Does Banking Consolidation Harm Households?

Abstract:

1 Introduction

Bank consolidation has reduced the number of active banks and thrifts by nearly 70 percent (from 14,417 to 4,379 institutions) between 1985 and 2020. Moreover, asset concentration has increased dramatically as well, with the number of banks controlling half of all banking assets falling from 91 to just 10 institutions over this period.¹ This consolidation has intensified regulatory and policymaker concerns that bank mergers erode competition and harm consumers.² These concerns are particularly salient for mortgage lending, the primary source (more than two-thirds) of American household debt.³ In this study, we directly examine bank consolidation and mortgage markets, finding mortgage rates, approval rates, and loan performance all remain stable after bank mergers, challenging the view that bank mergers systematically harm households by exercising market power.

Empirically, we identify the merger effect by employing a stacked panel event study difference-in-differences (DiD) design that compares the acquiring and target banks to control institutions within the same local markets before and after the merger. We find that changes in rates, approval rates, and delinquency rates in simple pre-post comparisons reflect portfolio composition effects rather than strategic responses to reduced competition. These composition effects arise naturally. For instance, when a large bank charging low rates acquires a community bank charging higher rates, post-merger rates mechanically rise. Examining rate distributions confirms this interpretation: acquiring banks’ rate distributions do not change around mergers.

Although measuring the welfare effects of consolidation is difficult in practice, we compile a unique dataset that allows us to jointly observe mortgage pricing, application outcomes, and subsequent loan performance at scale and over time. Our data allows us to distinguish true strategic responses from changes in borrower and bank characteristics. More specifically, we combine banking merger data for nearly 5,000 merger events with confidential borrower data for virtually all (44 million) U.S. mortgages initiated over nearly three decades, ensuring our results hold across diverse market conditions and regulatory regimes. This granularity allows us to examine post-merger prices, approval rates, and loan performance changes that help to separate strategic responses from composition effects.

At a high level, it is perhaps not surprising that we find no evidence of harm despite decades of bank consolidation. Despite extensive consolidation at the national level, the typical county hosts over 130 active lenders in any given quarter. Market shares are highly fragmented—the median lender holds just 0.4 percent—and Herfindahl-Hirschman Index (HHI) levels remain well below the 1,000 threshold used in deposit-market merger review to classify markets as unconcentrated. In fact, competition often intensifies around merger events: when large banks acquire community banks, the number of active lenders in affected counties rises from about 150 to 185 and county-level HHI declines from about 560 to 470–510, patterns inconsistent with strategies to enhance market power.

Our analysis of pre-merger characteristics reveals systematic sorting among merger types that also challenges the market-power story. We find that community banks acquired by large institutions look fundamentally different from those acquired by other community banks: community banks targeted by larger institutions keep 41 percent of their mortgages in portfolio—more than ten times the four percent rate of community–community targets. Moreover, community bank targets of larger institutions are more profitable and more mortgage-intensive (per employee), offer fewer Federal Housing Administration (FHA) mortgages and loans with higher rates than community–community targets. These differences point to distinct merger motives where large banks target small relationship-intensive lenders to access established customer bases while providing scale for securitization and funding. Community bank acquirers, by contrast, target relatively weaker peers, adding scale rather than opportunities to exercise market power.

Our findings should not be interpreted as evidence that all proposed bank mergers are harmless: they describe completed mergers, after regulatory review. Whether the absence of adverse mortgage-market effects reflects this regulatory filter, selection into markets with limited competitive concerns, or other merger-selection mechanisms is beyond the scope of our analysis.

Our study contributes to the literature on bank mergers and consumer welfare in three ways. First, we simultaneously examine pricing, credit access, and loan performance with the most comprehensive data to date (44 million mortgages spanning nearly 5,000 merger events over nearly thirty years) coupled with borrower-level characteristics to achieve internal validity that studies focusing on single outcomes cannot. Second, we offer novel evidence on local mortgage markets, documenting lender counts, market shares, and concentration metrics that reflect vigorous competition despite industry consolidation. Although policymakers have intensified scrutiny of bank mergers over market power concerns, our evidence suggests mortgage markets remain strongly competitive. Third, we characterize the pre-merger attributes of acquiring and target institutions, revealing systematic selection patterns consistent with efficiency considerations. Methodologically, our DiD design with bank-county-merger fixed effects isolates bank behaviors from mechanical composition effects, demonstrating that rate changes in simple pre-post comparisons reflect portfolio averaging rather than strategic pricing.

Our results are broadly consistent with studies documenting efficiency motives in mergers Sapienza, 2002; Focarelli and Panetta, 2003; Panetta et al., 2009; Erel, 2011; Akkus et al., 2016; Levine et al., 2020.⁴ Relatedly, Kovner et al., 2014 document economies of scale in bank operating costs and Minton et al., 2024 show that community bank consolidation can boost local investment.⁵ Although Huber, 2021 shows that larger banks can harm some borrowers in Germany and Garmaise and Moskowitz, 2006 link merger activity to tighter credit conditions, we find no adverse effects from U.S. bank mergers. Indeed, Park and Pennacchi, 2009 show that mergers can intensify loan competition. Allen et al., 2014 examine Canadian mergers and find price dispersion depends on local market structure—consistent with our finding that mergers in fragmented U.S. markets do not increase rates. Our evidence on business model sorting—large banks targeting portfolio lenders, community banks merging with high-cost peers—echoes Erel et al., 2012, who find efficiency motivations dominate merger selection.

Our work also contributes to the literature on consolidation and credit access. Our stable approval rates provide direct micro-level evidence consistent with aggregate patterns found by Jayaratne and Strahan, 1996, Strahan and Weston, 1998, Black and Strahan, 2002 and Beck et al., 2010. However, our findings contrast with Nguyen, 2019, who document that branch closures reduce credit supply—suggesting merger effects differ from physical contraction of bank presence.

The market structure evidence in the study complements Syverson, 2019 and De Loecker et al., 2020, who argue that concentration measures provide an imperfect guide to market power, and Saidi and Streitz, 2021, who show that competition and concentration measures need not map one-to-one into competitive outcomes. Benson et al., 2024 examine how geographic proximity affects antitrust assessments of bank mergers. Our analysis of rates, approval, and delinquency across a comprehensive sample of mergers provides evidence on actual borrower welfare effects beyond concentration measures. Banks may adjust their post-merger strategies differentially across markets. Our county-level analysis captures this heterogeneity by examining outcomes separately within each local market. Lastly, Corbae et al., 2026 model the post–Riegle-Neal merger wave and find that, while rising concentration can elevate loan markups, dominant banks’ geographic expansion—from roughly four to thirty-five states on average—also diversifies risk and is partly offset by non-bank competition. Their concentration measurement is deposit-based and their loan-side conclusions model-implied; we provide direct borrower-level evidence on mortgage rates, approvals, and performance, and show that the geographic-expansion mechanism is consistent with the intensified local competition we document around large–community mergers.

2 Data

We use confidential Home Mortgage Disclosure Act (HMDA) data, which reports the universe of mortgage applications and originations from regulated financial institutions.⁶ From over 350 million HMDA mortgage records, we extract detailed information on application outcomes, including application, origination and closing dates, as well as loan amounts, and originator identities.

We aggregate mortgage origination at the bank-parent company level using Robert Avery’s linking file to properly account for corporate structures and mergers.⁷

We merge confidential HMDA data with the McDash mortgage servicing dataset from ICE McDash Analytics (Intercontinental Exchange), which provides both loan-level contract terms (interest rates, loan-to-value (LTV) ratios, mortgage duration) and dynamic performance information throughout the loan life cycle. McDash covers mortgages originated between 1994 and 2023, with monthly observations tracking each loan from origination through resolution (e.g., refinancing, delinquency, etc.), allowing us to track delinquencies, foreclosures, prepayments, and other performance metrics, including whether a mortgage remains on the originator’s balance sheet (portfolio loans), has been securitized, or sold.

After linking confidential HMDA to McDash and filtering out incomplete records, we obtain approximately 62.6 million unique loans with comprehensive information on origination characteristics and subsequent performance.⁸ After applying all filters to ensure data quality, our final sample contains approximately 44.3 million unique mortgages from 1994 until 2023.

We collect bank merger information from the Federal Reserve’s National Information Center (NIC) database, which provides comprehensive coverage of bank acquisitions and regulatory approvals.⁹ The NIC data includes identities of acquiring and target institutions (Research, Statistics, Supervision, Discount (RSSD) identifiers), transaction announcement and completion dates, pre-merger asset sizes for both acquirer and target, and transaction types and regulatory approval status. We identify 4,829 distinct merger events occurring between 1996–2021 (two years after the start and two years before the end of our HMDA/McDash data availability) involving both horizontal mergers (bank-to-bank) and bank acquisitions of non-bank mortgage originators.

Figure 1 displays bank mergers over time, averaging 1,080 per year from 1996–2021. After restricting our analysis to mergers between HMDA filers only, our sample includes about 186 bank mergers per year on average (about 17 percent of all mergers).

Within our sample, we classify banks as community banks or large/regional banks by size, based on total assets following thresholds established by the Dodd-Frank Act in 2010 (with inflation adjustments). Community banks have total assets below $10 billion, regional banks have total assets between $10 billion and $100 billion, and large banks have total assets above $100 billion.¹⁰

For tractability in our empirical analysis, we consolidate large and regional banks into a single category. This grouping reduces computational complexity while maintaining meaningful size distinctions, as both large and regional banks differ substantially from community banks in terms of organizational structure, geographic footprint, and lending operations so we contrast between larger, diversified institutions and smaller, locally-focused community banks.¹¹

Table 1 presents the distribution of mergers by acquirer and target size. The banking consolidation landscape is dominated by horizontal mergers within size classes, with community–community mergers representing the largest category (2,642 cases, 55 percent of all mergers). Community banks are disproportionately acquisition targets, accounting for 90 percent of all targets (4,325 institutions), highlighting why concerns have been raised about consumer harm. Our final sample comprises 2,642 and 2,187 community and large bank acquirers, respectively. These mergers span 2,683 distinct counties.¹²

We also obtain bank financial information from the Federal Reserve’s merger-adjusted call reports (TINY database), which provide quarterly balance sheet and income statement data adjusted for merger activities. These data provide total assets and liabilities, income statement items including net interest income and non-interest income, loan portfolio composition, and various regulatory measures and capital ratios. The merger-adjusted data are essential for this analysis, as they account for restructuring and asset transfers around bank mergers, allowing us to measure bank characteristics more accurately and more consistently.

Our primary analytical sample is a merger-bank-county-quarter panel designed for stacked event study DiD estimation, tracking mortgage outcomes for acquiring banks, target banks, and control banks across a 17-quarter event window spanning merger completion.¹³ We include mortgages by target banks up to two quarters before the merger completion date to avoid potential misclassification issues (HMDA data reports some mortgages originated prior to the completion date under the acquirer’s identifier while the merger is in progress).

To analyze mortgage market competitiveness, we construct a county-quarter panel dataset aggregating all mortgages at the local market level to calculate market structure measures including lender counts, HHI, and individual lender market shares. For this analysis, we use the universe of loan originations in confidential HMDA, including both bank and non-bank lenders, allowing us to capture the full competitive landscape and address sample-selection biases. This data shows that local mortgage markets remain highly fragmented despite extensive industry consolidation. The typical county features more than 100 different lenders originating mortgages in any given quarter, resulting in HHI levels well below regulatory concentration thresholds.

We also construct bank-county-quarter data tracking each institution’s lending activity, market share, portfolio retention rates, and product composition, and bank-year data from merger-adjusted call reports providing balance sheet and income statement information. These data reveal that targets exhibit characteristics suggesting competitive weakness—such as relatively high expense ratios, lower profitability, and, for community banks acquired by large institutions, substantially higher portfolio retention rates.

3 Methodology

Our empirical strategy exploits the geographic and temporal variation in bank merger activity to identify effects on mortgage market outcomes. Banks operate across multiple counties, with large banks having a wide presence, whereas community banks focus on local markets.¹⁴ This geographic dispersion creates significant heterogeneity in acquirer-target overlap across counties during merger events, providing valuable variation in treatment intensity across banks, counties, and time periods. We concentrate our analysis on counties where the target institution had non-zero market share in the eight quarters before the merger, i.e., the markets most likely to experience changes in competitive dynamics.¹⁵

To examine the impact of mergers on the cost of credit to households, we first construct a residualized interest rate measure that isolates pricing variation unrelated to observable loan risk characteristics and market conditions. We calculate the residual interest rate \(r^*\) as follows:

\[\begin{align} r_{i,b,c,t} = \text{X}_{i,b,c,t} \beta \, + \, \alpha_{c,t} \, + \, r^*_{i,b,c,t} \ ,\tag{1} \end{align}\] where \(r_{i,b,c,t}\) is the raw mortgage rate for loan \(i\) issued by bank \(b\) in county \(c\) at time \(t\); \(X_{i,b,c,t}\) includes comprehensive loan characteristics such as applicant income, FICO credit scores (grouped into five-point bins), LTV ratios (grouped into integer percentage bins), mortgage amount, loan duration and term indicators, fixed rate indicator, home purchase indicator, loan type classifications (FHA, VA, conventional, etc.), and occupancy status (owner-occupied, investment, second home); \(\alpha_{c,t}\) are county \(\times\) year-quarter fixed effects that absorb all time-varying market-level factors affecting interest rates; and \(r^*_{i,b,c,t}\) is the residual rate (in percentage points) after controlling for these factors.¹⁶ By controlling for loan risk and market factors, we isolate the specific effects of the mergers from differences in mortgage attributes. With outcomes expressed at the merger–bank–county–quarter level, we then estimate dynamic treatment effects around merger completion using a stacked event-study DiD.

We examine three primary outcomes around merger events: residualized interest rates (\(r^*\)), approval rates, and delinquency rates. Our analysis combines nonparametric event study plots that illustrate the dynamic evolution of these variables around merger completion with parametric DiD specifications that quantify treatment effects.

Our parametric approach employs a stacked panel event study DiD design. The regression model is:

\[\begin{align} Y_{e,c,t,b} = \sum_{\tau=-8}^{8} \beta_{\tau} \cdot \mathbf{1}[t = t^* + \tau] \times \text{Acquirer}_{e,c,b} + \alpha_{b,c,e} + \gamma_{t,e} + \delta_{\tau} + \epsilon_{e,c,t,b} \ ,\tag{2} \end{align}\] where \(Y_{e,c,t,b}\) is the outcome variable (residual interest rate, approval rate, or delinquency rate) for merger event \(e\), county \(c\), calendar quarter \(t\), and bank \(b\). The coefficients of interest \(\beta_{\tau}\) measure the treatment effect at event-time quarter \(\tau\) relative to merger completion. The indicator \(\mathbf{1}[t = t^* + \tau]\) equals one when being \(\tau\) quarters from the merger completion date \(t^*\), and the treatment indicator \(\text{Acquirer}_{e,c,b}\) equals one for the treated (merged) entity and zero for control banks. In the stacked panel, pre-merger observations for the target bank are assigned to the acquirer’s identifier, so that the treated bank–county cell reflects the combined entity throughout the event window. The bank–county–merger fixed effect \(\alpha_{b,c,e}\) thus absorbs the level of the pro-forma merged entity, and \(\beta_{\tau}\) captures deviations from this combined baseline rather than shifts in portfolio composition at the merger date. We include calendar quarter-merger fixed effects \(\gamma_{t,e}\) that absorb common time trends within each merger event, and event-time fixed effects \(\delta_{\tau}\) that control for systematic patterns in outcomes relative to merger timing. The error term is denoted by \(\epsilon_{e,c,t,b}\).

This stacked event study design offers several advantages. First, it provides an intuitive counterfactual by comparing treated banks to control banks operating in the same markets during the same time periods. Second, it addresses concerns about negative weighting in staggered DiD designs with heterogeneous treatment effects Goodman-Bacon, 2021; Callaway and Sant’Anna, 2021. Third, the fixed effects structure ensures that our estimates are identified from within-event variation, and not from aggregate trends or cross-sectional differences between merger events.

If the underlying motive for the merger is to gain market power, the coefficients \(\beta_{\tau}\) for \(\tau \in (1, 2, \dots, 8)\) will be positive and statistically significant—indicating that households face higher interest rates after the merger.¹⁷

This empirical design follows the framework in Benson et al., 2024 and addresses several challenges to identification. The stacked DiD design draws identification from within-event variation, so that each merger serves as its own quasi-experiment. The event-specific fixed effects \(\gamma_{t,e}\) ensure that merger-specific demand or cost shocks do not confound estimates across events, and the bank-county-merger fixed effects \(\alpha_{b,c,e}\) allow each institution-market pair to have persistent, potentially unobservable differences from other pairs in the same event. By stacking event-specific balanced panels, we eliminate the negative weighting problem that arises when treatment timing and effect heterogeneity interact De Chaisemartin and D'Haultf\oeuille, 2020.

Our empirical design rests on important assumptions. The parallel trends assumption still must hold within each merger event: outcomes for treated banks must evolve similarly to those for control banks in the counterfactual absence of the merger. Our evidence supports this assumption: we find flat pre-merger coefficients \(\beta_{\tau}\) for \(\tau < 0\), indicating acquiring and control bank outcomes move in tandem before treatment. Since we exclude counties experiencing overlapping mergers and control banks involved in concurrent mergers, we avoid contamination from multiple treatments. Lastly, differences in bank size, geographic overlap, and strategic motives, mean that our estimates should be interpreted as average treatment effects across the events in our sample rather than as single uniform causal parameters.

Our approach relies on a transparent counterfactual: for each merger event, we compare the merging bank’s outcomes to those of all other banks originating mortgages in the same county and quarter. These lenders face the same concurrent local housing and credit conditions, so their behavior provides a natural benchmark for how pricing, approval decisions, and subsequent loan performance would have evolved absent the merger. Although no control group is perfect, our combination of within-county-quarter comparisons and the fixed effects in (2) delivers an economically intuitive test of whether mergers alter these key lending decisions.

4 Main results

Bank mergers could affect household welfare by changing the price of credit or the availability of credit. We therefore study three outcomes around merger completion: residualized mortgage rates, approval rates, and mortgage repayments. We present both average event-study plots (to visualize dynamics) and quantified results (from stacked event-study DiD specifications with rich fixed effects regressions).

4.1 Bank mergers and mortgage rates

We track residualized rates (\(r^*\), see equation (1)) that control for borrower and loan characteristics as well as market conditions over a 17-quarter window spanning eight quarters before and eight quarters after bank mergers. We focus on rates offered by three distinct groups: acquiring banks, target banks, and control banks.

We exploit the richness of our data to analyze mergers at a granular level, distinguishing between large and community banks, recognizing that the competitive effects and motivations for consolidation may differ systematically by bank size. This stratification is important because community banks and large banks often serve different market segments, employ different origination and underwriting technologies, and differ in loan retention/securitization Kovner et al., 2014; Minton et al., 2024; Jacewitz and Kupiec, 2012.¹⁸

Figure 2 presents the evolution of average residualized (or abnormal) interest rates around merger events. Each panel corresponds to a different type of merger based on the sizes of the acquiring and target institutions, and each displays three series: acquiring banks (red dashed line), target banks (green short-dashed line), and control banks (blue solid line). The vertical line at quarter zero denotes the merger completion date.

Panel (a) examines mergers between two community banks, the most common type in our sample. We observe three main patterns within community—community bank mergers. First, acquiring banks offer rates approximately five basis points lower than control banks prior to the merger. This difference suggests community bank acquirers price somewhat more competitively, although the difference is economically modest. Second, targeted community banks also offer slightly more competitive pre-merger rates than control banks (the difference averages three basis points). Third, rates across all three groups remain remarkably stable throughout the post-merger period. We find no evidence of divergence between acquirers and control banks, and no indication that merged community banks adjust pricing during the eight post-merger quarters.

Panel (b) presents a more complex picture when large banks acquire community banks. These mergers represent potentially significant shifts in market structure, as a local community bank gives way to a larger institution. During the pre-merger period, large bank acquirers offer slightly lower rates than control banks, mirroring the pattern in panel (a). However, community banks targeted by larger banks offer pre-merger rates that are 20-30 basis points higher than those of both large bank acquirers and control banks. Following these mergers, the average mortgage rate for the merged institution increases modestly, converging toward control bank rates. This modest rate increase reflects the mechanical effect of combining two lending institutions that offer different pre-merger rates and not market power.

Consider, for example, a large bank originating 1,000 mortgages per quarter at an average residualized rate of zero basis points acquiring a community bank originating 100 mortgages per quarter at +25 basis points. If both institutions maintain their pre-merger pricing strategies, the post-merger average rate increases mechanically by averaging the two portfolios. Indeed, the observed rate increase in panel (b) could reflect (i) a purely mechanical effect, (ii) that large banks exploit increased market power by raising their own rates, or (iii) that large banks are offering lower rates to target bank customers. We distinguish among these mechanisms with our DiD analysis below.¹⁹

Panel (c) plots interest rates around mergers when large banks acquire other large banks. These transactions represent economically significant consolidation given the size of the institutions involved in the merger. With the exception of the first two pre-merger quarters, large bank acquirers, targets, and control banks offer similar average pre-merger residualized rates. This similarity aligns with the hypothesis that large banks operate with similar pricing strategies regardless of acquirer or target status. Post-merger rates remain largely stable for all three groups, with a modest decline of approximately three to four basis points during the last four post-merger quarters.

The absence of any sustained increase in rates following large bank mergers suggests that these transactions do not systematically lead to the exercise of market power, as measured by changes in average mortgage rates. Still, the graphical evidence from Figure 2 carries important limitations since simple rate pre-/post- comparisons do not control for unobservable time-varying factors that may differentially affect treated and control banks. For example, if acquiring banks systematically enter mergers when facing particularly favorable local market conditions, post-merger rate stability might reflect these improving conditions rather than the absence of market power effects.

We present formal DiD estimates that address these concerns in Figure 3. Each panel displays treatment effect coefficients (\(\beta_\tau\)) measuring how acquiring bank residualized rates evolve relative to control banks after accounting for bank-county-merger and time fixed effects. This specification explicitly controls for the mechanical averaging problem by comparing changes within the same bank-county pairs over time.

Panel (a) of Figure 3 shows the stacked event-study DiD estimates, \(\beta_{\tau}\), for mergers between community banks. The pre-merger coefficients are small and statistically indistinguishable from zero, supporting the parallel-trends assumption. Post-merger, \(\beta_{\tau}\) remains close to zero and statistically insignificant through eight quarters. These dynamics corroborate the evidence in Figure 2: community-bank consolidations are not associated with systematic changes in mortgage rates. Any pre-merger level differences shown in Figure 2 panel (a) are absorbed by fixed effects, and we find no treatment effect on \(r^*\) after the merger.

Panel (b) addresses the identification challenge highlighted in Figure 2 panel (b), where targets priced mortgages substantially above acquirers, creating scope for misleading composition effects in simple averages. In our stacked DiD specification, pre-merger coefficients are largely insignificant—with the exception of quarters -8, -7, and -3—indicating that, conditional on bank-county-merger fixed effects, merged and control banks follow similar pre-trends. Post-merger estimates of \(\beta_{\tau}\) are not statistically significant, indicating no material effect on rates. Overall, we find no significant increase in \(r^*\) around bank mergers. These patterns are consistent with integration dynamics and not the exercise of market power.

Panel (c) presents the dynamic effects for mergers between large banks. As in the other cases, pre-merger \(\beta_{\tau}\) estimates do not display systematic trends. Following the merger, point estimates remain clustered near zero with a small, temporary rise in the first two quarters, followed by small negative (and generally insignificant) coefficients later in the window. Taken together with the level patterns in Figure 2, these event study results indicate that large-bank mergers do not result in persistent mortgage rate increases. If anything, longer-horizon estimates suggest modest decreases in post-merger rates.

Across all three merger types, our stacked panel DiD reveals no evidence of persistent, post-merger increases in residualized mortgage rates. The \(\beta_{\tau}\) profiles are centered near zero, pre-trends are flat, and we can rule out economically large effects. These results confirm that the level plots are confounded by acquirer–target composition effects, particularly in large–community bank mergers.²⁰ The combined evidence indicates that bank mergers do not systematically raise mortgage rates and any short-run changes are transitory and economically small.

Tables 2 and 3 decompose market and borrower attributes by merger type and provide further corroborating evidence. Table 2 shows modest post-merger increases in lender fees, ranging from $86 to $183, with the largest adjustment occurring in community–community mergers. These fee adjustments are economically small relative to total loan costs and coincide with changes in interest rate spreads that suggest a partial shift in revenue sources rather than an increase in overall borrowing costs. Interest rate spreads decline in community–community mergers (from 3.42 to 3.13 pp) and remain broadly stable in large–large mergers, though acquirers in large–community mergers show a modest increase (from 3.16 to 3.23 pp). These movements are not statistically significant.

To complement the summary evidence, we also estimate our stacked event-study DiD specification in equation (2) using lender fees as the dependent variable.²¹ Across all merger types, the post-merger treatment effect coefficients are small and statistically insignificant. These results confirm that any level changes in fees reflect compositional shifts rather than strategic price increases.

Table 3 indicates that target banks serve distinct borrower segments, as reflected in substantially higher portfolio retention rates (0.41 for community targets acquired by large banks versus 0.09 for acquirers) and lower FHA shares (0.11 versus 0.16 for acquirers). These patterns suggest that target institutions are more inclined to retain loans on their balance sheets rather than securitizing them through government programs. The differences are particularly pronounced comparing community–community and large–large mergers, underscoring that any merger effects vary meaningfully with institutional scale and business model alignment.

4.2 Bank mergers and mortgage approval rates

To examine whether and how bank mergers affect mortgage approval rates we employ a similar empirical strategy as in the previous subsection, examining nonparametric event studies coupled with a DiD analysis. Approval rates provide a distinct lens into merger effects compared to pricing: whereas interest rates reflect competitive intensity and cost pass-through, approval decisions capture changes in credit standards and risk appetite.²² Prior research documents that bank mergers can lead to lower approval rates, particularly for vulnerable borrower groups Ratnadiwakara and Yerramilli, 2022, making this dimension critical for assessing consumer welfare effects. As with interest rates, we examine approval rates by merger type, recognizing that community banks and large banks may exhibit fundamentally different lending cultures and underwriting approaches.

Figure 4 presents the evolution of average mortgage approval rates around merger events, with each panel corresponding to different acquirer–target combinations. As before, we display three series: acquiring banks (red dashed line), target banks (green short-dashed line), and control banks (blue solid line), with the vertical line at quarter zero marking merger completion.

Panel (a) examines mergers between community banks. Pre-merger approval rates are slightly higher for target community banks (89 percent) than for acquiring and control banks (87 percent), a pattern consistent with either relatively looser underwriting standards or differences in borrower composition. Post-merger approval rates decline modestly to roughly 87 percent, aligning with pre-merger approval rates of the acquirers. Approval rates among control banks remain highly stable throughout. The post-merger convergence toward acquirer approval levels is consistent with the imposition of acquirer underwriting standards on target institutions and pre-/post- differences remain economically small.

Panel (b) presents a more pronounced pattern for large bank acquisitions of community banks. Pre-merger approval rates for control banks are 87 percent, followed by large acquirers (82 percent) and community targets (78 percent). Post-merger approval rates at acquiring banks decline from 82 percent to about 78 percent, converging toward the pre-merger levels of the targets, whereas control banks remain stable. Although this pattern could be consistent with a tightening of lending standards by large banks after acquiring community institutions, it may also reflect changes in borrower composition or portfolio mix rather than shifts in underwriting per se. In any case, the difference between pre- and post-merger approval rates remains economically small.

Panel (c) examines mergers between large banks. Control banks display notably higher pre-merger approval rates (86 percent) than both acquirers and targets (76–77 percent). Post-merger changes are limited, with approval rates at acquiring banks remaining close to 75 percent. Taken together, Figure 4 provides little evidence of economically meaningful changes in approval behavior following bank mergers.

Of course, raw level comparisons may conflate merger effects with compositional or market-wide changes. To isolate any causal impact, Figure 5 presents formal DiD estimates that control for bank-county-merger fixed effects and time-varying shocks. Each panel displays treatment effect coefficients (\(\beta_\tau\)) measuring how acquiring bank approval rates evolve relative to control banks after accounting for compositional effects.

Panel (a) shows that for community bank mergers, our DiD estimates remain close to zero throughout the post-merger period, with tight confidence intervals that include zero at all horizons. This pattern is consistent with the interest rate evidence reported in Figure 3, Panel (a). Overall, these estimates indicate community bank mergers do not produce statistically significant changes in approval rates.

Panel (b) displays DiD estimates for large banks acquiring community banks, with post-merger point estimates trending slightly negative but mainly statistically insignificant. These results show that the decline in approval rates visible in our levels analysis results from factors other than merger-induced policy changes such as time-varying local economic conditions or an evolution in acquirer lending behavior that would have occurred absent the merger. Panel (c) examines large-large mergers and also reveals insignificant point estimates.

Taken together, our results suggest that bank mergers generate no significant systematic changes in approval rates once we properly account for compositional factors and time-varying heterogeneity across markets. Table 2 provides further confirmation: pre- and post-merger approval rates for acquirers and control banks differ very little across all merger types. Collectively, these patterns reinforce the conclusion that bank mergers do not systematically affect households’ access to mortgage financing.

4.3 Bank mergers and mortgage performance

We examine how bank mergers affect the quality of newly-originated mortgages by analyzing subsequent delinquency rates.²³ Although interest rates and approval decisions reflect lending terms at origination, delinquency rates capture ex-post loan performance, providing insight into whether mergers alter risk-taking behavior or underwriting quality. Figure 6 displays average delinquency rates for newly originated mortgages around merger events. Each panel shows the three comparison groups: acquiring banks (red dashed), target banks (green short-dashed), and control banks (blue solid), with quarter zero marking merger completion.

Panel (a) examines community bank mergers and reveals relatively stable delinquency patterns across all groups throughout our event window. All three groups maintain delinquency rates of approximately seven percent, with only modest quarter-to-quarter fluctuations. These co-movements suggest that the variations reflect broader market conditions rather than merger-specific effects. Moreover, post-merger delinquency rates for acquirer and control banks remain closely aligned. The stability across groups is consistent with our earlier findings that community bank mergers produce limited changes in pricing or credit standards.

Panel (b) presents a more intriguing pattern for large bank acquisitions of community banks. Large bank acquirers and control banks both maintain delinquency rates of approximately six percent throughout the event window, consistent with the averages reported in Table 2. Panel (c) displays mergers between large banks and shows considerable volatility throughout the event window, reflecting the small sample size. Table 2 indicates a modest increase in acquirer delinquency from seven to eight percent on average (though insignificant in our formal tests below).

As above, raw level comparisons may reflect time-varying market conditions rather than merger-induced changes. Figure 7 presents formal DiD delinquency estimates that control for bank-county-merger fixed effects and time-varying factors. Each panel displays treatment effect coefficients (\(\beta_\tau\)) measuring changes in acquiring bank delinquency rates relative to control banks. Panel (a) examines community bank mergers. Pre-merger coefficients remain centered near zero, supporting the parallel trends assumption. Post-merger estimates are similarly close to zero and statistically insignificant across all horizons, indicating no causal effect of community bank mergers on mortgage delinquency.

Panel (b) addresses large bank acquisitions of community banks. Despite the apparent pre-merger dynamics in levels shown in Figure 6, our DiD coefficients are small and statistically insignificant across all post-merger quarters. Lastly, Panel (c) examines mergers between large banks and also reveals statistically insignificant point estimates throughout the post-merger window. None of the post-merger coefficients are statistically significant, and we find no evidence of systematic post-merger drift in delinquency rates across merger types. Taken together, the evidence on delinquency reinforces our earlier findings regarding interest rates and approval rates: bank mergers do not systematically alter mortgage market outcomes.

5 How competitive are local mortgage markets?

Understanding the competitive structure of local mortgage markets is important for assessing the effects of bank mergers on household borrowing conditions. In more concentrated markets, consolidation is more likely to translate into increased pricing power, whereas in markets with many active lenders, the effects of individual mergers should be attenuated. In this section, we document that county-level mortgage markets are considerably less concentrated than commonly perceived.

Indeed, policy discussions surrounding bank mergers often presume that local mortgage markets are concentrated, with a handful of dominant lenders controlling credit access. Our analysis is, to our knowledge, the most comprehensive on mortgage market competition around bank mergers and challenges this view by revealing that local mortgage markets exhibit remarkably competitive intensity.²⁴

Figure 8 presents the distribution of lender counts and market concentration across county-quarters in our sample. Panel (a) documents an extraordinary degree of competitive intensity: the typical county-quarter hosts dozens of active mortgage lenders, with the distribution exhibiting substantial right skew. Although a small number of rural counties operate with fewer than 20 lenders (\(3.5\%\)), the average bank-county-quarter observation involves 164 (median 132) competing lenders—reflecting the competitive environment facing a typical mortgage origination (the unweighted county-quarter mean is 75, median 50; see Table OA.3 in the Online Appendix).²⁵ These figures dramatically exceed the levels typically assumed in discussions of banking competition and indicate that individual bank mergers occur against a backdrop of intense competition from numerous mortgage providers.

Panel (b) reports the HHI, the standard measure of market concentration used in antitrust analysis. The distribution of HHI values indicates that local mortgage markets are overwhelmingly unconcentrated by regulatory standards. The median HHI across bank-county-quarter observations is approximately 400, well below the 1,000 threshold that the U.S. Department of Justice and Federal Trade Commission use in deposit-market merger review to classify markets as unconcentrated.²⁶ The 75th percentile remains below 1,000, and only a small fraction of county-quarters exceed the 1,800 threshold for highly concentrated markets—primarily rural areas with limited lending activity.²⁷ Consistent with these patterns, the vast majority of U.S. mortgage originations occur in markets that are highly competitive by antitrust standards.

The highly competitive nature of local mortgage markets is also evident in the geographic distribution of lenders. Figure 9 provides a geographic map of the average number of active lenders across U.S. counties during our sample period. The map reveals substantial spatial heterogeneity: major metropolitan areas along the coasts and the industrial Midwest exhibit the darkest shading, indicating 150-200+ active lenders, whereas rural counties in the Great Plains and Mountain West show lighter shading with 25-75 lenders. Importantly, even the least competitive counties maintain dozens of active lenders.

Table 4, Panel A, provides granular detail on the distribution of lender market shares. The median lender captures only 0.39 percent of its county market, whereas the mean is 1.53 percent. This extreme right skew indicates that the typical lender is a marginal player in any given local market. Even at the 90th percentile, lenders control only 3.6 percent of their county markets. Banks maintain slightly higher market shares than the broader lender population at every percentile, reflecting competitive advantages from branch networks, deposit relationships, and regulatory scale economies but these differences are modest—and non-bank mortgage companies provide substantial competitive discipline.

Table 4, Panel B, documents the geographic scope of bank lending operations, revealing that most banks remain predominantly local institutions despite decades of consolidation in the sector. The median bank originates mortgages in only five counties per year-quarter, indicating highly localized operations concentrated around branch networks and established customer bases. The mean of 29 counties reflects substantial right skew driven by a small number of national institutions operating across hundreds of counties. At the 10th percentile, banks operate in just a single county, representing truly community-focused institutions, and even at the 90th percentile, banks originate in only 35 (about one percent) of U.S. counties.

This competitive landscape provides crucial context for interpreting our merger effect estimates: in markets with 100+ active competitors (and highly-competitive HHI levels), individual bank mergers are unlikely to generate substantial market power. We next examine whether merger activity materially changes local market structure itself.

Figure 10 presents event-study evidence on market structure changes around mergers (average number of lenders per county-quarter). The patterns reveal a surprising result: rather than increasing concentration, bank mergers are associated with increasing lender counts and declining mortgage concentration (HHI). For Large-Community mergers, the number of active lenders increases by 23 percent from approximately 150 pre-merger (quarter -8) to 185 by quarter +8, and HHI declines by nine to 16 percent from 560 to 470-510. Large-Large mergers show a larger 30 percent increase in lender counts and a 20–26 percent decline in concentration. Declines in HHI are observed even though the merger mechanically increases concentration by combining two institutions. The offsetting entry of other competitors dominates the direct consolidation effect since mergers occur disproportionately in expanding markets with positive net entry—see Berger et al., 1999; Hauswald and Marquez, 2006; Dimopoulos and Sacchetto, 2017; Corbae and D'Erasmo, 2021. The strong competition we document appears to discipline pricing behavior and prevents banks from exploiting potential market power following mergers.

Our combined evidence tells a coherent story: bank mergers occur in highly competitive markets where merged institutions face substantial discipline from numerous existing and entering lenders. The local mortgage market structure we document helps explain why consolidation does not translate into materially higher mortgage prices, lower approval rates, or worse loan performance for households.

6 Selection in bank mergers

The patterns documented above raise fundamental questions about why banks merge and which institutions select into mergers. Merger selection can help to interpret our findings, since characteristics of merging banks shape both the competitive effects of consolidation and the welfare implications for borrowers. If mergers involve financially distressed institutions seeking survival, the observed pricing and credit supply patterns may reflect efficiency gains and market discipline rather than the exercise of market power. Conversely, if strong institutions acquire profitable targets to consolidate market power, concerns about reduced competition become more salient.

Table 5 presents comprehensive financial ratios for both target and acquirer banks, organized by our three merger types. The difference columns—(1)\(-\)(2), (2)\(-\)(3), and (1)\(-\)(3)—report pairwise differences in means across merger types, showing that several of the economically important gaps are statistically significant. The target-side comparisons reveal pronounced heterogeneity. Target banks in community–community mergers are weak and operationally inefficient, with near-zero profitability (return on assets (ROA) of 0.001) and very high expense ratios (non-interest expenses account for 90 percent of total income). By contrast, community banks acquired by large institutions are significantly more profitable (ROA of 0.007), materially more efficient on both expense-ratio measures, and more productive per employee. These targets exhibit higher personnel-expense shares, consistent with more labor-intensive and relationship-focused lending models. These patterns suggest large banks select a distinct subset of healthier but operationally different community lenders.

The acquirer panel also points to systematic scale differences rather than distressed acquirers seeking rescue. Large-bank acquirers have slightly higher profitability, lower non-interest expense ratios, lower deposit reliance, lower data-expense shares, and substantially higher income per employee than community-bank acquirers. These patterns are consistent with large institutions bringing scale, funding capacity, and operational infrastructure to the merger. Operationally, community banks acquired by large institutions show substantially higher personnel expense ratios (over 36 percent of total expenses) relative to community–community targets (25.1 percent), suggesting these banks maintain more labor-intensive, relationship-focused operations. Differences in efficiency ratios between targets and acquirers further underscore potential cost rationalization motives. In community–community mergers, targets operate with non-interest expense to income ratios of 89.8 percent compared to 71.2 percent for acquirers, and total expense to income ratios of 95.2 percent compared to 75.9 percent. These disparities suggest scope for post-merger cost reductions.

Although acquirer profitability lies in a narrow range (ROA between 0.003 and 0.004), the broader acquirer characteristics indicate that acquirer weaknesses do not drive mergers.

The evidence in Table 5 collectively points toward merger selection consistent with competitive pressure and efficiency considerations rather than market power motivations. These findings indicate that consolidation concentrates in markets where acquirers already maintain presence rather than entering new territory. Community banks increasingly face competition from larger institutions with superior technology, funding advantages, and operational scale. Mergers provide a mechanism for community banks to survive in an evolving competitive landscape. Large banks selectively acquire community banks whose relationship-based lending models and local market presence complement the acquirer’s scale and infrastructure, creating potential synergies. This selection pattern comports with our main empirical findings that mergers result in stable mortgage rates and limited changes in credit supply. ²⁸

7 Conclusion

This study examines nearly 5,000 bank mergers over nearly three decades using confidential micro-level data linking mortgage applications to loan performance, providing the most comprehensive analysis of bank consolidation effects on household borrowing to date. We document that mortgage markets remain remarkably competitive despite extensive consolidation, with the typical county featuring over 130 active bank lenders and concentration levels well below regulatory thresholds. Our main finding is straightforward: bank mergers do not systematically raise mortgage rates or restrict credit access once we properly account for the mechanical averaging that occurs when institutions with different lending characteristics merge. When large banks acquire community banks that charge higher rates, the combined entity’s rates rise mechanically—but this reflects portfolio composition, not predatory pricing. Acquiring banks maintain pre-merger pricing strategies rather than exploiting reduced competition.

Pre-merger characteristics suggest efficiency-related motives: weaker and less efficient community targets tend to merge with other community banks, whereas large acquirers systematically select more profitable, relationship-intensive community lenders whose business models complement the acquirer’s originate-to-distribute infrastructure.

An important nuance bears emphasis. Although we find no evidence that mergers harm borrowers, the absence of rate decreases means we cannot conclude that any efficiency gains are passed through to households. Stable rates are consistent with several interpretations: merged banks may retain cost savings without raising prices, a form of rent retention that is economically distinct from the exercise of market power. Alternatively, mergers may be driven by considerations outside the mortgage market we study. Disentangling these channels and measuring longer-run merger-related efficiencies to borrowers is an important direction for future research.

As banking continues to evolve with technological change and the growth of fintech competitors, understanding how traditional bank consolidation affects consumers becomes increasingly important for designing regulatory frameworks that balance stability, efficiency, and competition objectives.

Variable definitions appendix

Mortgage Characteristics

Borrower Characteristics

Mortgage Costs and Fees

Application and Approval Metrics

Mortgage Performance

Market Structure Variables

Bank Classifications

Time Periods

Bank Financial Ratios

Figure 1: Number of bank mergers by year.
This figure illustrates the number of bank mergers by year from 1996-2021. The bars represent the number of bank mergers in our sample (left axis) involving banks that have filed HMDA reports. The solid line shows the total number of bank mergers (right axis). The percentages above each bar indicate the proportion of all mergers involving two HMDA filers that are included in our sample of bank mergers for each year. Source: NIC.

Accessible version

Figure 2: Average abnormal interest rates (r^*)
This figure plots the evolution of average residualized interest rates around merger events for acquiring banks (red dashed), target banks (green short-dashed), and control banks (blue solid). Each panel corresponds to a different merger type based on acquirer and target bank sizes. The vertical line at quarter zero denotes merger completion. Abnormal interest rates (r^*) are the residual mortgage interest rate (in percentage points) after controlling for loan characteristics and county-quarter fixed effects, estimated using regression (1). Source: Authors’ calculations using cHMDA and McDash Analytics (ICE) data.

Accessible version

Figure 3: Stacked panel DiD – interest rates
This figure presents DiD estimates of merger effects on residualized interest rates (r^*). Points represent treatment effect coefficients (β_τ) with 95% confidence intervals, showing how acquiring banks’ rates evolve relative to control banks at each quarter relative to merger completion. Each panel corresponds to a different merger type. The specification includes bank-county-merger and time fixed effects to control for compositional changes and time-varying shocks. r^* is the residual mortgage interest rate (in percentage points) after controlling for loan and borrower characteristics and county-quarter fixed effects. Source: Authors’ calculations using cHMDA and McDash Analytics (ICE) data.

Accessible version

Figure 4: Average mortgage approval rates
This figure plots the evolution of average mortgage approval rates around merger events for acquiring banks (red dashed), target banks (green short-dashed), and control banks (blue solid). Each panel corresponds to a different merger type based on acquirer and target bank sizes. The vertical line at quarter zero marks merger completion. Approval rate is the fraction of mortgage applications approved by a lender in a given county-quarter. Source: Authors’ calculations using cHMDA and McDash Analytics (ICE) data.

Accessible version

Figure 5: Stacked panel DiD – approval rates
This figure presents DiD estimates of merger effects on mortgage approval rates. Points represent treatment effect coefficients (β_τ) with 95% confidence intervals, showing how acquiring banks’ approval rates evolve relative to control banks at each quarter relative to merger completion. Each panel corresponds to a different merger type. Approval rate is the fraction of mortgage applications approved by the lender. Source: Authors’ calculations using cHMDA and McDash Analytics (ICE) data.

Accessible version

Figure 6: Average delinquency rates of newly originated mortgages
This figure plots the evolution of average delinquency rates for newly originated mortgages around merger events for acquiring banks (red dashed), target banks (green short-dashed), and control banks (blue solid). Each panel corresponds to a different merger type. The vertical line at quarter zero marks merger completion. Delinquency is defined as mortgages that become 90+ days delinquent or experience foreclosure, bankruptcy, or similar adverse outcomes at any point through the mortgage’s life. Source: Authors’ calculations using cHMDA and McDash Analytics (ICE) data.

Accessible version

Figure 7: Stacked panel DiD – delinquency rates
This figure presents DiD estimates of merger effects on delinquency rates for newly originated mortgages. Points represent treatment effect coefficients (β_τ) with 95% confidence intervals, showing how acquiring banks’ delinquency rates evolve relative to control banks at each quarter relative to merger completion. Each panel corresponds to a different merger type. Delinquency rate is the fraction of newly originated mortgages in a bank-county-quarter that become 90+ days delinquent or experience foreclosure, bankruptcy, or similar adverse outcomes at any point during the sample period. Source: Authors’ calculations using cHMDA and McDash Analytics (ICE) data.

Accessible version

Figure 8: Distribution of mortgage lenders and market concentration across county-quarters
(a) Number of lenders (county-quarter)
(b) HHI (county-quarter)

Accessible version

Figure 9: Geographic distribution of mortgage lenders across U.S. counties
This figure maps the number of active mortgage lenders across U.S. counties, during our sample period (averaged). Darker shading indicates a higher number of lenders. Number of lenders is the total number of distinct lenders (at the parent company level) originating at least one mortgage in a county-quarter. Source: Authors’ calculations using cHMDA data.

Accessible version

Figure 10: Evolution of market structure around bank mergers
(a) Number of lenders (county-quarter)
(b) HHI (county-quarter)

Accessible version

Table 1: Bank merger types by acquirer and target group

This table shows the distribution of bank mergers in our sample by acquirer and target size groups. Community banks have total assets below $10 billion; large banks (combining regional and large) have total assets above $10 billion. All thresholds are inflation-adjusted using the GDP deflator. Values show counts and percentages of mergers for each acquirer–target combination. The sample includes 4,829 mergers over 1996–2021.

Table 2: Market attributes by merger type

This table presents market and lending characteristics broken down by target type (Community versus Large) and acquirer type (Community versus Large), showing control, acquirer, and target banks pre- and post-merger (Pre-M/Post-M). Each cell reports the mean at the bank-county-quarter level, averaged over the 17-quarter event window (eight pre-merger quarters, the merger quarter, and eight post-merger quarters). Variable definitions: Int. rate (res.) is the residual interest rate (in percentage points) after controlling for loan characteristics and county-quarter fixed effects; Int. rate spread is the mortgage rate minus a benchmark rate (in percentage points); Approval rate is the fraction of applications approved (0–1 scale); Delinq. is the fraction of loans ever 90+ days delinquent or in foreclosure, bankruptcy, or similar adverse status (0–1 scale); Lender fees are origination fees in dollars; Approved loans is the average number of mortgages approved per bank-county-quarter; # of applications is the average number of applications received per bank-county-quarter; Bank Mkt share is the bank’s share of county-quarter mortgage originations (0–1 scale). Number of lenders is the average count of distinct lenders originating at least one mortgage in a county-quarter. Source: Authors’ calculations using cHMDA and McDash Analytics (ICE) data.

Acquirer: Community – Target: Community

Acquirer: Large – Target: Community

Acquirer: Large – Target: Large

Table 3: Borrower and mortgage attributes by merger type

This table presents borrower and loan characteristics by merger type, broken down by target type (Community versus Large) and acquirer type (Community versus Large), for control, acquirer, and target banks pre- and post-merger (Pre-M/Post-M). Each cell reports the mean at the bank-county-quarter level, averaged over the 17-quarter event window (eight pre-merger quarters, the merger quarter, and eight post-merger quarters). Variable definitions: Applicant income is the borrower’s annual gross income in thousands of dollars ($K); LTV ratio is the loan-to-value ratio in percent; FICO score is the borrower’s credit score at origination (300–850 scale); Fixed rate is an indicator equal to one for fixed-rate mortgages; Conventional is an indicator equal to one for loans not insured by government agencies; FHA loan is an indicator equal to one for Federal Housing Administration insured loans; Owner-occupied is an indicator equal to one if the property is the borrower’s primary residence; Portfolio is an indicator equal to one if the lender retains the loan rather than selling it on the secondary market. All indicators are expressed as shares (0–1 scale) when averaged. Source: Authors’ calculations using cHMDA and McDash Analytics (ICE) data.

Acquirer: Community – Target: Community

Acquirer: Large – Target: Community

Acquirer: Large – Target: Large

Table 4: Local mortgage market structure

This table reports summary statistics for local mortgage market structure. Panel A presents statistics at the county-bank-quarter level (the main unit of observation in our sample), where each bank’s activity within a county-quarter is one observation. Panel B presents statistics at the bank-quarter level. The sample covers 1994–2023. Variable definitions: Number of lenders is the count of distinct lenders (banks and non-banks) originating at least one mortgage in a county-quarter; Lender market share is a lender’s share of county-quarter originations (%); Top-3/5/10 share is the combined market share (%) of the three, five, or ten largest lenders in a county-quarter; Nonbank share is the combined market share (%) of all non-bank lenders; County-quarter HHI is the Herfindahl-Hirschman Index (sum of squared market shares \(\times\) 10,000); Lenders with \(\geq\)1%/\(\geq\)5% share is the count of lenders exceeding that market share threshold; Quarterly entry/exit rate is the percentage of lenders entering or exiting a county-quarter; Top lender persistence rate is the probability (%) that the leading lender retains its top position from the prior quarter; # of distinct markets as active lender (Panel B) is the number of counties in which a lender originates at least one mortgage in a given quarter. Source: Authors’ calculations using cHMDA data.

Panel A: County-bank-quarter level

Panel B: Bank-quarter level

Table 5: Financial characteristics of target and acquirer banks by merger type

This table presents Call Report financial ratios for target banks (top panel) and acquiring banks (bottom panel) by merger type: (1) Community–Community, (2) Large–Community, (3) Large–Large. The unit of observation is the bank-merger, and all ratios are computed for the quarter immediately preceding the merger. The difference columns—(1)\(-\)(2), (2)\(-\)(3), and (1)\(-\)(3)—report pairwise mean differences, and significance stars denote conventional univariate test levels. Variable definitions: Mortgages/Total Assets is the share of total assets held as mortgages; ROA is net income divided by total assets; Net Int. Income/Total Assets is net interest income divided by total assets; Non Interest Expense/Total Income is non-interest operating expenses divided by total income (higher = less efficient); Total Expense/Total Income is the overall cost-to-income ratio; Total Income/# of Employees is revenue per employee ($K); Mortgages/# of Employees is mortgage holdings per employee ($K); Personnel Expenses/Total Expenses and Personnel Expenses/Total Income measure labor cost intensity; Interest Expense/Total Liabilities and Interest Expense/Total Deposits measure funding costs; Delinquent Mortgages/Total Mortgages is the share of loans that become 90+ days delinquent or enter foreclosure, bankruptcy, or similar adverse status; Deposits/Total Assets measures deposit funding reliance; Deposits/Total Mortgages is the ratio of deposits to mortgage holdings; Total Debt/Total Assets is the leverage ratio; Total Data Expenses/Total Income measures technology spending relative to income. Source: Authors’ calculations using the Federal Reserve merger-adjusted Call Report (TINY) data.

Target financial ratios

Acquirer financial ratios

OA.1 Data construction appendix

This appendix provides technical details on data matching, sample construction, and variable definitions that supplement Section 2 of the main text.

OA.1.1 cHMDA–McDash matching procedure

The cHMDA and McDash datasets share no common loan identifiers. We link them using a probabilistic matching algorithm exploiting loan characteristics and timing.

Initial matching. We restrict attention to approved cHMDA applications and search for McDash loans originated in the same month and county with loan amounts within narrow tolerance bands. Before 2018, amounts must match within $500 (reflecting dollar-level recording precision). From 2018 onward, when amounts are recorded in thousands, we tighten the tolerance to $10 in the transformed scale.

Multi-factor ranking. When multiple candidates exist, we rank them by (i) the absolute difference between application and loan amounts and (ii) temporal proximity between application and closing dates. From 2018 onward, we prioritize amount differences over timing given improved recording precision.

Duplicate resolution. We resolve many-to-one and one-to-many linkages iteratively: sort all potential matches by rank score, then sequentially retain each match only if neither the application nor the loan has been previously assigned. Indicator variables flag records with multiple potential matches, enabling sensitivity analysis.

After resolution, we merge McDash loan characteristics—initial interest rate, LTV ratio, credit score, loan term, property type—and forward-looking performance indicators (90+ day delinquency, bankruptcy, foreclosure) onto matched records. The procedure yields approximately 75 million matched applications. After data quality filters (Section OA.1.5), the final matched dataset contains approximately 62.6 million loans.

Interest rate extraction. McDash records current interest rates that may reflect post-origination modifications. To recover the original contract rate, we identify the first non-missing rate observation for each loan within a three-year window of the closing date. This yields over 174 million mortgages with valid origination interest rates in the raw McDash data. Loans are also classified as GSE-backed or portfolio-held based on investor type indicators.

Lender identifier construction. cHMDA identifies lenders through systems that evolved over the sample period: earlier years use respondent identifiers from regulatory agencies; later years incorporate a universal legal entity identifier. These correspond to individual banking charters, so a single parent company may have multiple identifiers. Using Robert Avery’s crosswalk files, we map charter-level identifiers to parent bank holding companies, producing consolidated lender identifiers that change only when ownership genuinely transfers through mergers or acquisitions.

OA.1.2 Bank size classification

Bank size is measured by total assets in the quarter before the merger, following Dodd-Frank thresholds with GDP-deflator adjustments: community banks have assets below $10 billion, regional banks between $10 billion and $100 billion, and large banks above $100 billion.²⁹

For the main analysis, regional and large banks are consolidated into one category. Under the three-way classification, acquirers comprise 2,642 community, 1,924 regional, and 263 large banks; targets comprise 4,325 community, 490 regional, and 14 large banks, spanning 2,683 counties. Under the two-way classification used in the main text, the sample comprises 2,642 community and 2,187 large bank acquisitions.

OA.1.3 Merger event filters

We apply the following filters to identify mergers suitable for analysis (see also Section 2):

1. Active mortgage presence. Targets must maintain positive market share in at least one county during the two years preceding the acquisition.

2. Financial crisis exclusion. We exclude mergers completed during 2007Q1–2009Q4, as lending behavior during this period reflects extraordinary circumstances.

3. Acquirer pre-merger presence. We drop merger-county combinations where the acquirer lacks pre-merger market share, avoiding confounds from simultaneous market entry.

4. Overlapping mergers. We exclude counties experiencing multiple merger events within the 17-quarter event window (8 pre, merger quarter, 8 post), applied recursively. For each target-county pair, we search forward from the merger date; if a subsequent merger involving a bank active in that county occurs within eight quarters and involves targets above the county’s fiftieth-percentile market share, we exclude that county-merger combination.

5. Clean control group. Control banks must not experience any merger in the county during the 17-quarter window. Non-banks are excluded from the control group.

OA.1.4 Market share and concentration construction

For each lender-county-quarter, we compute:

• Market share: the fraction of approved loans (count) originated by a specific lender in a given county-quarter relative to total approved loans in that county-quarter.

• Market ranking: ordinal and percentile positions in the local distribution.

• Trailing four-quarter moving averages of shares and rankings to smooth temporary fluctuations—particularly valuable when characterizing pre-merger market positions.

County-quarter concentration is measured by the Herfindahl-Hirschman Index: \[\begin{equation} HHI_{c,t} = \sum_{i=1}^{N_{c,t}} (s_{i,c,t})^2 \times 10{,}000\tag{3} \end{equation}\] where \(s_{i,c,t}\) is the market share of lender \(i\) in county \(c\) at time \(t\) and \(N_{c,t}\) is the number of active lenders.

OA.1.5 Data quality filters

Geographic and institutional restrictions. We restrict the sample to the fifty states and the District of Columbia, drop observations with missing or invalid county codes, and exclude non-banks from the control group.

Loan-level filters. We require approved loan amounts to exceed $10,000, exclude interest rates below 0.5 percent, and drop records missing any of: interest rate, applicant income, LTV ratio, FICO score, loan amount, maturity, rate type, mortgage type, occupancy type, or repayment status. Continuous variables are winsorized at the 1st and 99th percentiles. The final sample contains approximately 44.3 million mortgages (1994–2023).

OA.1.6 Variable construction

Residualized interest rates (\(r^*\)). See Equation (1) in the main text. We regress raw mortgage rates on borrower and loan characteristics and county\(\times\)quarter fixed effects; the residual \(r^*\) (in percentage points) is our primary price measure. Several alternative specifications with different control sets and fixed effects structures are used for robustness. We also compute the spread between the mortgage rate and the Federal Funds rate and obtain qualitatively similar results.

Approval rates. The ratio of approved applications to total applications for each lender-county-quarter. Date formats are standardized across years to account for evolving reporting conventions.

Delinquency rates. The fraction of originated loans that become 90+ days delinquent or experience foreclosure or bankruptcy at any point during their life, computed at the lender-county-quarter level.

Panel datasets. We construct two complementary datasets:

1. A loan-level dataset where each observation is a single mortgage augmented with the lender’s market position and merger involvement.

2. A lender-county-quarter dataset in which loan-level variables are averaged within each cell. Both simple and residualized means are computed.

OA.2 Additional results

This section presents supplementary figures and tables referenced in the main text:

• OA.1: plots average lender fees for acquirers, targets, and control banks across the event window, separately by merger type, to document raw fee dynamics around mergers.

• OA.2: reports stacked difference-in-differences estimates of merger effects on lender fees, showing treatment effect coefficients at each quarter relative to merger completion.

• OA.3: presents pre-merger distributions of residualized mortgage rates for acquirers, targets, and control banks across merger types, as well as pre- versus post-merger comparisons for acquirers, highlighting the cross-sectional pricing differences that underlie composition effects in simple pre-post comparisons.

• OA.1: reports the minimum number of distinct banks underlying each figure and table in the paper, ensuring that no result is driven by a small number of institutions.

• OA.2: shows the full three-way distribution of mergers by acquirer and target size group (community, regional, large), complementing the two-way classification used in the main analysis.

• OA.3: provides county-quarter-level summary statistics on local mortgage market structure, including concentration measures, lender counts, entry/exit rates, and top-lender persistence.

Figure OA.1: Average lender fees around bank mergers
This figure plots the evolution of average lender fees around merger events for acquiring banks (red dashed), target banks (green short-dashed), and control banks (blue solid). Each panel corresponds to a different merger type based on acquirer and target bank sizes. The vertical line at quarter zero marks merger completion. Lender fees are fees charged by the lender for processing and underwriting the mortgage, excluding third-party costs.

Accessible version

Figure OA.2: Stacked panel DiD – Lender fees
This figure presents DiD estimates of merger effects on lender fees. Points represent treatment effect coefficients (β_τ) with 95% confidence intervals, showing how acquiring banks’ lender fees evolve relative to control banks at each quarter relative to merger completion. Each panel corresponds to a different merger type. The specification includes bank-county-merger and time fixed effects. Lender fees are fees charged by the lender for processing and underwriting the mortgage, excluding third-party costs.

Accessible version

Figure OA.3: Pre-merger and pre-vs-post-merger rate distributions across bank types
This figure presents empirical cumulative distribution functions (ECDFs) of residualized interest rates (r^*). Panels (a) and (c) compare the pre-merger distributions of control banks (blue dashed), target banks (red solid), and acquiring banks (green solid), separately by merger type. Panels (b) and (d) compare the pre-merger (blue dashed) and post-merger (red solid) distributions for acquiring banks only, separately by merger type. If acquirers changed their pricing strategies post-merger, the pre- and post-merger distributions in panels (b) and (d) would diverge. r^* is the residual mortgage interest rate (in percentage points) after controlling for loan and borrower characteristics and county-quarter fixed effects.

Accessible version

Table OA.1: Minimum number of distinct banks in each figure and table

This table reports the minimum number of distinct banks underlying each figure and table in the paper. For event study figures, the minimum is taken across all event-time quarters (\(-8\) to \(+8\)). For summary statistics tables, the minimum is reported separately by group (control, acquirer, target) and period (pre-merger, post-merger). Banks are counted at the parent company level.

Figures 2, 4, 6, OA.1: Levels event studies

Figures 3, 5, 7, OA.2: Stacked panel DID

Figure 10: Pre/post-merger market structure

Tables 2 & 3: Attributes by merger type

Table OA.2: Bank merger types by acquirer and target group

This table shows the distribution of bank mergers in our sample by acquirer and target size groups. Community banks have total assets below $10 billion, regional banks have total assets between $10 billion and $100 billion, and large banks have total assets above $100 billion. All thresholds are inflation-adjusted using the GDP deflator. Values show counts and percentages of mergers for each acquirer–target combination. The sample includes 4,829 mergers over 1996–2021.

Table OA.3: Local mortgage market structure: county-quarter level

Notes: This table reports summary statistics for local mortgage market structure at the county-quarter level, where each county-quarter is one observation. Variables include concentration measures (top-\(k\) lender shares, HHI), the nonbank market share, lender counts (total and by market share thresholds), quarterly entry and exit rates, and top lender persistence (the probability that the leading lender in a county retains its top position in the following quarter). The sample covers 1994–2023.

Footnotes