Where does the Stimulus go? Deep Generative Model for Commercial Banking Deposits

Ni ZhanarXiv 2021 · arXiv

Abstract

When the Federal Reserve implements Quantitative Easing (QE), trillions of dollars are injected into the banking system. However, the ultimate destination of these funds—whether they settle in the accounts of everyday individuals (retail) or large corporations (wholesale)—remains largely opaque due to limited reporting requirements. This paper introduces a Deep Generative Model (DGM) to "unmix" aggregate bank deposit data. Our findings reveal that QE primarily inflates wholesale deposits, contributing to a structural shift where the retail fraction of U.S. deposits has plummeted from 70% to 40% over the last twenty years.

Introduction

In response to the 2008 financial crisis and the COVID-19 pandemic, the Federal Reserve utilized Quantitative Easing (QE) to inject liquidity into the economy. While total bank deposits surged, the distribution of this wealth across different sectors of society remained a "black box." Banks are not required to report the specific split between retail and wholesale deposits, making it difficult for policymakers to assess the equitable distribution of stimulus measures. This research aims to peer inside this box using machine learning to estimate the hidden composition of the U.S. banking system.

Methodology

To address the lack of ground-truth labels, we approach the problem as an unsupervised learning task using a Deep Generative Model (DGM).

The Generative Assumption

We model a bank’s total deposits as a mixture of two distinct lognormal distributions. Retail accounts are characterized as smaller and more numerous, while wholesale accounts are fewer but represent massive individual balances. Lognormal distributions are specifically chosen to account for the "fat tails" inherent in wealth distribution, which traditional linear models often fail to capture.

Parameter Estimation and Loss Function

Neural networks are employed to map observable bank features—such as the number of physical offices and the percentage of accounts under the FDIC insurance limit—to the parameters ((\mu, \sigma)) of these distributions. The model is trained by minimizing the discrepancy between simulated bank metrics and actual reported FDIC data.

Time-series validation of the model's forward predictions against ground truth data.

To validate the model, we utilize time-series regression to compare predictions against known benchmarks and "pure" bank types (e.g., thrifts for retail and specialized custodial banks for wholesale).

Key Insights and Results

The 20-Year Structural Shift

The most significant finding is a massive structural transformation in the U.S. economy. Since 2000, the retail fraction of deposits has seen a steady decline. Our model suggests that while retail deposits once made up roughly 70% of the total, they now account for only about 40%.

Comparison of retail fraction estimates and the growth of retail vs. wholesale deposits over time (2000-2020).

QE is Not Neutral

The research demonstrates that QE is not a neutral tool. A $1.00 increase in Fed reserves leads to approximately a **$0.60 increase in wholesale deposits**, while the impact on retail deposits is statistically insignificant. In contrast, traditional commercial lending acts as a "great equalizer," with the resulting deposits distributed much more evenly between individuals and corporations.

COVID-19 and the "Dash for Cash"

The model successfully identified the "dash for cash" during the first quarter of 2020. As the pandemic took hold, large corporations immediately drew down credit lines and parked the funds in banks, leading to a massive spike in wholesale deposits. Traditional "rule of thumb" benchmarks (like the $500M branch threshold) largely failed to capture the magnitude of this shift, whereas the DGM provided a granular, responsive estimate.

Conclusion

By applying deep generative modeling to financial data, this paper provides evidence that the mechanisms of modern monetary policy disproportionately benefit large corporate entities over individual depositors. This lopsided growth in wholesale deposits suggests that QE may inadvertently exacerbate economic inequality. Future work should investigate how these trends vary across different bank sizes and the specific role of interest rate fluctuations in driving deposit migration.

Summary updated

Abstract

This paper examines deposits of individuals ("retail") and large companies ("wholesale") in the U.S. banking industry, and how these deposit types are impacted by macroeconomic factors, such as quantitative easing (QE). Actual data for deposits by holder are unavailable. We use a dataset on banks' financial information and probabilistic generative model to predict industry retail-wholesale deposit split from 2000 to 2020. Our model assumes account balances arise from separate retail and wholesale lognormal distributions and fit parameters of distributions by minimizing error between actual bank metrics and simulated metrics using the model's generative process. We use time-series regression to forward predict retail-wholesale deposits as function of loans, retail loans, and reserve balances at Fed banks. We find increase in reserves (representing QE) increases wholesale but not retail deposits, and increase in loans increase both wholesale and retail deposits evenly. The result shows that QE following the 2008 financial crisis benefited large companies more than average individuals, a relevant finding for economic decision making. In addition, this work benefits bank management strategy by providing forecasting capability for retail-wholesale deposits.
BibTeX
@misc{zhan2021where,
  author = {Ni Zhan},
  title = {{Where does the Stimulus go? Deep Generative Model for Commercial Banking Deposits}},
  year = {2021},
  eprint = {2101.09230},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2101.09230},
}
Figure