thanatchon36/stablerun

When Stablecoins Run: simulation outputs

Every simulation result behind the working paper When Stablecoins Run: Fire-Sale Spillovers to the Treasury Bill Market. An Agent-Based Stress Test (Thanatchon Chongmankhong, November 2026). The data are outputs of stablerun, a calibrated agent-based model of a payment-stablecoin run and its spillover to the US Treasury-bill market.

• Paper (PDF): https://github.com/thanatchon36/stablerun/raw/main/paper.pdf
• Code: https://github.com/thanatchon36/stablerun (python scripts/run_all.py --experiment all regenerates every file here)
• DOI: https://doi.org/10.5281/zenodo.23996044 (Zenodo; resolves to the latest release)

These are model outputs, not observations. Nothing here is market data. The institutional inputs (issuer sizes, reserve mixes) are defaults flagged VERIFY in config/params_default.yaml, and the model is a stress-testing tool, not a forecast.

Files

Config	Rows	Content
e1_paths	10,000	E1 baseline: one row per Monte Carlo path, 1,000 paths for each issuer (A, B) and scenario
e1_summary	10	E1 medians, 5th and 95th percentiles and probabilities per issuer and scenario
e2_tipping_point	60	E2: peak impact against issuer size (1, 2, 3, 5, 10× today), by scenario, impact form and redemption rails, 300 paths per cell
e2_crossings	12	E2: size multiple at which median peak impact first exceeds 10, 25 and 50 bp
e3_policy_grid	196	E3: cash buffer × redemption fee × maturity cap × Fed backstop, plus par against NAV redemption, 200 paths per cell
e4_contagion	6	E4: a run on Issuer A spills over to Issuer B's beliefs with weight beta
e5_sensitivity	30	E5: one-at-a-time sensitivity (tornado), low and high value of each parameter, 200 paths per run
benchmark_check	9	Simulated exit share with behavioural channels off against the closed-form threshold aggregate
calibration_grid	48	Calibration search over the behavioural parameters, with loss and fitted moments for both episodes
calibration_validation	5	Fitted moments against the USDC (March 2023) and USDT (May 2022) targets

config/params_default.yaml holds every parameter with its source and a confidence note. The E1 seed is 20261105.

Scenarios

Name	Meaning
control	No shock
loss05, loss10, loss20	Solvency run: 5, 10 or 20% of reserves exposed and lost
scare08 (E1), scare (E2 to E5)	Liquidity run: 8% of reserves exposed, resolved without loss on day 2, one day without processing (USDC 2023-like)

Issuer A is a large offshore issuer (supply USD 180 bn); Issuer B is a US-regulated issuer (supply USD 75 bn). In E2 the rails are calibrated (processing capacity from the calibration), fast (5% of supply per day) or instant (20% per day); impact is sqrt (square root, main case) or linear (upper bound). E3 and E5 use two configurations: scare_1x_calibrated (liquidity run at today's size) and loss10_5x_fast (solvency run at 5× size with fast rails).

Columns of e1_paths

The horizon is 30 business days. Dollar amounts are USD billions.

Column	Definition
issuer, scenario, path	Issuer, scenario and path index
min_price	Lowest secondary-market price of the token (par = 1)
R_day7, R_day14, R_end	Cumulative redemptions as a share of initial supply on day 7, day 14 and day 30
day_R10	First day on which cumulative redemptions reach 10% of supply (empty if never)
peak_Y_bp, Y_day7_bp, Y_end_bp	Three-month bill-yield impact in basis points: peak, day 7 and day 30
bills_sold	Bills sold outright before maturity
matured	Bills that matured over the horizon
withdrawn	Matured bills not reinvested: demand withdrawn from auctions
net_demand_withdrawn	Peak cumulative net demand withdrawn from the bill market
other_sold	Non-bill reserve assets sold
peak_inv_over_K	Peak dealer inventory relative to dealer capacity
peak_queue_days	Peak redemption queue in days of processing capacity
susp_days	Days on which redemptions were suspended
fs_loss	Issuer's fire-sale loss
impairment_loss	Realised reserve loss from the shock
nav_end	Net asset value per token on day 30
run, severe_run	Redemptions by day 30 of at least 20% and 50% of supply
Y_gt_10bp, Y_gt_25bp, Y_gt_50bp	Peak impact above 10, 25 and 50 bp

Summary tables use the same names with suffixes _med, _p5 and _p95 for the median and percentiles across paths, and run_prob, susp_any and P_Y_gt_* for shares of paths.

Loading

from datasets import load_dataset
paths = load_dataset("thanatchon36/stablerun", "e1_paths", split="train").to_pandas()
paths.groupby(["issuer", "scenario"]).peak_Y_bp.median()

License and citation

Data: CC BY 4.0. Code: MIT (GitHub repository).

@unpublished{chongmankhong2026stablerun,
  author = {Chongmankhong, Thanatchon},
  title  = {When Stablecoins Run: Fire-Sale Spillovers to the Treasury Bill Market. An Agent-Based Stress Test},
  note   = {Working paper and code repository},
  year   = {2026},
  doi    = {10.5281/zenodo.23996044},
  url    = {https://github.com/thanatchon36/stablerun}
}