Crafting Portable Excess Return by Investing in Liquid Commodity Futures
1. s
Universal Alpha Factory: Crafting Portable Excess
Return by Investing in Liquid Commodity Futures
European Alternative Investment Summit
a marcusevans summit series event
5-7 November 2008 | Fairmont Le Montreux Palace | Montreux | Switzerland
2. s
Disclaimer
The views and opinions expressed in this presentation are those of the authors
only, and do not necessarily represent the views and opinions of Siemens AG,
or any of its employees. The authors make no representations or warranty,
either expressed or implied, as to the accuracy or completeness of the
information contained in this presentation, nor is he recommending that this
presentation serves as the basis for any investment decision. This presentation
is prepared for the European Alternative Investment summit on 5-7 November
2008 in Fairmont Le Montreux Palace, Montreux, Switzerland only. Research
support from fin4cast is gratefully acknowledged.
Dr. Miroslav Mitev - Siemens AG Österreich, Siemens IT Solutions and Services,
Program and System Engineering, Fin4Cast, Gudrunstrasse 11, 1100 Vienna,
Austria, Phone: +43 (0) 517 07 46253, Fax: +43 (0) 517 07 56256, email:
info@fin4cast.com, www.fin4cast.com/indices.
11 08
2
3. s
Agenda
Definition of Beta and Alpha
Separating Alpha from Beta
Inter-dependences between different asset classes
Maximizing returns through commodity exposure
Generating Alpha from long & short exposure to
commodities using liquid futures
Measuring the effect of porting Alpha to core
investment
Conclusion and Q&A
11 08
3
4. s
Definition of Beta
In general Beta represents the market return (Risk Premium) of an
asset class
Depending on investor’s objectives the Beta could be defined as:
the return of the stock market (DJ Industrial Average Index)
the return of the bond market (U.S. Treasury Note)
the return of the commodity market (DJ AIG Commodity Index)
the return of the currency market (EUR/USD Exchange Rate)
the return of investor‘s liabilities (Liability Index = Zero Coupon Bonds)
Depending on the way investors take exposure to Beta we could
distinguish between:
Traditional Beta, i.e. the long exposure through buy and hold of futures,
ETFs, etc.
Alternative Beta, i.e. the rotation between the traditional betas and
taking advantage of short exposure (CS Tremont Hedge Fund Index)
11 08
4
9. s
Alternative Beta
11 08
Source: Thomson Reuters
9
10. s
Definition of Alpha
In general Alpha represents the excess return vs. a given benchmark
Per definition Alpha can not be replaced or explained by the existing
traditional and alternative Betas, i.e. it has a very low correlation to
Beta
Alpha can only be generated by taking active bets and is subject to
manager’s skills, i.e. Know-How and technology
Depending on investor’s objectives we can distinguish between:
Relative Alpha, i.e. the relative out-performance against a given
benchmark which is usually measured by the information ratio
Absolute Alpha, i.e. the absolute excess return above a pre-defined
threshold return usually measured by the Sharpe Ratio
An example for a commodity Alpha prepared for this presentation is
the fin4cast Commodity Index which benefits from long and short
positions in 13 liquid commodity futures
11 08
10
12. s
Beta and Alpha Sources
Source: Thomson Reuters
11 08
12
13. s
Separating Alpha from Beta
Yt = α + β * X t + ε t
Traditional beta:
Stock Market Return
Return Alpha = Skill = Residuals
Market Risk
Return of different
Alternative
Pure Asset Classes
Return Residuals
Alpha =
Beta
Traditional Beta
Yt = α + δ * At + β * X t + ε t
Alternative Beta:
Yt = α + β1 * X 1 t + β 2 * X 2 t + β 3 * X 3 t + β 4 * X 4 t + L + β k * X k t + ε t
Commodity Bonds Stocks Currency Hedge Funds Commodity
Alpha Risk Risk Risk Risk Risk
11 08
13
14. s
Interdependences between the asset classes (March 1999 – Sep 2008)
Rotated Matrix of the Principal Components a
Components
1 2 3
DJIA .797
10 year US
-.720
T-Note
CS HFI .565 .518
EURUSD .831
DJ AIGCI .642 .374
FIN4CAST .952
Method: Principal Components Analyse. Rotation:
Varimax with Kaiser-Normalisation.
a. The rotation converged after 7 iterations.
11 08 Multi – Correlation Coeffitien represents the average correlation to all other Betas and Alpha
14
Value Added Coeffitient = ABS (Sharpe Ratio/Multi-Correlation Coeffitien)
15. s
Interdependences between the asset classes (March 1999 – March 2003)
11 08
15
16. s
Interdependences between the asset classes (April 2003 – July 2007)
11 08
16
17. s
Agenda
Interdependences between the asset classes (July 2007 – September 2008)
11 08
17
18. s
Maximizing returns through commodity exposure
Agenda
11 08
18
19. s
Generating Alpha from long/short commodity exposure
Case study: fin4cast Commodity Index
benefiting from the most liquid commodity futures across
agriculture & live stock, metal and energy sectors by combining
long and short futures positions.
Eligible commodity futures:
Agriculture & Live Stock: Metal: Energy:
Corn (CBoT) Copper (COMEX) Natural Gas (NYMEX)
Soybean (CBoT) Gold (COMEX) Light Sweet Crude Oil
(NYMEX)
Wheat (CBoT) Silver (COMEX)
Coffee (NYBoT) Palladium (COMEX)
Cotton (NYBoT Platinum (COMEX)
Sugar (NYBoT)
Lean Hog (CME)
Live Cattle (CME)
11 08
19
20. s
Asset allocation as of 27th October 2008
11 08
20
21. s
Commodity long/short exposure YTD 2008
11 08
21
22. s
Performance attribution YTD 2008 (Agriculture)
11 08
22
23. s
Performance attribution YTD 2008 (Agriculture)
11 08
23
28. s
Measuring the effect of porting Alpha to the core investment
11 08
28
29. s
Thanky you very much for your attention!
Q&A
11 08
29
30. s
Appendix: Alpha-generation process
Forecasting
Selection of leading indicators
Evaluation of forecasts
Selection of forecasts
Portfolio construction
Trading
11 08
30
31. s
Modelbuilding & Forecasting Process
From Data Acquisition to Forecasts Generation
Data storage,
Data Input pre-selection Input Selection
processing &
Acquisition
cleaning
Criteria: Search Algorithms:
• Reuters • economical • Neighborhood search
• Thomson • statistical • Iterative improvement
Financial approaches
• Genetic Algorithm
Linear Models
Forecast Post analysis
• ARIMA/SARIMA
Comparative in sample and out of
• VAR/VARX
sample tests
• Factor Models
(Forecast Statistics)
• ARCH/GARCH
Evaluation
rejected
Estimation methods:
AOLS, WOLS, SUR, ML.
Forward tests
(Forecast Statistics) Non Linear Models
• Single & Multi Output MLP
Evaluation
rejected
Learning Algorithms
Forecasts • Steepest Descent
• Quick prop
11 08
31
32. s
Input Selection for the Mathematical Forecasting Models
Original Economical Technical Statistical Input Set Search Optimized
Input Set Criteria Analysis Analysis Algorithm Input Set
app.. 2000 app.. 800 app.. 3500 app.. 100 app.. 20
Time Series Time Series Time Series Time Series Time Series
Macro
gs
Correlation &
La
Economic
Stationarity Regression
Interest Analysis
Correlation
Rates
AN Algorithm
Dynamic
Price Data
Correlation Generic
Currency Algorithm
Normality
Rates
Economical
Granger
etc. Selection
Causality
Grading
Sensitivity
Stochastic max. 20 most
Analysis
Oscillators important
driving factors
Relative
Principal of the future
Differences
Component & returns of a pre-
(Exponential) Factor
specified asset,
Moving Analysis
e.g. S&P 500
Average
Cluster Future
11 08 etc. Reduction
32
33. s
Building & Evaluating of the Mathematical Forecasting
Models
Linear Modeling
Forecasts
Internal Selection of
Model &
Number of Factors and
Method
Inputs Forecast
Post-analysis
ARIMA/SARIMA
Optimized
Input Set VAR & VARX • Correlation
Factor Models • R2 &
ARCH/GARCH extended R2
• Hitrate
• Residual
Non Linear Modeling
Analysis
• Normality
Model &
Network Topology and Tests
Method
Parameter Tuning • etc.
Single Output MLP
Multi Output MLP
11 08
33
34. s
Selecting of the best Mathematical Forecasting Models
Use of
Model
In Sample Out of Sample Forward
Combination Models
500.000 Models 200.000 Models 50.000 Models
today live calculation of the mathematical models
1. Nov 2003
1. Jan 2000
(model compilation)
Evaluation of Selecting the Continuos
Postanalysis of accuracy
Model building
best
of forecasts accuracy of adjustment
• Building the basic model forecasting
min. 30 weeks forecasts and
Models
• linear vs. non linear
min. 4 weeks optimization
• stability of the model •Baysian
• can take several weeks
in real environment Model
• Adjusting and
to find optimal model
Averaging
Optimizing
•AIC & BIC
• real testing
Model
Combination
During the „Out-of-Sample“, „Forward“, and „Use of Model“ Process the mathematical
11 08
model is adjusted periodically to the changing market environment!
34
35. s
Portfolio Construction Process
From Forecasts Generation to Asset Allocation
Actual Portfolio Objective Function
Weights Maximize
φ(x) = pTx – ½ R xTQx – SC(x0, x)
Forecast for each Maximization of expected portfolio
asset return by simultaneous minimization
of expected portfolio risk and
Inputs for the Portfolio Construction
return forecasts implementation costs for the
Long/Short
respective coming period
directional forecasts
Asset Allocation
forecasts of the returns’
distribution
e.g.
Portfolio Optimization
Risk matrix + 15%
•Quadratic Optimization
- 20%
•Ranking
estimated variance-co-
- 10%
variance matrix (market
risk) + 30%
estimated residual
Constraints
diagonal matrix
(forecasting & model
risk) Market Neutrality, Long/Short,
Exposure, etc.
estimated slippage
(implementation risk) Min. or max. investment to a
single asset or an asset class
Combinatorial constraints
Risk aversion
Turn-over constraints
11 08
35
36. s
Strategy Implementation Process
From Asset Allocation to Order Execution & Portfolio Analysis
in-house or external Application Server
institutions 13 Portfolio Reconceliation, Portfolio
Proposed Asset Allocation &
1
Analysis & Risk Management
Consistency Checks
Confirmed weights &
number of contracts •Slippage Analysis
Internet
(128 Bit SSL) •Implementation Short Fall
Pre-Trade Analysis
•Return/Risk Analysis
2
•Stop-Loss
3
•If-than & Stress Tests
12
FIX Engine Scenarios
4 FIX 4.2 11
Private Network
Brokers
FIX Engine
Exchange(s)
reject
10
5
Consistency Checks
Confirmation
Orders
of the 9
6
Execution
Trading System
7
Interfaces
8
11 08
36
37. s
Commodity indices used in the presentation
Goldman Sachs Commodity Index: The S&P GSCI™ is a composite index of commodity sector returns
representing an unleveraged, long-only investment in commodity futures that is broadly diversified
across the spectrum of commodities (Energy 73.86%, Metals 8.73%, Agriculture 13.14%, Live Stock
4.26%) . The returns are calculated on a fully collateralized basis with full reinvestment. The combination
of these attributes provides investors with a representative and realistic picture of realizable returns
attainable in the commodities markets. Individual components qualify for inclusion in the S&P GSCI™ on
the basis of liquidity and are weighted by their respective world production quantities. The principles
behind the construction of the index are public and designed to allow easy and cost-efficient investment
implementation. Possible means of implementation include the purchase of S&P GSCI™ related
instruments, such as the S&P GSCI™ futures contract traded on the Chicago Mercantile Exchange (CME)
or over-the-counter derivatives, or the direct purchase of the underlying futures contracts.
The Dow Jones - AIG Commodity Index (DJ-AIGCI)® is composed of futures contracts on 19 physical
commodities. The component weightings are also determined by several rules designed to insure
diversified commodity exposure (Energy 33%, Metals 26.2%, Agriculture 30.3%, Live Stock 10.5%).
Investors may invest in the Dow Jones AIG Commodity Index buy purchasing futures contracts traded on
CBOT (Chicago Board of Trade). Alternatively, they may also purchase Pimco Commodity Real Return
Fund, which mimics the returns of the Dow Jones AIG Commodity Index.
The PHLX Gold and Silver Index is a capitalization-weighted index composed of the common stocks of
nine companies in the gold and silver mining index. The index is a product of the Philadelphia Stock
Exchange and began trading in January 1979 with an initial value of 100.
11 08
37
38. s
Biography
Dr. Miroslav Mitev
Siemens AG Österreich
Siemens IT Solutions and Services
PSE/fin4cast
Phone: +43 (0) 51707 46253
Fax: +43 (0) 51707 56465
Mobile: +43 (0) 676 9050903
Email: miroslav.mitev@siemens.com
Dr Miroslav Mitev is a managing director and head of quantitative research and strategy development at
Siemens/fin4cast. Dr Mitev is responsible for the development of innovative, systematic long-short investment
strategies for institutional investors world wide based on Siemens/fin4cast technology. After joining Siemens in
2001 Dr Mitev successfully formed a qualified team of 25 professionals which is continuously developing the
Siemens/fin4cast Technology and building mathematical forecasting models for a variety of financial
instruments like currency futures, commodity futures, stock index futures, bond futures, single stocks and hedge
fund indices. Dr Mitev is in charge of the Siemens/fin4cast’s research cooperation with various universities and is
actively involved in the scientific management of numerous master thesis and dissertations. Dr Mitev is a regular
speaker at international conventions on liability driven investing, asset management, hedge funds, portable
alpha, advanced quantitative studies, algo-trading and system research. Dr Mitev’s research is published on a
regular basis in international journals and presented on international scientific conferences. Prior to joining
Siemens Dr Mitev was at CA IB, the Investment Bank of Bank Austria Group, where he was in charge of the
quantitative research of the securities research division. Dr Mitev received a Master of Economics and Business
Administration with main focus on Investment Banking and Capital Markets. Dr Mitev also received a PhD in
Economics with main focus on Finance and Econometrics.
11 08
38