Qopius A.I Global Asset Allocation model
5 likes1,320 views
Qopius has developed an artificial intelligence asset allocation model and financial engine based on years of research in artificial intelligence and deep learning techniques. The model analyzes data from global financial markets to adjust exposures across asset classes and achieve absolute returns. It uses evolutionary algorithms and deep learning to generate predictions and combines numerous predictions to determine optimal outcomes. The model and engine demonstrate strong performance in backtests and paper trading, with returns that are uncorrelated to traditional asset classes in the long term. Qopius also provides an automated trading system to implement the model's strategies.
More Related Content
Similar to Qopius A.I Global Asset Allocation model (20)
Qopius A.I Global Asset Allocation model
- 1. FROM CUTTING EDGE A.I. RESEARCH TO INNOVATIVE FINANCIAL MODELS Qopius - Private & Confidential 1
- 2. CONTENT Qopius - Private & Confidential 1. QOPIUS ARTIFICIAL INTELLIGENCE ASSET ALLOCATION MODEL 2. QOPIUS FINANCE 2
- 3. QOPIUS A.I ASSET ALLOCATION MODEL
- 4. Qopius - Private & Confidential 4 ASSET ALLOCATION FOR AN EVOLVING WORLD A global and flexible Artificial Intelligence multi asset model Goal: Absolute return performance, uncorrelated in the long term to traditional asset classes. Approach: A multi-strategy model, with flexibility to adapt to different market environments over time, increasing efficiency and robustness. Tool: An Artificial Intelligence, quantitative and systematic system inspired by neurosciences designed to behave optimally in highly complex and unstable environments. Universe: The model daily adjust the exposure of the 40 most liquids financial instruments across all asset classes and geographic areas, depending upon deep price analysis, correlations, risks and market sentiment. Flexibility is key in todays rapidly transforming global economy.
- 5. Qopius - Private & Confidential 5 INVESTMENT UNIVERSE The universe contains the 36 most liquid instruments available, representing all asset classes and geographic areas. All the Asset are UCITS IV Compliants. 24 stock index futures. 6 long term (10 years) government bond futures. 3 short term (2/3 years) government bond futures. A Gold future. A commodities index (DJP ETF). A Volatility index future (VIX Index).
- 6. Qopius - Private & Confidential 6 RISK MANAGEMENT Two layers of risk control rules: 1- A Value-at-Risk (VaR) approach monitors and measures risk exposures at all times. The Fund is subjected to an absolute VaR limitation of 15% over a 1 month holding period with a confidence interval of 99%. 2- A maximum exposure constraint for each asset class. Asset class constraints Min Exposure Max Exposure Stocks indexes futures -40% 80% Sovereign Debt AAA futures 0% 60% Gold future 0% 8% DJP Commodities ETF 0% 8% Volatility futures -4% 4% Sum of the exposures 0% 100%
- 7. Qopius - Private & Confidential 7 DEVELOPMENT 7 years of asset management quantitative research combined with a cumulated 9 years of research in Artificial Intelligence and deep learning techniques. Backtest 9 Years of statistical backtest (2005-2013) The backtest period is not accounting the predictions of the Qopius Financial Engine which was in the learning and validation processes. Paper Trading 2 years (2014-2015) The period is a pure test with the full model. Any optimization neither over-fitting was possible during the period.
- 8. Qopius - Private & Confidential 8 QOPIUS FINANCIAL ENGINE PREDICTIONS For each instrument Qopius Financial Engine use deep learning techniques mutated by evolutionary algorithms and generates and combines numerous predictions in order to result in the desired prediction of the outcome. Predictions are verified and the performance of the predictive model is adjusted accordingly by refining its attentional control over the data acquisition layer and assigning high confidence on the most successful paths taken. The model adjusts progressively the exposure to global risk (beta) across asset classes depending on a state matrix. The matrix axis are the averages predictions of Equities and Government bonds components. Qopius I.A scoring Debt 88.8% Equities 37.5%
- 9. Qopius - Private & Confidential 9 ALLOCATION MATRIX A global state of the market is defined by the momentum of global equities and Long term Government debt : 1 : Flight to quality 2 : Inverse Flight to quality Cash is King Offensive Mode Defensive Mode Transition Mode Risk Appetite Global momentum Equities GlobalmomentumGovbonds 2 1
- 10. Qopius - Private & Confidential 10 BEHAVIORAL COMPONENTS Qopius Financial Engine use many sentiment based indicators to predict markets direction, like Twitter and news sentiment. Indeed, the model shifts fully from technical to behavioral analysis when the market is emotionally driven. The model use a Highly filtered Put/Call ratio to measure the excess in market sentiment and defines a Risk on/Risk off mode. When market sentiment is too high, risk is off to anticipate bull market reversal. When market sentiment is too low, risk is on to anticipate rebound after big sell off. Else the indicator is inactive. Measure of excess in market Sentiment Pessimism excess : Risk on No excess : Measure of trend excess Normal Condition: Qopius Financial Engine predictions Unusual trend : Long/Short market Neutral mode Optimism excess : Risk off
- 11. Qopius - Private & Confidential 11 2016 DAILY ASSET CLASS ALLOCATION
- 12. Qopius - Private & Confidential 12 BACKTEST PERFORMANCE
- 13. Qopius - Private & Confidential 13 PAPER TRADING PERFORMANCE
- 14. Qopius - Private & Confidential 14 MONTHLY PERFORMANCES Paper trading including trading fees Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Year 2016 0,81% 13,27% 11,65% 1,71% 0,95% 5,79% 41,53% 2015 7,10% 2,45% 1,69% 4,78% 0,02% 6,01% 4,86% 9,84% 7,04% 5,31% 3,54% 9,16% 81,91% 2014 2,13% 3,17% -3,81% 2,73% -0,23% 2,30% 0,25% 3,80% 0,97% 11,54% 2,51% 7,78% 37,58% Backtest including trading fees Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Year 2013 4,20% 5,55% 2,78% 3,25% -0,12% 7,97% 0,36% -2,14% 3,27% -0,71% -0,16% 1,13% 27,98% 2012 7,08% 2,07% -1,98% 0,45% 2,07% -1,09% -0,10% 1,92% 4,61% -1,75% 6,00% 1,95% 22,87% 2011 -1,05% -0,79% 0,98% 0,71% 0,25% 0,81% 3,08% 13,33% 6,59% -0,69% 2,57% -0,28% 27,62% 2010 -0,99% 0,76% 6,93% 0,07% 3,61% 0,30% 0,90% 0,75% 4,76% 4,48% -0,74% 3,04% 26,24% 2009 4,64% 1,71% -1,03% 8,52% 13,38% -0,22% 0,39% 7,91% 5,63% 0,03% 4,56% 6,14% 64,30% 2008 7,24% -0,43% 4,75% 1,26% 7,24% 4,99% 0,07% 0,69% 11,85% 38,06% 20,36% 3,24% 146,59% 2007 -0,62% 2,07% 10,67% 5,47% 5,97% -1,96% 3,11% 8,88% 4,27% 1,92% -1,35% -0,98% 43,37% 2006 5,88% 0,68% 0,92% 3,64% 3,36% 6,03% 2,30% 3,90% 1,05% 4,80% 3,08% 2,87% 45,86% 2005 -0,62% 4,62% -0,98% -0,79% 2,54% 3,66% 2,67% 1,11% 4,91% -0,94% 5,02% 1,02% 24,28%
- 15. Qopius - Private & Confidential 15 A NON CORRELATED MODEL TO TRADITIONAL ASSET CLASSES Good resilience when market is down and no correlation in the long term to traditional asset classes : Comparison to traditional asset classes S&P 500 US 10 YR Alpha 2.95% 2.84% Beta 0.09 0.15 Correlation 2.0% 7.5% Outperformance w/Benchmark is Positive (Monthly) 60.3% 54.0% Outperformance w/Benchmark is Negative (Monthly) 91.5% 92.5% These statistics include the backtest period(2004-2013) Statistics Compound Annual Return 51.3% Average Monthly Return 3.4% Largest Monthly Gain 38.1% Largest Monthly Loss (3.8%) % Positive Months 80% Average Positive Return 4.5% Average Negative Return (1.0%) % Negative Months 20% Worst 12 Months 5.6% Best 12 Months 150.6% Annualized Standard dev. 12.6% Sharpe Ratio (0,0%) 4.02 Sortino Ratio (0,0%) 10.78 Downside Deviation (0,0%) 1.30% Max Drawdown (daily) (7.8%) Months in Max Drawdown 3 Months To Recover 2
- 16. Qopius - Private & Confidential 16 QOPIUS FINANCIAL TRADING SYSTEM Qopius is offering a fully systematic and automated system. The data loading from Bloomberg, the computation via IBM cloud, the internal risk control and the trading via Bloomberg EMSX are fully integrated. The client will run all the process once a day via Excel in less than 15 minutes. Qopius Team will update the model once a month to insure that the predictive model is adjusted to the latest market conditions and correlations.
- 17. QOPIUS FINANCE
- 18. Qopius - Private & Confidential OUR MISSION Qopius proposes a new generation of Artificial Intelligence inspired by neurosciences designed to behave optimally in highly complex and unstable environments. 18 Like human beings, Artificial Intelligence learns by experience and instructions. Collecting data from its environment, it evolves to progressively achieve objectives through autonomous learning to perform better from one day to the other.
- 19. Qopius - Private & Confidential QOPIUS FINANCE TEAM 19 Antonin Bertin - CEO Qopius Technology / Fondateur Qopius Research, cognitive sciences and artificial intelligence - Paris (1 year) Ecole Normale Sup辿rieure (ULM) de Paris, Master in artificial intelligence. HEC Paris, Master in Management of new technologies Engineer - Telecom Paristech Amine Bennis - CEO Qopius Finance / Fondateur Chahine Capital : Fund Manager, portfolio combining European equity and Alternative Global Macro strategies - 550M AUM, Luxembourg (3 years) Financi竪re Arbevel : Quantitative global fund Manager, Wealth management advisor- Paris (3 years) OTC Financial : Asset Management Consultant - New York (1 year) AXA France : Quantitative Analyst - Paris (1 year) Engineer - Ecole Normale Sup辿rieure Telecom Bretagne Nicolas Barral - Business D辿veloppement - US Columbia Business School, MBA, New York City. Schlumberger, senior mechanical engineer and team leader - Oslo (6 years) UC Berkeley, Master in Mechanical Engineering Engineer - Arts et M辿tiers ParisTech Roy Moussa - Business D辿veloppement - Europe Schlumberger, team lead - Oslo (3 years) Best Buy, Commercial - Montreal (2 years) Engineer - Concordia University
- 20. Qopius - Private & Confidential QOPIUS TECHNOLOGY TEAM 20 Alexandre De Larrard - Research Statistical modeling AXA, Pricing and modelisation- Paris (1 year) Mazars. Audit - Paris (1 an) ENSAE, actuary certification Engineer - Telecom Paristech Benoit Boyadjis - Research Image recognition Expert in deep learning techniques of language and image processing, parallel computing, algorithms on attentional processing of information. Thales, PHD in artificial intelligence (2 years) Engineer - Telecom Paristech Vincent Lostanlen - Research Signal processing Expert deep wavelet representations and signal processing. Ecole Normale Sup辿rieure (ULM) Paris, PHD in artificial intelligence. (3 years) Engineer Telecom Paristech L辿opold CRESTEL - Research - Signal processing ATIAM IRCAM, PHD in artificial intelligence applied to acoustic and speech processing. (3 years) Expert in deep wavelet representations and signal processing. Engineer - Telecom Paristech
- 21. Qopius - Private & Confidential ADVANTAGES The research team of a big company with the agility of a start up Common ressources for development A Holistic approach A multidisciplinary approach 21 Qopius is using a unique cognitive architecture for a variety of challenges, the research team combines and mixes inside a common artificial intelligence engine the latest techniques of speech, image, and time series processing.
- 22. 22 QOPIUS FINANCIAL ENGINE Similar to the human brain, the Q-Engine is powered by components that are continuously learning, communicating and adapting to form a single unit that's robust in the face of surprise. The engine is constantly making prediction about future events and comparing that with the actual outcome to improve on its predictions. After the system has accomplished a task through the "Intelligent Behavior" module, it evaluates it's accuracy and transmits back the need (when necessary) for more relevant data sources. Qopius - Private & Confidential Selective Data Acquisition It powers the exploration of the system's environment Artificial Reality It builds its own representation of the environment Intelligent Behavior Takes decisions, acts and adapts to achieve various objectives or tasks. Although the Q-Engine develops autonomously, we have constructed a structure that can be unveiled and understood by the people it works with.
- 23. Qopius - Private & Confidential QOPIUS ENGINE TECHNIQUES 23 A few of the techniques behind the engine include deep learning, evolutionary algorithms and decision tree learning : Deep Learning Evolutionary Algorithms Q-Engine aims at : Searching Optimal Parameters in high dimensional space Integrating and fusing multimodal sources Combining collection of generated and external models
- 24. Qopius - Private & Confidential 24 CHARACTERISTICS PLASTICITY : The integration of new knowledge is done so that it does not disturb the existing one. The importance of each informational stream is constantly influences and weighted depending on their likelihood and usefulness. The Qopius engine has the ability to select modalities that give high performances at a time and inhibit the others. ARTIFICIAL MOTIVATION AND CURIOSITY : The Qopius Engine has the ability to discover by itself the objectives that will accelerate its development. VIGILANCE : The Engine always tries to predict the next information it will receive. Mis-predicted information receive high attention from the engine and lead to curiosity and learning. TRANSPARENCY : The structure of the engine has been imagined so that we can follow (and potentially influence) the development of the artificial intelligence.
- 25. Qopius - Private & Confidential 25 SUMMARY
- 26. FROM CUTTING EDGE A.I. RESEARCH TO INNOVATIVE SOLUTIONS Qopius - Private & Confidential 26