Fintech and Decision-Making


– Before I introduce
this final panel, I just want to
pause for a moment and offer a thank you,
which I haven’t done yet, to the hard work
of the Planning Committee. Their names are listed
in your program. The success of this conference
would not have been possible without their hard work. I’m deeply indebted to
their support of this effort and also our Event Team
for their assistance with the logistics. And I want to thank all of you
for your time and valuable insights
and continuous engagement throughout the day. It’s just fantastic
to have you here. It’s been a long day,
I know. So, here we are. Our final and last panel
of the day, our third research panel. At this panel, we’re going to
look at the potential benefits and unintended consequences
of using data and technology to inform financial
decision-making. So, how financial institutions,
consumers, investors, use and react to
new technologies, particularly
artificial intelligence and data aggregation, which has raised hopes
and concerns as we’ve already heard
throughout this conference for inclusion,
wealth accumulation, and financial well-being. Our chair for this discussion
is Penny Crosman, who is no stranger
to this topic. Penny is editor-at-large
at American Banker. So, whether you’ve
read her articles or listened to her podcasts, her deep knowledge of FinTech
and its role in banking becomes immediately apparent from her insightful questions
and analysis. So, thank you, Penny, so much
for adding your experience to this discussion,
and I’ll turn it over to you. – Thank you so much.
It’s great to be here. Thanks so much
for inviting me. These are really
interesting topics. I think especially the
research papers on this panel are sort of testing
some of the ideals and the myths around FinTech. You know, does it really
do all of the good that people think it should do,
and you know, the answer is always going to be
kind of, I think, you know,
in some ways, yes, and in some ways there may be
some room for improvement. So, I’m going to start with
just a little bit of a quick look at a few of
the trends in FinTech that these three researchers
have kind of dived into. So, the first one is the rise
of robo-advising. We saw the birth
of robo-advisors like for Betterment
and Wealthfront a few years ago, and I was actually surprised
last night to check the numbers to see that it’s currently, about 8.2 million Americans
use robo-advisors, and that was a little lower
than I thought. Betterment has about
400,000 users, and Wealthfront has
about 270,000 users. I did think those numbers
would be higher just based on how long
we’ve all been writing about and thinking about these things and how long these companies
have been around, but I think sometimes
these things take sort of a hockey stick
trajectory. Like digital banking, I think,
started off slowly and then kind of shot up. And according to a survey
conducted by Schwab, 58% of Americans expect to
use a robo-advisor by 2025. So, it could kind of escalate
like that. There have been concerns about
robots that occasionally arise. I did a podcast with John Taft,
the vice-chairman at Baird a little while ago,
and he talked about the fact that there’s been a huge kind of
movement of investing toward
passive funds rather than
actively managed funds over the past several years, and a lot of this has been
during a bull market. So, he worries about what
happens if we really hit a bear market. You know, could this compound
some sort of a meltdown? Could it lead to deeper
declines, faster declines? Could there be some sort of
scary scenario? Another red flag for
robo-advisors came up last week when it was reported that SoFi
sold customer shares out of their existing ETFs
and into a couple of ETFs that Schwab itself owns, and ostensibly
this was for tax benefits. But apparently the customers
were not asked about this, and they weren’t really told
until five days later, so possible policy concern
for the future and just something
that makes you think is, is this really in people’s
best interest. I personally
have always wondered, are robo-advisors
the right thing for everyone, and one of our researchers,
Nagpurnanand Prabhala, has looked into
that question pretty deeply and looked at a set of ETFs and how people’s
portfolios performed who use those ETFs versus
people who don’t use ETFs. So, we will hear
more about that which will be
really interesting. A second trend is the rise
of digital banking, and for incumbent banks
as well as FinTechs, you know, it’s no surprise
to anybody here, Bank of America now has
27.1 million regular digital banking users, and there are 6 million people
that use Erica, its AI-based
virtual assistant, and a lot of banks
also have been developing and rolling out very similar
AI-based assistants, giving you advice
and recommendations through an artificial
intelligence engine that reads and analyzes all of your current
and past transaction behavior. Wells Fargo has
23 million mobile users. Chase now has around
33 million, and among FinTechs we’re seeing
an increase in adoption also. Twenty million people use Mint. Moneyline has 4 million users. Chime has
close to 4 million users. Capital has 420,000 users. So, I know the banks used
to feel like oh, the FinTech, you know,
banking providers are not significant yet,
they don’t have market share, but they’re actually
getting there. So, that’s been kind of
an interesting thing, and we haven’t seen a lot of
dangers, I don’t think. But Bruce Carlin, has looked at
the question of, does the use of these apps
that help you monitor your financial
transactions and accounts, does that lead to
financial well-being which I think is a really
interesting question. He’s going to give us
some answers on that. A third trend is the increase
of money into venture capital. That’s obviously been
a big topic today. In 2018 US FinTechs raised
almost $12 billion from venture capital firms, and that’s an increase of
120% from the year before, but it’s notoriously difficult
for female start-ups founders to get the seed money. Last year, 2.2% of VC money, I think globally,
went to female founders. And the year before the number
was exactly the same, 2.2%. And, you know, we are going to
talk a little bit more about this later, but 91% of decisionmakers
at VC funds are female. So, that’s part of
the explanation for why this happens. There are many theories
about the extremely low rate of female-founder funding, but some of the theories
are around, you know, VC’s having
kind of a group think, you know,
of wanting to invest in people that look like them
or people they can relate to, and there are also theories
around women preferring to ask for loans versus pitching to
venture capitalists and there are a number
of other factors. But we do have Harvard Professor
Ramana Nanda, who has analyzed a piece of this
where he’s looked at things like product review sites and the way that venture capital
people are affected by things like
product reviews which tend to come more from– not only come more from men but also tend to
favor products– The VCs tend to gravitate
toward products that are popular with men
rather than women. So, he’s going to dive
into that aspect of it. So, hopefully, that sort of
sets the stage a little bit for these research papers
we’re going to hear that are all really interesting. Let me just briefly introduce
our three panelists. So, we have
Nagpurnanand Prabhala, who is professor of finance at Johns Hopkins
Carey Business School. His studies in the FinTech space
have looked at robo-advising in wealth management industry
and the use of soft information such as friendship networks
and peer-to-peer lending. We have Bruce Carlin,
who is a professor at UCLA, and a director of the
Financial Research Association. His work includes research
on consumer behavior in retail financial markets, the clarity of disclosures made
by financial institutions and consumer financial literacy, so all interesting
policy-touching areas. And Ramana Nanda is
the Sarofim-Rock professor of business administration, and codirector of
the Private Capital Project at Harvard Business School. His research examines
financing frictions facing new ventures. He aims to help entrepreneurs
with fundraising and to shed light
on how stakeholders can improve the odds
of selecting and commercializing on the most promising ideas
and technologies. And for our real person in
the trenches giving her take, her perspective on all of this, we’re fortunate to have
Ashley Nagle Eknaian, who is senior vice-president of
Eastern Bank where she is the chief
digital strategist and head of Eastern Labs. Eastern Labs is one of those
somewhat smaller banks that has done some really
innovative stuff. They have… you know, they
have their own innovation lab, they created a software
for making automated loans to small businesses in very short
timeframes, within minutes, and they actually spun out
that group into another company
called Numerated. But Eastern Labs continues
to innovate and create its own technology. Ashley is also the
Program Development Chair for Brandeis University’s
Masters in Digital Innovation for FinTech Program, which is the first program
of its kind in the country. So, with that I’m going to
turn things over to Nagpurnanand Prabhala. – Thank you. – And just a word,
for this panel, we’re going to have each person
give their research and then we’re going to have
a little bit of a discussion. I’m going to ask maybe
one follow-up question or two, with the researcher, and then have a little bit
of a discussion with Ashley about
how it resonates with her, and if during the course
of the discussion, one of you has a question please
come to one of the podiums and we will try to watch out
for you and call on you. Thank you. – Thank you very much to
the organizers for inviting us
to speak on this. So, I’m going to be talking
about robo-advising and talk a little bit
about the kind of robo-advising that we are concerned with
in this study. Then I’ll discuss something
about our own study, what we actually find, and then I will
get to the takeaway.. So, as background, you know,
household financial planning is a complicated
sort of creature as even the previous panel,
we had this discussion. There’re lots of things
that people need to worry about and the previous panel
has mentioned buying secondhand Ikea sofas. But there are many other things
over here that go on– how much cash you want to keep,
how you plan your taxes, what debt you take,
your credit card managing and lots of other things. The piece that we are
looking at is very small, one part of it,
which is wealth management, and even in that,
there are like a bunch of different pieces to it, which is investing
your stock portfolios, investing for retirement. We are just looking at
one piece of it and as we’ll, sort of,
see very shortly. One problem
with all of this is, not only is the problem
complicated, but households are not known to
be financially sophisticated. They make all kinds of mistakes and financial literacy
is kind of weak. It’s very interesting
that financial literacy is about as weak in the US
as in any emerging markets and a number of studies
have been done on these things. So, there’s a lot of room
for advice as a result, and advice hopefully
gets people to do better things. There is lots of room for advice
on any one of these pieces; you know, whether insurance,
what kind of insurance you buy, what mortgages to take, everywhere,
there’s room for advice. And the industry is right now, I would say it has
three characteristics. One is,
it’s human capital intensive; it’s mostly done through people. It’s very fragmented, and the last part of it
is about wealth management. Most of it is targeted at
people with substantial assets already under– large AUM. And the reason is that
because it is human intensive, you’ve got to pay the advisors
a lot of money. So, they look only at
wealthy populations. So, the answer to that
is robo-advising and a bunch of these
robo-advisors have cropped up over the last few years, and I’m not going to go through
each one of them. But basically in each panel
you’ll see that there’s a B2B component
and B2C component, and so one part of advising is you actually advise
the investors on how they can do better. And the second is
you create tools for the advisors themselves, and something again
that came up in the last panel, and it’s all over the place, all of these entities
that cropped up. So, what are we looking at
in our study? We are looking at
one piece of it, which is direct advisor,
B2C advisor, which is administered to
retail financial investors. The robo-advisor
we are looking at has the same information
as the human advisors. In fact in the brokerage
company that we are studying, all interactions
were through human advisors and they kind of
introduced the robo-advisor as an alternative, as an offering to some
of the investors. And why did they try to do
this type of automation? One is that it’s cheaper. Second, you know,
if you have a cheap tool, it allows you to expand your
reach to many more customers. The other thing is
that the execution– and this was what they stressed
was very simple– if you give advice, (the robo-advice tool
I’ll show you in a minute), you press a button
and it implements whatever advice you want, and all the trades that are
necessary get done at one click. One of the other things
that was mentioned is, that robo-advising
is not subject to any conflicts of interest
that advisors have. This I’m not 100% sure of, but this was
one of the selling points. The main question we
are asking is, does this work, what kind of effects it has,
and what we find is there are some expected effects
and some unexpected effects. And I will just quickly
go through the results. So, I just want to preface
the results by, you know, some survey data on why do
people like robo-advising; and you know, the top reasons that consumers cite are
convenience and simplicity. There are some sort of people
say that they are lower cost; and some sense
that robo-advising is not worse
than human advising. The one thing that you don’t see
in all of this is that robo-advice is capable
of producing better alpha or better returns. You don’t see that anywhere
at all. We find that consumers,
you know, don’t believe that robo-advisors
can give them anything more. So, in the study
that we are looking at, we are looking
at small investors who are not financially savvy. These people can make
systematic mistakes in their investments, some of which
we will talk about shortly. In this, the financial advising
is expensive, and they are potentially biased
because they are hired by the same brokerage company
that gives robo-advice. And robo-advising,
how can it help? One is that it’s
cheap and easy to use, and advisors are very costly, and long phone calls
before you do anything, when you talk to clients, and after you have
all of these long phone calls, they may or may not do
what you tell them to do. And even if they do
or do not do, they end up blaming you
for everything, the advisor. This is the kind of setting
in which robo-advice was administered to the client. So, this was a full-service
brokerage house in India. It has about
a million clients, and it’s a pretty big number,
many of these are active. In around July 2015, they introduced
a portfolio optimizer tool. So, what they were looking at
in their setting was that a lot of investors
were holding just one stock or two stocks,
a small number of stocks, and they were telling
the investors you are better off,
diversifying. And they, sort of,
gave the diversification tool. It’s a standard mean
variance optimization tool, the sort that’s used in
the U.S. with some shrinkage. It tells you
what is your best portfolio given a set of questions you
answer about your preferences, and if you choose to take up whatever
the portfolio advisor says, you press a button
and the trades are implemented. It also allows you
to experiment. You don’t need to
take up exactly what the robo-advisor tells you; you can play around with it
and figure out what other portfolios
you would like to have, and whatever you choose, in the end you just put in
your weights for your portfolio, click one button,
and the trades get done. So, this is kind of
what they put up and this is
very familiar to professors who teach finance classes. It’s a mean-variance frontier
and each portfolio is located somewhere inside it. So, they tell you where
your current portfolio is, where your proposed
portfolio will go to, then it’s up to you
to do what you want. So, what do we find? So, the first thing is that–
what surprised us is, this considerable amount of
take-up of robo-advising. People who are offered,
used it, experimented with it, and changed their portfolios
in response to that. And this was
a little surprising, because other evidence
before this shows that people are resistant
to robo-advice; people don’t use it as much. And as we just heard
even in the U.S., it’s not clear whether
8 million is a big number or a small number. But here it was used up and many of the small investors
who were undiversified actually end up taking up
diversification advice which is an interesting point
because many people believe that being undiversified may be an optimal choice of
investors. Well if it is, then if you
give them advice to diversify, they should continue to stick
where they are; they don’t. The most-diversified investors
sort of bring down the number of stocks
in our study. All investors have
the expected results; if you have more stocks
your volatility comes down, and it comes down substantially
for those with a low number of initial stocks. No surprises there. Investment returns
improve a little bit for the less-diversified
investors, and the consistent message
that you see out here is that the least-diversified
investors who had the worst positions actually benefit more
from robo-advising. Or let me put it another way,
they actually like to use, they don’t mind using
the robo-advice; they don’t mind getting
diversified in the process. What is it in for
the brokerage firm? So, initially you look at it,
you find that the fees that people pay go up
because they trade more and when you break up
all of these results. You have least sophisticated
investors who benefit most
as they’re supposed to. The news there is
not that they do benefit, the news is that they take up
the robo-advice that is offered to them. The more sophisticated
investors don’t benefit very much from
this robo-advice. The interesting story
out here, which led us to this question– so what is it in it
for the sophisticated investors, are they gaining
anything from here? So, we looked at the behavioral
biases that investors display, and we looked at three types
of behavioral biases, but you can think of them
as falling in two categories: Investors don’t make
buy decisions rationally; they make it subrationally. Investors also do sell
decisions subrationally; they sell
when they should not sell and what they should not sell, and they buy things
based on trend chasing. So, we looked at what happens to
all these sorts of biases. Now the experiment here– I just want to be very clear– you’ve given robo-advice,
they buy something, they hold it, and after that
they can do what they want. They can buy more,
they can sell more based on their own choices. It’s not what the algo
tells them from that point on. So, what we find out here is
that the biases of all sorts declined for investors. So, you give them an education
in mean-variance frontier, teach them something about
portfolio optimization and you find that, you know,
all their tendencies to behave sub-optimally
in the buy and sell decisions in the future come down. And this declines not only
for small investors with low number of stocks, but it also declines for
sophisticated investors also. And there’s a little bit of
things that the academics are concerned about. Causal identification–
we kind of identify the treatment effect by looking
at everybody who’s called. Some of them,
happen to have missed calls, some of them
don’t have missed calls, and we tracked
both of these investors and we find that all the effects
that we see in terms of
declining irrationality is concentrated in the people
who use the robo-advisor. So, you know,
where does this leave us? Traditional view
is that everybody knows what they want to do;
that’s the libertarian view. And the other view is
the Dick Thaler view that to give them a nudge
towards rationality; it works. And notably, by changing
default choices, you make people make sensible
default choices and their inertia
keeps them healthy. So, what we are suggesting is
that robo-advising, at least in the setting we have,
is somewhat better. It is an alternative; I don’t
want to say it is better. You give people the choice,
educate them, and then see what they do. And the important thing is that
you have easy implementation of whatever choices you make;
it seems to work. The last part of the study,
and this is a new part, is what happens to
human advising when you give robo-advising, and this is something that we
are continuing to investigate. So, some of the advisors say
that we don’t expect the robots to displace human
advisors anytime soon, and this is a view that others
have also expressed. And when you look at
using robot-like advisors in other areas,
there’s some research on when people use and
when people don’t use a robot. People are willing to
trust machines for things like Netflix. What movie should
I watch tonight? Whatever Netflix recommends
seems good to me and people behave like that. Same thing on Amazon
purchases also, and we have some discussion
of that. And– But things where decisions are
supposed to be more subjective; people trust machines less. A driverless car is number one, but people just don’t seem to
trust them that much, despite all the hoopla
that they have received. And the other thing
that you see is, they lose trust in these algos
very quickly for subjective decisions. So, what do we see
in our setting? Actually what we looked at
is we looked at the interactions between humans and clients
in two ways. We have data on calls,
call length, call transcripts, all kinds of data;
and trades after calls. So, this is
the initial evidence. What you find is that the demand
for human advising actually goes up for all those
who take up robo-advising. The top panel up there tells you
on client-initiated calls, they’re more likely to
initiate calls and they spend longer times
talking to advisors, and the lower panel
discusses the length of advisor-initiated calls. When advisors call humans,
their clients, after robo-advising, again,
the demand seems to go up. Their interactions go
much, much longer, they talk much longer, and we
are kind of now trying to flesh out these results
in some more detail to see what exactly
are they talking about, and actually does this translate
to some sort of trading action. When they talk to advisors
this time, are they more likely to use
whatever advice or not? This joint work with
Francesco and Alberto, my colleagues at Maryland. That’s all. [applause] – Thank you. One quick follow-up
for you is, do you think if you had done
this study in the U.S. with some of the robots here which are usually
focused on ETFs, and with the market here
which is generally people in their 30s and 40s
who invest in these; do you think you would
have had very similar results and conclusions? – I’m not sure.
It’s a good question to ask. The thing that I will say is
that the characteristics of the investors and especially
the portfolios they hold are identical for a start, to what individuals
hold out here. – That’s a good indicator. So, Ashley, you know,
as a banker, you know, you’re currently
with Eastern Bank. You were at State Street
before that, what’s your reaction
to the rise of the robots? How do you feel
about the effect it’s having on investors and on
the financial community? – Yeah, so I think–
When I think about the advent of robo-advisors
and this new technology it’s really democratizing–
not to use a buzzword– but democratizing
wealth management, giving people access
to the markets and tools; and a segment
of the population that really wouldn’t have
access to it, it’s pretty cool. I mean, this is what
you want technology to do in the Information Age; you want it to provide access. And I think we are seeing
when the robots first came out like the sky is falling, right. “Robots are
going to take over–” “The robots are coming, they are
going to take over the world; there’s not going to be
any human advisors anymore”. And now what you’re seeing
is sort of hybrid models. So, we go from sort of–
the pendulum swings, right. When they first come out, everyone
is sort of afraid of them. Then people say they could
never replace humans and now you see some
really great hybrid models where you have the robots doing
sort of the asset allocation and then as the research shows,
right, you have humans
sort of coming in for those more
complex questions, those more complex insights. So, I think it’s a really
good meshing of when, you know, technology can really automate
and help those advisors spend time on the questions
that robots can’t solve. Something simple
like asset allocation. I think you mentioned
a couple red flags, too, you know, and you kind of have
to step back to say, okay, well, you know, should everyone
be using robo-advisors just because
you have access to it? Is that really
the best solution for you? And I think as a bank, you know,
you start to think about, huh, well, you know,
and I think some of the VC panelists
said this beforehand, you know, if you’re a FinTech, you kind of come out
with one product; we’re going to do
a robo-advisor. But what’s next on your roadmap. You know, maybe you’re thinking
about account aggregation. Okay, you’re going to plug in
your bank accounts, maybe you’re going to
add some lending, maybe you’re giving people
a dashboard where they’re seeing
their whole financial picture, and you as bank,
you kind of get pushed down on the value chain, right,
and you get more separated from your customers in
the value and that experience that they’re having digitally. So, it’s a
really interesting space and something that we’re
certainly keeping our eye on. – You know, Betterment has said
they want to offer pretty much everything, everything financial a consumer
could possibly want, and also it was interesting,
I thought, that the need for human advisors
went up among the people
using the robot, which is good news
for investors, and probably not so good
for the Wall Street– the companies that think they’re
going to save a lot of money by not having to have a lot of
people offer hands-on advice. So, Bruce,
you’ve been very patient. – Oh.
– Let’s go to your research. – Okay, that sounds good. All right. So, everybody on that side of
the room is thinking about, does technology actually work. Does it help us? And everybody on that side
of the room is sitting there thinking to themselves,
well, if we give people access to more information, is that going to
make their lives better? And so we can
all talk about FinTech, but this is
a very hard problem to tackle because we have to
try to figure out with the growth
of this new technology, are people going to be
better off and make better decisions, okay, and that’s at the heart
of things. Empirically this is a very
hard thing to disentangle because there’s
a lot of moving parts all going on at the same time. Academics love to
call this endogeneity, but to pin down what’s happening
and, you know, whether the technology, and the adoption
of technology is useful, that’s a really hard thing
to do typically. Okay, so what we’re going to do
in this paper is actually use data from a
financial aggregation software and we’re going to show how technology
affects decision making. And, you know,
we are going to show that there’s some pretty
substantial effects on the financial health
of people, but we’re also going to document
some of the mechanisms by which this happens, and how people of different
generations behave, okay. And so, we’re going to use data
from Iceland, and so in that part
of the world, they collect data
on everything you do. Okay? And so, now what’s nice
about Iceland is, it’s a pretty homogeneous place. I mean, if you’ve watched
the World Cup soccer you know this to be the case,
okay. And so, there are
about 300,000 people including children living there. And so, it turns out that about
25% to 30% of the population, (of adult population), uses this financial
aggregator software. So, we can look at how changes
in that technology affect the population at large,
okay, and it turns out that people
who live in Iceland, they almost never use cash. So, we are able to monitor
not only what transactions that they make, the financial
penalties that they pay, their bank accounts and
the logins that they use, and which devises they use. And importantly,
as of November 2014, people had been using
a desktop to log in, and then
they would make decisions. As of November 2014,
all of a sudden, there was an app that was
exogenously introduced, and therefore, we can take
a look at what was the effect of the improved access
to this technology on information, decision making, how you use credit,
etcetera. Okay? And so that’s what we did. Now, this is the financial
aggregation app. It’s similar to Mint.com. Iceland is a little bit ahead in that the majority
of the population does use this, but this is growing
in the U.S. as well. So, just from the raw data,
as of 2014 in November, the frequency of
logging in jumped, and so, there is
a discontinuity where you see that access to information; all of a sudden
people are actually using it, okay, or at least
accessing it. Now what’s also important
about this app is, it only provides you access
to information. It does not provide you
easier transactions. So, any effects that we’re
going to see are going to be based on
having more info as opposed to, easier, you know,
opportunities in the market. Now, one of the things,
when you think about welfare, and I think one of the gentleman
at table two brought this up, is that you have irrational
behavior in the market, you have people with
different utility functions and so forth. One of the things that
we can sort of all agree on, no matter what
your utility function is, is that we don’t like to
pay interest, and we also don’t like to
pay penalties, okay. And so, if we can show
that access to this information actually causes those things
to go down; most people, unless you’ve
got a very odd utility function and you like fees,
your utility is going to go up. Okay?
And so just from the raw data, you can see right around
the introduction of this app, you see that the overdraft
interest fees go down and the late fees
start to go down, okay. And so, exogenously you
introduce this app to people who are already
using the technology and you not only see
higher logins, but now you start to see
penalties going down, and this suggests that there’s
a positive welfare effect, okay. And so, what we do in the paper
is formally, we perform a regression
discontinuity in time design. What that means is,
we look at people, individuals, over ,time and around the introduction
of this app we see how their behavior
changes, okay. And so, we don’t think
that the timing of the app has anything to do with
individual people, whether it was November,
December, January, but, since we can
look at individuals– we know the people who–
they’re logging in, we can actually introduce
characteristics called individual fixed effects to control for all kinds
of other things that might change their behavior around this introduction
of this app. Okay? And so with this in mind,
what we do is, we can now calibrate
the magnitude of this effect. And so what we show is
that every extra login was associated with roughly
$2 fewer penalties paid per month over
a two-year period of time. Now, $2,
may not seem like a lot but this is per login per month
over a 24-month period, and it’s a cross-section
of the population, a lot of people. This turns out to be
quite a lot of money, okay. Now, in terms of how these
penalties are split up, it turns out that a reduction
in overdraft interest accounted for most of
this effect, okay, and that other types of fees,
late fees, and things like that, were less important,
but still important. Now, this led us to look at– because we could see
in the data– how were people
using credit cards, how much were they paying off
their monthly balances. We could see this, and so, what we were able to
show was that the logins, the higher logins were,
in fact, associated with a reduction
in overdraft interest and an increase in
credit card use. And within this setting,
use of the credit card offers a 30- to 50-day float. So, ostensibly what people
were doing was they were more aware of what was in
their accounts, they were borrowing
on their credit cards to pay for things and lowering
their overdraft interest that they were paying,
leading to an overall reduction in financial penalties. So, we view this as
an important result. Now, if we start looking
across generations, this basically shows you–
this figure shows you the uptake of the technology by
different types of people of different generations. So, we divided people
into Millennials, Baby Boomers and Generation X.
Okay? And the Millennials
not surprisingly, took up the technology
the fastest, followed by the Gen Xers, followed by the Baby Boomers. And also of interest, was that men tended to take up
the technology faster than women,
but within category, Millennial women
were quite quick compared to Baby Boomer men,
okay. Now, if we look at the financial
penalties that are paid, the Baby Boomers
and the Gen Xers tended to incur
higher financial penalties than Millennials, okay. And so, they were
much more likely to have overdraft interest; they were more likely to incur
late fees, and NSF fees. Now at the same time,
the older generations– the Baby Boomers and the
Gen Xers– were more liquid. They had higher cash reserves
compared to Millennials. And so there’s a co-holding
puzzle that’s very important in consumer finance
where people have cash, and plenty of cash, but still
incur all of these, you know, late fees and high interest
payments and so forth, when they could use their cash
to just pay those things down. Okay? And so it turns out
in our setting that the older generations
were much more likely to have a co-holding distortion
than Millennials. And what we think
is probably happening, is that the technology
is affecting them differently. So, you know, where does this
kind of lead us? Well first of all, you know,
at least in this setting as we look at, you know,
financial aggregation software, access to more information
does appear to allow people to make better decisions,
and may even change the way they access credit
which is a good thing, okay. And now,
if we start to think forward, how are we going to use this,
you know? Well, it’ll affect people
of different generations differently, okay,
and so, perhaps, you know, we need to devise
different instruments for different generations, and perhaps improve the delivery
for older generations; because many people who’ve grown up with,
you know, reading books and magazines
and things like this, they’re not used to a lot of
these fancy gadgets. They don’t adopt them as quickly and perhaps they don’t use them
as efficiently. And so, introducing FinTech may not necessarily
just be around– let’s get things done faster,
quicker, easier access– and may also be around knowing
the target population and perhaps offering
a different product for people of different
generations and backgrounds. We’ve heard a little bit
of that before. The last thing is that,
you know, I’ve done a lot of work on the complexity
of financial products, and so, complexity
is a big issue. And people have
kind of raised this. I think Prabhala
also raised this as well, where you have, you know,
very complex decisions, and the aggregation
is a good idea, but as long as
everything is salient so people can actually not only
aggregate but distill, I think is gonna be important
for FinTech to have, you know, a big impact. So, that’s it.
– Awesome. Thank you. You know, I’m just sort of
curious– yes… [applause] – I’m kind of curious about
those Baby Boomers who had the higher cash flow but also had the higher
NSF fees. Are these– from a behavioral
psychology point of view, is this people who are kind of living beyond their means
in a way? You know, they have money
but they’re spending it all too quickly? – It’s hard to–
Wait a second, I think people
can still hear me. From the data, even though
it’s really detailed data, it’s unbelievably detailed,
we can’t disentangle that. So, we don’t know why,
but it’s been documented by a lot of previous people,
this co-holding puzzle, where even though
you have the cash, you still take out the debt. Now, you know, from–
in corporate finance, we– people like to think
of doing that because well,
debt is a tax shield and you can save on debt and
that’s a terrific thing to do, but from a personal standpoint
it’s not a great thing to do, so– but it looks like
it is exacerbated in the Baby Boomer population. – Also again, do you feel
that if you did this study in the U.S., you’d probably
see similar themes? – Yeah. I mean one of the things
you think about is, well, with
such a homogenous population like you have in Iceland, is that gonna translate into
the U.S. where we have a much more–
we’ve a melting pot– and of course,
from a research standpoint, having the homogeneous
population’s great. I personally think that this
would translate into the U.S., but the point about
the different generations accessing this differently, I think you would probably
get different people of different backgrounds
accessing this differently. And FinTech directed at
the different backgrounds I think is gonna be a winner. I think to really have the
welfare effects that we want. – Sure, and so, Ashley,
I wanted to ask you, what– in your experience,
are you seeing a connection between people who use
Eastern Bank’s mobile app regularly or a lot, and are those people
able to improve their financial health
as they go? – Yeah. That’s a hard question
to answer because we don’t have
the aggregation capabilities, and we’re not asking customers. You know, we basically
just know balance. However, more broadly
we do offer some financial literacy products
and services to help people kind of understand
the implications of debt and credit
and how things work. And I think a more educated
population where you have– you have the ability to have
these FinTechs come in and really solve a job
to be done, right? Like who likes paying fees,
as Bruce said. If people don’t understand
financial products and they’re getting more
and more complex as Bruce mentioned. So, if I’m a FinTech
and I’m coming in and saying, hey, I’m gonna help you,
right, I’m your friend, I’m a bank with a heart
because all of us bankers are heartless,
not me, but most. So, how do you, you know, how do you start to compete
with that, right, from a product development
perspective when you have a new,
digitally enabled product? You know, you’re educating
the customer, you’re helping them make
better financial decisions, and that’s something
that you care about and, you know, a bank
that’s not caring about that. So, whether or not you believe
that it’s your job to solve financial literacy,
or it’s an industry problem, or it’s a silo problem, which came up
with an earlier panel, I totally agree;
those are big questions. But you have to look at it
from the customer perspective. Is this a problem
that they are looking to solve, and if it is, they are going to
go to the solution provider whether that’s a FinTech
or bank that’s providing it that’s gonna give that
to them. So, I think a huge sort of
question mark for the industry and something that all,
you know, all banks and financial institutions
need to be, you know, looking out for, like does that
solve a customer need? And I think it does. – Sure. Thank you. Ramana, you’ve been
really patient. Ramana’s gonna take
a really interesting look at gender bias
in venture capital. So, please welcome Ramana. – Thank you.
– [applause] – So, thank you all
for inviting me, for sticking along
to the bitter end here. So, my presentation here today
is actually on the face of it neither about FinTech
nor about banking, but I hope that you will see
very clearly the connections between what I’m
gonna be finding, and a lot of themes that we’ve
been talking about today. So, the motivation for this
paper really stems from the fact that consumer reviews are increasingly
being used by financiers, as we’ve seen
in many of the earlier panels as signals that they can use
to make investment decisions. And that could be either
of the peer-to-peer sense or it could be
for FinTech lenders. Certainly, venture capital
which is the context that I have been studying
has begun to kind of think about measures of traction
as early stage indicators of whether or not
they should be making investment decisions, okay. Now the thing to note
is that people who review on these online platforms,
they select in, no one’s forcing them to review,
and they select in and they would typically
not represent the population that they are
supposed to represent in exactly the same proportion
as that population. Okay? So, in the context
that I’m gonna be studying, women are about
10% of the reviewers, but obviously 50%
of the consumers in the broader population. And so as we think about
the ways in which the preferences
of people are aggregated, sort of a naive preference
aggregation where we just say
we’ll just add up all the votes can lead to interesting
systematic biases in the way in which the reviews,
the aggregated reviews, end up being reported
by these platforms. So, obviously,
this is not a new thing. Entrepreneurs will go and pitch
to investors all the time, and investors have
their own preferences. But what’s particular
about online platforms I will argue, is that they
broadcast these signals widely and because of the fact
that these online platforms have these network-effect
properties that make them winner-take-all, the fact
that there is this sort of bias signal being developed
on one of these platforms means that if everyone is using
the same traction metrics to make their
investment decisions, you can have a systematic bias
that creeps in in a way that wouldn’t happen
if everyone was having individual meetings
with different people who had different questions
and sets of preferences. Okay? So, that’s
kind of the motivation here. What’s our setting? Our setting is an online
platform called Product Hunt. This is an important venue
for launching and beta-testing technology-related products. It’s largely used by start-ups but tech companies
also launch their new products, so you will see, you know,
Stripe or LinkedIn, or Facebook and others
put new products onto platform. And as I mentioned, there’s
certainly anecdotal evidence that VCs particularly
at the early stage are looking for indications
of tractions. So, an entrepreneur shows up
and says I was ranked number two,
on Product Hunt the day that I launched, that’s a, you know,
that’s a sign that– that they actually
have a product that was valued by consumers, that showed
that there was traction and is worthy of them
taking the meeting to do further diligence, okay. And so what we are gonna do
is we’re gonna look at products that were launched
on Product Hunt. And then, we’re gonna measure
the follow on VC financing that these start-ups received
within about six months of launching on the platform,
okay. And what we’re gonna do
is examine the degree to which products that
are relatively more appealing to women consumers
are faring relative to those that are more
appealing to male consumers, and how that varies based
on the number of men and women who are showing up
on the platform, okay. So, what’s the data?
We’re gonna use three years of data
for about 40,000 products; so that’s about 250 products
a week, and we’re gonna restrict it
to technology products which are web apps,
mobile apps, hardware. So, we’re gonna be taking, for those of you who know
Product Hunt, we’re gonna be taking books
and podcasts out of the set because of the certain features that they are not
very frequently reviewed. Okay? Now what we have is individual
person-level identifiers, so you need to login to
the website to be able to vote. So, we know who you are
and we know the time of the vote which will allow us
to also get a sense of when you logged on
to the website, and at the moment
we’re gonna be using that to infer which products
you looked at, but didn’t vote on, based on products
that had launched prior to your login
that you were, you know, you could’ve seen and voted on
but you chose not to. Okay? Now we have– there is an interesting feature
of this website. There is a focus on community, people are encouraged
to kind of put their names– and indeed in a large fraction
of the situations we actually have the names
of the people who are engaged,
rather than, you know, RN305 which is obviously
certain accounts which are gonna be
hard to code the gender on, but we have three sets
of constituents. We’re gonna have the voters; so these are the members
of the community who will upvote or not a certain product. We have the makers;
that’s the last bullet point. These are what you could
think of as the entrepreneurs, and then there’s this
middle category that is gonna be important
for our analysis. These are hunters; so in order it be voted on,
on Product Hunt, someone has to hunt
your product. Okay? And we will– and we’re gonna
think of these hunters as sort of early indications
of consumers who value these products, who find them interesting. And by sort of looking at
the gender of the hunters, we’re gonna try to
get a sense before the votes actually happen whether this particular product
would have been more or less appealing
to women consumers. Now obviously
this is noisy, right. There’s gonna be
some male hunters who are gonna hunt products
that would be more appealing to women and vice-versa, but I’m gonna hope
to convince you that there’s some signal here
that’s gonna be useful for our analysis, okay. So, just to give you some
high-level descriptives, you know, as I mentioned
there’s about 40,000 products; about 90% of them were
hunted by male hunters, about 10% were hunted
by females. Both sets of products received
about the same number of views. Both sets of products
were featured at about the same percentage. So, Product Hunt
has the opportunity to curate certain products, and they were, you know,
curating them and featuring them
at the same rate, and obviously we’re gonna be
controlling for that in our analysis. In about half the cases,
the maker, is actually listed
on the website. Now what’s interesting is, conditional on
having the maker listed, you can see that
female-hunted products have a much higher share
of female makers, okay. So, there is some–
already you can begin to see that there is
some homophily here. The kinds of products
that are made by females are perhaps more interesting
to female hunters. But note that 50%
of the products that these females are hunting don’t have
any female entrepreneur. So, there’s a lot of variation
in our data. Males tend to upvote
male-hunted products a little bit more than females, and females tend to upvote
female-hunted products more than males. And that’s perhaps not
surprising if you begin to, you know,
sort of see the fact that the gender
of the hunter is correlated with what these consumers
might find interesting. And that’s
sort of the variation that I’m gonna be
using in my analysis. Okay. So, just a very quick– I’m not gonna get into
the fancy econometrics here, but essentially,
what I wanna do is, I wanna try to control
for the attractiveness of a product by setting
how many, you know, how many people
voted for that product; and I’m gonna study the
fixed harshness of a reviewer by looking at how many times
I tend to upvote, conditional on
the sets of products I see. So, if I’m a particularly
lenient voter, I can control for that because,
remember, I know the identity
of the voter. If I’m a particularly
harsh voter, I can control for that. And what I’m gonna do
is I’m gonna take those sort of fixed effects out,
and what I’m gonna have left is an individual-specific
assessment for a given product, that controls
for my fixed harshness and for the fixed attractiveness
of that product. And I’m gonna
aggregate that up by whether or not the voters are
males, or they are female. And you can see here on average,
males and females are very, very close
in their assessment across all sets of products,
okay. So, these residuals
are essentially, you know, centered around zero. On average were not too harsh
or not too lenient, on average our assessments
are about zero. And the chart on the right
is essentially documenting the sort of, the difference
between the male and the female reviewer scores
on average, okay. But this sort of average
is actually masking a lot of heterogeneity. Okay? So, what you can see here
is that the left bar, which is sort of the all,
the average again, is tiny but you have a very,
very, very large difference that’s showing up
for female-hunted products. And a similar opposite effect
is showing up for male-hunted products. But remember,
because female-hunted products are about 10% of the sample, that very, very large
negative number is being multiplied
by only 10% and that relatively small
positive number is being multiplied by 90% to get the approximately
zero overall effect. Okay? And so this is essentially
the variation that we’re gonna be studying, which is in some ways
what we’re saying is, the female-hunted products are
liked a lot more by females relative to males, and male-hunted products
are liked a little bit more by males relative. Okay? Now, you know, I don’t wanna– one can easily kind of
get into stereotypes here, and I’m gonna
do that for a second. So, these are some–
among the top male relatively most-interesting
products for males, okay. And we have Soylent here
which is, you know– for those of you
who know these tech bros, who are kind of like,
so into doing their tech coding that they don’t even wanna
take a break to have a meal and they’re gonna have,
you know, Soylent. And that’s kind of–
that’s so– so that’s one of the products
that are over here. You can see the kinds of things
they care about are really a lot about technology,
about coding, about developing apps. So, products that are more
appealing to women voters, obviously there are
certain products that are just, you know,
very useful for women consumers, period. But what’s interesting to note
about this product is that this was a product
that was developed by all male entrepreneurs, okay. So, this is not always
going to be about the gender
of the entrepreneur. This is gonna be
about what consumers actually find interesting. And you’ll see here that some
of these are not obvious things that you might have thought, oh, this is going
to be super, you know, interesting
for a woman consumer relative to a male consumer. But it seems
as if the kinds of things that are showing up here
are much more related to travel, related to health
and other features, not as much focus on coding.
Okay? So, I wanted to just emphasize
one quick thing which is that this difference
that we’re finding actually shows up
even when you restrict it to only male makers, okay. So, this is not just about
the gender of the entrepreneur; this is something
about the consumers as proxied by whether this is
hunted by male or female, that really seems to be
showing up as we think about
the differences and the preferences across
these voters. Okay? So, I don’t have time today
to kind of show you these follow-on effects, but what we find
is that on days when there is a smaller share
of women on the platform, female-hunted products
are gonna get fewer upvotes. And then we also find on days
when there is a smaller share of women
on the platform, start-ups with products
that were hunted by women are less likely
to subsequently raise VC six months– within six months
of being featured. Okay? And there’s a whole set of
questions around is there reverse causality, is there admitted variable bias,
and so on and so forth. Where this is at–
still at an early stage, so I’m not gonna tell you
conclusively that I have causal effects here, but I certainly
find the patterns to be quite interesting and
we’re kind of working through those findings right now. Okay. So, the summary
of the findings so far that I’ve presented are that there are
the strong gender differences in the types of
technology products that are appealing to reviewers
on Product Hunt. These differences appear
linked to the consumers rather than just to
the gender of the entrepreneurs. And the aggregation
of preferences appears to lead to these,
you know, biases, where products that are
hunted by women are doing less well on days
when there’s fewer women voters who are showing up on
the platform. Okay? Now, putting it all together
I sort of want to leave you with three broad thoughts here,
okay. So, the first is, we have
relatively few gatekeepers who are responsible
for kind of financing and commercializing new ideas
and technologies and they’re going to be
using signals to make their investment
decisions, and those signals are going to
be obviously, you know, noisy signals. But it is important to realize,
that, you know, you can have bias that arises– the signals that they’re using
can have bias in them purely from preference
aggregation without any discrimination
or malicious intent, and the kinds of preference
aggregation here where we’re talking about
just adding up uploads, is a very common way in which
a lot of rankings happen whether it’s on, you know,
Yelp, or Reddit, or other places like that,
and so we kind of have to think, the more we start using
online platforms, the more we want to start
thinking a little bit carefully about how we’re aggregating
these preferences of reviews. Okay? I think it provides
a little bit of a nuance to perspective
on online platforms. They obviously democratize
access as we’ve talked about, but also can impact
who gets funded and hence, kind of the direction
of innovation. And then finally, you know, we talk a lot about the gender
of the entrepreneurs who are getting,
or not getting financed, and obviously
that is a big concern, but it’s sort of,
worth thinking about the fact that ultimately,
if there’s some homophily between entrepreneurs
and consumers; if women entrepreneurs
are bringing products to the market that are
important for women consumers, then the consequences
that we have of this imbalance in terms of who’s getting funded is even bigger than just
the entrepreneurs themselves. And so obviously that
has sort of much broader and deeper welfare
consequences. Okay, so with that,
thank you very much. [applause] – Do you see the production
review sites making any kind of effort to recruit more female reviewers
or product hunters? – So, actually, no. The folks who run this, well, and I think one of the things
that we are going to be working through together is to try to understand
how best we can, you know, think about what this means
for the community, because, you know, you
don’t want to force people to come on to the platform. People come on
based on their interests. That’s one of the nice things
about these platforms, and it’s not clear– I think
as we were talking before– it’s not clear
that you want to even have a different aggregation rule,
right, because one of the things
that is nice about this is that it’s simple
and it’s transparent and you kind of say, look, we understand that there’s
going to be noise in this process; buyer beware, let’s make you
aware of these things. But if we take
a different approach to say we are going to
try to tell you the truth, we are going to try to deal
with this set of biases, well, what about other biases that have not been
factored in here; what about– you know,
how do we think about those. And the more we start trying to
turn it into a black box, perhaps the less good
this is doing to the users. So, I think it’s
an interesting question. I don’t have the answers, but it’s something
that I’m hoping to kind of work through with the team
to think about how– you know, to understand
how they’re thinking about it and perhaps examine
different ways to deal with it. – All right. I’m going to ask Ashley
a question or two, but then we’re going to go
to the audience for questions in just a minute. Ashley, I know you work
with FinTech entrepreneurs. I don’t know if you work
with any that have products designed for women. I sort of doubt it
because there aren’t that many financial products
designed for women, but what are you seeing
in this whole space in terms of products, you know,
FinTechs, tech companies with products for women
getting more funding and also this sort of
imbalance of funding going toward male entrepreneurs
versus female? – Yes. So, I mean I think
the 2.2% funding stat just makes me sad, right. Like, there’s obviously
something going on, and when you look at it broadly,
you try to think about, okay, well the stats
you mentioned, right, 91% of VCE firms are male; there’s got to be
some group thing, there’s got to be some bias towards people
that look like you. And you think,
okay, well, you know, they can hire more women, they can put more women
on boards that they have, they can install more CEOs
that are women, and those are all good things. Then you sort of see this idea
with Product Hunt, and you say, ah, technology,
that’s the answer right? Like let’s let consumers
choose, right. You think that would be a
more fair, fairer approach. And I think what Ramana’s
research has shown is that, you know, technology itself
is unbiased, but when have people putting
the models together, when you have the aggregation
capabilities, a human is behind that. So, you really do need to
think through those sort of unintended consequences,
or the buyer beware to say, okay, this is what, you know, aggregating in this way
is going to do to that data. And I think that sort of
filters through whether you’re talking about,
you know, funding for female entrepreneurs or just financial products
in general. When you talk about models,
you can kind of take this and sort of extrapolate it
and say okay, well, any type of model
that you’re doing, there’s humans or
programmers behind that, that they’re putting their
own unconscious bias behind, and how do you look for that,
how do you test for that, how do you monitor it, how do you let people know
and make that transparent? You don’t want things
in a black box, but how do you bring
that transparency to light and how do you make sure
that your outcomes are what is intended, and you’re not throwing
off some weird signal, right. Does it end up that
you are no longer you know,
funding people most in need? You know, are you discriminating
against minorities? You know, are you not giving
mortgages to Millennials? You know, you have to sort of
think about all these things, especially in terms of,
you know, financial products and services
getting more complex and getting more
“AI-powered”. So, I don’t have stats
in terms of, like, women with financial products. I do know a couple founders,
and you know, it is a struggle, right, even though
they’re rock stars. But I think generally, you know,
you see Ramana’s research really kind of extrapolating
towards the use of models and how do we deal with this
as an industry, and how do regulators
deal with it, and how do you make sure
that the customer isn’t harmed in a way. – Look beneath the surface
a little bit at some of these numbers. Any questions
from the audience? All right, I guess you guys
have covered it all. Yes, it has been
a very cerebral day, a very long day. Well, I’ll just wrap this up
with a few final questions, and if anybody has a question,
come on up. But for you,
and I’ll use your last name because I’ll probably butcher
your first name, Prabhala. What would you like to see
when you look at the robo-advising space
looking forward? Are there any changes
you would like to see the providers of robo-advisors
make perhaps, in terms of being more clear
about who would best benefit from their product? Or anything,
is there anything you would like to see them
do differently? – So, the one part is
transparency of what goes on
in the black box, and it’s kind of
a subtle question because when you work
with human advisors also, you don’t know how exactly
they arrive at their decisions. And so, but the same thing
arises in robo-advising because there are obvious
conflicts of interest in the way humans
are compensated and the question is, are these robots
conflict free or not. So, some degree of transparency
is probably necessary, but I don’t know
what or how to get it. – That makes sense. And for you, Bruce,
what, if anything, would you like to see
providers of these kinds of apps that provide account access
and account information, bank account
information to consumers to kind of resolve some of the
things you talked about about the people who ended up
paying higher fees and such? – I mean, I think the name of
the game, is simplicity. I mean people are daunted
by financial products– borrowing, mortgaging,
investing, and even that screenshot
that I showed you of all their accounts
all coming into one screen, that’s nice, but the fact is,
is that people I thought that one of the
findings in Prabhala’s research that I found to be
most like invigorating was the fact that the people
who were participating with the robots, this was increasing their search
for advice. And so in some way, the robot
was making them feel like they could participate
in a reduced form or simple way and then they learned,
and they got better, and they could then look
for advice and participate. And I think a lot of people
are reticent to participate in financial markets,
maybe for good reason, but I think as we make things
less complicated so that they can access them, I think that’s going to lead
to better decisions. – That makes sense, and Ramana,
what would you like to say to venture capitalists? I assume you’d like them
to see your research, and what would you like them
to think about and maybe take away from that? – You know, I think the thing
that’s sort of interesting to factor in for VCs is a little bit what Ashley was
talking about in terms of we have preferences
amongst ourselves and there’s information, and that the information
then comes with a little bit of a preference. And trying to think about
how your own personal preference is different from
the informational content that is there in the signal, and being
just a bit more aware of that I think is probably
a helpful thing for all of us as we rely more and more
on kind of just one signal that is in a agglomerated index of all kinds of things
that are under the hood and kind of
going with the same line around transparency
and simplicity. There is sometimes
a tradeoff here, right, which is the Product Hunt
website has a very, very simple
aggregation rule; they just add up the upvotes, and it’s also pretty
transparent. But on the other hand,
if you’re not aware about it, it’s perhaps
not the best kind of close-to-the-truth signal. But any attempts that they
might make to try to fix that will always leave
some residual amount of, you know, corrupt signal there, and so how do we think about
that tradeoff between simplicity
and kind of truth is a more metaquestion
that I think we should all be thinking about
as we rely more and more on this hard information
that’s coming out from these platforms. – Thank you. Ashley, any last thought
about all of these technologies and how they’re
changing the way people make their decisions? I think we talked about
a lot of the good and some not so good. You know, I know it would be
hard to sum it all up in one sentence, but, how do you feel
the overall impact of this and how that’s affecting
people like your customers? – I think the overall impact
is good, right. You’re giving people
more options, right. We’re making sure that we’re
giving them digital experiences in financial services
that they are used to seeing in sort of their personal lives,
right, with other sorts of tools
that they use. So, it’s more normalized
and I think, you know, there’s a lot of
opportunity here for banks to do some really,
really cool stuff and take advantage of this. And all of the research
that I think we’ve heard today is sort of teasing that out,
right. So, it’s just a really
exciting time, I think, to be in financial services. – All right, and Jason Henrichs
has a question. – [indistinct] Technology, you know,
at least starts amoral, but it can develop its own
discrimination tendencies, you know, over time
as you’re both all discussing. I’m curious
what you think about this, when trying to drive outcomes,
there’s a level of morality, you know,
that comes into this in terms of what is the good and the welfare
for the individual. That came up in one
of the prior panels in the question around payday
lending in that: I’m curious what the panelists
think about that. Who sets what that is
and how much do we actually give that determinant
to the individual because that also– giving them
tools is one thing, asking them to make other
choices is another thing. But it’s also a slippery slope
where they can be their own worst enemy sometimes. – Yeah, great question. – So, the man with the tie
asks a good question. The last time I wore a tie
was when I was in D.C. giving a talk three years ago,
so I will tell you that. All right.
Well, you’re asking a question that is incredibly hard, and the thing is, whose welfare
should we be maximizing? And so academics have
thought a lot about this because, you know, you think
about maximizing the welfare of the rational person who makes
the right decisions. But then there could be somebody
with a different like, utility, for things in life,
and maybe, they like other things. And so I think that, you know,
we really can’t take a position, where we’re, you know,
imposing, you know, a certain value system
on everybody. But there are certain things
that we can do. So, like for example, one of the things
that I referred to in my talk was that we were trying
to minimize their transactions costs,
or minimize their fees. So, that would increase
their capacity to purchase or save
in other ways. So, I think that as policymakers
we can look at ways in which we can affect
most people in the best way, but I don’t think we’re
ever going to have that optimal
welfare consideration because we can’t impose utility
functions on people. – I agree with that,
but I think it kind of interacts in sort of
interesting ways with the incentives of
the players involved. So, I was at conference at
HBS recently on AI, and there was someone
from the FinTech industry who runs
a fraud prevention company. And they have to build models
to determine whether there’s
credit card fraud in a certain transaction,
okay. And the way in which they win
contracts is, that they are given
a test dataset by the banks and that same test dataset
is given to five people who bid for the contract. They all run models
and the model that’s “the best”, in terms of predicting fraud
will be the one that will win. Okay? But going back to
all these questions around statistical
discrimination that we talked about before, it is very easy to not
overtly bring in certain factors that would,
you know, be illegal. But you can easily
fall into the trap of engaging in statistical
discrimination not even for any,
you know, obvious reason. You’re just– you’re kind of
data mining and it turns out that certain things
turn out to be valuable. So, that’s a slippery slope, and he talked a lot
about being transparent, around what do you have
in your models. And a lot of our AI models
now as we move from, you know, random forest
to deep learning and neural networks, it’s really very hard to know exactly what kinds of
correlations these models are picking up. And so there is
actually a very hard time in trying to figure out
exactly the “morality” behind these decisions. So, I think it’s a complex
and important issue not one that we can solve, certainly not at 5:30,
at end of the conference. But a good question, anyway. – Yeah.
I mean, do we have time? – We have time?
– Do we? Okay. One more? One more. Okay, a quick one. – Quick comment
on what Bruce said, and this is
a very important point. We can try our hardest
to do all the things to improve service
and reduce fees for customers, but there are some things
that we don’t control and that’s the irrational part. And I’ll give you case
in point, very simple. In the last recession, when the level of
money anxiety was high, consumers, depositors,
shifted money from their CDs, to liquid accounts. And they lost billions of
dollars in interest, in yield, because of the shift. So, the shift was irrational, but they felt that they needed
the money to be more accessible to them which is a fallacy. But that’s how they felt. So, with
all the good intentions and with all the–
really the attempt to help banking customers,
at certain conditions, they will make
irrational decisions and still lose money. That’s a good point
that Bruce brought up. – There’s always someone
to kick you. – Thank you.
[chuckles] – All right. Well, thank you
so much to the panelists. This was great research,
and thank you, Ashley, for all the feedback. [applause] – Okay.
Thank you to this panel. And one final thanks
to all our speakers and panelists today, the Planning Committee,
the Event Team, and for you guys
for sticking it out with us for such a long day. We really appreciate
the engagement that you all provided. Hope you enjoyed the conference,
and I would invite you for a brief reception
in our lobby now. Thank you again. [applause]

Leave a Reply

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