Financial Stability Implications of New Technology
Notes from the Vault
Larry D. Wall
The financial system has been evolving in response to changing financial and information technology over the last several decades. Recent advances in areas such as big data, machine learning, and distributed ledgers have the potential to accelerate the pace of change significantly. These technological advances could result in a stronger, more resilient financial system; however, these changes also have the potential to make the system less stable. These topics and more were addressed in a recent academic workshop, Financial Stability Implications of New Technology. The Atlanta Fed hosted the event, which was cosponsored by the Center for the Economic Analysis of Risk at Georgia State University. This post reviews many of the papers and presentations from the workshop dealing with a variety of regulatory policy issues. An accompanying macroblog post will review the panel on cryptocurrency and central bank e-money.
Liquidity in automated markets
The long-run trend in financial markets has been to replace trading between humans with trading between machines. In many markets, machines acting within a fraction of a second make most buy and sell decisions. This has led to considerable interest in the actions of these so-called high-frequency traders (HFTs) that supply liquidity to financial markets—temporarily bridging the gap between when a buy order reaches the exchange and when an offsetting sell order is received (or vice versa). The desire for speed from HFTs is so great that firms have their computers installed adjacent to the exchange to get their orders posted ahead of other traders. Similarly, traders that seek to exploit differences across markets located in different places (such as the cash and futures markets) demand the highest-speed communications equipment to allow information to be quickly communicated, often within milliseconds.
A large fraction of the academic analysis of high-speed markets, especially equity markets, has focused on HFTs and their role in supplying liquidity. However, the large buy-side firms (such as mutual fund families) also use computer-driven trading to obtain the best prices. University of Illinois, Urbana-Champaign professors Sida Li, Xin Wang, and Mao Ye develop a theoretical model in which the buy-side firms are somewhat slower than HFT firms, but the buy-side is motivated to transact based on private valuations of a stock. Their paper "Who Supplies Liquidity, and When?" finds that if the tick size is sufficiently small relative to the variability of the stock, these buy-side firms find it more profitable to step inside the bid-ask spread quoted by HFT firms and become liquidity suppliers. Arguably, a downside of the buy-side firms providing liquidity is that HFT firms quote much wider bid-ask spreads on their limit orders. As a result, a sudden jump in the net demand for a stock is more likely to cause a mini-flash crash, when spreads are low.
Federal Reserve Board economist Dobrislav Dobrev and AQR Capital Management managing director Ernst Schaumburg exploit the development of high-speed communication and trading to help determine which exchanges lead in price setting and which tend to be followers for different instruments that are priced on the same underlying security or commodity. "High-Frequency Cross-Market Trading: Model Free Measurement and Applications" takes advantage of the rather consistent lag between when a trade happens on one exchange and the earliest point at which that trade would be known at another exchange. Their analysis confirms prior beliefs that the S&P futures market leads the New York Stock Exchange cash market. However, they also find that the CME's Treasury futures market is about as likely to lag as it is to lead BrokerTec's cash market.
Changing technology and mortgage market risk
The market for residential mortgages and mortgage-backed securities has also undergone a variety of changes. Two papers presented at the workshop address the extent to which technological developments contributed to the 2007–09 financial crisis. Economists Lara Loewenstein from the Cleveland Fed and Christopher L. Foote and Paul S. Willen from the Boston Fed study the extent to which the development of automated underwriting models starting in the 1990s may have contributed to a spike in mortgage defaults in the early 2000s. "Technological Innovation in Mortgage Underwriting" observes that statistical estimates of mortgage default models in the 1990s found debt-to-income ratios had little predictive power; as a consequence, this ratio became less important in the automated-mortgage underwriting systems. The reduced emphasis on borrowers' groups debt-to-income ratio did not lead to a spike in mortgage defaults in the 1990s, but the authors speculate the reduced emphasis may have facilitated increased demand by mortgage borrowers based on inflated expectations of housing prices in the early 2000s.
During the onset of the recent financial crisis, a significant concern was the extent to which the prices of credit default swaps (CDSs) written on mortgage-backed securities accurately reflected their riskiness. Single-name CDS contracts (written on a particular mortgage-backed security) and, even more so, CDS contracts written on an index of single-name CDSs (the ABX indices) attracted considerable attention in 2007–08 as an indicator of a coming wave of mortgage defaults and foreclosures. However, market practitioners often criticized CDSs on mortgage-backed securities as trading at excessively pessimistic prices. The relationship of the cash and CDS markets is the subject of a study by professor Michael B. Imerman of Claremont University, professors Joseph R. Mason and Rajesh P. Narayanan from Louisiana State University, and PhD candidate Meredith E. Rhodes, also from Louisiana State. "Market Dynamics among the ABX Index, Credit Default Swaps, and Mortgage-Backed Bonds" used data releases on the performance of the underlying mortgages (remittance reports) from that period. They find that the cash market for residential mortgage-backed securities (RMBSs) most accurately reflected fundamental values. CDS and CDS index prices departed from these fundamentals, especially in the period between remittance report dates.
Interest in the database technology underlying bitcoin, called the blockchain, has grown considerably in the last several years. Two papers presented at the conference address two very different economic issues associated with blockchain technology: how to value blockchain tokens (including cryptocurrencies such as bitcoin) and how blockchains could be used to increase the efficiency of corporate audits.
Professors Lin William Cong of the University of Chicago, Ye Li of Ohio State University, and Neng Wang of Columbia University develop a theoretical model of the pricing of "(crypto) tokens on (blockchain-based) platforms." These tokens could be a cryptocurrency or a utility token needed to access some service on the platform. In the paper "Tokenomics: Dynamic Adoption and Valuation," the authors consider how the use of tokens can increase platform usage. If the platform does not use a token (in other words, payment for usage is in dollars), potential users consider only the immediate benefits they may obtain from using the platform. However, if the platform issues a token, then potential users value not only the immediate benefits but also the expected appreciation of the value of the token, leading to an increase in the number of users. Moreover, this increase in the number of users may be multiplied to the extent that the platform benefits from network effects.1
The second paper, "Auditing and Blockchains: Pricing, Misstatements, and Regulation," examines the potential for blockchain technology to improve the efficiency of corporate audits. Auditing firms currently manually confirm a sample of a corporation's transactions to verify the transactions' authenticity. However, this manual sampling is costly and, because it is incomplete, leaves open the possibility that some transactions are fake. The paper—by professors Baozhong Yang and Sean Cao of Georgia State University, along with Lin William Cong—considers an environment in which transactions between firms are given a universal identifier and made available to accounting firms. In their model, auditing firms could verify the universe of transactions between firms that make the data available to their auditing firm at (near) zero cost, reducing reporting errors. Assuming that adoption of this technology requires the payment of a one-time fixed cost, the authors find that either all auditors will find it cost-efficient to adopt the technology or all auditors will decide the costs of adoption exceed the benefits.
Research on financial innovation
Financial innovation is not new. Over 30 years ago, Nobel Prize–winning economist Merton Miller (1986) observed that the "word revolution is entirely appropriate" to describe the changes in financial institutions and instruments over the period from approximately 1966 to 1986. The conference dinner speaker, New York University professor Lawrence J. White, reviewed the state of the academic literature on financial innovation. White observed that when he and my Atlanta Fed colleague W. Scott Frame surveyed the academic literature in a 2004 paper, they found considerable evidence of further financial innovation but only a tiny number of papers relative to a vast industrial organization literature on other types of innovation. More recently, White noted a "vast expansion of the literature on financial innovations" arising in part because of recent innovations such as fintech. However, White observed that we still "know little about how and why financial innovations are initially developed, and by whom."
Online lending markets
The development of peer-to-peer lending was initially promoted as an activity that could replace bank lending to consumers with direct lending between consumers, just as Airbnb disintermediates hotels. However, as Atlanta Fed economist W. Scott Frame observed in a 2015 post, peer-to-peer lending morphed into marketplace lending, as personal lending platforms found it more efficient to obtain most of their funding from institutional investors. Papers at the workshop addressed developments in online lending from a variety of perspectives.
The ongoing development of online lending was chronicled in a paper by professors Tetyana Balyuk of Emory University and Sergei Davydenko of the University of Toronto. "Reintermediation in Fintech: Evidence from Online Lending" updates Frame's work to show that institutional investors remain the dominant source of funding for online lending (approximately 90 percent of funding in early 2018). However, the authors find the market has continued to evolve so that these large institutions no longer help price the loans and in most cases do not conduct independent analysis of them. Instead, institutional investors tend to rely on the platform to perform these functions. Looking more carefully at Prosper's loan evaluation, the authors also find that its ability to predict defaults has steadily improved since 2013 and is now "much more informative than FICO" scores by themselves.
Given that online (or fintech) lenders compete with commercial banks in consumer lending, a natural question is the relative credit efficiency of fintech lenders. A paper by professors Joseph P. Hughes from Rutgers University, Choon-Geol Moon from Hanyang University, and Philadelphia Fed economist Julapa Jagtiani addresses the efficiency of one of the largest online lenders in the United States, LendingClub, relative to commercial banks broken into several size categories. "Consumer Lending Efficiency: Commercial Banks versus a Fintech Lender" measures lending (in)efficiency as the nonperforming loan ratio of unsecured consumer loans. Using a methodology similar to that used in bank cost and profit efficiency studies, the paper seeks to identify best lending practices, given some other important determinants of loan performance such as the lenders' volume of lending, the average rate on the loans, and market conditions. The paper finds that LendingClub was similar to the largest of five size categories of banks in that it both takes on a large amount of credit risk but is relatively efficient in measuring the credit riskiness of that group.
Consumers in marketplace lending
Over 70 percent of the marketplace loans at the two largest marketplace lending platforms, Prosper and LendingClub, are given to borrowers whose stated purpose is "expensive debt consolidation." This use of the funds raises an important question in personal finance: are individuals using these loans to pay down expensive loans and ultimately reduce their debt load? Alternatively, are the loans merely facilitating an increase in some consumers' debt levels—either going immediately to finance increased consumption or doing so over a longer horizon? Two papers at the conference looked at this issue from different perspectives.
Georgia Institute of Technology professor Sudheer Chava and PhD candidate Nikhil Paradkar follow a sample of individual consumers after they borrow from a marketplace lender using anonymized data from a credit bureau. "Winners and Losers of Marketplace Lending: Evidence from Borrower Credit Dynamics" finds that, on average, consumers use marketplace loans to reduce their credit card balances substantially and this results in a jump in their credit score. This jump in credit scores is then accompanied by an increase in credit card limits for all three of the ratings group they study: prime, near prime, and subprime borrowers. However, the proportionately largest increase in limits came for the subprime (riskiest) borrowers. After paying down their credit card debt, Chava and Paradkar find that average consumer debt starts increasing for all groups, with average credit card balances for subprime borrowers increasing over the next three quarters back to the levels observed before their borrowing funds from the marketplace lenders. The end result is a jump in the probability of default for subprime borrowers over the three quarters after borrowing from the marketplace lender but essentially no change in default rates for higher-rated borrowers.
Professors Piotr Danisewicz of the University of Bristol and Ilaf Elard of the Shanghai University of International Business and Economics analyze what happens when the flow of funds to marketplace borrowers is suddenly disrupted. "The Real Effects of Financial Technology: Marketplace Lending and Personal Bankruptcy" analyzes the effect of the disruption in lending to Connecticut and New York as a result of a U.S. Circuit Court of Appeals finding in Madden versus Midland Funding, which raised doubts about the enforceability of marketplace loans. They find that this reduction in marketplace lending led to an increase in personal bankruptcy in those two states, especially for low-income households.
The seemingly contradictory findings between the two studies generated a number of comments from workshop attendees. The papers' discussant, Lauren Lambie-Hanson from the Philadelphia Fed, and other attendees suggested a number of ways to strengthen the analysis in the two papers. Assuming further work continues to support each study's conclusion, there was some speculation that the results may reflect timing differences with marketplace lending, allowing some borrowers who were likely to default in any case to postpone bankruptcy. If so, we should see an uptick in bankruptcies for Chava and Paradkar's sample of marketplace borrowers. In any case, the apparent contradiction supports the case for further research in this area.
Banks and new technology
The final session looked at three different issues related to the relationship of commercial banks to new technology. The first paper by professor Dalida Kadyrzhanova from Georgia State University, along with economists Giovanni Dell'Ariccia and Lev Ratnovski from the International Monetary Fund and Camelia Minoiu from the Federal Reserve Board, analyzes the growing importance of intangible (knowledge) capital on bank lending. "Bank Lending in the Knowledge Economy" observes that businesses typically provide tangible capital that can be readily resold (such as equipment and buildings) as collateral for their bank loans. However, as knowledge has become more important, a larger fraction of firms' economic value comes from intangible capital that cannot be used as collateral for a loan. The authors analyze the growth of intangible capital at the level of the metropolitan statistical area (MSA) and find that faster growth in intangible capital explains approximately 25 percent to 40 percent of the decline in the share of bank lending devoted to commercial and industrial loans. This reduced commercial lending is also accompanied by an approximately equal increase in real estate lending by banks.
Bank of Italy economists Matteo Accornero and Mirko Moscatelli analyze whether social media sentiment could provide an informative signal on retail depositors' level of trust in their banks. "Listening to the Buzz: Social Media Sentiment and Retail Depositors' Trust" analyzed tweets on Twitter to determine the sentiments being expressed by depositors about individual banks. The results of the study suggest that the volume of negative sentiment expressed in tweets helped predict reductions in retail deposits even after taking account of other variables such as the banks' capital and bad loans.
The last paper by professors Santiago Carbó-Valverde and Pedro J. Cuadros-Solas from Colegio Universitario de Estudios Financieros and Francisco Rodríguez-Fernández from the University of Granada analyzes the adoption of digital technology. "How Do Bank Customers Go Digital? A Random Forest Approach" analyzes Spanish consumers' response to a survey on digital banking using a machine learning procedure called random forest. The authors find that the process of going digital starts with consumers using online access to obtain basic information about their banking accounts (such as account balances) and to transfer funds between accounts. Over time, their usage expands as they become more aware of what can be done online and more comfortable that online access to banking services is safe.
The financial system has been experiencing ongoing innovation over at least the last 50 years, with the pace of innovation arguably accelerating in recent years. Informed public policy decision making requires both an understanding of how past innovation has affected the financial system as well as an informed view about how new technology may spur further changes. The studies presented at a recent workshop at the Atlanta Fed contribute both to our understanding of past developments and provide some insights to likely future developments.
Frame, W. Scott and Lawrence J. White (2004). "Empirical Studies of Financial Innovation: Lots of Talk, Little Action?" Journal of Economic Literature 42: 116–144. Behind a paywall at https://www.jstor.org/stable/pdf/3217038.pdf.
Miller, Merton H. (1986). "Financial Innovation: The Last Twenty Years and the Next." Journal of Financial and Quantitative Analysis 21: 459–471. Behind a paywall at https://www.jstor.org/stable/2330693?seq=1#metadata_info_tab_contents.
Larry D. Wall is executive director of the Center for Financial Innovation and Stability at the Atlanta Fed. The author thanks Scott Frame and Brian Robertson for helpful comments. The view expressed here are the author's and not necessarily those of the Federal Reserve Bank of Atlanta or the Federal Reserve System. If you wish to comment on this post, please email firstname.lastname@example.org.
1 Network effects exist to the extent that the value a person derives from a platform is increasing in the number of users of that platform. For example, no one would place much value on a sharing economy platform such as Uber or Lyft if it had only one user, but many people place considerable value as these platforms attract more buyers and sellers.