FinTech, AI, Machine Learning in Finance

ML, AI, DL in FinTech
Sanjiv R. Das
Santa Clara University
http://srdas.github.io/Papers/fintech.pdf
CFA, San Francisco, January 17, 2018

What is FinTech?
● FinTech refers to various financial technologies used to automate
processes in the financial sector, from routine, manual tasks to
non-routine, cognitive decision-making.
● FinTech may be characterized by technological change in three broad
areas of finance: (This framework was proposed by my colleague,
Professor Seoyoung Kim.)
1. raising capital,
2. allocating and investing capital,
3. transferring capital.
● My definition: "FinTech is any technology that eliminates or reduces the
cost of the middleman in finance."
There is now a growing interest and literature: - http://lfe.mit.edu/research/fintech/
- Risk and Risk Management in the Credit Card Industry (Florentin Butaru, Qingqing Chen, Brian Clark, Sanmay Das,
Andrew Lo, Akhtar Siddique), Journal of Banking and Finance 72(2016), 218–239.

The Costs of Financial Intermediation
Philippon (2016)

FinTech Landscape
● 1400 FinTech companies with $33 billion in
funding.
● Losses from credit card fraud are $31
billion a year.
“Using Big Data to Detect Financial Fraud Aided by FinTech Methods” - S. Srinivasan, Texas
Southern U.
● 2017, Q1: Over 100 FinTech startups with $3.2 billion in funding.
https://assets.kpmg.com/content/dam/kpmg/xx/pdf/2017/04/pulse-of-fintech-q1.pdf
FinTech adoption rates:

FinTech Startups by Year
https://www.newconstructs.com/big-banks-will-win-the-fintech-revolution/
https://www.accenture.com/_acnmedia/PDF-57/Accenture-Fintech-Did-Someone-Cance
l-The-Revolution.pdf

FinTech Framework
1. Machine Learning, AI, and Deep Learning.
2. Network Models.
3. Personal and Consumer Finance.
4. Nowcasting.
5. Cybersecurity.
6. Fraud Detection.
7. Payment and Funding Systems.
8. Automated and High-Frequency Trading.
9. Blockchain and Cryptocurrencies.
10. Text Analytics.
Examples
This is implicit : Banks will soon be technology companies and will need to invest
heavily in R&D Tech

Game Changers
● Mathematical innovations,
computing architectures
(e.g., LSTMs)
● Hardware
● Big Data

AI and the Technological Singularity
https://hackernoon.com/why-ai-is-now-on-the-menu-at-dinner-even-with-my-non-te
ch-friends-44c666348de4
Transformation
vs
Change
Kurzweil claims that AI will
pass the Turing test in 2029,
and the singularity will come in
2045.
https://futurism.com/kurzweil-claims
-that-the-singularity-will-happen-by-
2045/

Huge Demand for AI Talent
https://www.nytimes.com/2017/10/22/technology/artificial-intelligence-experts-salaries.html

Definition of AI
● Intelligence exhibited by machines
● Narrow or Weak AI: “Expert systems that match or exceed human
intelligence in a narrowly defined area, but not in broader areas” ( Dvorsky G.,
2013) e.g. Siri.
● Artificial General Intelligence: An artificial neural network not preprogrammed
with fixed rules. Rewire itself to reflect patterns in the data, adaptable to its
environment, in which (hopefully) advanced skills emerge organically.
● “Humans don’t learn to understand language by memorizing dictionaries and
grammar books, so why should we possibly expect our computers to do so?” (
LEWIS-KRAUS G, 2016).
● And, Super AI?
http://io9.gizmodo.com/how-much-longer-before-our-first-ai-catastrophe-464043243 ( Dvorsky G.,
2013)https://mobile.nytimes.com/2016/12/14/magazine/the-great-ai-awakening.html?_r=0&referer= ( LEWIS-KRAUS G, 2016)

Two Types of AI
● Rule-Based AI
● Data-driven AI
● Example: Checkers. Albert Samuel @IBM began writing
code for a checkers game program in 1949. In 1956, the
program was demonstrated to the public on live television.
In 1962, the computer beat checkers master player Robert
Nealey, and IBM’s stocks rose 15 percent overnight.
● Rule-based AI can never be more intelligent than its
creators, but data-driven AI can!

AlphaGoZero https://www.quantamagazine.org/artificial-intelligence-learns-to-learn-entirely-on-its-own-20171018/

Machine Learning
● Machine learning is the subfield of computer science that gives computers the
ability to learn without being explicitly programmed." (Arthur Samuel)
● Definition (Tom Mitchell): "A computer program is said to learn from
experience E with respect to some class of tasks T and performance measure
P if its performance at tasks in T, as measured by P, improves with
experience E."
● Supervised vs Unsupervised.

Insurance
https://www.lemonade.com/

Healthcare
https://www.remedymedical.com/
https://www.nature.com/articles/srep26094

● Bridgewater Associates: World’s largest hedge fund has a project to automate
decision-making to save time and eliminate human emotion volatility.
● Goldman Sachs: Two out of the 600 equity traders left. Found that four
traders can be replaced by one computer engineer.
Deep Learning in Finance

● Transactions: By 2020 at least five percent of all economic transactions will
be handled by autonomous software. AI will process payment functions and
learn from customer behaviours, through Intelligent Payment Management
(IPM).
● Savings: AI will help consumers make daily financial decisions and monitor
spending. New Personal Financial Management apps use contextual
awareness, which measures spending habits and online footprints to create
personalised advice. Combining pooled financial data with end-user control to
offer tailor-made services is a classic AI solution.

● Cross-selling: Categorization-as-a-Service (Caas), for understanding
customer transactions for cross selling.
● Fraud prevention: Mining user data to detect abnormal behavior, anomalies,
and unusual transactions.

● Mizuho Financial Group sent Pepper, its humanoid robot into its Tokyo branch
to handle customer inquiries. Partnering with IBM to enable Pepper to
understand human emotions, and build interaction into apps.
● RBS is trialing Luvo AI, a customer service assistant to interact with staff and
customers.

● AXA (insurer) has an app-based bot called Xtra, it engages in bespoke
conversations with customers about healthy living.
● AI is used in peer2peer lending.
● https://www.bloomberg.com/graphics/2017-wall-street-robots/
These Are the Wall Street Jobs Being Replaced By
Robotshttps://www.bloomberg.com/graphics/2017-wall-street-robots/

Hedge Funds use Machine Learning
Blackrock: replacing human stock
pickers with machine algorithms.
Sentient Inc: Hedge fund run entirely
using AI. Secret algo with adaptive
learning. Uses thousands of
machines.
Numerai: Hedge fund makes trades
by aggregating trading algorithms
submitted by anonymous
contributors, prizes are awarded in
cryptocurrency.
Emma: Evolved a hedge fund using a
software that writes news articles.
Very little data about the track record of these hedge funds, as
the business remains secretive.
Investor reluctance to turn over money completely to a machine.

Genesis: Breast Cancer Detection
● We use a data set of breast cancer cell measurements and assess
whether we can predict whether a tumor is malignant or benign.
● The data used comprises nine cell measurements such as cell
thickness, dimension, etc., and from these measurements we wish
to attain a high level of accuracy in classification of cells.
● A system such as this can potentially be the first line of diagnosis
and may be able to improve the doctor's diagnosis, in a Bayesian
sense, as there are now two readings of the cell, by the deep
learning net and the doctor.
● The data set is the Wisconsin breast cancer data.

Cancer Data
https://futurism.com/machine-learning-identifies-breast-lesions-likely-to-become-cancer/

TensorFlow and One-Hot Encoding

Canonical Example: Digit Recognition
https://www.cs.toronto.edu/~kriz/cifar.html
● Futile to run a
multinomial
regression on
784 variables
● Extensible to
many finance
problems

Amazon Rekognition https://console.aws.amazon.com/rekognition/home?region=us-east-1#/label-detection

Deep Learning is Pattern Recognition

The Mathematics of Deep Learning (http://srdas.github.io/DLBook/)

Subset of the Net
Net input
Activation function
Sigmoid function

Activation Functions https://en.wikipedia.org/wiki/Activation_function

Loss Function
Cross
Entropy
Quadratic
Loss

Gradient Descent
Total
Gradient
Stochastic
Batch
Gradient

TensorFlow Playground
http://playground.tensorflow.org/

Learning the Black-Scholes Equation (Culkin & Das, 2017)

Fit the Model
In-Sample Out-of-Sample

Predicting the Direction of the S&P500
Culkin, Das, Mokashi (2017)

Organize data and fit the model

Predictions
TensorFlow
H20.deeplearning : lookback 30 days, forward predict
30 days, total experiments 115.
● More than 50% right 72.1% of the time
● Overall accuracy 54.4%

Everyone’s trying to deep learn market prediction
https://www.youtube.com/watch?v=ftMq5ps503w

Two Curses of Predictive Analytics
Non-stationarity
● Strong: Joint distribution of all variables (Y,X) remains the
same over time.
● Weak: only mean and autocorrelation need to be same over
time. If we are predicting the first moment, then this works.
Randomness
● If noise swamps the signal, then we get poor predictions.

Homomorphic Encryption
● Machine learning on encrypted data
● Homomorphic encryption
(https://en.wikipedia.org/wiki/Homomorphic_encryption) is a form of
encryption that allows computations to be carried out on ciphertext, thus
generating an encrypted result which, when decrypted, matches the result of
operations performed on the plaintext.
● Homomorphism applies under different mathematical functions. Different
schemes support various mathematical operations.

AI as a Service
● Commoditized Platforms: Amazon's AWS Machine
Learning VMs.
● Commoditized Services: e.g., Image Rekognition.
Document analysis.
● Commoditized Models: Model zoos. NLP offerings.
● Encrypted Crowdsourcing of AI. e.g., Numerai. Managing
data and privacy.
● Federated Machine Learning. Distributed Learning and
data privacy.

Other applications of deep learning in finance
● Forecasting the VIX curve.
● Predicting Credit Card Default, see Khandani,
Lee, Lo (2016).
● Portfolio optimization.
● Text analytics.
● Text generation (e.g., Narrative Science).

What happens to financial sector employment?

The Atomic Level of Work
David Beyer
What tasks get
automated first?
AI researchers salaries go through the roof:
https://www.nytimes.com/2017/10/22/technology/artificial-intelligence-experts-salaries.html?hp&action=click&pgtype=Homepage&clickSource=story-heading&module=second-column-r
egion&region=top-news&WT.nav=top-news

Text Analytics
● Text analytics: Das (2014) http://srdas.github.io/Papers/Das_TextAnalyticsInFinance.pdf;
Jegadeesh and Wu (2013); Loughran and McDonald (2014).
● Number of risk words predicts earnings next quarter to be
lower.
● Lower readability predicts lower next quarter performance.
● Larger annual report MD&A predicts lower next quarter
performance.
● Size of filing to SEC server predicts worse performance.
● Using Word Embeddings. (word2vec from Google)
● Topic analysis, Blei, Ng, and Jordan (2003). Opens up new
areas of risk.

RegTech
Zero-Revelation Linguistic Regulation: Detecting Risk Through
Corporate Emails and News (Das, Kim, Kothari 2016)
● Financials are often delayed indicators of corporate quality.
● Internal discussion may be used as an early warning system for
upcoming corporate malaise.
● Emails have the potential to predict such events.
● Software can analyze vast quantities of textual data not amenable to
human processing.
● Corporate senior management may also use these analyses to
better predict and manage impending crisis for their firms.
● The approach requires zero revelation of emails.

Enron: Returns and Characteristics

Enron Movie (by Jim Callahan) http://srdas.github.io/Presentations/JimCall
ahan_enron-sm.mov

Value Drivers in FinTech
● Using Theory to develop models to apply to Big
Data.
● Question/problems are primary, data is secondary,
in the success of FinTech ventures.
● Simplicity, transparency of models fosters
implementability.
● Analytics per se is multidisciplinary.
● Disparate data is the norm.
● Significant investment in hardware and talent.

Pitfalls to avoid
● GIGO: Garbage in, garbage out. See [Alexander et al (2017)
(http://srdas.github.io/Papers/big.2016.0074_FINAL.pdf).
● IO (Information Overload): Collecting too much data and not using it correctly.
Use theoretical models.
● BiNB (Bigger is Not Better): Big data leads to bigger errors if misused. Taleb
critique. TDA (topological data analysis, Ayasdi https://www.ayasdi.com/,
Simility https://simility.com/.
● CCC: Confusing correlation with causality. Tighter review cycles for predictive
models.
● $$$: May involve expensive infrastructure. Go all in.
● TiP (Trust is Paramount): Privacy issues. Implement trust through technology.
● CS (Customer Satisfaction): Excessive misdirected automation leading to
poor client service. Robo-advising, chatbots. Use Design Thinking for
consumer centric technology.

Estimating the effects of technology
● (Roy) Amara’s Law: “We tend to overestimate the effect of a technology in
the short run and underestimate the effect in the long run.”
● Arthur C. Clarke’s Three Laws:
a. When a distinguished but elderly scientist states that something is
possible, he is almost certainly right. When he states that something
is impossible, he is very probably wrong.
b. The only way of discovering the limits of the possible is to venture a
little way past them into the impossible.
c. Any sufficiently advanced technology is indistinguishable from magic.

Readings on AI and Deep Learning
1. http://srdas.github.io/DLBook/
2. https://medium.com/@katherinebailey/hashtag-artificial-intelligence-47ff35e6a9cc
3. https://www.linkedin.com/pulse/ai-deep-learning-explained-simply-fabio-ciucci
4. https://hackernoon.com/why-ai-is-now-on-the-menu-at-dinner-even-with-my-non-tech-friends-44c666348de4
5. https://futurism.com/kurzweil-claims-that-the-singularity-will-happen-by-2045/
6. https://www.quantamagazine.org/artificial-intelligence-learns-to-learn-entirely-on-its-own-20171018/
7. https://www.technologyreview.com/s/603984/googles-ai-explosion-in-one-chart/?utm_campaign=add_this&utm_source=twitter
&utm_medium=post
8. http://www.zerohedge.com/news/2017-02-13/goldman-had-600-cash-equity-traders-2000-it-now-has-2
9. https://futurism.com/an-ai-completed-360000-hours-of-finance-work-in-just-seconds/
10. https://www.theguardian.com/technology/2016/dec/22/bridgewater-associates-ai-artificial-intelligence-management
11. https://www.mizuhobank.com/mizuho_fintech/news/pepper/index.html
12. http://www.businessinsider.com/royal-bank-of-scotland-launches-ai-chatbot-luvo-using-ibm-watson-2016-9?r=UK&IR=T&IR=T
13. https://www.the-digital-insurer.com/dia/xtra-by-axa-ai-driven-personal-wellness-coaching-app/
14. http://www.nanalyze.com/2017/04/ai-fintech-startups-loans-new-credit/
15. https://www.americanbanker.com/news/this-is-how-financial-services-chatbots-are-going-to-evolve
16. http://fortune.com/2017/03/30/blackrock-robots-layoffs-artificial-intelligence-ai-hedge-fund/
17. https://en.wikipedia.org/wiki/Sentient_Technologies
18. https://medium.com/numerai/an-ai-hedge-fund-goes-live-on-ethereum-a80470c6b681
19. https://venturebeat.com/2017/05/06/ai-powered-trading-raises-new-questions/
20. http://www.wired.co.uk/article/how-ai-is-transforming-the-future-of-fintech
21. https://medium.com/numerai/encrypted-data-for-efficient-markets-fffbe9743ba8
22. https://www.bloomberg.com/news/features/2017-09-27/the-massive-hedge-fund-betting-on-ai
23. https://www.wired.com/2017/02/ai-threat-isnt-skynet-end-middle-class/?mbid=social_twitter_onsiteshare
24. https://www.wired.com/2008/06/pb-theory/
25. https://www.gartner.com/smarterwithgartner/gartner-predicts-our-digital-future/

FinTech, AI, Machine Learning in Finance

More Related Content

What's hot (20)

Similar to FinTech, AI, Machine Learning in Finance (20)

Recently uploaded (20)

FinTech, AI, Machine Learning in Finance