Notes from the Vault
W. Scott Frame, Larry D. Wall, and Lawrence J. White
January 2019

Technological changes arising from advances in telecommunications, information technology, and financial practice have had a dramatic effect on financial intermediation. This technological progress has spurred financial innovations that have altered many financial products, services, production processes, and organizational structures. We recently posted an Atlanta Fed working paper, "Technological Change and Financial Innovation in Banking: Some Implications for Fintech," which surveys some of the work done by economists to understand better technology-driven financial innovation. This post draws on that working paper to highlight why research on financial innovation is important and discusses some data limitations that inhibit such research.

Contribution to private and public policy
Financial innovation poses challenges to both the private and public sector. Those in the private sector must first decide which innovations are worth developing and then how best to implement those that are created. Those in the public sector need to understand the social welfare implications of different types of innovation, and the ways in which innovations are implemented. Economic research can be of help with these questions.

The very broad question of whether financial innovation overall is socially beneficial is almost impossible to answer. However, questions about the welfare effects of particular innovations are an ongoing area of research. One pre-crisis innovation that subsequently gained an especially bad reputation is subprime mortgage lending. Strictly defined, a subprime mortgage is home financing for a borrower with less than a prime credit score. However, this term is often used in a more generic sense of higher-risk mortgage loans, which include those with high loan-to-value ratios and/or those where the borrowers provided little or no documentation to support their stated income. The growth of subprime mortgage lending was largely driven by technological advances that allowed for more sophisticated analysis of how these loans could be expected to perform based on borrower and loan characteristics, and also facilitated risk-based loan pricing.

Although subprime mortgages were widely accepted prior to the financial crisis, they were subsequently condemned as exploiting unsophisticated borrowers and contributing to the large losses borne by lenders and investors. The case that subprime mortgage loans were exploiting borrowers was based on readily observable differences in default rates between subprime and conventional mortgages. However, our working paper points to several studies that find that specific, unique features of subprime mortgages were plausibly beneficial (on net) to borrowers.

A second hypothesis was that bad mortgage loans were being originated because the financial institutions subsequently sold them to investors in mortgage-backed securities. This hypothesis also links two readily observable facts in a causal relationship, but that link is not supported by research that found securitized and nonsecuritized loans to perform about the same after controlling for observable information about the borrower, loan, and local economic conditions.

The hypothesis best supported by existing research is that both subprime borrowers and lenders had unreasonably high expectations about house price appreciation that would allow both parties to benefit from the mortgage. When house price appreciation slowed and then declined, such assumptions became unviable, resulting in many borrowers ultimately defaulting and investors taking losses. Thus, while post-crisis economic research has refuted some of the initial condemnations of subprime mortgage lending, it also suggests that much of the pre-crisis growth in subprime mortgage lending depended upon excessive optimism about home price appreciation.

The working paper also reports on several studies relating to a relatively recent innovation: marketplace lending to retail borrowers. Marketplace lenders were originally referred to as "peer-to-peer lenders," as they provided an online platform where individuals seeking to borrow could obtain funds from other individuals who had funds to lend. However, these online intermediaries evolved into broader marketplace lenders, as institutional investors funding large portfolios fueled their growth. This migration raised the policy question of whether marketplace lending was—in any economically meaningful way—different from unsecured consumer lending offered by banks. And, from the private perspective, how optimally should marketplace lenders manage information flows?

Loans from marketplace lenders are economically similar to some types of bank lending. For example, although the mechanisms are somewhat different, unsecured consumer lending via a marketplace lending platform is very similar to credit card lending insofar as both rely heavily on their ability to sell loans to institutional investors. Additionally, at one time, marketplace lenders followed credit card banks in marketing their loans to investors based on borrower credit scores. Credit scores have the advantages of being cheap, readily observable, and well understood by both lenders and investors. However, marketplace lenders have evolved to rely more heavily on their own credit risk models, which seem to be increasingly based on machine learning techniques applied to credit scores and other types of data. A recent Philadelphia Fed working paper by Jagtiani and Lemieux (2018) and a paper by Balyuk and Davydenko (2018) have found that the credit assessments made by marketplace lenders are becoming both less correlated with credit scores and more accurate.

The question for marketplace lenders is whether they should provide enough information to allow investors with superior analytic skills to select their own loans. A paper by Vallee and Zeng (2018) finds that providing too much information to investors may actually reduce the volume of approved loans, as it creates the risk of adverse selection for less sophisticated investors. Consistent with that model, marketplace lenders have substantially reduced the types of information provided to investors over time and encouraged them to rely instead on a marketplace lender’s credit evaluation.

Data limitations
Studies of financial innovation can be divided into prospective and retrospective studies. Prospective studies make assumptions about the future of innovation and seek to model the likely outcomes. Retrospective studies analyze innovations that have already been adopted. Both types of studies face potentially significant data issues.

Prospective studies face the unavoidable problem that we lack hard data about the future. This can lead some observers to question whether a specific innovation is worth studying at all. Many seemingly promising innovations fail completely, and others have only relatively minor effects. If we assume that an innovation seems likely to become important, a second issue that arises is that some assumptions about how the innovation will operate in practice need to be made. These assumptions need not be totally accurate—economic models are necessarily simplifications of the real world—but they should be sufficiently close to provide useful insights.

Despite the issues that arise with prospective studies, a number of economists have determined that such analysis will prove worthwhile for some recent innovations, including blockchain ledgers. Blockchains are a specific type of distributed database substantially more tamper-resistant than other types of databases. The original blockchain was developed to facilitate the creation of bitcoin, a cryptocurrency that does not rely on a trusted third party. Since the original proposal, many computer scientists, investors, and some major technology firms have concluded that blockchains have a wide variety of potential applications beyond cryptocurrency.

The widespread support for applying blockchains to financial problems has persuaded some economists of the merits of carefully studying the technology’s potential impact. For example, a paper by Catalini and Gans (2017) provides a model in which blockchains could result in the creation of new markets by providing a lower-cost substitute for a trusted third party. In terms of public policy, a recent paper by Cong and He (2018) finds that blockchains could also be used to facilitate collusion by allowing firms to learn more about their competitors’ sales.

The potential for data issues to impede quality economic analysis is not limited to prospective studies, but can even serve as a barrier to learning about established innovations. An outstanding example of where data limitations could easily have been a problem is small business credit scoring. Small business loans have historically been underwritten by loan officers based on their knowledge of the business, its owner, and the local market. However, banks came to understand that a loan to a small business owner had a risk profile very similar to that of a consumer loan to that same individual. This insight was applied to the loan underwriting process by some banks earlier than others, which created a unique opportunity to evaluate how the use of credit scores changed small business lending. However, there was no requirement that U.S. banks disclose any information on how they evaluated small business lending, which could easily have precluded research into small business credit scoring. Fortunately, one of us (Scott Frame) worked with a former Atlanta Fed economist, Aruna Srinivasan, to conduct a survey of bank’s usage of credit scoring for small business lending. That survey allowed researchers to address a variety of questions, including: (a) which banks were more likely to be earlier adopters? (b) how were loan pricing and loan quality affected? (c) did banks begin to lend over greater distances? and (d) how was credit availability affected overall and for low- and moderate-income areas?

The financial services industry has long been characterized by technologically driven innovation, and the pace of change seems to be accelerating. Economic research on these innovations can contribute to better business and public policy decisions. However, an important obstacle to such research is the unavoidable lack of data about the future impact of innovation and, in many cases, the lack of sufficient high-quality data on what has happened in the recent past.

W. Scott Frame is a senior adviser at the Federal Reserve Bank of Atlanta. He can be reached at Larry Wall is the executive director of the Center for Financial Innovation and Stability at the Federal Reserve Bank of Atlanta. He can be reached at Lawrence J. White is the Robert Kavesh Professor of Economics at New York University. He can be reached at The views expressed here are the authors’ 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