Subscribe
Share
Share
Embed
Where is consumer finance headed? Dr. Nikhil Paradkar, assistant professor in finance at the Terry College of Business at the University of Georgia, uses data and machine learning to study how changes in the financial sector impact consumers’ credit availability. Learn what he discovered in this interview with Jeff Dugger, principal data scientist and university research director at Equifax.
A podcast where innovative business leaders discuss data: how to think about, how to use it and how it can help us all make better business decisions every day. As they tell their stories of trials and triumphs, you’ll gain key insights to leverage in your own day-to-day operations.
Jeff Dugger (00:08):
Welcome to another episode of Data Dialogues. I'm Jeff Dugger, your host today. I'm excited to be joined by Dr. Nikhil Paradkar, assistant professor in finance at the Terry College of Business at the University of Georgia. Today, we're going to talk with Nikhil about some consumer finance research that may help indicate where consumer finance is headed. He has also conducted some very interesting research on corporate buzzwords, innovation and company earnings calls. I can't wait to dive into that. Welcome Nikhil and thanks for joining us.
Nikhil Paradkar(00:39):
Thanks for having me.
Jeff Dugger (00:42):
Nikhil, your focus area at UGA’s Terry college of business spans financial intermediation, FinTech, household finance, and machine learning. Can you tell us a little bit more about your work?
Nikhil Paradkar (00:52):
Yes, of course. Very broadly, my research interests lie at the intersection of banking and household finance. Now, specifically I investigate how changes in the financial sector owing either to regulation, technological advancements, or crises impact the availability of credit to households. More recently, I've also explored applying machine learning techniques to finance research.
Jeff Dugger (01:17):
So Nikhil you mentioned regulation. Let's talk about some of your research you recently presented to the consumer financial protection bureau or CFPB on bank funding shocks. Tell us a little bit more about that if you would.
Nikhil Paradkar (01:31):
Yes. I presented this study to the CFPB in December, 2019. So in this study, my co-authors and I explore how banks respond to liquidity shocks by adjusting their supply of credit and consumer credit card markets. Now specifically about a decade ago, uninsured short-term wholesale funding was an extremely important source of funding for banks. However, this funding source just suddenly dried up in 2008. This dry up actually ended up being quite a serious shock to bank liquidity, mainly because banks were unable to substitute away from short-term wholesale funding. So specifically what we found is that banks ended up responding to this liquidity shock by cutting back on credit limits issued on consumer credit cards.
Nikhil Paradkar (02:37):
Now, interestingly, the transmission of the shock to consumers was not uniform. The largest credit limit cuts in percentage terms were levied on consumers that had relatively poor credit fundamentals. Basically individuals with lower credit scores or higher credit utilization. Importantly, these larger cuts also translated into reduced credit card balances. For example, if you take consumers with very high credit scores or very low utilization ratios before this liquidity shock, they really did not reduce their credit card borrowing when they experienced a reduction in credit limits. However, if you took consumers with low credit scores and high utilization ratios before the shock, these people experienced a large reduction in credit limits. And for these people, every dollar reduction in credit card limits translated up to almost a 71 cent decline in credit card borrowing.
Jeff Dugger (03:40):
One thing I would imagine is that this could have an impact on a person's credit score if their credit limit is dropped. Did you see that?
Nikhil Paradkar (03:50):
So that's actually a hypothesis that we were testing. We did expect this to have a negative impact on consumers’ credit scores. A consumer's credit utilization ratio plays a significant role in determining credit scores. But for the people that experienced large reductions in credit limits, they also reduced their balances. And as a result, their utilization ratio remained unchanged and the effect of their credit score was generally very small. If anything, the people that experienced that reduction in credit scores were people that had extremely high credit scores before as well. They experienced a reduction of one to two points, but for them, that reduction did not matter. However, an interesting thing we did find is that for the people that experienced a negative effect in terms of credit limits and subsequently their credit card balances, was that these negative effects appear to persist for quite a long time. What my co-authors and I found that was rather surprising is that consumers that were more exposed to their banks liquidity shocks, had relatively lower credit card balances for up to 10 years after the initial pass through of the funding shock to them.
Jeff Dugger (05:14):
So I imagine if that happens in a financial crisis, that would in turn have some impact on the financial recovery since people are spending less.
Nikhil Paradkar (05:23):
Exactly. So that's the main interesting takeaway from this paper, which was that the largest reduction in credit limits fell on individuals that had the highest marginal propensity to borrow. And as a result, you know, it did reduce credit card borrowing. Now that being said, the caveat here is that we are specifically studying one channel, which is the credit card channel. So as a result, we cannot really say much when it comes to conspicuous consumption, specifically consumption of the type of autos. But what we can talk about is, you know, everyday consumption such as groceries, paying for utilities, paying for clothing, et cetera. Everyday consumption was maybe negatively impacted because of the pass through of the shock.
Jeff Dugger (06:16):
And I would imagine for our listeners who may not be familiar with why banks may cut back on the amount of funds they make available for people on the revolving accounts is that they have to maintain a certain level of reserves to cover deposits, et cetera.
Nikhil Paradkar (06:36):
That's right. Essentially in this case, what banks were trying to do was there was a crisis going on in 2008 and they were simply trying to reduce their risk exposure. And the one way to do that is to maybe limit the supply of credit to the riskiest population who also happen to be the people that have the highest marginal propensity to borrow.
Jeff Dugger (06:59):
So taking all this together, how was this work received by the CFPB and what was their particular interest in it?
Nikhil Paradkar (07:07):
So the particular interest comes from the fact that there was a unique data advantage that we had. Specifically you know, there were hypotheses earlier that these effects can take place. Effectively this is the first paper that linked a bank liquidity shock to consumer credit card. Specifically because we had the data available for it. Now, the key challenge that people have is that when you study credit card lending from traditional sources of data it's very hard to disentangle supply and demand effects. The data advantage that we had was that we focused on individuals that had multiple credit cards.
Nikhil Paradkar (08:12):
So specifically what that does is we study one individual that has multiple credit cards, where these cards are issued by banks that are differently exposed to the funding shock. Now, what that helps us do is it helps us isolate credit demand. Specifically, in a sense, the consumer is held fixed. We are able to completely identify a supply effect from the bank's perspective and are isolated perfectly to this liquidity shock. Or are better than we could have using traditional data sources. So from the CFPB perspective, they were very excited to see these results. Mainly because they got a quantification of some of these hypotheses that they had come to know. They may have expected some of these findings, but it was still good to see how a dollar reduction in limits translated into a dollar change in balances. And how that relationship held across different consumer segments and all along the credit score spectrum or the credit utilization ratio spectrum.
Jeff Dugger (09:15):
This is all very fascinating application of economics theory to the real world. It's very interesting.
Nikhil Paradkar (09:24):
Thank you.
Jeff Dugger (09:25):
So another area that you study regarding household finances, not just the intersection of household finance and the banking system, but also household finance and FinTech. So tell us a little bit more about that and the places where you see machine learning being able to help consumers. In particular, tell us about your paper impact of marketplace lending on consumers' future borrowing capacities and borrowing outcomes. What did you learn about that and what modeling techniques did you use?
Nikhil Paradkar (09:56):
Yes. so this paper was broadly motivated by the rise of several FinTech disruptors, including marketplace lending platforms and consumer credit markets. Now, these FinTech platforms compete with banks in providing credit to consumers. The goal of these FinTech platforms is to necessarily use alternative data. Basically data that's not used by traditional banking models in making their credit decisions. Moreover, these platforms also publish the repayment histories of their borrowers, and then they use this information in future lending decisions. And the idea is that by publishing the three payment histories, they can potentially discipline borrowers. Now these special aspects of FinTech platforms could potentially challenge the traditional banking model and affect how consumers access credit. Now in this paper, what we try to do is we try to compare how borrowing from FinTech platforms versus traditional banks impacts consumers, credit scores, and their default propensities. My coauthors.
Nikhil (11:11):
And I interpret these effects in terms of the FinTech platform's ability to reduce information frictions between lenders and borrowers. As an example, if FinTech lenders are better able to screen - that is identify good quality borrowers and discipline their borrowers compared to banks - then FinTech borrowers should have fewer defaults and higher credit scores compared to observably similar bank borrowers. Now, obviously to do this comparison we need to have some sort of matching strategy. The modeling technique that we used was a K nearest matching approach where we match each FinTech platform borrower to a traditional bank borrower. The match was actually pretty thorough in the sense that it was created such that both types of borrowers have identical credit and income profiles. Moreover, the borrowers are selected such that they reside in the same five digit zip code. And, we also conditioned on the loan product that they originate.
Nikhil Paradkar (12:17):
We made sure that both types of borrowers originate the same loan product, which is an unsecured installment loan, and that this loan product originated in the same month for both types of borrowers. What we ended up finding is that despite being identical in the months leading up to the origination of their respective loans, FinTech borrowers had lower credit scores and higher default propensities in the long run compared to bank borrowers. And as a result, our results seem to indicate that FinTech platforms are less able to mitigate information frictions with respect to their borrowers than traditional banks.
Jeff Dugger (12:59):
Very interesting. And in this case, it seems like the machine learning technique you used was a fairly straightforward clustering technique to try to bring your subjects closely in line for each particular bank and the FinTech lenders, correct?
Nikhil Paradkar (13:17):
Yes. So it was just a straightforward K nearest neighbor matching approach. It's fairly standard practice. But the good thing that we were able to do with this approach was that we were able to match consumers, not just on levels, which is you know, at a specific point in time, but also on trends. Basically you know, the MPO borrower was matched to the bank borrower such that they both had similar trends in credit scores. If the MPL borrower was experiencing increase in credit scores in the month leading up to MPO loan origination, if the FinTech borrower was experiencing increasing credit scores in the month leading up to FinTech loan origination, then the bank borrower that was matched to this FinTech borrower had a similar trend as well. And that matching approach provides a significant upgrade over some of the existing techniques that were observed in the existing literature.
Jeff Dugger (14:19):
I may be asking you to speculate here, but how could a marketplace lender use information such as this to improve their lending?
Nikhil Paradkar (14:33):
That's a difficult question to answer. Specifically it kind of depends on the loan product that they are specializing in. But I think the one thing that the FinTech platform should maybe understand is that what our paper seems to suggest is that there is a selection of higher risk borrowers on such FinTech platforms that is not captured on an observable dimension. And the reason they can say that tentatively is because our matching technique was identifying FinTech borrowers that were identical to bank borrowers on observable terms. So the matching can only take place on observable characteristics, but there's something on the observable that results in FinTech borrowers defaulting more. And though that could occur because there is, you know, a selection of lower or higher risk borrowers onto FinTech platforms or something about the FinTech platforms’ loan product itself encourages such borrowers to default more or causes them to default more. It's just something that FinTech platforms can best be aware of when they supply credit to consumers.
Jeff Dugger (16:08):
So there may be some sort of latent variable that hasn't been discovered yet that could impact the kinds of people who are selected by FinTech.
Nikhil Paradkar (16:19):
I think that's the best way to put it, yes. That there is probably some kind of a latent unobservable variable that either causes borrowers to approach FinTech platforms with the intention of accessing credit through FinTech sources. Or there's some component of the FinTech loan itself that results in FinTech borrowers defaulting more. And it's very hard to disentangle these two effects, but the paper that we have kind of sheds light and provides suggestive evidence along both dimensions that both problems may be at play here.
Jeff Dugger (16:59):
So it sounds like this is an area that's wide open for a lot more research.
Nikhil Paradkar (17:06):
Exactly. I think there is a lot more work to be done and nailing down specifically what it is exactly that causes FinTech borrowers to default more. Is it really the borrowers themselves? Or is it something about the product that's being offered? But you know, that will require a substantially different type of data and the ability to differentiate between these two effects, which is obviously an interesting avenue for other researchers to pursue.
Jeff Dugger (17:47):
So having covered some of these more serious topics, let's turn to something a little more fun now. You have very interesting research. You have very interesting research you did on buzzwords innovations and quarterly earnings calls. And you applied machine learning to this research, right?
Nikhil Paradkar (18:07):
Yes. That's actually right. In this paper, my coauthors and I apply some machine learning techniques to finance research. So this is the first for me. So specifically what my coauthors and I study is whether the discussion of emerging technologies by firms management and their earnings conference calls conveys credible information to investors. Now, as such emerging technologies are characterized by their novelty. So as such these emerging technologies are characterized by the novelty of both their origin, as well as their applications. Emerging technologies are also categorized by their relatively fast growth as well as their potential to maybe exert considerable impact on business and society. The key issue here though, is that this impact is expected to be experienced not immediately, but in the nebulous future.
Nikhil Paradkar (19:21):
And as a result, emerging technologies also entail significant uncertainty. So as such, you know, the existing literature is unclear about how the stock market would possibly respond to a firm's discussion of emerging technologies. Now, as an example, there was actually one paper that was published in the early two thousands that found that during the dot-com era, companies that changed their corporate names to internet related.com names experienced an immediate stock price reaction. Immediate, positive stock price reaction. Now this reaction was positive, even if the firm itself had no specific focus on technology or anything to do with the internet broadly. On the other hand, there is an existing feature research that also finds that as such the stock market tends to be quite slow when it comes to recognizing the benefits of any kind of R and D investments that firms may undertake. You know because of this uncertainty and how the stock market could really respond to emerging technologies, we started exploring this research question. But before we can actually analyze how the stock market could respond to the discussion of emerging technologies, the main challenge that we face is that of identifying these emerging technology phrases themselves. And now that happens because by definition, emerging technologies are continuously evolving over time. Moreover, the scientific or technical jargon of emerging technologies may not actually be used word for word by firms' executives or the business press, which may instead rely on some shorthand, nicknames or synonyms. For example, you have electronic paper, which is the official emerging technology, but people call it e-paper. And as a result, we know we need to be agnostic when it comes to identifying these phrases. Now we mitigate these concerns by using a newly developed language model, which is the bi-directional encoder representations from transformers or birds. Using this language model, we create a dictionary of emerging technology, buzzwords. And importantly, this dictionary will identify not just the emerging technologies as they materialize over time, but it's holistic in the sense that it contains both the jargon that's associated with emerging technologies, as well as their common everyday nicknames and synonyms.
Nikhil Paradkar (22:13):
Now using this dictionary, we analyze the stock market response to the discussion of emerging technology phrases. We find that firms' discussion of emerging technologies does generate an immediate stock price reaction. However, we find that this effect or our evidence suggests that this effect appears to be rational because such firms actually have mentioned emerging technologies in their earnings conference calls to subsequently increase their R and D investments. And they have patents granted in the three-year window following the discussion of these emerging technologies. Where the patents that are granted to these firms are related to the specific technology that they mentioned in their earnings conference call. Overall, our results seem to suggest that the discussion of emerging technologies in earnings calls is not just hype, but it's something that conveys credible information to investors.
Jeff Dugger (23:14):
So you all will have to, excuse me, while I go off and write my own AI algorithm to boost my stock portfolio based on your research then.
Nikhil Paradkar (23:23):
Yes. So this result was a little surprising. The reason for that was we had this earlier paper where we, which I just talked about, says that there is a positive price response when firms associate themselves with the latest trends. In the case of the two thousands, it was firms changing their names to dotcom names. But that paper finds that it was maybe not a rational response because all firms experienced that positive response, even though their business had nothing to do with the internet broadly. But in our case, we find that firms actually do go. In our case, we find that firms that mentioned emerging technologies will actually go and innovate in, in the relevant technology domains. So in the case of your AI specifically, it will be great. I think for short-term trading strategies. Highly risky, but I think we have now two papers that seem to suggest that you will generate positive returns in the short term.
Jeff Dugger (24:27):
I think it goes without saying that we are not financial advisors, so don't base any stock picks on what we are telling you. But this sounds like a very fascinating application of emerging technologies in and of itself with natural language processing.
Nikhil Paradkar (24:45):
Thank you. We are currently revising the draft of this paper hoping to present it at conferences soon.
Jeff Dugger (24:55):
Well, I look forward to getting a copy of that. That would be very interesting. So Nikhil thank you so much for joining us today and for sharing your rich knowledge with us. If our listeners are interested in more information, where can they find you?
Nikhil Paradkar (25:09):
You can find me on the University of Georgia's Terry college of businesses directory. I'm an assistant professor in finance there. So my information, my contact information, my email address is all up on the Terry College of Business's website.
Jeff Dugger (25:27):
Well, thanks again for joining us today. If you haven't yet, I invite you to listen to another episode I recorded with Dr. Jennifer Priestley from Kennesaw State University. We talked about the different disciplines within data science and how university programs collaborate with businesses to build better data scientists. If you would like to be notified of future episodes, hit the subscribe button, wherever you are listening. And if you liked what you heard, please leave us a review.