Advanced Analytics for Investment Firms and Machine Learning

1© Cloudera, Inc. All rights reserved.
Advanced Analytics for
Investment Firms
Faster Time to Value and Higher Returns
for Your Analytics Projects
Dr. Richard Harmon
Sean Anderson

Types of Analytics
Dr. Richard Harmon
Financial Services Industry Leader
Cloudera
Machine learning and advanced analytics for
higher investment returns
Develop better models faster with Data
Science Workbench
Sean Anderson
Sr. Product Marketing Manager, Data
Science and Machine Learning
Cloudera

Our relationship with data is changing
Data is now a strategic asset, how you use it is a key differentiator

• What makes “big data” big?
⎼ Volume?
⎼ Variety?
⎼ Velocity?
• Data becomes big when we take one or
more large data sets and start to analyze
relationships between observations
Big Data and Data Products

Wrangling Big Data is Time Consuming
Data preparation accounts for about 80% of the work of data scientists
(Source: CrowdFlower)

METHODS
DATA
Data Mining
Machine Learning
Large Scale
Data Analysis
OPTIMIZATION
Optimization
Techniques
SIMULATION
High Performance
Computing
Spatial
Computation
Run-Time
Monitoring
Numerical
Simulation
NATURE INSPIRED
Evolutionary
Algorithms
Swarm Insects
Gene Regulatory
Networks
Reasoning
Compositional
ReasoningStrategic
Reasoning
Artificial
Intelligence
FORMAL METHODS
Model
Validation
Program
Analysis
DECISION MAKING
New Institutional
Economics
Incentive
Schemes
Game
Theory
MODELING
Agent-Based
Modeling
Complex System
Methods
Multi-Scale
Modeling Dynamical
Modeling
Social Aspects
Social
Computing
Human Systems
An Analytical Methods Viewpoint based upon the Approach Used

An Overview of the Analytics Space from a AI Perspective

A Machine Learning Viewpoint with a Use Case Mapping

Storage
• Archival
• Traceability
Batch
• ETL
• Data
Validation
• Reg Reporting
Interactive
• Risk
Aggregation
• Stress Testing
HPC
• Risk Modeling
• Backtesting
Streaming &
Real Time
• Mkt
Surveillance
• Best execution
Evolution
Risk & Regulatory Compliance Use Cases on Big Data
Cloudera Data
Science
Workbench

Why is Hadoop Ideal for Data Science?
• High volume, low-cost shared storage = more data, more kinds of data
• Parallel compute, local to the data = more experiments, better results
• Scalable, fault tolerant = easy scale out, not just scale up
• Flexible, multipurpose data platform = easier path to production
• Superior flexibility and price/performance to any other data platform

Cloudera Enterprise Data Hub
Handle real-time
data ingest from
diverse sources
Governance and
Security
Data Streams
Deployment Flexibility
Machine Learning
Capabilities
Diverse Analytical
Options
Combine Data from Different Sources
Data Mgmt. Hub
Scale easily & Cost
effectively
Batch or Real- time
Data Streams
Data Sources
Data Sources
Data Storage &
Processing
Reporting, Analytics &
Auditing
Data Ingest
Other
Data Governance (Data Lineage, Data Protection)

Why Cloudera Enterprise
Making Hadoop Fast, Easy, and Secure
Hadoop:
• One place for unlimited data
• Unified data access
Cloudera makes it:
• Fast for business
• Easy to manage
• Secure without compromise

Machine Learning in
Investment Management

Advanced Analytics Use Cases
Patterns in Market Data
Satellite Image Analysis
Simulation for
Strategy Modeling
News Feeds for
Economic Forecasting
Market Sentiment Localized Pricing
Risk Assessment Compliance

Interconnected Risk Indicator: CoVaR (Conditional Value at Risk)
CoVaR: Conditional Value at Risk
• Dr. Tobias Adrian (NY FED) & Prof. Markus Brunnermeier (Princeton Univ) –
American Economic Review, Vol 106, no. 7, July 2016
• Measures “tail-event” linkages
• It is the VaR (Value at Risk) of the whole financial sector conditional on one or more
institutions being in distress.
• CoVaR - measures interconnectedness, spillover effects, risk transmission,
deleveraging, etc.

CoVaR Estimation: Application of ML for Improved Model Fit
Quantile Regression vs. Kernel Quantile Regression
Kernel Quantile Regression (KQR-CoVaR)
Robust
Quantile Regression
Noisy
Adrian (NY FED) & Brunnermeier (Princeton Univ) Harmon (Cloudera), Bittman, Ruttenberg (SAP)

Improved Economic Forecasting:
Predicting US Non-Farm Payroll (NFP)
Abby Levenberg, Stephen Pulman, Karo Moilanen, Edwin Simpson, and Stephen Roberts, “Predicting Economic Indicators from Web
Text Using Sentiment Composition”, International Journal of Computer and Communication Engineering, Vol. 3, No. 2, March 2014.
(http://www.ijcce.org/papers/302-E3007.pdf)

Modeling The Real World – Simulation & Optimization
Agent-Based Models (ABMs)
• Agent-based models (ABM) consist of heterogeneous agents
that are allowed to freely and randomly interact in the
interests of maximizing their own goals.
• ABMs are uniquely able to include evolving intelligence that
allows agents to learn and adjust as new information comes
into the system and how the system evolves in response to
this new information.
• ABM’s are ideally suited to a model behavioral heterogeneity,
and the dynamics that result from that heterogeneity.

Mortgage Prepayment Reduced Sensitivity: (Burnout)
In US RMBS market, the phenomenon known as interest rate burnout
arises from the changing composition of homeowners in a fixed
mortgage pool over time. Many prepayment models applied various
techniques to track heterogeneous borrower behavior within mortgage
pools.
JPM created multiple dimensions of homeowner incentives - essentially
the homeowner must be "ready”, "willing”, and "able” to refinance.
This early form of ABM was a key way to model heterogeneous
behavior when individual loan data was not access for a pool of
mortgages.
Option Adjusted Spread (OAS) Analysis
OAS is a widely-used computational simulation valuation model for
RMBS. This requires a Monte Carlo based simulation of the distribution
of future cash flows under thousands of different interest rate paths.
Residential MBS Investment: Mortgage Prepayment Modeling

What is Endogenous Risk?
- The conventional view was that markets were huge and anybody trading in the market was
small by comparison – that you could neglect the impact of what the firm was doing.
- In most empirical risk models, risk is determined by exposure to a group of factors – markets,
industries, countries, exchange rates, ... and using volatility as a metric for uncertainty.
- While these models allow for uncertainty about those exposures, but often ignore
endogenous influences such as how market participants react to each other.
- Endogenous risk refers to the risk that comes from market participants reacting to each other
rather than reacting to outside forces.
Endogenous Risk – A New Challenge for Investment Managers
Source: “Avoiding crowds: BlackRock leads push to model 'endogenous' risk”, Risk.Net, 2016.
”A market becoming more correlated more quickly is a market becoming increasingly hard to predict.”

Feed-Back Loops & Tipping Points:
Jean-Philippe Bouchaud of Capital Fund Management (CFM) thinks investors with even relatively
small positions can be trapped in a feedback loop where they cause asset prices to fall by selling,
and are forced to liquidate more assets as a result.
The logic here is that firms should look at holdings relative to securities available in the market,
rather than compared with the whole universe of issued securities.
CFM uses an agent-based modeling approach that takes into account the size of holdings by
market participants to help evaluate the tipping point at which they risk creating their own
deleveraging vortex.
Endogenous Risk – New Challenge for Investment Managers (Con’t)
Source: “Avoiding crowds: BlackRock leads push to model 'endogenous' risk”, Risk.Net, 2016.

Polling Questions #1
What kind of advanced analytics do you use for your investment strategy?
1. Machine Learning
2. Natural Language Processing
3. Simulations/Agent-Based
4. Multiple Approaches
5. None

Alternative Data Sources
& “Quantamental Investing”

What is Alternative Data?
"Alternative data draws from non-traditional data sources,
so that when you apply analytics to the data, they yield
additional insights that complement the information you
receive from traditional sources."
S&P Global CIO Krishna Nathan
(CIO, January 3, 2017)

“Quantamental” Investing – “Big Data” driven investment approach
- Quantamental managers combine the bottom-up stock-picking skills of fundamental
investors with the use of computing power and big-data sets to test their hypotheses.
- Quantamental investing has been driven in part by the rise of so-called alternative
data. Fund managers can now study everything from social-media data (to predict
footfall in a location or sentiment around a new movie) to the number of cars in a mall
parking lot.
Merging Quantitative Analysis, AI, and Fundamental Research
Some $30 billion in assets (about 11 percent of active equity funds) will be targeted,
with $6 billion rebranded BlackRock Advantage funds. These funds focus on quantitative
and other strategies that adopt a more rules-based approach to investing.
“At BlackRock, Machines Are Rising Over Managers to Pick Stocks” – NY Times, March 28, 2017

Alternative Data Sources - Examples
Satellite Data - These are companies that
utilize image data from orbiting satellites to
do things like measure the number of cars in
Walmart parking lots or farm health based on
the color of crops.
Web/App/Social Media Data – These are
companies which mine social media or use data
firehoses from the web/mobile to understand
what’s happening in the world or how people are
interacting with their devices.
Weather Data – These are companies which are
developing weather models and utilizing more
sensors to get better localized data or improve
weather forecasting.
Location/Foot Traffic Data – Companies
that use different means to understand
where consumers are going by measuring
foot traffic via check-ins, video analysis, etc.
Local Prices – These companies can see what’s
happening to prices and inflation by
aggregating data from ground-level sources.
Alternative Credit - Companies developing new
credit models that utilize sources of alternative
data (like mobile usage).
Credit Card Transactions – These are
companies that use anonymous
aggregate transaction data to understand
trends in consumer purchasing habits.
Alternative Data Monetizers/Aggregators –
These are companies who pay for access to
individual data streams which become more
valuable in a bundle, and then sell those
packages to investors.

Alternative Data Source Vendors

Alternative Data Example: Best Predictor of Chipotle Sales
On April 12, Jeff Glueck, the
CEO of Foursquare, published
a post predicting that
Chipotle's first-quarter sales
would be down nearly 30%.
That was based on foot-traffic
statistics built from explicit
check-ins and implicit visits
from Foursquare and Swarm
app users who enable
background location.
The chart below shows the share of visits to Chipotle restaurants in comparison to visits to ALL restaurants in the
United States. In the 2015–2016 winter, visits to Chipotle restaurants declined more significantly than in 2014–2015.

Advanced Text Analytics: Asset Management Use Case
Partners
Big Data Enabled Investment Management Process:
Goal: To provide new data sources and analytic tools to streamline and
improve the portfolio investment process.
Benefits:
- Centralized, secure & scalable investment analysis data lake
- Streamlined information management and analysis process
- Flexibility for portfolio managers to utilize any tools and any data sources
- New insights from sentiment/emotional analysis of sell-side analyst
reports and news feeds
- Expanded usage of ML-based analytics for macroeconomic, risk and
market analysis.
- Automated market, portfolio and event monitoring and alerts
- Part of a wider digital transformation initiative

Open data science in the enterprise
IT
drive adoption while maintaining compliance
Data Scientist
explore, experiment, iterate

Our goal: An open platform for data science at scale
Help more data scientists
use the power of Hadoop
Use a powerful, familiar
environment with direct access to
Hadoop data and compute
Data Scientist
Data Engineer
Make it easy and secure to
add new users, use cases
Offer secure self-service analytics
and a faster path to production on
common, affordable infrastructure
Enterprise Architect
Hadoop Admin

Introducing Cloudera Data Science Workbench
Self-service data science for the enterprise
Accelerates data science from
development to production with:
• Secure self-service environments
for data scientists to work against
Cloudera clusters
• Support for Python, R, and Scala,
plus project dependency isolation
for multiple library versions
• Workflow automation, version
control, collaboration and sharing

Polling Question #2
How fast do you need to ingest your data?
1. sec
2. ms
3. hours
4. min
5. don’t know

Demo

Q&A

Thank you

Advanced Analytics for Investment Firms and Machine Learning

Recommended

More Related Content

What's hot (20)

Similar to Advanced Analytics for Investment Firms and Machine Learning (20)

More from Cloudera, Inc. (20)

Recently uploaded (20)

Advanced Analytics for Investment Firms and Machine Learning