FPA of San Antonio & South Texas -Richard Roche
05/22/2019
1
Rick Roche, CAIA2013-19. Managing Director –
Little Harbor Advisors, LLC
2019. Roche Invest AI, LLC
38-YR industry veteran
Kidder, Peabody & Smith Barney financial advisor
Series 3, 7, 63 & 65 licenses
Chartered Alternative Invt. Analyst; Member-Society of Quantitative Analyst
Compulsive Counter; Light Quant; Data-Science Curator
Trained 74,500 financial pros
(c) 2019 Little Harbor Advisors, LLC 1
Model Behavior – AI/ML & BIG DataAction Agenda
Follow the Money!
Machine Learning Origin, Applications & Diffusion
Quants: Then & Now! People, Players & AI/ML Practitioners
‘Revenge of Nerds” ML & Big Data Asset Wealth Managers
Quanta-mental Diffusion
Digital Disrupters-Amazoned?
AI-ML & Big Data Drawbacks
Steps for CFPs to Enhance Job Security/Stay Relevant!
(c) 2019 Little Harbor Advisors, LLC 2
Which is Bigger? Do the Math.
(c) 2019 Little Harbor Advisors, LLC 3
Quant HFs $932Bn-AUM
$88.5 Trillion-Dec 2017; “The New Great Game in No. American Asset Mgmt”, Nov 2018 by McKinsey
1
2
3
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
2
Put Quant AuM in Perspective
(c) 2019 Little Harbor Advisors, LLC 4
“Quant Investing: Bridging the Divide” Morgan Stanley, October 1, 2017
AI Birth: When, Where, Why?
(c) 2019 Little Harbor Advisors, LLC 5
AI-Machine Learning Overview
(c) 2019 Little Harbor Advisors, LLC 6
1955. Dartmouth John McCarthy. Artificial Intelligence (AI) - Power of machine to mimic intelligent human behavior, incl., if-then rules, logic and statistical methods
1950’s Machine Learning. Subset of AI, algorithms focused on finding patterns & making predictions
Computer programs itself - learnsby analyzing large data or novel sets.
When there are too many “ifs” to code, Machine Learning to rescue…
4
5
6
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
3
(c) 2019 Little Harbor Advisors, LLC 7
May 11, 1997. Game 6 1952 – 1962 Beat CT Champ
Feb 2011. Final Jeopardy Watson Mar 2016. DeepMind vs. Go Champ
Playing Games to Major Disruptor!ML detect patterns and make predictions by processing data, rather than explicit programming
ML avoid elaborate programming by defining rules rely on Inputs
ML algos adapt in response to new data & experiences to “learn” or improve efficacy over time!
Machine Learning intersection of Computer Science & Statistics to find patterns in datasets
ML heavily weighted towards predictive modeling vs. inference!
Uses probability to make product recommendations, speech recog…
2016 Due Diligence Material 8
“How can we build computer systems that automatically improve with
experience…” , Tom Mitchell, PhD former Chair, Machine Learning Dept. CMU
ML- Prediction Machines
(c) 2019 Little Harbor Advisors, LLC 9
2015 - 67,318 papers on ML in SSRN over 12-months
2017 – 14,000 asset mgmt papers - BIG Data/Analytics
ML = Cheap Prediction
When prices fall, use more!
Use information you have to generate info you don’t have
Cheap Predictions = + Demand Human Judgment/Expertise!
7
8
9
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
4
Humans & Machines Cohabitate
(c) 2019 Little Harbor Advisors, LLC 10
Machine Learners – Asset Mgmt.
(c) 2019 Little Harbor Advisors, LLC 11
“Big Data & AI Strategies: Machine Learning & Alternative Data Approach to Investing: Lead authors M. Kolanovic & R. Krishnamachari, JP Morgan, May 2017
1 1 5 4 3
7 5 3 5 3
5 5 9 0 6
3 5 2 0 0
Features (Attributes) & Labels
10
11
12
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
5
ML Model Behavior
(c) 2019 Little Harbor Advisors, LLC 13
Classifier network. Algo classifies new observed data into categories.
Regression trained with goal of level estimation/numeric value
Clustering. Assign observations into subset “cluster” where data are similar, e.g. Netflix/AMZN
(c) 2019 Little Harbor Advisors, LLC 14
(c) 2019 Little Harbor Advisors, LLC 15
13
14
15
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
6
Deep Learning – ANN/DNNDNN has 3 layers: 1) Input; 2) Hidden Layer(s); 3) Output
Analyzes data by passing thru multiple layers (neurons) -software-based calculators –that learn increasingly complex features of data
Extract data features passing multiple filters in over-lapping segments data/weight data.
ANN infer patterns in data
and make predictions
Widely-used in image, voice recognition; NLP-text analysis
(c) 2019 Little Harbor Advisors, LLC 16
Biologically inspired programming attempts to mimic human brain
Algo Trading: In a Blink of the Eye
(c) 2019 Little Harbor Advisors, LLC 17
300-400 milliseconds-Eye Blink
HFTs trade in 10 microseconds
HFT 40,000 Back-2-Back trades
New Frontier – Nanosecond (bn)
HFTs trade on “news analytics” on DJ New Wire w/in 5-seconds
RenTec Atomic Clock Patent (bn)
Stocks. JPM est. 5-10% stock vol. by fundamental traders¹
~60% of equity trades by HFT
805 futures products; 1.5 Bn-Trades. 5% trades in Pit
86.7% of E-mini SPX is ATS; ES 2nd most active; # 1 EuroDollar
Very Few End-2-End Portfolio Design
(c) 2019 Little Harbor Advisors, LLC 18
“Artificial Intelligence – Chances & Challenges in Quantitative Asset Management”, Aquilla Capital Systematic Trading Group, Hamburg 2018.
16
17
18
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
7
AI-ML After Decades of Promise – Why NowMoore’s Law. Cheaper/faster; 1-Gig storage $277 to $0.79
Nvidia GPU chips. 70Xs faster than CPUs – deep learning
Cloud Computing – lowered AI barrier of entry for crowd
2016 1.49Bn Smartphones Sold! 40% Adults own SmtPhs
1,459 Satellites Dec-16/IoT Smart Sensors 2Xs every 2 yrs.
BIG Data Cambrian Explosion (500+ Alt Data Providers)(c) 2019 Little Harbor Advisors, LLC 19(Machine | Platform | Crowd, McAfee & Brynjolfsson, 2017/JPM “Big Data”)
World’s 1st Quant Fund Manager
MIT Math & UC-Irvine Math & Quant Finance Professor
Expert in Game Theory/Stats
Used IBM 704 & Kelly criterion to count cards/place casino bets
1962. Beat The Dealer: Winning Strategy for the Game of 21
Co-invent 1st wearable computer with Claude Shannon of MIT
Early 1980’s pioneered Statistical Arbitrage with Gerry Bamberger
In 2010 book, The Quants, he’s called The Godfather of Quants
(c) 2019 Little Harbor Advisors, LLC 20
Market Mathematician/Magician
1967. Beat The Market: Scientific Stk. Mkt. System. Sheen Kassouf
1969. Convertible Hedge Associates - Princeton/Newport Partners
C0-founded 1st “Market-Neutral” Derivatives Hedge Fund/converts
1969-1989. PNP Return 15.1%; STDV 4% and Sharpe ratio of 3+
1969. Shared his option formulas w/Fisher Black & Myron Scholes. Advanced Thorpe’s delta hedging strategy assuming no arbitrage
200 Long/200 Short; O beta; Mkt Neutral; Kelly-sized Bets ½ of 1%
Market Wizards -- “..Two best records…ever complied in trading…”(c) 2019 Little Harbor Advisors, LLC 21
19
20
21
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
8
The Black Box Revealed
(c) 2017 Little Harbor Advisors, LLC 22
The Evolution of Quantitative Investment Models
Goals. Evaluate wide swath of securities, build portfolio more efficient & dispassionately than subjective models
Quants use widely shared Value & MOM alpha signals, 3rd party optimizers, risk mgmt system & execution algos
Quant Quake-Aug 2007 showed risk of copy-cat quants
Today, advanced Quants use Data-Driven Strats (AI/ML)
Quant Models -- Then & Now
(c) 2019 Little Harbor Advisors, LLC 23
1st Gen Generic Quant Models used computers/algos for yrs.!
Systematic Traders early quant
Most use small data/price data
Most/many static algorithms, subject to model (alpha) decay
Execution algorithm dominate Equity trading/Mgd. Futures
Next Gen Quants use Big Data to “train” machines w/ML
“Data-Driven” alpha models neither common or easily used
ML applied parameter setting; avoid overfitting a model
ML algos seek patterns/CORR: Relative value trades; NLP to read text for Invt. Sentiment
Revenge of the Nerds – AI/ML
(c) 2019 Little Harbor Advisors, LLC 24
22
23
24
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
9
Renaissance Technologies LLC (RenTec)Founded 1982. $131Bn-AuM
James Simon, PhD Math, NSA code breaker/Math Champion
Use Big Data/advanced math models long before other HFs.
Cryptography/Hidden Markov models seek short-term signals
Medallion averaged 71.8% ann. rtn-before fees-1994 to 1H2014
RenTec controls 4 Hedge Fund Medallion closed! 99% owned
Trend-following roots. Use mgd futures to express ideas
(c) 2019 Little Harbor Advisors, LLC 25
"From 2001 thru 2013, the fund’s worst year was a 21 percent gain, after … fees. Medallion reaped 98.2% gain in 2008, the year the S&P 500 Index lost 38.5%." Rubin & Collins. 06/16/15, Bloomberg
RenTec Medallion v. Buffett/BRK1988. RenTec acquires Kepler Financial Mgmt - StatArb firm
1991. RT raids IBM’s Speech Recognition/Translation Team
Strong connection bwtn speech recognition & market signals!
NLP maps speech to predict
next sound; adapted to securities
Short-term trading of futures-easier predict S-T movements
Peter Brown, Co-Head-RenTec
“We have 10,000 processors … constantly grinding away looking for signals”¹
(c) 2019 Little Harbor Advisors, LLC 26
17.1% BRK vs 40.6% Medallion (2.4x)BRK 6 down yrs. vs Medallion 1 yr.
2008 BRK down; Medallion up 98.5% Medallion only 2-yr earn > 21% return
DE Shaw & Co./GroupGlobal Invt & Technology Firm
PhD/computer science Stanford
1986. Morgan’s APT hired D.E. Shaw, parallel processing expert
1988. Quit MS; Open shop over communist bookstore/Village
Statistical Arbitrage pioneer
Dr. Shaw infamous for keeping DE Shaw’s secret sauce Secret!
David E Shaw is Chief Scientist
$50B of invt capital (Decl 2018)
250 Technologists at DE Shaw²¹ The Quants, Scott Patterson, 2010, p. 43
²”2016 Tech 50: Neil Katz”, Instl Invt., 2016
(c) 2019 Little Harbor Advisors, LLC 27
“Our goal is to look at the intersection of computers and
capital, and find as many interesting and profitable things to do in that intersection as we can.”
– David E. Shaw, PhD
25
26
27
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
10
Bridgewater Associates, LP
(c) 2019 Little Harbor Advisors, LLC 28Accessed from https://jobs.bwater.com/explore/technology_lounge on 07/04/17
About Ray Dalio: Paul Volker admits to feeling that “ he has a bigger staff, and produces more relevant statistics and analyses, than the Federal Reserve.” – Paul Volker in “Man and Machine”, The Economist, 03/10/12
Dave Ferrucci, PhD Senior TechnologistIBM Fellow led 30 A.I. researchers who built DeepQA Project. In 2011, Watson beat the world’s best Jeopardy! players.
Founded 1975. $163Bn-AuM
Dalio obsessed with cognitive biases/Principles-Algorithms!
Believes in expert systems like
“Systematized Intelligence Lab” to automate day-2-day management
Dalio doesn’t like machine learning moniker (prefers AI)
Bridgewater models informed by “Economic Machine” works
Combine Data Science, Software Engineering + Product Mgmt
Bridgewater more a technology company than it is a hedge fund.
Two Sigma – Applied Data Science
(c) 2019 Little Harbor Advisors, LLC 29
Founded 2001. $78Bn-AuM (2018) Co-Founder, David Siegel, PhD, in
computer science from MIT Co-Founder, John Overdeck, Intl
Math Olympiad, MS-Stats-Stanford Co-Founders worked at D.E Shaw 1,100+ employees. 2/3rds in R&D “Harness data from thousands of
diverse sources” Guided by the Scientific MethodHired, Mike Shuster, Google Brain,
to consolidate AI research. Shuster was Google Sr. Scientist for 12-yrs
Host hackathons, sponsor contests, programming challenges to recruit future associates!
(c) 2019 Little Harbor Advisors, LLC 30
28
29
30
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
11
Quant-amental Synergy
(c) 2019 Little Harbor Advisors, LLC 31
Fundamental Analysis
QuantitativeComputations
AI/ML Investing Gathers Steam
Mar-2017, BLK moved $30B (11% active equity assets) to “scientific active equity” funds – combining BLK quants & its stock-pickers
BLK let go PMs, re-christen BlackRock Advantage/lowering fees
BLK’s Mark Wiseman told Bloomberg, “We can more efficiently deliver alpha at a better cost with automated processes” (3/28/17)
BLK’s $100Bn quant shop-- Systematic Active Equities (SAE)- SFO
1985. SAE’s fund, Alpha Tilts. Today, 80 quant PMs incl. 30 PhDs
2018. BLK opens “Lab for AI” in Palo Alto, CA; “Data Science Core”(c) 2019 Little Harbor Advisors, LLC 32
BlackRock, Inc. – BLKBlackRock has $6.3 TR-AuMBLK is Fund/ETF Lead Steer
PIMCO Going All-In on AI?Bond King, Bill Gross, to Machine Learning Algos
1966, Gross @ Duke, car accident; read Ed Thorpe!
Gross’ Masters thesis on convertible bonds/Thorpe
Grow FI Analytics Team; Grow workforce 10%/250
Hire U-TX technologists to
lower costs/Predict Inflation
New Slate PIMCO FI funds driven by Machine Learner
(c) 2019 Little Harbor Advisors, LLC 33
Minds & MachinesPIMCO $ 1.77 Trillion (03/18)One of Top 3 Largest FI MgrsUse Bond Market Math & ML
to search Private Debt bets
31
32
33
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
12
(c) 2019 Little Harbor Advisors, LLC 34
Tech Innovation Estimation Errors
"We always overestimate the change that will occur in the next two years and underestimate
the change that will occur in the next ten." -- Bill Gates, The Road Ahead (1996)
(c) 2019 Little Harbor Advisors, LLC 35
BIG Data Drawback/Side Effects
(c) 2019 Little Harbor Advisors, LLC 36
34
35
36
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
13
Jobs at Risk of ML/AI Disruption
(c) 2019 Little Harbor Advisors, LLC 37
Up Next, FinTech Disruption
(c) 2019 Little Harbor Advisors, LLC 38
Chatbots as Digital Disrupters
(c) 2019 Little Harbor Advisors, LLC 39
37
38
39
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
14
HNWI interest in AWM by BigTech?
(c) 2019 Little Harbor Advisors, LLC 40
“World Wealth Report 2017”, Capgemini, Figure 20, 2Q 2017, p 22.
BigTech on Doorstep
(c) 2019 Little Harbor Advisors, LLC 41
Schwab Intelligent Portfolios Premium
(c) 2019 Little Harbor Advisors, LLC 42
Commoditization
40
41
42
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
15
BigTech – Big Threat
(c) 2019 Little Harbor Advisors, LLC 43
Fire Hydrant – Drowning in DataTsunami of New BIG Data
90% of digital data today was created w/in last 2-YRs¹
Avg. 4,000 brokerage Rpts. per day/36,000 pp/53 languages³/MS 80,000 Rpt²
5,000 public company docs filed daily/150,000 words in a 10K Annual filing w/SEC³
Earnings Season – up to 400 conference calls per day³
(c) 2019 Little Harbor Advisors, LLC 44
¹“Big Data & AI Strategies: Machine Learning & Alternative Data Approach to Investing”, Lead authors M. Kolanovic & R. Krishnamachari, JP Morgan, May 2017;
³“Tone At The Top? Quantifying Mgmt Presentation”, Luo’s QES Research, Wolfe Research Multi-Dimensional Alpha, Jan 23, 2018, page 6 & 9-10.
Sources: JP Morgan; IBM; EMC; ³“Finding Big Alpha in Big Data” The Evolution of Active Investing”, BLK’s Scientific Active Equity, March 2015²
Need to Speed Read – Use NLP
(c) 2019 Little Harbor Advisors, LLC 45
At its core, artificial intelligence is “based on doing ridiculous amounts of multiplication problems very fast!” – John Wihbey, M.S., Prof. at Northeastern/Author
43
44
45
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
16
Lexicons 4 Central Banks/Stks.
(c) 2019 Little Harbor Advisors, LLC 46
Social Media - Crowded Trade?
(c) 2019 Little Harbor Advisors, LLC 47
Eye in the Sky/Satellite Alpha
(c) 2019 Little Harbor Advisors, LLC 48
46
47
48
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
17
Data Sets for Alpha Generation
(c) 2019 Little Harbor Advisors, LLC 49
Estimates that only 0.5% of data currently analyzed.~80% of Big Data in unstructured/unsupervised data sets(source: Patrick Wolfe, PhD UCL Big Data Institute, 2013)
From Checkmate to TeammateCentaur amplifies Human Freestyle Chess Tournament
1) Rely solely on AI program; 2) Human only; 3) Human + pc
Final Four were all “Centaurs”
Winners two amateur – soccer coach & d-base administrator
Used three AI programs run one consumer grade pcs
Today, best chess AI programs consistently lose to Centaurs
Winners are amateur players who understand & integrate AI
Human-Machine pairings Excel
(c) 2019 Little Harbor Advisors, LLC 50
June 2005. Online “Freestyle Chess” Match
won by dark horse, ZackS
Human + Machine Intelligence
Racing with the Machine beats Racing against the Machine
Artificial intelligence won’t replace human intelligence
“Assisted Intelligence” – automation of repetitive tasks
“Augmented Intelligence”. Human + Machine collaborate
Even advanced FinTech still requires human intervention!(c) 2019 Little Harbor Advisors, LLC 51
49
50
51
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
18
(c) 2019 Little Harbor Advisors, LLC 52
Advisory Model: 360ºFunctional View
(c) 2019 Little Harbor Advisors, LLC 53
Hybrid Advice: What to Automate?
Quantify Your Value to Client
(C) 2008-15 Roche Securities Sales 54
“Quantifying Your Value to Your Clients”, Vanguard Advisor’s Alpha”, Kinniry, Mar 2014
52
53
54
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
19
How to Quantify Advisor Alpha
(C) 2008-15 Roche Securities Sales 55“The Evolution of Vanguard Advisor’s Alpha: From Portfolios to People”, January 2018
Minds & Machines: The Rest of the Story
Augmented Intelligence Key TakeawaysResearch Quant Funds, if it has low or – CORR to equities & up more often
Diversifier to Qual./Index
Automate Portf. Rebalance & Performance Reporting
Communicate Your Value!
Learn a new language – in Programming. Python or R
Test-drive Algo trading at Open-source Platform!
(c) 2019 Little Harbor Advisors, LLC 56
Not either or proposition. Combine human intelligence with machine.** FinTech can help engineer better invt. outcomes for your clients** Enhance portfolio construction** Automate client servicing tasks to free-up time for behavioral coaching
Rick’s Recommended Reads
(c) 2019 Little Harbor Advisors, LLC 57
55
56
57
FPA of San Antonio & South Texas -Richard Roche
05/22/2019
20
Author/Speaker DisclosureThe information contained in this
presentation was obtained from sources
believed to be reliable, but its accuracy
cannot be guaranteed. The information
communicated is NOT intended to
constitute individual investment advice
and is not designed to meet your personal
financial situation. Opinions expressed
herein are those of the presenter, Rick
Roche, and not of Little Harbor Advisors.
Such opinions are subject to change
without notice. Information may become
outdated & there is no obligation to
update any such information. Any errors
are the sole responsibility of Rick Roche.
58(c) 2019 Little Harbor Advisors, LLC
Sources & Firms CitedRick Roche (and Little Harbor Advisors)
have NO affiliations with the industry
sources and firms cited in this
presentation. In addition, Rick Roche has
not received any compensation for
mentioning services or products. His
citation of said firm or service is not meant
to be an implied endorsement. Rick
Roche and Little Harbor Advisors are
not affiliated with any of the firms cited.
Opinions expressed in this presentation
are subject to change without notice.
Information may become outdated and
there’s no obligation to update any such
information.
59(c) 2019 Little Harbor Advisors, LLC
58
59
Top Related