The Deposit Magnet: Quantifying the Drivers of Bank Geographic Spread

A Spatial Econometric Analysis of Commercial Bank Branching Strategies

Spatial Econometrics R Programming Causal Inference Financial Services

Executive Summary

Core Finding: Banks' spatial expansion strategies are overwhelmingly driven by market opportunity (deposit volume) rather than competitive structure (market concentration), with market size explaining 68% of variance in geographic lending radius across 393 MSAs.

Dataset Scale

58,123 branch records
3,207 unique banks
393 MSAs

Methodology

Spatial econometric modeling
Haversine distance calculation
HHI concentration analysis

Key Tools

R (tidyverse, geosphere)
Statistical inference
Data visualization

Research Question

What determines a bank's geographic presence—its "lending radius"—in a given market? This project tests two competing hypotheses:

Understanding these dynamics has significant implications for financial inclusion policy, competitive strategy, and addressing banking deserts in underserved communities.

Data Engineering & Feature Construction

Data Source & Cleaning

The analysis utilizes publicly available Federal Reserve branch-level deposit data, comprehensively cleaned and engineered for spatial analysis:

Initial Dataset: 75,000+ branch observations with 40+ variables

Final Analytical Dataset: 58,123 branches (77% retention) after outlier removal and geographic validation

Key Feature Engineering

1. Dependent Variable: Geographic Lending Radius

Calculated as the mean pairwise distance between all branches operated by a single bank within an MSA using the Haversine formula:

\[ d = 2r \arcsin\left(\sqrt{\sin^2\left(\frac{\phi_2-\phi_1}{2}\right) + \cos(\phi_1)\cos(\phi_2)\sin^2\left(\frac{\lambda_2-\lambda_1}{2}\right)}\right) \]

Where φ represents latitude, λ represents longitude, and r is Earth's radius (6,371 km)

2. Independent Variable: Market Concentration (HHI)

Herfindahl-Hirschman Index calculated at the MSA level:

\[ HHI = \sum_{i=1}^{N} s_i^2 \]

Where si is the market share (by deposits) of bank i in the MSA

3. Control Variable: Market Size

Log-transformed total deposit volume per MSA to normalize right-skewed distribution:

\[ \text{Market Size} = \ln(\text{Total Deposits}) \]
Scatter plot showing no relationship between HHI and lending radius

Figure 2: Market Concentration (HHI) vs. Lending Radius - No systematic relationship (p = 0.502)

Market size vs Lending Radius scatter plot

Figure 3: Market size drives geographic spread (R² = 0.68)

Statistical Modeling & Inference

Model Specification

Ordinary Least Squares (OLS) regression with robust standard errors:

\[ \text{Lending Radius}_{im} = \beta_0 + \beta_1 \text{HHI}_m + \beta_2 \ln(\text{Total Deposits}_m) + \epsilon_{im} \]

Where i indexes banks and m indexes MSAs

Regression Results

Variable Coefficient Std. Error t-statistic p-value
Intercept -45.23 3.41 -13.26 < 0.001 ***
HHI (Market Concentration) 0.0012 0.0018 0.67 0.502
ln(Total Deposits) 5.87 0.34 17.26 < 0.001 ***

Model R² = 0.683, Adjusted R² = 0.681, N = 3,207 bank-MSA combinations
*** indicates significance at p < 0.001

Key Findings & Interpretation

Finding 1: Competition Doesn't Drive Spatial Strategy

The HHI coefficient is statistically insignificant (p = 0.502) and economically negligible (β = 0.0012), indicating that banks do not systematically alter their geographic spread in response to market concentration.

Finding 2: Market Size Dominates Branch Dispersion

A one-unit increase in log(deposits)—approximately 2.7x increase in market size—corresponds to a 5.87-mile increase in average lending radius, significant at p < 0.001. This effect is both statistically and economically substantial.

Finding 3: High Explanatory Power

The model explains 68.3% of variance in lending radius using just two variables, with market size as the primary driver. This suggests spatial expansion follows economic opportunity rather than competitive positioning.

Model Validation & Robustness

Animated visualization showing flat relationship between HHI and lending radius

Figure 4: HHI vs Lending Radius - Animated demonstration of null relationship (β = 0.0012, p = 0.502)

Side-by-side comparison of HHI null effect versus market size strong effect

Figure 5: Hypothesis Testing Comparison - Competition rejected vs Market Size confirmed

Robustness Checks Performed

  • Heteroskedasticity-robust standard errors (reported above)
  • Outlier sensitivity analysis (Cook's distance threshold at 4/n)
  • Alternative functional forms (quadratic, log-log specifications)
  • Geographic fixed effects (regional dummy variables)

Strategic & Policy Implications

For Banking Strategy

Banks should prioritize branch expansion decisions based on market deposit volumes rather than competitive positioning. This suggests a first-mover advantage in high-deposit markets rather than defensive responses to competitor moves.

For Regulatory Policy

Findings challenge the effectiveness of competition-focused branch policies. Regulators seeking to improve banking access in underserved areas should focus on economic development and deposit growth rather than micro-managing competitive structure.

Ethical Considerations

⚠ Critical Caveat: The market-size finding could be misinterpreted to justify neglecting low-deposit communities, potentially exacerbating banking deserts. Policy must balance economic efficiency with spatial equity through targeted incentives for rural and low-income area expansion.

Technical Skills Demonstrated

Data Wrangling

  • Large-scale dataset cleaning (75K+ observations)
  • Missing data imputation strategies
  • Geographic data validation
  • Feature engineering from raw data

Statistical Analysis

  • OLS regression with robust inference
  • Hypothesis testing (t-tests, F-tests)
  • Model diagnostics & validation
  • Causal interpretation frameworks

Spatial Analysis

  • Haversine distance calculations
  • MSA-level aggregation
  • Geographic clustering analysis
  • Market delineation methods

Tools & Languages

  • R (tidyverse, geosphere, lm)
  • R Markdown for reproducible research
  • Data visualization (ggplot2)
  • Statistical computing

Academic Context & Citations

This analysis builds on recent spatial banking literature:

Begenau, A., Oberfield, E., Rossi-Hansberg, E., & Wenning, D. (2024). Banks in Space. National Bureau of Economic Research Working Paper 32262.

Bouakez, H., Côté, J., & D'Souza, C. (2020). A Spatial Model of Bank Branches in Canada. Bank of Canada Staff Working Paper 2020-4.

Project Resources

📄 Full Report (PDF) 💻 R Code 📊 Presentation 🔗 GitHub Repository