Selecting Investments Using Artificial Intelligence

Transcription

1 Selecting Investments Using Artificial Intelligence

2 2 Predictive Algorithm I Know First provides daily investment forecast based on an advanced self-learning algorithm Technology based on artificial intelligence & machine learning, incorporating elements of artificial neural networks and genetic algorithms, developed to analyze and predict financial markets. The algorithm is self-learning, adaptable and scalable, is applied to discover best investment opportunities or as a decision support system of an existing investment process Tracks over 3000 assets (Stocks, World Indexes, ETFs, Interest rates) and growing Customized algorithmic solutions (extended tailored access to algorithmic predictions, integration of additional markets/securities) Clients: Family Offices, WM Companies, Advisors, Retail Clients grown by 400% during Genetic Algorithms (GA) I Know First predicts 3000 securities daily Artificial Intelligence (AI) Joint Venture: I Know First is launching a fund this year with a financial agency in Israel

3 3 I Know First Team Dr. Lipa Roitman - Co-Founder and CTO: Over 20 years of research and experience in artificial intelligence and machine learning fields Concept of IKF s algorithm has crystallized following years of his prior research into the nature of chaotic systems Head of IKF s R&D Team Ph.D. from the Weizmann Institute of Science Yaron Golgher - Co-Founder and CEO Previously division manager at OIC with over 15 years of experience in managing and leading consulting projects for industrial and financial institutions. EMBA from Ben Gurion University, B.Sc. in Industrial Engineering from Tel-Aviv University & R&D Team: professionals with backgrounds in computer science, applied mathematics, and finance Operations Team: finance and marketing

4 How It Works Daily Prediction Process 4 Daily data added to our 15 years historical file Run a learning & prediction cycle with new combined data. Daily predictions for each stock, currency, commodity, etc.. The results are constantly improving as the algorithm learns from its successes and failures

5 5 Tier 1: Algorithmic Forecast Daily Market Heatmap ordered by signal strength Every day the algorithm generates heat maps demonstrating the overall direction of the markets in the 6 time frames. The algorithm outputs two indicators: Predicted movement of the asset the fitness function (simplified: a correlation based quality measure of the signal) Key to identify and focus on most predictable markets and securities, enhancing the overall performance

6 Performance Systematic Trading (S&P 500 universe) 6 Daily trading - rankings based on the short-term signals, filtered predictability $ $ $ $ $ $ $ $ $ $ $ $ $9.500 $9.000 S&P % % % % % +9.08% SR: β < 0.6 α: Tier 2 Custom Strategy Rules Strong Signal S&P500_equity IKF_Top20_pure IKF_Top20_acc_cons_filter IKF_Top20_consist_streak IKF Top 20_pure_cons_comb IKF Top 20_trend_avg_signal High Predictability Level

7 Performance (S&P 500 universe) signal strength (relative on a given day) w eakest 10% m iddle 80% strongest 10% all S&P500 constituents avg daily close-to-close return 0.012% 0.054% 0.139% 0.122% annualized (252 business days) 3.19% 14.59% 41.75% 35.94% Predictability filter Focusing on a higher level of predictability further improves the returns for stronger signals: signal strength (relative on a given day) w eakest 10% m iddle 80% strongest 10% all S&P500 constituents avg daily close-to-close return 0.005% 0.074% 0.250% 0.122% annualized (252 business days) 1.29% 20.50% 87.63% 35.94% Good foundation for systematic trading 7

8 8 Bottom-up approach: Investment analysis starts with and focuses on individual stocks IKF s algorithmic predictions are integrated: a) To discover additional opportunities and/or b) To perform algorithmic screening in parallel Forecast Utilization: Different Investment Approaches Top-down approach: merged By Industry and major World Indexes forecasts within (macro-) economic analysis identifying most promising markets/industries (i.e. sub-universes) PTBV P/E EPS Growth Debt/ Equity Fundamentally healthy, reasonably priced sub-universe + SWN MUR REGN PXD MNST CNX CHK ALXN PCLN OKE CPB PRGO GAS PCG AAL KMB DPS CLX TE HRB Algorithmic screen (pattern recognition in historical trading data) focus on those sub-universes and go deeper from there, integrating the individual stocks forecasts (as DSS or opportunities identifier) Constructing Final Portfolio

9 9 I Know First Customized Solutions for Financial Institutions From out of 3000 financial assets that I Know First is tracking, the predictions universe can be tailored according to client s investment focus. Various filtering criteria: Local markets/exchanges, e.g. JSE All Share, Nigerian SE 30, Nairobi SE 20 etc. Asset classes - Stocks, ETFs, Interest Rates etc. Sector/Industry specific forecasts, merged signals Market Capitalization, Liquidity Risk (volatility) Dividend paying stocks Fundamental key ratios, e.g. P/E, P/S, EPS Growth etc. Reported insider trades Custom universe, received from client can be integrated if a) enough historical data and b) sufficient predictability level We work closely with clients and partners to customize the prediction table for best performance with your portfolio, with hardware designed to keep the system optimized to their trading universe.

10 Business Model and Solutions We Offer IKF Predictive Algorithm Prediction Services Fund Management Other Industries Extended access to algorithmic predictions Institutional Clients Customization, Integration of additional securities Best Picks (predefined subuniverses) Retail Clients Robo-Advisory Platform In partnership with institutional clients Involved in launching IKF s own fund Incoming Requests for: - Automotive (lots of sensors data -> Real time prevention of engine/parts failures; sales forecasts) - Financing (loan risk based on socioeconomic factors, age, payment patterns etc.) - Insurance (e.g. car insurance policies based on driving patterns) 10

11 I Know First Predictive Algorithm 11 Possible Cooperation Opportunities Local Partner Tier 1: retail or customized forecast Tier 2: rules-based trading Daily Forecast (short- and long term) Update Development Joint Venture Trades Recommendations* Distribution/Sales Financial Advisors/Research (internal use) Robo-advisory Platform Fund Management, Trading Local Partner Fields of Activity/Services * Daily trading or 5d Buy & Hold (1/5 of Portfolio rebalanced each day)

12 12 Thank you! Q&A and few more details

13 Performance Systematic Trading (S&P 500 universe) 13 Daily trading - rankings based on the short-term signals, filtered predictability $ $ $ $ $ $ $ $ $ $ $ $ $9.500 $9.000 S&P % % % % % +9.08% SR: β < 0.6 α: Tier 2 Custom Strategy Rules Strong Signal S&P500_equity IKF_Top20_pure IKF_Top20_acc_cons_filter IKF_Top20_consist_streak IKF Top 20_pure_cons_comb IKF Top 20_trend_avg_signal High Predictability Level

14 14 Key results YTD and July 2016 (S&P 500 universe) Key results YTD (since launching on 7 th Jan 2016 and until the end of July 2016): overall return IKF between +22.1% and +45.0% vs. S&P % with a high Sharpe ratio between 1.74 and 5.13 Key results July 2016 (pilot project): Table A: Predictability filtered signals worst 10% middle 80% strongest 10% strongest 5% all signals<>0 avg real return avg daily close-to-close return % 0.149% 0.263% 0.428% 0.140% 0.205% annualized (252 business days) 2.25% 9.19% 41.68% 52.48% 11.36% 28.39% * avg real return the average daily close-to-close return of the S&P 500 stocks in July Table B: strategies performance ( = pilot project): - overall return IKF between +5.13% and (8.42%) vs. S&P %. S&P 500 IKF_Top20_pure IKF_Top20_acc_cons_filter IKF_Top20_consist_streak IKF_Top20_pure_cons_comb IKF_Top20_trend_avg_s IKF_Top20_acc_filtered_alt* Total return (Jul'16) 3.56% 7.11% 5.13% 7.55% 6.05% 6.94% 8.42% # trades Hit Ratio (% of winning trades) 59.25% 57.54% 56.68% 55.50% 60.75% 57.21% average trade return (C-to-C) 0.35% 0.45% 0.39% 0.30% 0.34% 0.40% W/L return ratio

15 $ $ $ $ $ $ $9.800 $9.600 July 2016 Daily Stock Selection IKF_Top20_pure & IKF_acc_cons_filter, (S&P 500 universe) Daily Selection Process - July 2016, I Know First vs. S&P 500 Index % % %; +6.94% % % % S&P 500 $9.400 S&P500_equity IKF_Top20_acc_cons_filter IKF_Top20_consist_streak IKF_Top20_trend_avg_s IKF_Top20_pure IKF_Top20_same_direct_consist_filter_all_time_frames* IKF_Top20_pure_cons_comb

16 Short-Term Trading: Realistic Back Tests (incl. Slippage & Commissions) (S&P 500 universe) 16

17 5d Buy & Hold (1/5 of Portfolio rebalanced each day) incl. slippage & commissions, S&P 500 universe Table: statistics 5 different 5days Buy & Hold equity lines for , I Know First S&P500 Index IKF_5d_B&H avg final return 11.59% 27.04% median final return 12.20% 27.85% std dev final return 1.92% 6.27% max final return 13.09% 34.83% min final return 8.86% 18.42% S&P % S&P 500 equal weight %

18 Performance improvement through intraday adjustments incl. slippage & commissions, S&P 500 universe $ Daily trading - IKF consistent streak Close-to-Close vs. Open-to-Close $ $ $ % % $ $ $ $ $ $ S&P % Consistent intraday adjustment $ $ IKF_Equity_No_Slippage/No_Commission(O-to-C) IKF_Equity_Top20_consist_streak(C-to-C) S&P500_equity

19 19 Tier 1: Algorithmic Forecast Daily Market Heatmap Every day the algorithm generates heat maps demonstrating the overall direction of the markets in the 6 time frames. The algorithm outputs a predicted trend as an absolute number (not a percentage) known as signal strength Table is ordered by signal strength. Example: A bullish asset would be indicated by a green buy signal at the top of the table. A bearish asset would be denominated by a red sell prediction at the bottom of the table.

20 Algorithmic Forecast Signal & Predictability XOMA returned 61.45% in 1 month from this forecast Two indicators: Signal Predicted movement of the asset Predictability Indicator the fitness function (simplified: a correlation based quality measure of the signal) Key to identify and focus on most predictable markets and securities, enhancing the overall performance 20

21 Algorithmic Forecast Performance 1 Month Example 21