Leda Braga: Data science and its role in investment strategy


[APPLAUSE]
>>Thank you. Good morning, everyone. Thank you, Margaux,
for the lovely invitation. Thank you, Gunther, Walter
as well, who, who helped to invite me. I don’t know
if Gunther is around, I haven’t seen him.
Good, yeah. So it’s a fantastic opportunity
to be here speaking to you. You know, I have a tendency to
try to rationalize things in my mind and I, a few years ago
I rationalized Stanford to myself. And I thought Stanford
is this place where anything is possible. [LAUGH] And if
you, and if you get a number of intellectually capable
people, hard working people, to believe they are in a place
where anything is possible, that is a very powerful thing
to do. And so to me, it’s a pleasure, it’s a privilege
to be contributing to this conference today. And [COUGH]
you know, in preparing for the conference, in in doing
some homework, I looked at the previous speakers and
the previous schedules for the other previous
two conferences. And I realized that I’m the first
financial industry speaker in this conference. And so
I’m gonna spend a little bit of time taking you through
what we do as investment managers that we are and
a bit about our firm. But I, I have a feeling
that there’s a reason why they haven’t been many
financial industries speakers. It’s because if you take us
back to that percentage that Margaux was talking about,
the percentage of women in the industry, I’ll bet with
you the financial industry is right at the bottom. It’s
gonna be small single digits, I suspect. And so, you know,
that’s something to change. And hopefully, in today’s
talk you’ll get a feeling for what we do and how you can
make a difference in data science in this industry. And
so, I also realized that a lot of the speakers here come from
industry giants like Google, Facebook, Intel, Microsoft.
I work for an industry that is new to the conference, the
financial industry, and for a much smaller company. So I’m
gonna spend a couple minutes taking you through who I
work for and what I do for a living. So I, as you
can see from the program, I represent a firm called
Systematica Investments. We are an alternative
asset manager, and we have been in charge of
funds, systematic strategies, investment funds that go back
to 2004. And so we have a long track record. I’m saying we
are an alternative investment manager. There’s more to come
on that in a minute. But basically, out of the pool of
people out there that manage assets, that manage money for
other people, the alternative slice of that industry
is the state of the art, the very advanced piece of the
industry. We are allowed to trade very liberally, a lot
of different instruments. We trade in and out of of
securities very rapidly, very, very, the turnover
is very high, we deploy a lot of leverage.
And so we are somewhat the, the leading edge of
the investment management industry. We have about $9
billion under management in our firm. And we are quite
small in terms of staff. We, we employ 108 people across
five locations worldwide. But, we are diverse. Across
these 108 people we have 26 nationalities. So that is,
that is something to be proud of. Many of us have PhDs
in science and engineering. I’m a PhD in engineering. And I was an academic for
several years before I joined the financial industry
in the early 90s. There’s a little bit
of a picture there, just to put some faces,
to the name. So that’s who we are, and
and we, in our mission statement we openly talk
about employing science and technology to achieve returns
in investment management. Now then I thought, you know,
this is a data conference. Let, let the data speak,
right? So I actually computed the word cloud on on
three staff questions. Three questions that we pose
to staff. What do we do for a living? And you can see
the word cloud there, we make money for
clients, pretty good.>>[LAUGH]>>So this is 98% participation
by staff by the way. So this is a good word cloud.
Why do you work for Systematica? Well,
clearly the environment plays, plays a big part, and
the culture, and the team. And finally, describe the culture
of Systematica? And I’m very surprised by this where
the word family comes up, and the culture, and excellence,
which is a value of our company. So, so that’s who we
are. So that’s Sytematica. So we are an alternative
asset managers. I said there’s gonna be
a little more about this. And I’m going to talk about
systematic investment management. So first of all,
let’s take a step back to think about what investment
management is. So investment management is the
professional administration of various securities and other
assets to achieve specified return goals and specified
investment goals for clients. What, so what, what do we do? We, we deploy the capitals of
the world, right? And, and where do these capitals come
from? Well, there’s a lot of big pools of money out there
that need to be managed. You start with pension funds,
so all of our pension money needs to be managed.
If you think about insurance companies, insurance companies
are in the business of collecting premiums and
stashing up the money in preparation for the need
to pay out upon some events. And so this money that the
insurance companies put aside needs to be managed. Sovereign
wealth funds holds the wealth of various countries and they
need to be managed. And so each one of these clients will
have different preferences for time horizon, commitments
that they have against those monies that they need managed.
So some investors have a longer
time horizon than others. Some investors need a lot of
return, some investors have restrictions on what they can
and cannot invest. And our job is then to, to organize
the portfolios that suits these investment objectives
and we do that through funds. We set up funds and investors
then come and invest in funds. So that’s what the investment
management industry does. [COUGH] There’s about $80
trillion of professionally managed money worldwide,
$80 trillion. And, and how is this pie split?
So as I said to you, the majority of this money is
managed in a fairly direct and straightforward way. So much of this money is
deployed in buying stock markets. I just buy
the stock markets and I hope that they will grant
me the return that I need. Or I buy a portfolio of
treasuries of government bonds and I hope that that will give
me the return that I need. The alternative
asset management, the alternative slice of the
investment management pie is about 10% of that 8 trillion.
So 80 trillion is the big pie, 10% of the 80 trillion,
about 8 trillion, is called alternative
asset management. And then within the alternative
slice, there are three, parts. There’s real estate and
infrastructure, private equity, and hedge funds.
So real estate infrastructure, private equity, hedge funds.
Systematica lives in the third of the 8 trillion 10% that
is called hedge funds. And as I say we are the state
of the art, we are the more innovative
part of asset management. We trade in and out of things
with little restriction, we deploy leverage. We, we, we
have the freedom to innovate. And so, the, the activity
itself in terms of how we deal technically
with the activity, there are two avenues of work.
One is, what we call, signal generation and the other one is portfolio
construction. So every asset manager will talk to you
about signal generation and portfolio construction. Signal
generation is the process whereby we decide Which
securities or which assets we have a view on. This is a good
buy, this is a good sale. This is, this is going to go
up, this is going to go down. And so that is the forecasting
part of the job. Then the portfolio construction is
once I’ve selected the number of securities that I’m
interested in trading and having positions in, then how
do I size those positions? How much of each and
how quickly do I go in and out of them? And, and that
is the build of the exercise where we really say,
well, wait a minute. What kind of parameters
am I working to here? What kind of risk does
my client require? And what is, what kind of time
horizon does the client have? So, so, so, that,
that’s what we do. And, and the, the game is
in terms of data science, the systematic approach to
investment management is the one that is really of
live data science, right? And, and the because we
live in a world where all decision-making is becoming
data-driven, the, the press has really become very
interested in systematic asset management. The thing is that
15 years ago when I joined this side of the industry,
you wouldn’t see any press in systematic trading. Nowadays
you see it all the time. There’s headlines all
the time. Now if you actually look historically, the, the,
the big famous investors in the world behave like they
the scientists. I don’t know if you care to listen to what
Warren Buffett had to say. I mean, Warren Buffett is
perhaps an, an early time data scientist. He’s, he’s
a disciplined man who looks at the data to make his decisions
and uses the data and monitors the data to support
his ongoing decisions. So, so, so I think, you know, there is a lot of press there
but people are accepting systematic trading as a
strategy. Now what is it that stocks systematic trading
becoming more widespread and more dominant? I think there’s
amount of algorithm aversion. So what I’m saying is I’m
gonna get to how we go about dealing with data and
the challenges that we face. But before that just to put it
in context, just telling you, look, you know,
it’s a very good approach. We look at data to make our
investment decisions, but there’s a lot of algorithm
aversion to, to fight against, to battle with. So some of
the things that we hear, I like systematic but systematic
funds, they tend to be all the same. But I cannot really
understand the strategies. It’s all rocket science.
It’s black box stuff. It’s less transparent than
discretionary trading. And, and then you have to think
discretionary trading is the, the part of the investment
managers where, where the, the investment
process is dominated by human decision-making without
the formality of data science necessarily. And so you know,
one of the, the slides that I’ve got here is just to
say, what does the data say? So if I’m here defending the
data science driven approach to investment management,
is that a superior approach, is that an inferior approach?
How does that compare to the, the history of asset
management which up to now has been predominantly
discretionary? And so this is a study, I’m,
I’m quoting a study by, by a set of a peers
in the industry that that computed the returns of
over 9,000 hedge funds from one of the key industry
databases the, the Hedge Fund
Research Database. And those returns span the period,
an almost 20-year long period between 1996 and 2014. And
you can see the results there. What these guys did was they
picked up the 9,000 funds, and they used some natural
language processing algorithm to classify them as
discretionary or systematic using the description given in
the database. And then they computed averages and some
statistics on it. And you can see that they ended up with
four groups, systematic macro funds, discretionary macro
funds, systematic equity and discretionary equity. And
you can see that the returns, the average return, and
this is excess return. This is return over
the risk free rate. The average return is similar
across all four categories. And then another exercise
that they did was to try to explain these returns
using common factors. And common factors have to be
securities or types of portfolios that are easy, that
are obvious, that are easy to, to deploy. And so they did
about of analysis on that, a bit of regression there
to try to explain how much, to assess how much of the
returns of each category could be explained by simple
factors. And again, then what you’re left with
once you’ve explained some of the returns using the simple
factors, we’re gonna go alpha, we’re gonna go skill.
We’re gonna say, look, this is really what these
funds have delivered. And so there’s an alpha
role there. And by the time you’ve accounted
for the easy to explain piece of the return, the returns
look even more similar. And then finally, if you normalize
by the return volatility, so how much risk did I take
in stomaching those returns, then the ratio at the bottom
is quite comparable. So what is the slide actually
here to say? The slide is here just to give you the message
that the systematic approach to investment management
is about the same, maybe a little bit better than the
discretionary approach. So, so there’s a lot of, of future
promise in this approach. And so it’s okay. So later you
tell me that you, you work for this company. It’s state of
the art investment management. You are about
deploying the pools of capital of the world. And,
and you tell me that up to now the world has been
predominantly doing that job in a discretionary manner by
opinions and by people looking at data in a perhaps in
a less objective way. So, so what is the difference then?
What is the difference between the systematic investment
management approach and the more traditional discretionary
approach? Well, this, this graph is about this,
this difference. And, and the first thing that I need to
say is that this difference is going away. The two
approaches are merging, and in future everybody will talk
about the systematic approach because the discretionary
guys have realized that, that they are at an
information disadvantage. So they’re trying to join us and,
and integrate their, their world with ours. But in
any case, historically this is the main difference, the level
of diversification. So that if you listed all
the trading opportunities in, in order of trade conviction,
the discretionary guys tend to trade to the right of us
systematicful. So that means that they have fewer trades on
and higher conviction on their trading. We, by contrast,
have lots of trades on and not so much conviction on
each trade that we have. And that is the,
the main difference. Now what is the implication of
that? You know, I, I, this is a, a highly science-oriented
forum here, so I don’t need to explain this to you, but
I can tell you that, that most people out there don’t
think about these things. I used to ask this question to
motivate the thinking about diversification. I used to say
to people, listen, you know, if I tell you that you’re
going to toss a coin for your life and
if you get heads, you survive. If you get tails, you die.
And I, I offer you one throw on a very highly biased
coin towards heads, cuz I actually I’m lying,
saying I want you to live. Or I offer you a truckload of
throws on a less biased coin, which one do you take?
And, and most people take the, the highly biased coin.
But that’s not the answer. You guys are technical,
you know that. So, so the game here is this. How
biased does the coin have to be towards heads to achieve
a positive outcome, to achieve on average more
heads than tails, say with 80% probability? And the answer is
if you have a lot of throws of that coin then you only need
a little bit of bias, right, because you’ve got central
limit working for you. So, so that is the game that we play.
The systematic cloud tends have a more diverse sets of
trades on. And, and we don’t focus quite so much on getting
each trade right. So in other words, by diversifying the
choice of trade opportunity, the systematic approach makes
the investment process less reliant on the random
nature of forecasting, and more reliant on the risk
control in the portfolio construction. Okay that is,
that’s the key insight. Now, having said that,
if you think about investment management, the glamorous part
of investment management is forecasting? This is what
gets people the headlines. Like if you call a market
right, if you, if you, if you call a big event, that
is what is glamorous about it. But, but our approach is very
robust and very scalable and very auditable as well.
Okay, so having said that, and you’re now pretty sold
on the idea that, that systematic
trading is very good, is very auditable, it’s,
it’s a great approach and it’s going to grow to be the
majority of, of the assets on the management out there. Then
let me get a bit more prag, more practical about
what we actually do on a, on a day to day basis. So
in terms of signal generation, that’s the forecasting
problem, right? So
there we’re trying to look for factors that might tell us
if a particular company or a currency or the stock market
of a particular country is gonna do better or
worse in the future, so we’re trying to forecast.
What do we do? We look at also up to data,
price and volume data are key, we always start with price and
volume. We perform a lot of regressions, you know, various
flavors of regressions. Natural language processing
is a great discipline for us because it enables
us to parse data and unformatted, in unformatted
ways, so news and events and company filings and, and, and,
and general press material. Occasionally, we work in
the frequency domain to try to assess a signal generation.
And then at the portfolio construction side of things,
what do we do there? There once you’ve decided
which securities you have a view on, how much of
each one do you wanna own, how are you going to
construct that portfolio? What are some of
the techniques there? We use volatility
estimation techniques. The, the problem of portfolio
construction is a constrained optimization problem. So, so
we deploy all the techniques there. We do a lot of matrix
manipulation because you could think of the stock market as
this multi dimensional world described by all the stocks
in your traded universe. And then there’s another problem
that we all deal with, is well which is the problem
of executing the trade. So, once you’ve decided you
need to buy this many shares, then how do you go about
buying them so that you don’t move the market and you
achieve good execution prices? And I, I’ll take you
through that in a minute. So now, so let’s get concrete
now and let me take you through some examples.
So execution algorithms. So this is an area where data
science really rules. And, and, and, and I think
execution algorithms were the early part of data science
in, in investment management, because everybody, everybody
likes automated execution. Everybody likes algorithmic
execution. In particular, the regulators feel very
comfortable with it because they can check how, how things
are being done and, and, and we have a very direct way to
monitor our participation in the market. So there’s a slide
there just to talk a little bit about how you can go about
executing a trade throughout a day. So, if you’ve got
1,000 shares to buy, you can do what we call a VIEWAP
execution, a volume weighted average price execution. And
so what you do then is you, you take a little slice
of every bit of volume that goes through,
throughout the day, and you try to complete your
transaction while staying in parallel with the volume
traded. And so VIEWAP is a very safe and, and
no brainier way to execute. But it doesn’t really optimize
any advantage you might have of the information of, of
intra-day behavior. So what do we do? We study intra-day
volume trends. We try to establish whether volumes
are higher in the morning, if there’s seasonality in the
day, and you try to time your flows to take advantage of
that intra-day seasonality. We also look at the order
book information and, and the order book is this, great big list of people who
want to buy and to sell. And so there’s a bunch of
people who want to buy, and each one will, was prepared
to buy at a certain price. And a bunch of people that are
prepared to sell, each one is prepared to sell at a certain
price. Analyzing the dynamics of that order book like how
many people do I have in each side and do I have a lot of
people wanting to sell and not a lot of people
wanting to buy? And other people wanting to buy
very spread out in price? Are they arriving faster
into the order book chat room than the people that
want to sell? All that is very useful information about the
price dynamics of the day, and we use all that as short
term signals for execution. So that execution. Then
a couple of comments on how big data can be used to, for
investment. I mean here, this is just a slide to, to
highlight the fact that when it comes to exotic
data provision, there is a, a dynamic in
the market place where by the first level
processing of all these exotic data is typically
done by a small company and there’s a lot of startups
in that field. So people that will deal with
the processing imagery on shopping mall car
park activity or the shadows of reservoirs of,
of oil or of gas to extract information.
We don’t tend to do that first pass information, we tend to
buy that pre-processed data, but there’s a lot of small
companies doing that and we interact with them all.
And so, you know, a couple examples of how
we can use big data in investment. So suppose I have
the following investment thesis. Large stock moves
are legitimate when backed up by professional participation.
And they are ephemeral, they are bound to disappear if
they are backed up by retail participation, so that’s
the thesis, right? And so I can then assess every time
a stock moves up by a lot or down by a lot, I can ask
myself, is that a legitimate move or is that a move
that will disappear and I should bet against it. Well, one way to evaluate that is
to look at the activity on professional and retail
interfaces. And so we tend to buy that data, we buy it from
technology companies. Some of it we processed ourselves,
some of it we buy and we, we do post processing of,
of that data. But trying to assess whether the retail
has been active entails for example, monitoring a number
of Wikipedia pages, an, and all the mapping that goes
with that. You know, if you are interested in Apple stock,
which Wikipedia page are you likely to look at? Well,
there’s several that will, that will relate
to Apple stock. And so all that data
mapping and, and, and relationships between data
needs to be addressed. And so this is just a chart of an
investment return, you know, the scale is not really
relevant, but this is just showing that, that applying
this investment thesis, using some exotic data to assess the
level of professional activity versus the level of retail
activity does work. That, that produce
a gaining strategy. [COUGH] There’s another
example. Classification and grouping of companies.
You know, there’s lots of phenomena out there that,
that stock markets do display that have to do with groups of
companies. And in particular, certain groups of companies
tend to move together. And if a particular company
deviates from the group, The, the price action tends to be
of mean reversion to the, the, the, the, the group, the pack
again. And traditionally, you look at sector classifications
to group these companies. Sector classifications are
somewhat limited apart from anything, because given
a sector, a company either belongs or doesn’t belong.
And so an innovative way to do sector classification is to do
natural language processing on text-based information,
news, or company filings. And group the companies
according to how they describe themselves or how
they are spoken about. And so again this is a chart, the
scale is not very relevant. But this is just showing
the same signal applied on a traditional company grouping
and an alternative company grouping and you can see that
the performance improves. So okay, so
systematic trading, use of big data,
execution algorithms, enhancing investment signals
using exotic data. The other question that I often get
is artificial intelligence. How is that changing
the landscape? And then again the dream there is that of
autonomous investing, right? Everybody thinks that,
you know, with the learning
algorithms I can throw all the data at
the algorithms and the algorithms will tell
me how to make money. Well unfortunately, the
biggest crime, the biggest sin of, of systematic trading is
over fitting. And a lot of these learning algorithms are
very, very rich models, they, they’re very rich in
parameters and in structure. The financial data is by
contrast quite sparse and limited. And
if you think about, financial stocks there’s maybe
4,000, 5,000 investable stocks to a certain size. And you get
one price point per day in a daily time series, you might
get a tick level times series of prices, but that investment
horizon is quite tight, not, not quite so easy to,
to profit from. And so the idea that you can throw
the data at an algorithm and the algorithm will manage
money for you is still very far. I, I think the, the more
likely scenario is these scenarios that I’ve described
where an investment thesis is derived from economic
observation, and then we look at the data
to enrich the expression of that investment thesis and
to, and to make it work. And then finally, I wanted
to talk to you about ESG investing. And I, I, I think, I hope I can leave you with a
good message on this. I mean, ESG is a big topic, it’s
coming to everybody’s mindset. ESG stands for
Environment Social and Governance considerations in
investment. And more and more people are becoming aware that
how they deploy their monies, which companies they invest
in, sends a message about what they think is sustainable,
what they think is right, what they think is proper
governance. In particular, the UN has published years
ago their global goals. And from the UN’s global
goals the principles for responsible investment came
out and systematically is a signal authority to the UN
PRI. And what do we do in that matter? Several things that
we do. So, for example eh, in the, in the, in,
in the terms of screening. So that’s the first step of ESG
investment is to say, look, I am an investor. I want you to
build a portfolio for me, but I don’t want you to include
stocks that do certain things. I don’t want you to include
tobacco or I don’t want you to include gambling stocks in,
in, in the mix. So then what we can do with
the systematic approach is we can actually rebalance the
portfolio to compensate for those exclusions, and, and we
can perform the exclusions on a systematic basis. Then in
terms of alpha generation. The practice of finding
companies that behave well, that have good governance
should give you some alpha sources. And
then finally, impact, so can you let the companies know
that you are investing more in them because they have good
ESG practices? And so, then, perhaps the most important
slide of this presentation. I was watching some of Margo’s
videos about the conference and she was saying something
that Maria said that, you know, you want to join
the data science community and make a difference, because so
many important decisions are being made on
data driven basis. So there’s $80 trillion of
professionally managed assets in the world. Diversity
plays a big part in that, you need to be at that table.
And I just wanted to quote for you a, a, a statistic that I
found, that in 2015 following the Paris Climate meeting.
The two years that followed the Paris Climate meeting saw
a 55% annualized growth rate in assets deployed under
a sustainable mandate. So now we have a lot of
diversity coming to the fore. Through ESG investment and through disclosure rules from
the, the various regulators. So what is that going to do
to the investment landscape? And then finally,
a couple of fun final slides. Something to say in this
forum. If you’re contemplating a career in
investment management, if you like data science,
exotic data. And looking for investment management,
it would appear that the data supports a good
track record for women. So the, the,
the database providers that do analysis on female run funds
report an outperformance. I have to say this,
the sample is quite small and the time scale is short but,
but it’s a positive->>[LAUGH]>>But it’s also positive message any way. And
then finally just to close, I just want you to go home,
remembering that systematic investment management is data
science applied to investment. Think of Warren Buffett as
a data scientist from now on, and this approach is at least
on a par with the human approach, but perhaps more
scalable and more disciplined. There’s a large element
of randomness in markets, relatively sparse data, so
learning algorithms have limited use. You have
to really watch out for overfitting.
Overfitting is a big risk. $80+ trillion of managed
assets globally. So this bit of data science
has a lot of power. If you want to change the
world you’ve got $80 trillion there to change the world
with and ESG investing is the future. So let’s try to
join forces in shaping it, and that’s what I had.
>>[APPLAUSE]>>We have just two minutes or so for some questions. So while you’re thinking
about a question let me start it off. First of all, I
just wanted to say I love that statistic on women
outperforming men.>>[LAUGH]>>Because it just shows you how fudgeable data is, right?
>>[LAUGH]>>You can always put it in the way that benefits
you the most.>>That’s true.>>Have you thought about high frequency trading or do you
have any, any thought on that, it’s very popular right now?
>>Yes, so, so we do high frequency trading in the
context of executing trades. So, so, so, if you decide us
to buy agent shares going to market entails looking
at tick level data. We’ve had some success
of neural nets in that application but limited so
far. The problem with high frequency trading is the
capacity. So $80 trillion is a very large amount of money.
I mean, institutional investors have very large
tickets and the horizon for investment needs to be longer.
>>Right, makes sense. Okay, question, yeah?
>>Thank you for talk, I really enjoyed it.
I have a question. Do you feel like the business
insight or financial background are prerequisite
in your area or you feel the data science part is more
of a heavy lifting compared to the business insight itself?
>>I don’t see you, can you raise your hand?
>>Up here. [LAUGH]
>>There, good. So look I’m an engineer
by background. I was never particularly interested
in financial markets. You know, I’ll be honest about
it. I was interested in, I was an academic. You know,
I, I, I had my position and I was teaching away and
doing research. I, I think, you know,
as part of the job, of course, you do become aware, but, but,
you know, you, you as a good professional you want to
understand the environment that you’re in. And so I’ve
done plenty of studying of, you know, macroeconomics and,
and, and, and news, and, and, and other bits that
affect markets. But in answer to your question,
if I’m anything to go by, then no, you know, I think
a passion for data science and an interest in making
a difference. Again, I think, you know, there’s something
to think about is this, look, you know, if you
want to change the world, bank your money in
the right places and if you think about investment
management as this activity whereby the pools of capital
of the world get directed, gosh, that is so powerful. And
if, if that is going to become completely data driven over
time, then you can’t miss that opportunity. You’ve got
to join in and, and, and have your say.
>>Grateful, thanks very much, Lila.

Leave a Reply

Your email address will not be published. Required fields are marked *