U.S. Banks' Artificial Intelligence and Small Business Lending: Evidence from the Census Bureau's Annual Business Survey
February 2025
Working Paper Number:
CES-25-07
Abstract
Utilizing confidential microdata from the Census Bureau's new technology survey (technology module of the Annual Business Survey), we shed light on U.S. banks' use of artificial intelligence (AI) and its effect on their small business lending. We find that the percentage of banks using AI increases from 14% in 2017 to 43% in 2019. Linking banks' AI use to their small business lending, we find that banks with greater AI usage lend significantly more to distant borrowers, about whom they have less soft information. Using an instrumental variable based on banks' proximity to AI vendors, we show that AI's effect is likely causal. In contrast, we do not find similar effects for cloud systems, other types of software, or hardware surveyed by Census, highlighting AI's uniqueness. Moreover, AI's effect on distant lending is more pronounced in poorer areas and areas with less bank presence. Last, we find that banks with greater AI usage experience lower default rates among distant borrowers and charge these borrowers lower interest rates, suggesting that AI helps banks identify creditworthy borrowers at loan origination. Overall, our evidence suggests that AI helps banks reduce information asymmetry with borrowers, thereby enabling them to extend credit over greater distances.
Document Tags and Keywords
Keywords
Keywords are automatically generated using KeyBERT, a powerful and innovative
keyword extraction tool that utilizes BERT embeddings to ensure high-quality and contextually relevant
keywords.
By analyzing the content of working papers, KeyBERT identifies terms and phrases that capture the essence of the
text, highlighting the most significant topics and trends. This approach not only enhances searchability but
provides connections that go beyond potentially domain-specific author-defined keywords.
:
enterprise,
microdata,
survey,
financial,
financing,
borrower,
lending,
loan,
bank,
lender,
borrowing,
debt,
gdp,
credit,
banking
Tags
Tags are automatically generated using a pretrained language model from spaCy, which excels at
several tasks, including entity tagging.
The model is able to label words and phrases by part-of-speech,
including "organizations." By filtering for frequent words and phrases labeled as "organizations", papers are
identified to contain references to specific institutions, datasets, and other organizations.
:
Internal Revenue Service,
Small Business Administration,
Federal Reserve Bank,
Federal Reserve System,
Employer Identification Numbers,
Paycheck Protection Program,
Census Bureau Disclosure Review Board,
Federal Statistical Research Data Center,
National Center for Science and Engineering Statistics,
Annual Business Survey
Similar Working Papers
Similarity between working papers are determined by an unsupervised neural
network model
know as Doc2Vec.
Doc2Vec is a model that represents entire documents as fixed-length vectors, allowing for the
capture of semantic meaning in a way that relates to the context of words within the document. The model learns to
associate a unique vector with each document while simultaneously learning word vectors, enabling tasks such as
document classification, clustering, and similarity detection by preserving the order and structure of words. The
document vectors are compared using cosine similarity/distance to determine the most similar working papers.
Papers identified with 🔥 are in the top 20% of similarity.
The 10 most similar working papers to the working paper 'U.S. Banks' Artificial Intelligence and Small Business Lending: Evidence from the Census Bureau's Annual Business Survey' are listed below in order of similarity.
-
Working PaperSpillovers From Costly Credit
March 2013
Working Paper Number:
CES-13-11
Recent research on the effects of credit access among low- and moderate-income households finds that high-cost payday loans exacerbate, rather than alleviate, financial distress for a subset of borrowers (Melzer 2011; Skiba and Tobacman 2011). In this study I find that others, outside the borrowing household, bear a portion of these costs too: households with payday loan access are 20% more likely to use food assistance benefits and 10% less likely to make child support payments required of non-resident parents. These findings suggest that as borrowers accommodate interest and principal payments on payday loan debt, they prioritize loan payments over other liabilities like child support payments and they turn to transfer programs like food stamps to supplement the household's resources. To establish this finding, the analysis uses a measure of payday loan access that is robust to the concern that lender location decisions and state policies governing payday lending are endogenous relative to household financial condition. The analysis also confirms that the effect is absent in the mid-1990s, prior to the spread of payday lending, and that the effect grows over time, in parallel with the growth of payday lending.View Full Paper PDF
-
Working PaperAfter the Storm: How Emergency Liquidity Helps Small Businesses Following Natural Disasters
April 2024
Working Paper Number:
CES-24-20
Does emergency credit prevent long-term financial distress? We study the causal effects of government-provided recovery loans to small businesses following natural disasters. The rapid financial injection might enable viable firms to survive and grow or might hobble precarious firms with more risk and interest obligations. We show that the loans reduce exit and bankruptcy, increase employment and revenue, unlock private credit, and reduce delinquency. These effects, especially the crowding-in of private credit, appear to reflect resolving uncertainty about repair. We do not find capital reallocation away from neighboring firms and see some evidence of positive spillovers on local entry.View Full Paper PDF
-
Working PaperThe Rise of Industrial AI in America: Microfoundations of the Productivity J-curve(s)
April 2025
Working Paper Number:
CES-25-27
We examine the prevalence and productivity dynamics of artificial intelligence (AI) in American manufacturing. Working with the Census Bureau to collect detailed large-scale data for 2017 and 2021, we focus on AI-related technologies with industrial applications. We find causal evidence of J-curve-shaped returns, where short-term performance losses precede longer-term gains. Consistent with costly adjustment taking place within core production processes, industrial AI use increases work-in-progress inventory, investment in industrial robots, and labor shedding, while harming productivity and profitability in the short run. These losses are unevenly distributed, concentrating among older businesses while being mitigated by growth-oriented business strategies and within-firm spillovers. Dynamics, however, matter: earlier (pre-2017) adopters exhibit stronger growth over time, conditional on survival. Notably, among older establishments, abandonment of structured production-management practices accounts for roughly one-third of these losses, revealing a specific channel through which intangible factors shape AI's impact. Taken together, these results provide novel evidence on the microfoundations of technology J-curves, identifying mechanisms and illuminating how and why they differ across firm types. These findings extend our understanding of modern General Purpose Technologies, explaining why their economic impact'exemplified here by AI'may initially disappoint, particularly in contexts dominated by older, established firms.View Full Paper PDF
-
Working PaperAI Adoption in America: Who, What, and Where
September 2023
Working Paper Number:
CES-23-48R
We study the early adoption and diffusion of five AI-related technologies (automated-guided vehicles, machine learning, machine vision, natural language processing, and voice recognition) as documented in the 2018 Annual Business Survey of 850,000 firms across the United States. We find that fewer than 6% of firms used any of the AI-related technologies we measure, though most very large firms reported at least some AI use. Weighted by employment, average adoption was just over 18%. AI use in production, while varying considerably by industry, nevertheless was found in every sector of the economy and clustered with emerging technologies such as cloud computing and robotics. Among dynamic young firms, AI use was highest alongside more educated, more-experienced, and younger owners, including owners motivated by bringing new ideas to market or helping the community. AI adoption was also more common alongside indicators of high-growth entrepreneurship, including venture capital funding, recent product and process innovation, and growth-oriented business strategies. Early adoption was far from evenly distributed: a handful of 'superstar' cities and emerging hubs led startups' adoption of AI. These patterns of early AI use foreshadow economic and social impacts far beyond this limited initial diffusion, with the possibility of a growing 'AI divide' if early patterns persist.View Full Paper PDF
-
Working PaperBrighter Prospects? Assessing the Franchise Advantage using Census Data
January 2017
Working Paper Number:
CES-17-21
This paper uses Census micro data to examine how starting a business as a franchise rather than an independent business affects its survival and growth prospects. We first consider the factors that influence the business owner's decision about being franchised, and then use different empirical approaches to correct for selection bias in our performance analyses. We find that franchised businesses on average benefit from higher survival rates and faster initial growth relative to independent businesses. However, the effects are not large and, conditional on first-year survival, the differences basically disappear. We briefly discuss potential mechanisms to explain these results. U.S. Census Bureau. All results have been reviewed to ensure that no confidential information is disclosed. Support for this research at the Michigan Census Research Data Center is gratefully acknowledged.View Full Paper PDF
-
Working PaperJob Creation, Small vs. Large vs. Young, and the SBA
September 2015
Working Paper Number:
CES-15-24
Analyzing a list of all Small Business Administration (SBA) loans in 1991 to 2009 linked with annual information on all U.S. employers from 1976 to 2012, we apply detailed matching and regression methods to estimate the variation in SBA loan effects on job creation and firm survival across firm age and size groups. The estimated number of jobs created per million dollars of loans within the small business sector generally increases with size and decreases in age. The results suggest that the growth of small, mature firms is least financially constrained, and that faster growing firms experience the greatest financial constraints to growth. The estimated association between survival and loan amount is larger for younger and smaller firms facing the 'valley of death.'View Full Paper PDF
-
Working PaperLocally Owned Bank Commuting Zone Concentration and Employer Start-Ups in Metropolitan, Micropolitan and Non-Core Rural Commuting Zones from 1970-2010
August 2018
Working Paper Number:
CES-18-34
Access to financial capital is vital for the sustainability of the local business sector in metropolitan and nonmetropolitan communities. Recent research on the restructuring of the financial industry from local owned banks to interstate conglomerates has raised questions about the impact on rural economies. In this paper, we begin our exploration of the Market Concentration Hypothesis and the Local Bank Hypothesis. The former proposes that there is a negative relationship between the percent of banks that are locally owned in the local economy and the rate of business births and continuations, and a positive effect on business deaths, while that latter proposes that there is a positive relationship between the percent of banks that are locally owned in the local economy and the rate of business births and continuations, and a negative effect on business deaths. To examine these hypotheses, we examine the impact of bank ownership concentration (percent of banks that are locally owned in a commuting zone) on business establishment births and deaths in metropolitan, micropolitan and non-core rural commuting zones. We employ panel regression models for the 1980-2010 time frame, demonstrating robustness to several specifications and spatial spillover effects. We find that local bank concentration is positively related to business dynamism in rural commuting zones, providing support to the importance of relational lending in rural areas, while finding support for the importance of market concentration in urban areas. The implications of this research are important for rural sociology, regional economics, and finance.View Full Paper PDF
-
Working PaperCredit Access in the United States
July 2025
Working Paper Number:
CES-25-45
We construct new population-level linked administrative data to study households' access to credit in the United States. These data reveal large differences in credit access by race, class, and hometown. By age 25, Black individuals, those who grew up in low-income families, and those who grew up in certain areas (including the Southeast and Appalachia) have significantly lower credit scores than other groups. Consistent with lower scores generating credit constraints, these individuals have smaller balances, more credit inquiries, higher credit card utilization rates, and greater use of alternative higher-cost forms of credit. Tests for alternative definitions of algorithmic bias in credit scores yield results in opposite directions. From a calibration perspective, group-level differences in credit scores understate differences in delinquency: conditional on a given credit score, Black individuals and those from low-income families fall delinquent at relatively higher rates. From a balance perspective, these groups receive lower credit scores even when comparing those with the same future repayment behavior. Addressing both of these biases and expanding credit access to groups with lower credit scores requires addressing group-level differences in delinquency rates. These delinquencies emerge soon after individuals access credit in their early twenties, often due to missed payments on credit cards, student loans, and other bills. Comprehensive measures of individuals' income profiles, income volatility, and observed wealth explain only a small portion of these repayment gaps. In contrast, we find that the large variation in repayment across hometowns mostly reflects the causal effect of childhood exposure to these places. Places that promote upward income mobility also promote repayment and expand credit access even conditional on income, suggesting that common place-level factors may drive behaviors in both credit and labor markets. We discuss suggestive evidence for several mechanisms that drive our results, including the role of social and cultural capital. We conclude that gaps in credit access by race, class, and hometown have roots in childhood environments.View Full Paper PDF
-
Working PaperDo SBA Loans Create Jobs? Estimates from Universal Panel Data and Longitudinal Matching Methods
September 2012
Working Paper Number:
CES-12-27
This pape reports estimates of the effects of the Small Business Administration (SBA) 7(a) and 504 loan programs on employment. The database links a complete list of all SBA loans in these programs to universal data on all employers in the U.S. economy from 1976 to 2010. Our method is to estimate firm fixed effect regressions using matched control groups for the SBA loan recipients we have constructed by matching exactly on firm age, industry, year, and pre-loan size, plus kernel-based matching on propensity scores estimated as a function of four years of employment history and other variables. The results imply positive average effects on loan recipient employment of about 25 percent or 3 jobs at the mean. Including loan amount, we find little or no impact of loan receipt per se, but an increase of about 5.4 jobs for each million dollars of loans. When focusing on loan recipients and control firms located in high-growth counties (average growth of 22 percent), places where most small firms should have excellent growth potential, we find similar effects, implying that the estimates are not driven by differential demand conditions across firms. Results are also similar regardless of distance of control from recipient firms, suggesting only a very small role for displacement effects. In all these cases, the results pass a "pre-program" specification test, where controls and treated firms look similar in the pre-loan period. Other specifications, such as those using only matching or only regression imply somewhat higher effects, but they fail the pre-program test.View Full Paper PDF
-
Working PaperPersonal Bankruptcy Law and Entrepreneurship
January 2017
Working Paper Number:
CES-17-42R
We study the effect of debtor protection on firm entry and exit dynamics. We find that more lenient personal bankruptcy laws lead to higher firm entry, especially in sectors with low entry barriers. We also find that debtor protection increases firm exit rates and that this effect is independent of firm age. Our results overall indicate that changes in debtor protection affect firm dynamics.View Full Paper PDF
