cube Systematic Alpha Fund

A Machine-Learning Based Systematic Fund

January 2016
Table of contents

Rcube Asset Management

Rcube Systematic Alpha Fund Overview

Track record & Conclusion

Appendix: The case for using Machine Learning in

systematic trading

2
 RCUBE

Investment Process
Rcube Asset Management
Rcube Asset Management

Rcube stands for Research, Returns and Risk management

Started in 2011 as an investment research provider

Team of 6 experienced professionals

Currently managing two funds totaling $58m AUM

Rcube Global Macro Fund UCITS (launched in February 2014)
Rcube Systematic Alpha Fund (launched in May 2015)

4
Rcube Team
Founding Partners

Asset Management
Partners: Cyril Castelli (CEO), Paul Buigues

Risk management,
Systematic Global Macro Business
middle office,
funds funds Development
compliance

Paul Buigues Cyril Castelli Morgan Rossi Kati Kukkasniemi

Remi Takase Paul Buigues Agama Conseil Max Kamir

Systematic Alpha Investment Team

Paul Buigues, CIO Systematic Strategies, PM

Paul is a founding partner and CIO of systematic strategies at Rcube Asset Management. Paul has 17
years of experience in systematic trading (in credit, equities and global macro). For the past few years
Paul has focused on applying machine learning to systematic trading. Previously, Paul founded
Fimaxis, a quantitative investment research firm. Formerly, Paul was a fund manager at ADI, where he
managed up to $1Bn of assets in quantitative longonly high yield and credit arbitrage funds.

Paul holds an MSc in Engineering and an MSc in International Finance from HEC, Paris.

Remi Takase, PM

In addition to co-managing the strategy, Remi is responsible for coding the machine learning
algorithms used in the Systematic Alpha Fund. Before joining Rcube, Remi had gained experience in
machine learning and semantic analysis at Proxem SAS.

Remi holds an MSc in Engineering from ENSIIE, an MSc in Financial Engineering from the University
of Evry, and is certified by the AMF.

6
 RCUBE

Investment Process
Rcube Systematic Alpha Fund Overview
Quick facts

Investment Strategy Systematic trading based on Machine Learning

Expected return range 10-15% per year over the risk-free rate

Target volatility range 10-15% annualized

Investment Universe 13 highly liquid futures (in equities, rates, forex, commodities)

Fees (A shares) Management fees: 1.5% / performance fees: 15%

Fund type Cayman fund

Fund inception May 1st, 2015

Track record length 9 months since launch (plus 14 months as a sub-portfolio)

8
Fund description
The fund is a portfolio of tens of thousands of trading systems,
generated, selected and managed by machine learning algorithms.
Trading systems are based on:

Prices/returns
of the traded
Economic Other
assets
asset itself:
<0.1%

data: prices /
returns:
80% 20%

By design, each trading system has a neutral long-term average

exposure to the assets it trades (no structural beta => pure alpha)
9
Investment process

Step 1
Step 2
Step 3
Trading system
Trading system
generation & Execution
allocation
selection

For each of the assets Allocation to trading Daily execution of the

we trade, thousands of systems, subject to the aggregate portfolio of
trading systems are portfolios aggregate trading systems
selected volatility target

10
 Step 1: Trading system generation & selection
S&P 500 Estoxx 50 MSCI EM US 10Y DE 10Y EUR GBP JPY CAD AUD Gold
For each of the
13 assets we Copper Crude Oil
trade, machine
learning algos
generate and ...
select thousands
of trading
systems. Examples of trading systems generated by machine learning algorithms:
Asset: S&P 500 Asset: Eurostoxx 50 Asset: Gold
Input: SLOS Survey Input: German 10Y yields Input: Consumer Sentiment

LONG
Exposure
to the asset
SHORT

In-sample Sharpe = 0.44 Sharpe = 0.67 Sharpe = 0.64

performance
and Sharpe
ratio
11
 Step 2: Trading system allocation
S&P 500 Estoxx 50 MSCI EM US 10Y DE 10Y EUR GBP JPY CAD AUD Gold
13 assets
Copper Crude Oil

Thousands of
trading systems per
asset

S
Trading systems are aggregated according to:
Trading system - their past risk-adjusted performance
aggregation - their expected persistence

One aggregate
trading system for
each asset (9).

The allocation to aggregate trading systems is subject to the portfolios volatility target.
12
Step 3: Execution

Every night, trading systems are Typical daily trading activity (for $ 25m):
generated and aggregated. Current # of Target # of
Asset code Delta
At 9am EST, we calculate new target contracts contracts

positions for each asset.

ES1 Index 56 52 -4
VG1 Index 193 189 -4
MES1 Index -10 -10 0
After checking for potential errors, we TY1 Comdty
RX1 Comdty
-63
79
-53
86
10
7
trade the delta between current fund DX1 Curncy 88 88 0

positions and target positions (using GC1 Comdty

CL1 Comdty
-22
-21
-22
-21
0
0
execution algos). HG1 Comdty -23 -25 -2

We trade on average around 10% of the portfolios nominal size every

day.
Note: Trading the aggregation of thousands of trading systems vs.
trading individual systems dramatically lowers trading costs
13
Forward test procedure

Step 1
Step 2 Step 3
Trading system
Trading system Execution
generation &
allocation
selection

Rerun the process at times t, t+1, t+2, T

300 trillion operations needed to complete a full 15 year forward test.

14
Risk management

Risk management is embedded in the investment process (maximum

volatility target: 15% annualized)
Risk is also managed independently by our risk manager (using
HedgeGuard Financial Software)
If the NAV experiences a 10% drawdown, we cut the portfolio for a
minimum period of one week.

Discretionary interventions: When we believe that an assets former

drivers are not going to be relevant for a certain time, we can
discretionarily adjust (or eliminate) the strategys target position for the
asset. This can take place in the event of:
Major geopolitical events
Changes in the nature of an asset (e.g.: pegging)

15
Trading system diversification
Correlations between beta-neutral trading systems are considerably lower than
correlations between assets.
Reducing the average correlation between trading systems is the key success
factor for the Systematic Alpha strategy.

r=0

r=0.01

r=0.05

r=0.20
r=0.50

16
 RCUBE

Investment Process
Track record & Conclusion
 Fund performance since inception (May 1st, 2015)
Return since inception: 11.26% (15.29% annualized) Annualized volatility: 9.90%

RCUBE SYSTEMATIC ALPHA FUND NAV VS. SG CTA INDEX

1200

1150

1100

1050

1000

950
4/30/2015 5/31/2015 6/30/2015 7/31/2015 8/31/2015 9/30/2015 10/31/2015 11/30/2015 12/31/2015 1/31/2016

RSAF SG CTA Index

18
19
20
Rcube Systematic Alpha Fund: Conclusion

An innovative investment process based on the collective intelligence

of tens of thousands of virtual traders

23 month live track record

Expected Sharpe ratio > 1 (return and vol between 10 and 15%)

Investor relations: Kati Kukkasniemi [email protected]

21
General Fund Information

Fund name: Rcube Systematic Alpha Fund

Launch Date: May 1st, 2015
Prime Broker: Credit Suisse
Administrator: Bank of New York
Legal Advisor: Walkers
Minimum Investment: 150 000 $ (or equivalent in )
Liquidity: Monthly
Subscription / Redemption Notice: 10 Business Days
Domicile: Cayman Islands
Fees: 1.5% Management fees / 15% Performance fees
High-Water mark

22
 RCUBE

Investment Process
Appendix: The case for Machine Learning-based trading
What is Machine Learning-based trading ?

Machine Learning Systematic Trading

+
Giving computers Giving computers
the ability to learn the ability to trade

=
Machine Learning-based Trading
Giving computers the ability to learn how to trade
(and to trade accordingly)

24
 The discretionary / systematic trading continuum

0% systematic 25% systematic 50% systematic 75% systematic 95% systematic

Discretionary Systematic with
Pure Traditional Machine
trading with discretionary
discretionary systematic Learning-based
quantitative overrides (or
trading trading trading
inputs vice versa)

Machine Learning: the most extreme systematic trading approach.

25
Traditional systematic trading

75% systematic

Trading
Validated
system

Trading System designer Backtest Execution

Not validated

System abandoned due to weak live performances

Trading systems are the assets of traditional systematic trading.

Trading those assets can be as emotional as discretionary trading.
26
Machine learning-based trading
95% systematic

ML algorithms
Selected TS
Data universe

Automated TS TS
Coder / data scientist discovery & selection execution

New market observations=> learning

When ML is used, trading systems are discovered and selected by

algorithms. Humans only provide computer code and data sets.
27
Man vs. Machine: Trading System discovery speed

Manual discovery Automated discovery

TS Code

Manual validation

Trading systems can take 1 000 000

days to develop trading systems / hour

28
 Man vs. Machine: validation accuracy
Manual trading system selection Automated trading system selection

Selection point

Performance
Performance

Selection points

Time Time

Due to selection bias, a group of A continuous and automated

manually selected trading systems selection of trading systems enables
will underperform when traded live forward testing (no selection bias)
vs. backtests. Forward test results are close to live
Validation is therefore impossible performances (WYSIWYG)
(or at least misleading)
29
 Machine Learning: two alternative approaches
A small number of complex models A large number of simple models
- Each model has many inputs - Each model only has a few inputs
- Models are based on non-parametric - Models are based on simple and robust
techniques such as Neural Networks: statistical techniques

Input 1
Input 2
Input Output
Input 3
Input 4
Input 5

This approach is valid in fields
where rules do not change over Simple models are structurally

time. much more persistent than
complex ones.
It generally does not work in
finance, which is a complex => We opted for this approach
adaptive system.
30
Recap: Successful human traders vs. Machine Learning

Successful traders Machine Learning

Quick-witted
Disciplined
Unemotional
Patient
Thinks probabilistically
Able to adapt
Under continuous
Have a critical mind improvement

31
Machine Learning outperforms most human pros in:

Chess Giraffe (Lai, 2015)

Quiz competitions IBM Watson (Jeopardy)

Reading facial expressions MIT Affectiva

Criminal recidivism prediction (Maloof, 1999)

Why not in systematic trading?

32