A. Sample Construction

We use the LPC DealScan data to measure banks’ specialization in corporate lending. Using link tables from Schwert (Reference Schwert2018) and Gomez, Landier, Sraer, and Thesmar (Reference Gomez, Landier, Sraer and Thesmar2021), we merge DealScan lenders with bank call report data.Footnote ⁴ We also use the link table from Chava and Roberts (Reference Chava and Roberts2008) to match borrowers with their accounting and industry data from Compustat.

Data on banks’ branch characteristics (e.g., name, address, BHC, and deposits) are from the Summary of Deposits, which covers the universe of banks’ depository branches annually from 1994. Small business lending is measured using the Community and Reinvestment Act small business loans database from the Federal Financial Institutions Examination Council, which reports the number and volume of loans originated by each reporting bank at the county level since 1996.

Mortgage data are collected from the HMDA database. We follow the prior literature and drop nonconventional loans and loans for manufactured housing and multifamily dwellings to remove the impact of government subsidies on banks’ lending decisions.Footnote ⁵ We also exclude other nonstandard mortgages, such as mortgages for home improvement and nonowner-occupied dwellings. We further exclude counties in which a bank has fewer than five mortgage applications per year to ensure that our results are not driven by outliers.Footnote ⁶ We merge the HMDA data with banks in the call reports by matching agency-specific IDs in HMDA (e.g., Federal Reserve RSSD-ID, FDIC Certificate Number, and OCC Charter Number) to RSSD IDs.

We complement the HMDA data with information on monthly loan-level performance from three sources: Fannie Mae and Freddie Mac single-family loan-level data sets and McDash loan-level data. The Fannie Mae data cover the fixed-rate single-family mortgage loans acquired by Fannie Mae from January 2000 to December 2022, with the origination year starting from 1999. The Freddie Mac data cover approximately 52.2-million fixed-rate single-family mortgage loans originated between Jan. 1, 1999 and Sept. 30, 2022 that are acquired by Freddie Mac. The McDash data are a proprietary database compiled by Black Knight, which tracks the dynamic performance of both agency and non-agency loans. Depending on the years, the McDash data cover 60%–80% of the U.S. mortgage market. Important to our study, all three data sets include a rich set of information not available in HMDA, including borrowers’ credit scores, LTV ratios, interest rates, and ex post monthly loan performance. We follow Chu, Ma, and Zhang (Reference Chu, Ma and Zhang2022) and match the three data sets to HMDA. The matched government-sponsored enterprise (GSE) mortgage sample is based on Fannie Mae and Freddie Mac data, whereas the non-GSE mortgage sample is based on the McDash data. We combine GSE and non-GSE mortgages and focus on the period from 1999 to 2017.

To quantify the distribution of employment across industries within a county, we use data from the Quarterly Census of Employment and Wages (QCEW), which provides annual employment figures from 1990 to 2018 for all 6-digit NAICS industries across more than 3,000 U.S. counties. In addition, data on county-to-county distances and county-level characteristics, such as income, housing price index, population, race, and age, are obtained from the Bureau of Economic Analysis, the Federal Housing Finance Agency, and the NBER database. County-level mortgage delinquency rates are from the Consumer Financial Protection Bureau (CFPB), and the county-to-county Social Connectedness Index (SCI), based on Facebook friendship links in 2016, is from Bailey, Cao, Kuchler, Stroebel, and Wong (Reference Bailey, Cao, Kuchler, Stroebel and Wong2018). Data on establishment-level locations and employment for U.S. firms are from the Your Economy Time Series, provided by the Business Dynamics Research Consortium.

B. Measuring a Bank’s Industry Specialization

We measure each bank’s industry specialization using DealScan data. Borrowers in DealScan are relatively large firms, and interactions with them enable banks to acquire advanced and comprehensive industry knowledge. We use origination dates and maturities to create a panel that tracks each bank’s lending portfolio at any given time.

Most of the loans in DealScan are syndicated and thus have multiple lenders. However, only lead lenders assume the monitoring responsibilities (Sufi (Reference Sufi2007), Gustafson, Ivanov, and Meisenzahl (Reference Gustafson, Ivanov and Meisenzahl2021)). In addition, the lead lenders have stronger incentives and better opportunities than participating lenders to acquire information about the borrowers and accumulate industry expertise. As a result, lending specialization matters more for lead lenders than for participating lenders (Blickle et al. (Reference Blickle, Parlatore and Saunders2025)). Lead lenders are also less likely to sell all of their loan shares in the secondary market (Irani, Iyer, Meisenzahl, and Peydro (Reference Irani, Iyer, Meisenzahl and Peydro2021)). We therefore focus on the lead lenders of syndicated loans.Footnote ⁷

We assume that the lead lenders commit all capital in a loan because the allocation of loan shares is missing for most loans in DealScan, and the lead lenders obtain industry knowledge by monitoring the total loan amount rather than their own capital (e.g., Giannetti and Saidi (Reference Giannetti and Saidi2019), Saidi and Streitz (Reference Saidi and Streitz2021)). For loans with multiple lead lenders, we divide the loan amount equally among all lead lenders.Footnote ⁸

We aggregate banks’ outstanding loans at the 3-digit NAICS industry level each year. Our choice of the 3-digit NAICS code level ensures sufficient precision of industry breakdowns and a reasonable number of firms and loans in each industry. We exclude firms in the financial industry. Following Paravisini, Rappoport, and Schnabl (Reference Paravisini, Rappoport and Schnabl2023), we classify a bank as specialized in an industry if its loan share in that industry is an outlier relative to the portfolio shares of other banks:

(1) $$ {Specialization}_{i,t}^b=\left\{\begin{array}{cc}1& {L}_{i,t}^b\geqq {L}_{i,t}^{\ast },\\ {}0& otherwise,\end{array}\right. $$

where $ b $ denotes bank, $ i $ denotes industry, and $ t $ denotes year. $ {L}_{i,t}^b=\frac{Loan_{i,t}^b}{\sum_{i=1}^I{Loan}_{i,t}^b} $ is bank $ b $ ’s portfolio share of syndicated loans toward industry $ i $ in the list of industries from 1 to $ I $ , at time $ t $ . $ {L}_{i,t}^{\ast } $ is the threshold to identify outliers in the distribution of $ {L}_{i,t}^b $ among all banks in industry $ i $ . For each industry, the threshold is the 75th percentile plus 1.5 times the interquartile range of the distribution of all banks’ portfolio shares in the industry (Hodge and Austin (Reference Hodge and Austin2004)).

There are at least two advantages to measuring lending specialization in a relative way. First, this method accounts for the heterogeneity in the size of different banks and industries. Specifically, scaling a bank’s loans to a given industry by the bank’s total loans makes the measure impervious to bank size. Comparing different banks’ loan shares within the same industry makes the measure impervious to industry size. Second, as we will discuss later, we include county-by-year fixed effects in our main empirical specifications to compare different banks’ mortgage lending in the same county. A relative measure enables us to focus on banks’ relative industry advantages in a county.

C. Measuring a County’s Industry Specialization

We use the employment information provided by the QCEW to identify key industries in a county. We exclude employment by government-owned entities and the financial industry and aggregate employment at the 3-digit NAICS level. An average county has 59 3-digit NAICS industries.Footnote ⁹ Figure 1 shows the employment shares by the top 20 industries in a county, which range from 1.12% to 19.35%. We classify industries that provide at least 5% of jobs in a county as the county’s specialized industries. Our choice of 5% ensures that an industry has a material impact on the local economy and household income. In total, these industries provide about 58% of jobs in an average county.

FIGURE 1 Average Employment Share by the Top-20 Industries in a County

Figure 1 shows the average employment share by the top-20 industries in a county.

D. Measuring the Industry Expertise Channel

Using industry specialization measures for each bank and county, we classify a bank and a county as connected through the industry expertise channel if the bank has one or more specialized industries that provide at least 5% of jobs in the county. Banks can use their industry expertise to better screen eligible mortgage borrowers and monitor their income risks, allowing them to extend more mortgage credits to local residents. By construction, variations in this channel come mostly from changes in the bank’s loan portfolio and the distribution of portfolio shares of other banks in the industry.Footnote ¹⁰

E. Sample and Summary Statistics

We aggregate mortgage data at the bank-by-county-by-year level. The sample consists of 78 unique banks with mortgage business in 3,165 counties from 1999 to 2017.Footnote ¹¹ Table 1 reports the summary statistics of the variables used in our empirical analyses. Panel A presents the county-level statistics, Panel B presents the bank-level statistics, Panel C presents the HMDA-based main sample at the bank-county level, and Panel D presents the matched bank-county-level sample between HMDA and monthly loan-level performance from the Fannie Mae, the Freddie Mac, and the McDash data sets. The sample period is 1999–2017, except that the county-level mortgage delinquency in Panel A is only available from 2008 to 2017.

TABLE 1 Summary Statistics

The average asset size of the banks in our sample is 174-billion USD, and the median is 51-billion USD, indicating that our sample predominantly covers large banks. The number of mortgages a bank approves in a county has a mean of 88.0 and a median of 19.0. The standard deviation is 193.8, suggesting large variations across bank-county pairs. The mean dollar volume (in millions) of approved mortgages is 14.4, and the median is 2.3. The average number-based mortgage approval rate is 74.6%, and the average volume-based mortgage approval rate is 75.6%. A total of 16.5% of the 316,552 bank-county pairs are connected through the industry expertise channel.

A. The Number and Volume of Approved Mortgages

We conjecture that industry expertise enhances banks’ abilities to assess household income risk and therefore reduces information frictions in mortgage decisions. The lower information asymmetry mitigates credit rationing, leading to more credit supply. We test this conjecture using the following empirical specification:

(2) $$ {Y}_{bct}={\pi}_{ct}+{\mu}_{bs}+\beta \hskip0.32em Industry\hskip0.32em {Expertise}_{bct}+\delta {\boldsymbol{X}}_{bct}+{\varepsilon}_{bct}, $$

where $ b $ denotes the bank, $ c $ denotes the home county of the borrower, $ s $ denotes the home state of the borrower, and $ t $ denotes the year. $ {Y}_{bct} $ is the natural logarithm of the number or dollar volume (in millions) of the mortgages bank $ b $ approves to borrowers in county $ c $ in year $ t $ . $ Industry\;{Expertise}_{bct} $ is a dummy variable equal to 1 for a bank-county pair if there exists at least one industry in which the bank $ b $ specializes and provides at least 5% of jobs in county $ c $ in year $ t $ . $ {X}_{bct} $ is a vector of controls, including the average LTI ratio of all mortgage applicants, the percentage of male applicants, the percentage of minority applicants, the natural logarithm of 1 plus the number of branches a bank has in the county, the natural logarithm of the geographic distance between the headquarters county of a bank and the borrower’s home county, the natural logarithm of 1 plus the number of small business loans that a bank originates in the borrower’s home county, the average fraction of mortgages retained in the balance sheets in the borrower’s home county in the past 3 years, the natural logarithm of bank assets, total loans scaled by assets, deposits scaled by assets, commercial and industrial (C&I) loans scaled by total loans, RE loans scaled by total loans, return on assets, and total liquidity scaled by assets. $ {\pi}_{ct} $ is county-by-year fixed effects, which allows us to compare different banks’ mortgage lending in the same county. $ {\mu}_{bs} $ is bank-by-state fixed effects, which controls for hidden links between banks and states, such as political rent-seeking (Chu and Zhang (Reference Chu and Zhang2022)).

We present the results of estimating equation (2) in Table 2. In column 1, the coefficient estimate on Industry Expertise is positive and statistically significant, indicating that industry expertise increases banks’ mortgage lending. The significance remains after adding mortgage-level or bank-level controls in columns 2 and 3. We use county-by-year fixed effects to replace borrower’s home county and year fixed effects in column 4, and bank-by-state fixed effects to replace bank fixed effects in column 5. The results continue to hold, and the estimated effect is economically significant. The result in column 5 suggests that industry expertise increases banks’ mortgage lending by 6.3%. Columns 6–10 repeat the analyses using the dollar volume of approved mortgages as the dependent variable. The findings are consistent with those in columns 1–5.

TABLE 2 Mortgage Lending Through the Industry Expertise Channel: Number and Volume

B. Robustness

In Supplementary Material Appendix IA.II, we conduct a series of additional tests to demonstrate the robustness of our results. First, we reconstruct the industry expertise channel by accounting for each borrowing firm’s market position, assuming that lending to industry leaders enhances banks’ expertise. Second, we develop two continuous measures that capture the intensity of connections between banks and counties through the channel: one based on the share of residents working in bank-specialized industries that provide at least 5% of county jobs, and the other based on the share working in any bank-specialized industry. Third, we revise our expertise measure as the difference between a bank’s industry loan share and the threshold $ {L}_{i,t}^{\ast } $ used to identify outliers in equation (1). Fourth, we include bank-by-year and bank-by-county fixed effects to control for time-varying bank characteristics and time-invariant bank-county links. The results remain robust. Fifth, we use both linear regressions and the fixed effects Poisson model (Cohn, Liu, and Wardlaw (Reference Cohn, Liu and Wardlaw2022) to address concerns with log-transformed dependent variables; the results are statistically significant and economically stronger. Sixth, we follow Blickle, Fleckenstein, Hillenbrand, and Saunders (Reference Blickle, Fleckenstein, Hillenbrand and Saunders2022) to estimate loan shares and reconstruct the industry expertise measure. Our findings remain robust.

C. Mortgage Approval Rates

Although the results hold after controlling for county-by-year and bank-by-state fixed effects, they could still be driven by demand-side factors. For example, certain households may prefer to borrow from a bank due to brand preferences or access to mobile apps. To alleviate this concern, we examine banks’ mortgage approval decisions conditional on received applications using approval rates, defined as approved mortgages (by number or volume) divided by applications received, as the dependent variable. The results are reported in Table 3. We find that, conditional on received applications, industry expertise increases both number- and volume-based approval rates by 40 BPS, implying that the findings in Table 2 are unlikely to be driven by demand-side factors.

TABLE 3 Mortgage Lending Through the Industry Expertise Channel: Approval Rates

Overall, the results in Section III support the conjecture that the industry expertise channel mitigates information asymmetry and, hence, credit rationing.

A. Industry Growth, Household Income, and Mortgage Delinquency

A prerequisite for the industry expertise channel is that the conditions of the key industries in a county are useful in assessing borrower credit quality (i.e., banks can use their industry expertise to assess local borrowers’ income dynamics and mortgage default probabilities). We test whether this is true using the following empirical model:

(3) $$ {Y}_{ct}={\theta}_c+{\tau}_t+\beta \hskip0.4em Sales\hskip0.4em {Growth}_{ct}+\delta {\boldsymbol{X}}_{ct}+{\varepsilon}_{ct}, $$

where $ c $ denotes the county and $ t $ denotes the year. $ {Y}_{ct} $ is the dependent variable, representing either the income growth rate or the annual change in the mortgage delinquency rate. $ Sales\;{Growth}_{ct} $ is the standardized employment-weighted industry sales growth rate in a county. The weights are the fractions of local residents working in a given industry. The sales growth rates for each industry are estimated using the sales of all U.S. public firms in that industry. $ {\boldsymbol{X}}_{ct} $ is a vector of county-level controls, including the natural logarithm of the population, the percentage of the population over 65, the percentage of the male population, the percentage of the minority population, and the percentage of the population with a bachelor’s degree or above. $ {\theta}_c $ is the county fixed effects, and $ {\tau}_t $ is the year fixed effects.

The results are reported in Table 4. The dependent variable in columns 1–3 is the income growth rate. The coefficient estimate on sales growth is positive and statistically significant, suggesting that faster industry growth is associated with greater growth in household income. The correlation is also economically significant. In column 3, a 1-standard-deviation increase in sales growth is associated with a 14.9% increase in household income growth. In columns 4–6, we examine the mortgage delinquency rate, an indicator of mortgage performance.Footnote ¹² Consistent with our expectation, the coefficient estimate on sales growth is negative and statistically significant, suggesting that faster industry growth is associated with lower mortgage delinquency rates.

TABLE 4 Industry Growth, Household Income, and Mortgage Delinquency

Overall, the findings suggest that growth in a county’s key industries is positively correlated with the county’s household income growth and negatively correlated with the county’s mortgage delinquency rate. The evidence builds the foundation for the key argument in this article: industry expertise enables banks to predict local household income dynamics and, therefore, mortgage default risks after origination.

B. Information Asymmetry

We then investigate the effect of information asymmetry on the industry expertise channel in mortgage lending. We start with the geographic distance between banks’ headquarters and mortgage borrowers. Previous studies show that long geographic distance erodes banks’ ability to acquire information, creating significant barriers for banks to reach distant borrowers (Agarwal and Hauswald (Reference Agarwal and Hauswald2010)). We expect that industry expertise mitigates information barriers and enables banks to extend mortgage credits to distant borrowers. We test this prediction in columns 1 and 3 of Table 5. Consistent with prior studies, mortgage credit declines with the distance between the banks’ headquarters and borrowers. More importantly, the effect of industry expertise increases with distance, more than doubling for a 1-standard-deviation increase. This suggests that industry expertise mitigates distance-related information frictions between banks and mortgage borrowers.

TABLE 5 Information Asymmetry and the Industry Expertise Channel

We also examine how soft information in the channel interacts with soft information banks collected from other sources. To this end, we use social networks as a proxy for alternative soft information in columns 2 and 4 of Table 5. Consistent with Rehbein and Rother (Reference Rehbein and Rother2020), social connections between a bank’s headquarters county and a borrower’s home county significantly increase banks’ mortgage lending. However, social connections decrease banks’ reliance on industry expertise. In column 2, a 1-standard-deviation increase in SCI is associated with a 31.3% decrease in the effect of industry expertise. The evidence further suggests that industry expertise offers additional soft information that can substitute for that from social connections.

C. Borrower Risk

Credit rationing caused by information asymmetry should be more severe for ex ante riskier borrowers. We therefore expect that the impact of industry expertise should be stronger for riskier borrowers. Our first proxy for borrower risk is the local house price volatility (Gerardi et al. (Reference Gerardi, Herkenhoff, Ohanian and Willen2018)). We report the results in columns 1 and 3 of Table 6. The coefficient estimates on the interaction term between Industry Expertise and HP Volatility, the standardized county-level housing price volatility, are positive and statistically significant, indicating that banks rely more on industry expertise when local house prices are more volatile. In column 1, the effect of industry expertise on mortgage lending increases from 6.7% to 11.4% for a 1-standard-deviation increase in house price volatility.

TABLE 6 Borrower Risk and the Industry Expertise Channel

Our second proxy for borrower risk is the LTI ratio. A higher LTI ratio indicates higher mortgage leverage and higher borrowing constraints. The results in columns 2 and 4 of Table 6 show that the effect of industry expertise is stronger for borrowers with higher LTI ratios. The estimate in column 2 suggests that the effect of industry expertise increases from 5.3% to 11.1% for a 1-standard-deviation increase in the LTI ratio.

D. Bank Size and Real Estate Loan Share

We also explore how banks’ asset sizes affect their use of the industry expertise channel in mortgage lending. Despite the limited size variation among the 78 banks in our sample, the results in columns 1 and 3 of Table 7 show that the effect of industry expertise is stronger for larger banks. This suggests that large banks rely more on industry expertise, consistent with prior findings that smaller, more localized banks have a comparative advantage in collecting soft information through other channels (e.g., Berger et al. (Reference Berger, Miller, Petersen, Rajan and Stein2005), Loutskina and Strahan (Reference Loutskina and Strahan2011)).

TABLE 7 Bank Size and Real Estate Loan Share and the Industry Expertise Channel

Additionally, we explore heterogeneities in banks’ business models, focusing on the importance of RE lending within their loan portfolios. We expect the effects to be more pronounced among banks with higher shares of RE loans, as their revenues are more closely tied to mortgage performance, and they are less aggressive in shifting risks through securitization. The results in columns 2 and 4 of Table 7 support our conjecture.

E. Soft Information in Mortgage Contracts

To provide more direct evidence, we test soft information contained in mortgage contracts by examining whether mortgages originated through the industry expertise channel are less standardized (i.e., greater dispersion in contractual terms). This is because better information allows banks to better distinguish between “good” and “bad” borrowers (Cornell and Welch (Reference Cornell and Welch1996), Rajan, Seru, and Vig (Reference Rajan, Seru and Vig2015)). As a result, banks can grant mortgages with favorable terms to “good” borrowers and mortgages with strict terms to “bad” borrowers. In contrast, when banks lack sufficient information, they rely on the quality of average borrowers and offer similar mortgage terms to all.

We construct four variables to capture the dispersion in terms of approved mortgage contracts: the natural logarithm of the standard deviations of the loan amounts, LTI ratios, interest rates, and LTV ratios (Fisman et al. (Reference Fisman, Paravisini and Vig2017), Lim and Nguyen (Reference Lim and Nguyen2021)). We present the results in Table 8. Consistent with our prediction, mortgages originated through the industry expertise channel have less standardized contractual terms. The standard deviations of loan amounts, LTI ratios, interest rates, and LTV ratios are 0.6%, 0.5%, 2.1%, and 2.2% higher for mortgages originated through the channel.Footnote ¹³

TABLE 8 Dispersion in Mortgage Contractual Terms

F. Conventional and Government-Insured Mortgages

Government-insured mortgages (i.e., FHA and VA loans) are less subject to credit rationing (Duca and Rosenthal (Reference Duca and Rosenthal1991), Ambrose, Pennington-Cross, and Yezer (Reference Ambrose, Pennington-Cross and Yezer2002)). Banks should therefore originate more conventional mortgages relative to government-insured mortgages in counties connected by the industry expertise channel if it mitigates credit rationing. To test this, we re-estimate equation (2) by extending the HMDA-based mortgage sample to include government-insured mortgages and replacing the dependent variable with the percentage of conventional loans originated in a county. The results are presented in Table 9, with columns 1–3 for the number-based percentage of conventional mortgages and columns 4–6 for the volume-based percentage. The coefficient estimates on Industry Expertise are all positive and significant, suggesting that banks increase conventional mortgages relative to government-insured mortgages in counties connected by the industry expertise channel, consistent with the argument that banks’ industry expertise mitigates credit rationing.

TABLE 9 The Percentage of Conventional Mortgages

G. Mortgage Performance

Lastly, we examine the effect of industry expertise on ex post mortgage performance. If it helps banks better screen applicants and monitor income risk, we expect improved mortgage outcomes. To test the performance implications, we focus on mortgage delinquency and foreclosure rates using the matched sample between HMDA and monthly loan-level performance from the Fannie Mae, Freddie Mac, and McDash data sets. Specifically, we track each mortgage’s monthly payment records to identify whether a mortgage ever had a 60-day-plus delinquency, a 90-day-plus delinquency, or a foreclosure. We aggregate loan-level data to the bank-county-year level and construct three outcome variables: Delinquency 60 Days, defined as the share of mortgages more than 60 days past due on monthly payments; Delinquency 90 Days, the share more than 90 days past due; and Foreclosure, the percentage of mortgages that have entered foreclosure proceedings.

The results are reported in Table 10. The coefficient estimate on Industry Expertise suggests a negative effect of industry expertise on subsequent mortgage delinquency and foreclosure rates. On average, mortgages originated by banks with industry expertise have 4.1% lower 60-day-plus delinquency rates, 4.0% lower 90-day-plus delinquency rates, and 4.8% lower foreclosure rates, respectively.

TABLE 10 Mortgage Delinquency and Foreclosure

In summary, our analyses show that banks increasingly rely on the industry expertise channel when borrower information is scarce or borrowers are riskier. Mortgages originated through the channel embed more soft information, as reflected in more dispersed terms. Banks also issue more conventional (vs. government-insured) loans in connected counties, consistent with reduced credit rationing. Finally, the channel is associated with lower delinquency and foreclosure rates. Together, these findings provide strong evidence for the information mechanism underlying the industry expertise channel.Footnote ¹⁴

A. Industry Distress

We first design a difference-in-differences test using unexpected industry distress, a sharp downturn in an industry accompanied by significant uncertainty and operational strain. This distress can negatively affect household income, particularly for those employed in the affected industry or living in the counties where it is concentrated. In severe cases, households can face layoffs and complete income loss.

Relevant industry expertise enables banks to better assess the duration and severity of industry distress and its implications for mortgage risk. As a result, these banks can more accurately price borrowers’ income risks and mitigate defaults (e.g., by timely selling high-risk mortgages to third parties). Consequently, the positive impact of industry expertise on mortgage lending should be stronger in distressed industries.Footnote ¹⁵ More importantly, industry-wide shocks are plausibly exogenous for any given bank, county, or mortgage borrower, mitigating the issues of omitted variables and reverse causality (Giannetti and Saidi (Reference Giannetti and Saidi2019), Babina (Reference Babina2020)).

We measure industry distress following previous studies (Opler and Titman (Reference Opler and Titman1994), Babina (Reference Babina2020)). Specifically, we classify a 3-digit NAICS industry as distressed in a year if the industry-level 2-year sales growth is negative and the industry-level 2-year stock return is less than −10% from the beginning of that year. For robustness checks, we also use two additional stock return thresholds: −20% and −30%. We then compare the effects of industry distress on mortgage lending with differential ex ante industry specializations using the following model:

(4) $$ {\displaystyle \begin{array}{c}{Y}_{bct}={\pi}_{ct}+{\mu}_{bs}+{\tau}_{bt}+{\beta}_1\hskip0.32em Industry\hskip0.32em {Expertise}_{bct-2}\times {Distress}_{bct-1}\\ {}\hskip1.32em +{\beta}_2\hskip0.32em Industry\hskip0.32em {Expertise}_{bct-2}+\delta {\boldsymbol{X}}_{bct-2}+{\varepsilon}_{bct},\end{array}} $$

where $ b $ denotes the bank, $ c $ denotes the borrower’s home county, $ s $ denotes the borrower’s home state, and $ t $ denotes the year. $ {Y}_{bct} $ is the dependent variable: the natural logarithm of the number or the dollar volume of mortgages (in millions) bank $ b $ approves to borrowers in county $ c $ in year $ t $ . $ Industry\;{Expertise}_{bct-2} $ is a dummy variable equal to 1 for a bank-county pair if there exists at least one industry in which bank $ b $ specializes and provides at least 5% jobs in county $ c $ , measured at $ t-2 $ . $ {Distress}_{bct-1} $ is a dummy variable that equals 1 for a bank-county pair if distress happens in any of the industries in which bank $ b $ specializes and provides at least 5% of jobs in county $ j $ , measured at $ t-1 $ .Footnote ¹⁶ In addition to county-by-year fixed effects $ {\pi}_{ct} $ and bank-by-state fixed effects $ {\mu}_{bs} $ , we also add bank-by-year fixed effects $ {\tau}_{bt} $ to account for the potential negative effects of industry distress on bank capital.

Table 11 presents the results. Columns 1 and 4 use a return threshold of −10%, columns 2 and 5 use −20%, and columns 3 and 6 use −30%. The coefficient estimates on the interaction term between Industry Expertise and Distress are positive and statistically significant, suggesting that banks rely more on their industry expertise in distress periods. In column 1, the effect of industry expertise on mortgage lending rises from 2% in non-distress periods to 6.4% during distress. The incremental effect grows with distress severity (rising from 4.4% to 5.6% when tightening the return threshold from −10% to −30%, a 27% increase). Similar patterns are observed for mortgage volumes.

TABLE 11 Industry Distress and the Industry Expertise Channel

B. The 2008 Financial Crisis

We use the 2008 financial crisis as an alternative shock, given that mortgages and housing markets were central to the recession. From 2007 to 2009, national house prices fell by more than 10%, and average delinquency rates on single-family mortgages rose from 1.84% (2004–2007) to 7.04% (2008–2009).Footnote ¹⁷ These widespread defaults resulted in substantial losses for banks. A key driver was the fraudulent overstatement of income in mortgage applications (Mian and Sufi (Reference Mian and Sufi2017)). Banks should be more cautious in screening mortgage borrowers during and after the crisis. Therefore, industry expertise should become more valuable in mortgage underwriting.

We design a difference-in-differences test to assess how the crisis influenced banks’ use of industry expertise in mortgage lending from 2004 to 2010, using the following model:

(5) $$ {\displaystyle \begin{array}{c}{Y}_{bc t}={\pi}_{ct}+{\mu}_{bs}+{\tau}_{bt}+\hskip0.32em {\beta}_1\hskip0.32em Industry\hskip0.32em {Expertise}_{bc2003}\times {Crisis}_t\\ {}\hskip1.3em +\hskip0.32em {\beta}_2\hskip0.32em Industry\hskip0.32em {Expertise}_{bc2003}+\delta {\boldsymbol{X}}_{bc t}+{\varepsilon}_{bc t},\end{array}} $$

where $ b $ denotes the bank, $ c $ denotes the borrower’s home county, $ s $ denotes the borrower’s home state, and $ t $ denotes the year.Footnote ¹⁸ $ {Y}_{bct} $ is the dependent variable measuring mortgage lending, and $ Industry\hskip0.3em {Expertise}_{bc2003} $ captures the industry expertise channel as measured in 2003. $ Crisis $ is an indicator variable equal to 0 for the period of 2004–2007, and 1 for the period of 2008–2010. $ {\pi}_{ct} $ denotes the county-by-year fixed effects, $ {\mu}_{bs} $ denotes the bank-by-state fixed effects, and $ {\tau}_{bt} $ denotes the bank-by-year fixed effects.

Table 12 presents the results. Columns 1 and 3 show that industry expertise positively affects mortgage lending prior to the crisis. Banks’ reliance on the channel increases from 3.8% to 12.3% in column 1, and from 4.7% to 12% in column 3. In columns 2 and 4, we break down the Crisis dummy into year dummies. The year 2007 is the base year and thus omitted. The coefficient estimates on the interaction terms Industry Expertise $ \times $ Year 2004, Industry Expertise $ \times $ Year 2005, and Industry Expertise $ \times $ Year 2006 are not statistically significant, suggesting that the effect of industry expertise on mortgage lending is stable before the crisis. In 2009, the effect of industry expertise increases by 14.2% and 12.6% in columns 2 and 4. The effect slightly decreases in 2010 after the peak of the crisis, but it is still positive and significant. Figure 2 shows the dynamics of the coefficient estimates.

TABLE 12 The 2008 Financial Crisis and the Industry Expertise Channel

FIGURE 2 The 2008 Financial Crisis and the Industry Expertise Channel

Figure 2 shows the dynamic treatment effects of the 2008 financial crisis on banks’ use of the industry expertise channel in mortgage lending. Graphs A and B show the effects on the number and volume of approved mortgages, respectively. The regression results behind the graphs are reported in columns 2 and 4 of Table 12.

In summary, the findings suggest that industry expertise becomes more valuable during periods of greater uncertainty, when income risk is more pronounced. Importantly, these tests help address endogeneity concerns, supporting a causal interpretation of the industry expertise effect on mortgage lending.