In a crisis digital money

How to use it?

1. Choose the parameters, then press Setup .
2. Run / stop to proceed. The crisis news arrives from step 10 onwards.
3. Keep the same random seed, change one parameter, and set up the model again.

Amounts are model units; one step is not a day or an hour. All starting figures and behavioural coefficients are illustrative.

Three comparisons

1. CBDC on / off: compare the payment options under the same crisis, seed and cash capacity.
2. Limit 10 / 50 / unlimited: keep the other settings fixed. Look at the deposit graph, the shift to cash and the households at the limit.
3. Social influence 0 / 0.8: examine how neighbours' decisions spread at the same crisis intensity.

What is happening inside?

1. Initial state: each household receives 80 to 120 units of deposits and a different decision threshold. The share of bank deposits you select starts as reserves, and the remainder as fixed loan assets.
2. News and neighbours: the crisis news takes effect from step 10. The household considers the intensity of the news, its neighbours who converted in the previous step, and bank stress.
3. Choice: CBDC access, the gap to the limit and the privacy/redress scenario affect the target instrument. If CBDC is not chosen, cash becomes the alternative.
4. Realisation: The decision does not mean the conversion has taken place. The bank reserve, the CBDC limit and the cash service quota limit the amount.
5. Feedback: conversion reduces both deposits and reserves. When the reserve-to-deposit ratio falls below 8%, the bank is treated as stressed; this information feeds into the decision at the next step.

What does each indicator say?

Where does the gap in a chart come from?

This version assumes faster transfers to CBDC than to cash. Comparing CBDC on/off changes both instrument availability and transfer speed. The mechanism audit below tests this difference separately under equal transfer conditions.

Because no new liquidity arrives from the central bank, total conversion is limited by the initial reserve. A higher CBDC limit therefore does not always produce more conversion. When interpreting the chart, the behavioural rule and this accounting constraint should be read together.

A 120-step horizon may end the cash channel early: in a separate validation experiment with the same start, when the run was extended to 480 steps, cash outflows also reached the reserve ceiling. Because bank assignment and the retail network are tied to the same index, social spread within banks should be tested separately.

Model assumptions

Households take into account the neighbours' behaviour in the previous step and bank stress. Conversion is limited by bank reserves. Once the limit is reached, cash remains as the alternative. In this version, digital conversion is assumed to be faster than cash service; cash capacity is a separate control.

The interest rate switch only changes the attractiveness of preferences; it does not accrue interest. Privacy and redress options are hypothetical designs.

Credit assets are held fixed. Deposit insurance, central bank liquidity support, asset sales and interbank transfers are not present in this version. Model results should be read together with these assumptions.

Rules or behaviour?

The new mechanism audit matches 40 experimental cells on the same eight initial populations: 320 NetLogo runs. We equalised cash-transfer conditions, separately removed the addition of CBDC appeal to withdrawal propensity, and extended the horizon from 120 to 480 steps.

At a high limit, aggregate outflow can be identical while its allocation between cash and CBDC differs. A limit of 10 also changes short-run speed and asset composition. Initial reserves set the ceiling here; a more comprehensive financial-system model may produce different results.

Download the 320-run data · Code, raw outputs and reproduction bundle

Method and replication

At every step, the model checks the conservation of household wealth, central bank monetary liabilities and bank balance sheets. The 150-condition limit × crisis scan was run with 10 different seeds. The code, parameter table and experiment definition are included in the downloadable method file.

The model code is MIT licensed. NetLogo and the web runtime are subject to their own licences. Model licence.