Published June 2, 2026 | Version v1
Peer review Open

A dynamic programming framework for promotional deposit arbitrage with eligibility constraints: Numerical illustrations from offer-level calibration

Description

Figure Descriptions

Figure 1: Promotional and Normal APYs in the Six-Bank Calibration

A grouped bar chart comparing the promotional APY (blue bars) and normal/post-promotion APY (magenta bars) across six U.S. digital banks: SoFi, Ally, Marcus, Discover, Varo, and CIT Bank. The figure illustrates the yield spread that creates the arbitrage opportunity, with promotional rates ranging from 4.35% to 5.20% and normal rates from 3.00% to 3.60%. Varo offers the highest promotional rate (5.20%) but the lowest normal rate (3.00%), while CIT Bank shows the smallest promotional spread.

Figure 2: Signup Bonuses and Promotional Deposit Caps in the Six-Bank Calibration

A dual-axis chart displaying signup bonuses as blue bars (left axis, in dollars) and deposit caps as a black line with markers (right axis, in thousands of dollars). SoFi offers the highest bonus ($300) with a $50,000 cap, while CIT Bank has the highest cap ($250,000) with a $175 bonus. The figure highlights the inverse relationship between bonus size and deposit capacity across institutions.

Figure 3: Computed Wealth Trajectories over 36 Months

A line plot showing wealth accumulation paths under five benchmark strategies starting from $50,000 initial capital over a 36-month horizon. Strategies include: Static HYSA (blue), Calendar Rotation (magenta), Calendar + Bonus (teal), Bellman Proxy (purple), and Bonus-Aware Bellman Proxy (green). The Bonus-Aware Bellman Proxy achieves the highest terminal wealth (~$56,273), while the static benchmark reaches ~$54,389, demonstrating the value of eligibility-aware optimization.

Figure 4: Reported Monte Carlo Premiums by Strategy

A bar chart displaying the promotional arbitrage premium (percentage gain over static HYSA) across five strategies based on 50,000-path Monte Carlo simulations. The regime-switching optimal strategy yields the highest premium at 14.65%, followed by Promo + Bonus at 12.06%, Optimal Deterministic at 7.59%, and Simple Calendar Rotation at 4.48%. Error bars reflect simulation standard errors.

Figure 5: Sensitivity of the Promotional Arbitrage Premium to Key Parameters

A tornado-style sensitivity diagram showing how the arbitrage premium (%) varies under low, base, and high scenarios for six parameters: tax rate, bonus size, switching cost, promotional volatility, requalification period, and inflation. The premium is most sensitive to bonus size (range: ~11% to ~29%) and requalification length, and least sensitive to inflation. Base-case premium is anchored at approximately 20%.

Figure 6: Eligibility-State Transitions

A state-transition diagram illustrating the three eligibility states for each bank: Ineligible/Waiting (0), Eligible for Promotion (1), and Active Promotion (2). Arrows indicate transitions triggered by account opening/switching, promotional expiration, and requalification period lapse. This figure visualizes the mode-dependent feasibility structure that distinguishes the model from standard switching problems.

Data Description for Zenodo Repository

The replication dataset (data_V3.xlsx) contains seven worksheets supporting the numerical calibration, simulation, and sensitivity analysis reported in the manuscript. All monetary values are in U.S. dollars unless otherwise specified.

Worksheet

Description

Records

Key Variables

Fed_Rate_Cycles

Federal Funds Effective Rate and CPI inflation from January 2020 through May 2026, retrieved from FRED (DFF series). Used to calibrate the opportunity-cost rate and macroeconomic context.

24

Period, Fed_Funds_Rate_Pct, Inflation_CPI_Pct

Bank_Promo_Rates

Six-bank promotional deposit product calibration extracted from Bankrate and Marcus.com (2025). Defines the core parameters for the dynamic programming model.

6

Bank, Promo_APY_Pct, Normal_APY_Pct, Promo_Window_Months, Requal_Months, Deposit_Cap_Dollars, Bonus_Dollars

Sim_Parameters

Baseline simulation parameters including initial capital ($50,000), 3-year horizon, 22% tax rate, 2-day ACH delay, $250,000 FDIC limit, and 50,000 Monte Carlo paths.

14

Parameter, Value, Unit, Source

MC_Results

Monte Carlo simulation outcomes for six strategies: Static HYSA, Simple Rotation, Optimal DP, Promo + Bonus, Regime-Switching, and Behavioral.

6

Strategy, Mean_Terminal_Wealth, Std_Dev, Pi_Pct, Interest_Contribution_Pct, Bonus_Contribution_Pct, Cost_Contribution_Pct

Bootstrap_CIs

Bias-corrected bootstrap confidence intervals (B = 10,000) for the arbitrage premium of four key strategies.

4

Strategy, Point_Estimate, SE, CI_95_Lower, CI_95_Upper, B

Behavioral_Frictions

Sensitivity analysis incorporating behavioral frictions: switching fatigue, attention costs, status quo bias, and loss aversion.

6

Friction_Type, Parameter, Pi_Pct, Optimal_Switches, Welfare_Loss_Dollars

Offer_Level_Summary

Descriptive statistics from an expanded offer-level dataset (N = 847 promotional deposit offers) spanning January 2024 to May 2026.

12

Statistic, Value

Data Sources

Federal Reserve Economic Data (FRED), Bankrate, Marcus by Goldman Sachs, FDIC Quarterly Banking Profile, IRS Publication 17, NACHA Operating Rules, Doctor of Credit. Retrieval date: 2026-05-26.

Usage

These data support replication of the dynamic programming calibration, Monte Carlo validation, bootstrap inference, and sensitivity analyses reported in the manuscript. The six-bank subset (Bank_Promo_Rates) is the primary calibration input; the offer-level summary provides external validity for parameter ranges.

Files

fig1_offer_rates.png

Files (898.4 kB)

Name Size Download all
md5:5faf7e4ce85e94e1557083cf13f2cc6d
10.9 kB Download
md5:bdaff6bc937bb4e292b30dd661148336
115.6 kB Preview Download
md5:9ea34357f24811ab93f736f804286843
176.0 kB Preview Download
md5:9a9ff8bd324e5cfeff924628a1a8de50
228.7 kB Preview Download
md5:6c18c8fe60ac2182c2f18e05ee2e00b5
192.4 kB Preview Download
md5:f0728191675988e959ac920eacf4d1ae
102.8 kB Preview Download
md5:4f5828ae6dc563bded4612986b89d3e3
72.0 kB Preview Download