long-term
mutual fund and ETF Assets (1993 Through July 2009)
%
18
16
14
12
10
0
93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 Memo:
Institutions
Index Funds ETFs Indexed
Share 08
Source: Investment Company Institute, Greenwich Associates, Empirical Research Partners Analysis and Estimate.
Alpha decomposition: Illustrative example (3)
Index arbitrage: an anomaly that has been increasing in size
S&P 500: Relative performance of additions versus deletions
between announcement and effective dates
20%
14.7%
15%
Additions
Deletion
10%
5.0%
5% 3.9%
0%
-2.8%
-5%
-5.9%
-10% -8.5%
Announcem ent Date Announcem ent date to Event Date Post-Event Date
Source: The price response to S&P 500 index additions and deletions: Evidence of asymmetry and a new explanation. Chen,
Noronha and Singal, The Journal of Finance, 2004 (data analysis from July 1962 to December 2000)
Alpha decomposition can fight cap-weighted indexation
● Why are we allowing such huge pools of assets to be moved to passive strategies when
we can offer better risk-adjusted solutions?
● Cap-weighted indexation is a poor solution: it has the known problem of overweighting
overpriced securities – and even equal-weighted portfolios can generate better risk-
adjusted returns.
● In order to be able to pitch our proposition to asset allocators and fight indexation, we
should simplify and decompose our strategies – and therefore be much more
transparent about the alpha opportunity we are presenting.
This is not about a “new passive” approach
● Implementing any of the above-mentioned strategies is far from trivial. It requires
extensive modelling to understand risk-adjusted returns, hit-rates, drawdowns and
turnover associated with each of them.
● They can often be implemented in a number of different ways (sector neutral, beta
neutral, maximum weights for single stocks).
● Only Quants can perform the necessary modelling to successfully implement these
strategies. And all of them should represent a better solution than classical indexation.
Breaking the paradigm: decomposing our alpha
Alpha decomposition, the story so far…
Delivering Alpha Decomposition
Multi-factor Quant Multi-factor Quant Multi-factor Quant Multi-factor Quant
Models Models Models Models
Global Income Global Income Global Income
Funds Funds Funds
Fundamental Fundamental
Strategies Strategies
Increasing depth Index Arbitrage
of research
Minimum Variance
Quant will continue to develop into a multitude of decomposed alpha
strategies
Decomposing multi-factor models into new products
Fundam. Income
Funds funds
Index Multi-factor Quant Minimum
Arbitrage Models Variance
Long-Only and Long-Short Funds
Trend Structured
Following? Solutions
Quality
Strategies?
Quantitative Research
Trend Following High Quality Structured Solutions
Recap on the main points discussed today
● Multi-factor models will continue to be the core of our proposition as quants, and each of
us will continue to try to differentiate ourselves in this space
● But there is more to our proposition than multi-factor models. We should fight cap-
weighted indexation
● We are the only ones that can provide a credible alternative to this trend
● For doing that, we should be prepared to decompose our alpha and work together with
the asset allocation process
