Tip:
Highlight text to annotate it
X
Naomi - click here and write a clever title
Transcription of interview with Eric Siegel on April 15, 2013.
Douglas Goldstein, CFP�, Financial Planner & Investment Advisor
Eric Siegel is an expert in data mining and predictive analytics. He just came out with
a new book called Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie
or Die. He also founded an organization called Predictive Analytics World.
Douglas Goldstein, financial planner & investment advisor, interviewed Siegel on Arutz Sheva
Radio.
Douglas Goldstein: What do you do?
Eric Siegel: Predictive Analytics is a technology that learns from data to predict the behavior
of individual people. A lot of people think of prediction as sort of forecasting and that
might be predicting the general direction of an economy or how many ice cream cones
the company is going to sell next quarter. Predictive Analytics is making predictions
for each individual person, customer, patients, donor for non-profits, constituents, potentially
investor or potentially other organizational elements such as an individual stock price.
So it wouldn�t be how many ice cream cones are going to be sold but will this individual
be likely to buy an ice cream cone.
Douglas Goldstein: Data mining meaning you have somehow collect a huge pool of information
and by analyzing it, you then try to figure out what an individual within that data will
do at anyone point?
Eric Siegel: You can think of data as a lot of hyper and how much data is growing or how
much there is. That�s what it is, recording of things that have happened so if you learned
from that experience to make prediction by way of certain kinds of regression, decision
trees, neural networks, some of these buzz words people may have heard. This is a very
broad discipline that applies across not just finance but all sort of industry sectors within
government, law enforcement, for political campaigns making predictions about what direction
people are going to go for voting which you can think of as a marketing application, universities,
educational applications, but in one way or another, it�s about predicting behavior,
what people are going to do whether they�re going to click, buy, lie or die, whether they�re
going to steal, whether the thrive dropped out of school, make a donation, vote one direction
or another. As you can imagine, those types of predictions or behaviors have all source
of applications but the value ultimately comes one way or another in driving and improving
all of the operational decisions, all the actions organizations take in the way they
interact and provides services to the world and in the case of finance, it�s going to
be about making buy and sell decisions.
Douglas Goldstein: You might look for example at a brokerage firm and see how their clients
tend to react after getting statements whether they buy or sell after that?
Eric Siegel: I don�t know if you�d think of that as a financial application because
then you�re talking about the marketing of that brokerage but sure, just like any
other company, the brokerage does marketing and they want to determine for example which
of our customers are we at risk of losing, who�s going to defect or cancel their usage
of our product and our services and that�s something that money comes in with cell phone
companies and most of the big ones do this. They target the retention efforts and the
expenses for retention that discounts that are intended to retain the phone calls made.
Those resources are targeted very carefully according to where they can best serve by
making a prediction which customer are we at risk of losing. That�s one of the min
marketing applications across sectors of predictive analytics.
Companies have an incredible amount of data about their customers. Usually this kind of
thing can be done to a great degree just using the internal data that already exists. Data
are just of sort of like the residue of the function and operations of organization. It�s
being collected. It�s aggregating and it has this whole source of secondary purpose
which is you can learn from it. So organizations are sitting on this data and they could potentially
make great use of it to make prediction.
Douglas Goldstein: Do you need a whole lot of data or can someone look at his own household
budget or his own portfolio and is that enough data to make any statistically significant
analysis?
Eric Siegel: For most individual investors even if one or another, they�ve accumulated
a lot of data and you can download a great deal of data in terms of the behavior of certain
instruments, but usually the application of actual prediction and acting on prediction
is something reserved more for organizations rather than the individual. Now there are
exceptions and in fact, the first chapter of the book kicks off with the story about
a colleague of mine. He�s very much a long term expert in predictive analytics and he
actually when he was more junior, maybe about 15 years ago, he invested all of his and his
wife�s personal money in his own stock market prediction system.
There are certainly cases where the experts who know how to do this stuff can use it as
an individual but usually organizations, the one has got all the data and organizations,
the one that is executing so many operational decisions a day where there�s a benefit
by driving those decisions from the individual predictions per person or in finance, sometimes
it�s per instrument.
Douglas Goldstein: What is Predictive
Analytics World?
Eric Siegel: Predictive Analytics World is the leading vendor in neutral conference.
It focuses on commercial deployment so it�s not researching development or academic conference.
There are 7 a year. In fact, there�s one in London coming at the end of this year,
October 23-24 and you can go to www.predicitiveanalyticsworld.com to find out about that. It�s a great place
to glance and to look at the programs of these conferences. They are laid out in detail and
you�ll see all the different organizations, well-known brand name organizations that are
executing on prediction in this way.
Douglas Goldstein: We often use a type of statistical simulation called a Monte Carlo
simulation where we gather whole lot of market information and we run many simulations of
possible market returns in order to come up with a practical decision of what to do today.
Based on what you�re saying, it sounds like these are not the same.
Eric Siegel: Monte Carlo simulations are usually more about what would I refer to as forecasting
where you sort of looking and computing the probable aggregate overall outcomes rather
than that lower level of granularity per individual person or per element. This is different.
This is sort of moment to moment buy and sell decisions or in the case of marketing, who
to contact and not contact, in the case of presidential campaign in the United States
and the Obama campaign did this, who to contact, who to call, whose door to knock on, where
all this individual low-level decisions make a difference so is this stock going to go
up over the next 5 hours or is it going to go down. It�s not something that sort of
drive a broader high level strategy about diversification. It�s going to actually
potentially drive one of these funds about what your making those decisions.
Douglas Goldstein: Are there money managers who in fact use this type of predictive analytics
to make minute to minute decisions or day to day decisions?
Eric Siegel: Making these kinds of predictions and it�s basically about predicting better
than guessing. In general, you can�t predict a future in an accurate way but you very much
can tip the balance an odd in your favor by predicting better than guessing. By putting
the odds and playing the numbers game in a more effective way, you can make a difference.
The colleague I mentioned, whose story is written in the first chapter of the book,
he likes to say that black box trading or algorithmic trading driven by these means
is potentially the most difficult and challenging although of course also potentially most profitable
arena in which to apply predictive analytics.
Douglas Goldstein: How does a decision tree
apply to someone in your field doing predictive analytics?
Eric Siegel: Decision tree is not necessarily something that comes from a machine, learning
from data but it can be, so you can take all sorts of things. There�s a Humerus Decision
Tree online that went viral deciding if you drop some food on the floor, should you eat
it anyway. It depends did your parents see it and did your parents see you drop the food,
is there bacon involved, how expensive was it. You can make these decisions and in decision
tree, you start from the top, you answer yes or no questions and you go left and right.
It�s another way basically to sort of put a bunch of rules or business rules. If the
stock market 30-day window average is this much higher than a 60-day moving average and
a certain relative price to the S&P 500 etcetera and-and-and then we think that there�s a
3% greater chance than average that things are going in a good direction over the next
so many hours. That�s kind of very kind arcane and-and-and rule so that�s what decision
trees embody and you can capture those rules and tweak them and optimize them automatically
from data. I don�t know whether chess players actually use decision trees that were created
and optimize over real historical data. I think there maybe ways to do that and maybe
part of the way computers play chess but it�s very much one of the more popular methods
in predictive analytics applications.
I
actually have a story from the same guy I mentioned, the colleague who used his personal
investment where he discovered one of his colleagues had done very much the same thing.
I also started on Apple II Plus hacking when I was a kid with 48k of memory. The Apple
two plus comes with a decision tree program but it�s not one that learns from mass amounts
of data. It learns as you play the game Guess The Animals or 20 Questions so the computer
ask you yes or no questions and every time you tell the animal that�s never heard of
before, you add a new question to help it differentiate so it�s actually in its memory
building a decision tree.
Douglas Goldstein: Is there a website or way to people can keep track of your work?
Eric Siegel: The website for the book is www.thepredictionbook.com that will actually bring you information about
The Predictive Analytics World Conference and we�ve got a bunch of excerpts and videos
and stuff about the book on there.
Douglas Goldstein, CFP�, is the director of Profile Investment Services and the host
of the Goldstein on Gelt radio show (Monday nights at 7:00 PM on www.israelnationalradio.com.
He is a licensed financial professional both in the U.S. and Israel. Securities offered
through Portfolio Resources Group, Inc., Member FINRA, SIPC, MSRB, NFA, SIFMA. Accounts carried
by National Financial Services LLC. Member NYSE/SIPC, a Fidelity Investments company.
His book Building Wealth in Israel is available in bookstores, on the web, or can be ordered
at: www.profile-financial.com (02) 624-2788 or (03) 524-0942.
Disclaimer: This document is a transcription and/or an educational article. While it is
believed to be current and accurate, divergence from the original is to be expected. The original
podcast can be heard at https://sites.google.com/site/goldsteinradioshows/. All information on this website is purely
information and should not be used as the sole basis for making financial decisions.
The opinions rendered herein are those of the guests, and not necessarily those of Douglas
Goldstein, Profile Investment Services, Ltd., or Israel National News. Readers should consult
with a professional financial advisor before making any financial decisions. Please see
the complete disclaimer at https://sites.google.com/site/goldsteinradioshows/.