Examining the Sensitivity of Regional Banks to Macroeconomic Shocks

August 04, 2026

Examining the Sensitivity of Regional Banks to Macroeconomic Shocks

Faith Achugamonu, Elena Afanasyeva, Tim Schmidt-Eisenlohr, and Matthew P. Seay¹

Introduction

Following the Global Financial Crisis (GFC), banking supervision and regulation became more stringent for largest banks, particularly systemically important institutions and those with at least $100 billion in consolidated total assets. However, the 2023 stress period following the default of the Silicon Valley Bank (SVB) highlighted that problems at regional banks, which we define as banks between $10 billion and $100 billion in assets, may also cause broader banking system stress.

These developments raise important questions surrounding regional bank vulnerabilities and their implications for financial stability. In this Note, we ask: How well do conventional top-down stress testing models reflect financial performance of regional banks during macro-financial stress?² If regional banks respond differently to shocks due to their business models, this top-down approach may mischaracterize their vulnerabilities. In addition, top-down models typically estimate relationships between economic conditions and bank performance using data for all banks, but because large banks represent most of industry assets, they dominate the statistical results. To address this limitation, we estimate a stress testing model that is restricted to regional banks only and compare it to a baseline model for all banks.

Our analysis reveals that differences in business models lead to notably different balance sheet exposures of regional banks compared with large banks. We also find that regional banks are generally less sensitive to broad macro-financial stressors than their larger peers. At the same time, regional banks are likely more vulnerable to shocks to low-cost deposit funding and to certain sectors of the commercial real estate market given their large exposures to both. Of note, our approach does not capture vulnerabilities to region-specific economic shocks, which could weigh on individual regional bank performance but are unlikely to rise to the level of systemic concern.

In the remainder of this Note, we compare large and regional bank business models and use the Forward-Looking Analysis of Risk Events (FLARE) stress testing model (detailed in subsequent sections) to evaluate the macro sensitivity and forecasting accuracy for regional banks under each modeling approach. Lastly, we test the banks with hypothetical stress scenarios to demonstrate implications of using a regional bank model for vulnerability assessment.

What are the Major Differences in Regional Bank and Large Bank Business Models?

In this Note, we define regional banks as any firm with consolidated assets of at least $10 billion but less than $100 billion throughout our sample period.³ Large banks are firms with assets exceeding $100 billion during at least one quarter over the same period.⁴

As a first step, we compare balance sheet and income statement information across these two bank groups. Table 1 documents the corresponding means for the main pre-provision net revenue (PPNR) and other balance sheet components relative to total assets as well as the loan portfolios relative to total loans for the last pre-COVID quarter ( 2019Q4). Our sample for 2019Q4 includes 143 regional banks and 41 large banks. Based on these statistics, regional banks operate a classic commercial banking model - taking deposits (75.7 percent of assets) and making loans (69.3 percent of assets). The business model of large banks appears more diversified, as they have a lower share of deposits (60.1 percent) and a lower share of loans (50.3 percent), suggesting more capital markets activity, trading, and wholesale funding.

In terms of revenue structure, regional banks rely on their interest income more than large banks (4.3 percent vs. 3.9 percent of assets). Large banks take in relatively more fees and trading revenue (their fee income measures 0.6 percent of total assets compared with 0.4 percent for regional banks, while large banks' other non-interest income is 1.2 percent against a lower 0.7 percent for regionals), reflecting their higher participation in investment banking, asset management, and transactional services.

On the expense side, large banks have relatively higher interest expenses (1.3 percent vs. 0.9 percent). This result is likely due to a smaller retail deposit base, more wholesale funding exposure and ultimately a different deposit mix (less core deposits) on large banks' balance sheets.

The most striking difference between regional banks and large banks is in their credit risk profile. Regionals banks have substantial commercial real estate (CRE) loan exposure measuring 41.6 percent of their total loans vs. 16.8 percent for large banks. Regional banks focus on local CRE - predominantly fixed-rate products with higher duration. In contrast, large banks are relatively more exposed to consumer lending (10.8 percent vs 2.1 percent), reflecting their higher credit card portfolios and more exposure to auto loans. Taken together, these business model differences suggest a lower pass-through of interest rate changes to the asset side of the balance sheets of regional banks, supporting their profitability and resilience in recession scenarios with falling yields. At the same time this balance sheet structure can also imply a larger interest rate risk exposure for regional banks, which may generate sizable fair value losses when interest rates rise. ⁵

Table 1. Comparing Income Statements and Balance Sheets: Regional Banks vs. Large Banks as of 2019:Q4

Geographic diversity is another key dimension differentiating large and regional bank business models. According to FDIC's summary of deposits data from 2025, the median large bank operates in approximately 14 states (the largest operating in 48 states and D.C.), while the median regional bank operates in about 4 states (and about 20 percent of regionals operate in only one state). Taken together, these facts point to less diversification, on average, by asset composition and geography among regional banks.

In the next few sections of this Note, we introduce the FLARE stress testing model and explore how differences in bank business models may impact goodness of fit and out-of-sample performance.

The FLARE Model and How it is Used in this Analysis

FLARE is a top-down, bank-level stress testing model that projects the banking system's PPNR, loan losses, and capital using public financial statement data and macroeconomic variables. Most financial information is sourced from FR Y-9C and Call Reports. The macroeconomic variables represent a subset of the 16 domestic variables used in the annual stress test exercise scenarios. Components of PPNR and losses by loan type are individually estimated using regression models that incorporate autoregressive (AR) terms, financial market indicators, macroeconomic variables, and bank-specific control variables. Our estimation sample generally spans from 1997 forward. See Correia et. al 2022 for more information on FLARE.⁶

In contrast to prior publicly released analyses, we run the model to estimate loadings on two distinct samples: an "All Banks" model, and an only "Regional Banks" model. Based on the estimated sensitivities, the model forecasts each bank's income, expenses, loan losses, and resulting capital changes. Before proceeding further, we first address the basic question of applicability of a top-down model to analyzing regional banks.

Do Regional Banks' Concentrated Individual Footprints Imply Top-Down Models Will Miss Systemic Risks?

To answer this question, we use Summary of Deposits (SOD) data from the FDIC to understand geographical diversification for regional banks (based on the branch-weighted location of their deposits). Figure 1 shows some concentration of regional bank deposits in select regions, but overall indicates that the sector's deposits are broadly distributed across the U.S. The sum of shares for the top 5 states is around 34 percent, which closely reflects the share in U.S. GDP of these 5 states. Standard measures of industry concentration, such as the Herfindahl-Hirschman Index (HHI), also point to broad geographic diversification.⁷

This diversification of the regional bank sector has important financial stability implications. While individual regional banks may face local economic shocks that top-down models fail to capture in isolation, the sector's wide geographic footprint suggests broad macro stress could simultaneously impact regional banks across the U.S. Thus, applying top-down models like FLARE should appropriately capture the most systemic risks - correlated losses associated with broader macro stress - even if region-specific vulnerabilities or heterogeneity are not fully reflected in the modeling approach.

How Does the Fit Compare across the "All Banks" and "Regional Banks" Models?

To compare the fit of the "Regional Banks" model with our baseline "All Banks" model, we follow two approaches. First, we compare the regression output across these two models, assessing how well the models explain the relationship between balance sheet items and macro and financial variables equation by equation. Second, we study the out-of-sample forecasting accuracy of the two models for PPNR, Net Charge-Offs (NCOs) and Common Equity Tier 1 (CET1) ratios for the aggregate of all regional banks.

We begin by re-estimating the FLARE model on two separate samples: regional banks and all banks, comparing the in-sample fit by equation. Table 2 documents the R-square metrics, ordering the equations to reflect the importance (as a share of risk-weighted assets) of the corresponding component on the balance sheet of regional banks. The in-sample fit measure, R-square, of the "Regional Banks" model are either comparable or in several cases somewhat better than those of the "All Banks" model, including the most quantitatively important components, such as interest income and interest expense.

Table 2. In-Sample Fit Measures across Models by Component Equation

We next compare the out-of-sample performance of the "Regional Banks" and "All Banks" models. To this end, we conduct forecasting exercises, in both tranquil and stressful times, using the actual paths of the aggregate macroeconomic and financial variables as scenario trajectories. For the first episode, we examine the GFC by initiating our forecast in 2008Q2 - the final quarter before the Lehman Brothers bankruptcy in September 2008. As examples of tranquil times, we consider two pre-COVID episodes - the first one with the jump-off in 2016Q4 and the second one with the jump-off in 2017Q4, reporting average forecasting metrics across them. Finally, we also compute model projections using the last quarter preceding the SVB failure (2022Q4) as the jump-off, augmenting this exercise with an additional stress ingredient - the funding shock (see Achugamonu et al. 2026 for more details). We introduce the funding shock to better reflect the stress profile characteristic of that episode.⁸From the macroeconomic point of view, the SVB episode was not characterized by a substantial real activity slowdown or by a remarkable uptick in stock market volatility, especially when aggregated to quarterly frequency. Yet in this high-interest-rate environment a subset of banks experienced growing deposit outflows amid sizable fair value losses on their loans and securities. The funding shock enhancement of the FLARE model is designed to capture these developments.⁹

Table 3 illustrates the root mean squared errors (RMSEs) for these episodes across three aggregated variable categories: PPNR (PPNR/total assets), net charge-off (NCO) rates (NCO*4/loans), and CET1 ratios (CET1/RWA). Results in the columns with the header "Regional" correspond to the projections obtained for the sample of regional banks using the corresponding estimated coefficients for this sample. Results in the columns "All Banks" correspond to the projections obtained for the sample of regional banks using the coefficients estimated for the sample of all banks. Accordingly, the comparison between these two columns allows us to assess the relative goodness of fit of the regional model measured out-of-sample.

The differences in out-of-sample forecasting performance are most pronounced in the crisis time, such as GFC, when macroeconomic and financial conditions are in substantial distress (Table 3). In macroeconomically calmer times (such as pre-COVID episodes and in the quarters after the SVB distress episode), the overall performance of the regional model is still better, albeit not for all variables. In particular, in these cases NCO rates are projected equally well by both models, with a slight (although not statistically significant) advantage in favor of the "All Banks" model.¹⁰ Overall, though, the "Regional Banks" model produces predominantly lower forecast errors for regional banks than the "All Banks" model.

Table 3. RMSEs across Various Episodes: "Regional Banks" Model vs "All Banks" Model

When running a regional version of the FLARE model, we do not replace macroeconomic regressors with regional economic indicators. Apart from the satisfactory in-sample and out-of-sample performance of the regional bank model as discussed above, there are two additional reasons for this choice. First, we are interested in modeling the propagation of system-wide stress, which is more closely tied to macroeconomic aggregates. Second, because regional banks are geographically diverse, it is less likely for a local shock to become systemic and unlikely that the model will fail to capture broader correlated losses that result from stress.

How Material are Differences in Regression Coefficients in the "All Banks" vs. "Regional Banks" Models?

Comparing the key regression coefficients across models, we find that regional banks are typically less sensitive to aggregate macroeconomic and financial conditions than large banks. This result holds for the majority of PPNR components and NCO rates as well as across most macro-financial variables.¹³ Table 4 provides an illustrative example that covers both sides of bank balance sheets - interest income and interest expenses. The "Regional Banks" model coefficient on the 3-month Treasury yield is substantially lower relative to its "All Banks" model counterpart in both regressions. The differences are economically sizeable and statistically significant, as indicated by the corresponding p-values in the last columns of Table 4. These results highlight a lower short-run sensitivity of regional banks to aggregate macroeconomic and financial conditions. A large portion of these results survive both in economical and statistical sense, when a long-run multiplier is considered instead.¹⁴

Table 4. Coefficient Comparison Example: "Regional Banks" Model vs. "All Banks" Model

Although the overall tendency points to smaller sensitivity of regional banks to aggregate macroeconomic and financial variables, there is one important exception: NCO rates of regional banks in certain CRE segments (construction and multifamily loans, in particular) are more sensitive to aggregate CRE prices or changes in the unemployment rate. This higher sensitivity, combined with a greater exposure to CRE of regional banks, indicates that regionals are more vulnerable to deterioration in select CRE segments.

Abstracting from the exception of CRE, why might regional banks be less sensitive to aggregate macroeconomic and financial conditions? One possible answer points to the differences in business models. Regional banks generally carry more duration on their assets and finance a greater fraction of their liabilities with low-cost deposits than large banks, which is consistent with a lower pass-through of short-term interest rate changes on both sides of the balance sheet (Drechsler et al. 2017). In line with this mechanism, regional banks are indeed on average more profitable than large banks.¹⁵ That said, such a business model may also imply more interest rate risk exposure (see Drechsler et al. 2023).

The mechanism outlined above does not speak to the lower sensitivity of the majority of NCO rates to aggregate CRE and housing prices, though. Housing prices are subject to local rather than aggregate cycles (see Del Negro and Otrok, for example). Hence, regional banks' variables could be less reactive to aggregate developments than local developments. ¹⁶ A related argument applies to aggregate CRE prices. Regional banks have granular, relationship-based information about local CRE markets that aggregate price indices likely miss. Accordingly, they know which specific properties or borrowers are distressed and therefore rely less on noisy aggregate signals (see Stein (2002) for the theoretical argument).

A second distinctive feature resulting from our estimation of regional banks is a higher persistence of their PPNR components and NCO rates. Regional banks exhibit both higher persistence and lower aggregate sensitivity, consistent with a business model driven more by stable, local relationships than by aggregate financial conditions. As a consequence, and as discussed below, the "Regional Banks" model will produce flatter and more persistent projections of income and loan losses, relative to projections based on the "All Banks" model.

How Material are the Impacts for Regional Bank Revenues, Loan Losses, and Capital under Hypothetical Macro-Financial Stress Scenarios?

Thus far, we have established differences in estimated coefficients and the qualitative implications of these results. In this section, we ask what these differences mean in quantitative terms for stress test projections of PPNR, NCOs, and bank capital under the "Regional Banks" and "All Banks" models.

To compute these projections, we train each model on the 1997Q1-2023Q4 sample and project PPNR, NCOs, and bank capital for the next 9 quarters under the 2024 severely adverse scenario path from the Board's annual stress test exercise. The goal of this exercise is to cleanly isolate the effects of the "Regional Banks" model coefficients on the resulting projections and to quantify the differences relative to the "All Banks" model (in which large banks dominate the coefficient estimation). To this end, we report projections for regional bank revenues, expenses, and loan losses using coefficient loadings from both models, respectively.

Projections using the "Regional Banks" model are shown in the dashed red line of Figure 2, and projections of regional bank PPNR using coefficients from the "All Banks" model are depicted in solid black. PPNR projections using the "Regional Banks" model are substantially higher, relative to projections using the "All Banks" model. This result is consistent with the higher persistence in regional banks' PPNR and NCOs and the lower sensitivity of their interest income and interest expenses to changes in Treasury yields.

We observe a qualitatively similar result for the NCO projections (Figure 3). Regional banks' aggregate NCOs using the "Regional Banks" model loadings increase by less than projections using the "All Banks" model. We note that using the "All Banks" model results in higher NCO projections than the severe levels observed during the Global Financial Crisis (GFC).¹⁸

Our results highlight two sources of resilience for regional banks - a relatively more stable PPNR and a more stable NCO rate. What are the resulting implications for the capital projection? Consistent with these results, the CET1 ratio in the "Regional Banks" model drops by substantially less than that in the "All Banks" model (Figure 4). The maximum difference between CET1 ratio projections under the "Regional Banks" model and the "All Banks" model amounts to nearly a percentage point by the end of the projection horizon. PPNR projections account for about two thirds of the overall change in the CET1 projections, and the NCO projections account for about one third of the overall change. Summing up, attributing "All Banks" coefficients to regional banks can lead to a sizeable overestimation of capital losses in stress test scenario analysis.

Conclusions

In this Note, we described the results of a FLARE estimation exercise for regional banks. The exercise revealed that regional banks have substantially higher CRE exposures, are less sensitive to aggregate macroeconomic and financial conditions, and have more persistence in PPNR components. These differences between regional banks and large banks amount to non-trivial implications for stress testing. Attributing "All Banks" coefficients to regional banks can lead to a quantitatively sizeable overestimation of capital losses for regional banks in standard top-down stress test analysis.

References

"2024 Stress Test Scenarios." Board of Governors of the Federal Reserve System. Accessed 2 Apr. 2026.

"June 2024 Exploratory Analysis." Board of Governors of the Federal Reserve System. Accessed 2 Apr. 2026.

Achugamonu, Faith, Tim Schmidt-Eisenlohr, and Matthew P. Seay (2026). "Assessing Bank Resilience to a Funding Shock," FEDS Notes. Washington: Board of Governors of the Federal Reserve System, February 17, 2026.

Armantier, Olivier, Marco Cipriani, and Asani Sarkar (2024), "Discount window stigma after the global financial crisis." FRB of New York Staff Report.

Correia, Sergio, Matthew P. Seay, and Cindy M. Vojtech (2022), "Updated primer on the forward-looking analysis of risk events (flare) model: A top-down stress test model." FEDS Working Paper.

Del Negro, Marco and Christopher Otrok (2007), "99 Luftballons: Monetary Policy and the House Pirce Boom Across U.S. States," Journal of Monetary Economics 54(7): 1962-1985.

Drechsler, Itamar, Alexi Savov, and Philipp Schnabl (2017), "The Deposits Channel of Monetary Policy," The Quarterly Journal of Economics 132(4): 1819 - 1876.

Drechsler, Itamar, Alexi Savov, Philipp Schnabl, and Olivier Wang (2023), "Deposit franchise runs." National Bureau of Economic Research.

Hinzen, Franz, Felipe Severino, and Stijn Van Nieuverburgh (2026), " Too-Many-to-Ignore: Regional Banks and CRE Risks." Working Paper.

Hirtle, Beverly, Anna Kovner, James Vickery, and Meru Bhanot (2016), "Assessing financial stability: The capital and loss assessment under stress scenarios (class) model." Journal of Banking & Finance, 69, S35-S55.

Stein, Jeremy C., (2002) "Information Production and Capital Allocation: Decentralized versus Hierarchical Firms" 57(5): 1891-1921.

Appendix

Table A. Comparing Income Statements and Balance Sheets: Regional Banks vs. Large Banks as of 2023Q4