DemonetSim-TH: simulated demonetization of Thailand's 1,000-baht note

Every row is a simulation. The model's inputs are approximations compiled without access to live sources and parameter ranges that are assumptions. The outputs describe the model, not Thailand: verify the inputs against primary sources, in particular Bank of Thailand banknote statistics, before citing them or using them for policy.

Data for the preprint Can Demonetization Defeat Grey Capital? Evidence, a Calibrated Agent-Based Model, and a Research Agenda for Retiring Thailand's 1,000-Baht Note by Thanatchon Chongmankhong (read online · PDF · ฉบับภาษาไทย, ธนัชชนม์ จงมั่นคง). DemonetSim-TH is an agent-based model in which illicit holders choose among deposit, structuring through money mules, conversion into gold, FX or crypto, amnesty and abandonment under capacity-constrained laundering, while honest holders face a liquidity gap governed by replacement logistics. It is calibrated to India's 2016 demonetization by rejection ABC on the 99.3% return rate. Code: https://github.com/thanatchon36/demonetsim-th; archived on Zenodo: 10.5281/zenodo.24050061.

Independent personal research. Does not represent the views of any employer or institution.

Files

path	content
results/mc_results.csv	Monte Carlo draws: 600 for each Thai scenario S1–S4 and 600 ABC-accepted India draws (IN16), 3,000 rows. Each row holds the sampled parameters, the honest-side outcomes, the fate of illicit cash, the channel choices and the return rate
results/scenario_summary.csv	per scenario: median, 25th/75th and 10th/90th percentiles and mean of the disruption index, the bite and each fate of illicit cash
results/india_prior_draws.csv	4,000 draws from the India 2016 prior, same columns as mc_results.csv without liq_index
results/india_abc_accepted.csv	the 1,214 prior draws whose simulated return rate is within 99.3% ± 0.3 pp
results/india_abc_posterior_summary.csv	count, mean, standard deviation and percentiles of the accepted draws
results/india_calibration_curve.csv	simulated return rate against the illicit share (2-pp bins): median, 10th and 90th percentiles and the ABC acceptance rate
results/sensitivity_s1.csv	median bite and laundered share in S1 at an illicit share of 10%, by bank scrutiny (0.3, 0.5, 0.8) and laundering capacity (% of the stock over the 50-day window); 80 draws per cell
results/prcc_s1.json	partial rank correlation coefficients of 13 parameters with the bite (hit), the laundered share (laund) and the disruption index (liq), scenario S1
results/summary_stats.txt	PRCC ranking, bootstrap standard errors of the medians (1,000 resamples) and ABC prior/posterior percentiles
results/denominations.csv	19 large banknotes: face value, GDP per capita and exchange rate (approximate), status, value as % of GDP per capita, USD value and days of average income
results/bulk_10m_thb.csv	number of pieces, weight (kg) and volume (litres) of THB 10 million in five banknotes, gold and a stablecoin
results/logistics_summary.csv	5th, 50th and 95th percentiles (THB) of the stock of 1,000s, notes removed and added, printing costs, seigniorage shift and leakage, and ATM conversion cost
results/fig*.png	the six diagnostic plots written by the model
inputs/scenarios.csv	the scenario definitions: design flags and the lower (_lo) and upper (_hi) bounds of every uniform prior
inputs/holders.csv	the six holder groups for Thailand (TH) and India (IN): share of the demonetized stock, transaction weight, bank access, digital readiness, cash dependence, and the conversion time constant
SHA256SUMS	checksums of every file above

Columns of the simulation draws

Parameters (sampled per draw): r_day replacement currency per day and pre_supply replacement ready at day 0 (shares of the stock), R0 share replaced when legal tender lapses, kyc_bank and kyc_dealer scrutiny at banks and at gold, FX and crypto dealers (0–1), fee_struct and fee_conv cost of structuring and of conversion, q_struct and q_conv relative detectability, cap_struct_total and cap_conv_total laundering and conversion capacity (shares of the total stock), amnesty_tax, illicit_share share of the demonetized stock held illicitly, black_frac share of illicit holders with criminal rather than tax-evaded money.

Outcomes: gap_days liquidity gap, liq_raw disruption and liq_index the same indexed to the mean India prior draw (= 100); honest_lost and cross_border_lost honest wealth not converted (shares of the stock); burden_vulnerable_informal share of disruption borne by vulnerable households and informal businesses. The fate of illicit cash, as shares of the illicit stock summing to one: extinguished (abandoned), captured (seized, taxed or fined), regularized (kept after a penalty), laundered (entered banks undetected) and displaced (moved into other stores of value or paid as fees); hit = extinguished + captured is the bite. ch_* are the value shares of the channel each agent ended in, and return_rate the share of demonetized value returned.

Scenarios: IN16 India 2016 (calibration), S1 shock with a 50-day window, S2 announced 90-day deadline, S3 five-year phase-out, S4 phase-out with an anti-money-laundering package.

Reproducing

git clone https://github.com/thanatchon36/demonetsim-th && cd demonetsim-th
make setup && make simulate analysis      # seed 7, 600 draws per scenario
python scripts/export_hf.py               # rebuilds this dataset in build/hf/

Licence

CC BY 4.0 (LICENSE). The code that produced the data is MIT licensed.

Citation

@misc{chongmankhong2026demonetization,
  author = {Chongmankhong, Thanatchon},
  title  = {Can Demonetization Defeat Grey Capital? Evidence, a Calibrated Agent-Based Model,
            and a Research Agenda for Retiring {Thailand}'s 1,000-Baht Note},
  year   = {2026},
  doi    = {10.5281/zenodo.24050061},
  url    = {https://thanatchon36.github.io/demonetsim-th/paper/},
  note   = {Preprint}
}