# Digital money in a crisis ODD description of prototype v0.1, NetLogo 7.0.4. The model is theoretical and uncalibrated. Amounts are abstract model units and steps have no day/hour correspondence. The code uses the MIT licence; NetLogo and its web engine retain their own licences. ## 1 Purpose Explore how CBDC access, a holding limit, social influence and a crisis signal interact with deposit withdrawals and bank liquidity stress. Cash-only comparisons change transfer frictions in the original model and must be interpreted accordingly. No empirical Turkish predictions are claimed. ## 2 Entities and state variables Households hold deposits, cash and CBDC, belong to a bank, have a panic threshold, four illustrative perception scores, a previous-step transfer indicator and a target instrument. Initial deposits are uniform from 80 to 120; panic thresholds are uniform from 0.35 to 0.80; each perception score is uniform from zero to one. These are not observed or validated SPAR scores. Banks hold reserves and fixed loans, aggregate deposits, a stress indicator and per-step cash-service capacity. Initially reserves plus loans equal deposits. There is no capital, asset sale, deposit insurance, interbank funding or central-bank liquidity support. The central bank is represented implicitly through reserves, cash and CBDC liabilities; the aggregate monetary liability is conserved. The household network starts as a degree-four ring; each edge is rewired with probability 0.05 without self-links or duplicate links. Spatial drawing positions have no behavioural effect. Bank assignment and the original ring share an agent index, which limits bank-specific contagion analysis. ## 3 Process overview and scheduling 1. Reset per-step unserved demand and replenish bank cash-service quotas. 2. Compute every household’s target and withdrawal tendency from previous-step bank stress and neighbour transfers. 3. Serve households in a deterministic identity order with a rotating starting position. All transfers consume the household’s bank reserve. CBDC is also subject to the remaining individual holding limit; cash is also subject to the bank’s service quota. 4. Carry forward the new transfer indicator, reaggregate bank deposits and update stress when reserves are below 8% of current deposits. 5. Check accounting identities and advance the tick and plots. ## 4 Design concepts Aggregate withdrawals and stress arise from individual transfers; stress feeds back into next-step decisions. Agents adapt to neighbour activity, bank stress and instrument appeal, without learning or optimisation. Interactions occur through social links and shared bank reserves. Randomness is used for initial conditions and network construction, not for additional decision noise during the run. A fixed seed reproduces the initial population and network. There is no collective decision organisation or spatial transaction cost. Observed outputs include total deposits, cash and CBDC, stressed-bank count, the share of households at their limit, per-step unserved transfer demand and CBDC wealth shares by high/low surveillance concern. The cap indicator at a zero limit is not a CBDC usage indicator. Repeated unserved requests cannot be summed as distinct wealth losses. ## 5 Initialisation and inputs Setup resets balances, network and plots using the chosen random seed. Startup prepares the initial view. No external data or participant answers are read or collected. | Parameter | Range | Default | Meaning | |---|---|---|---| | Households | 100–1000, step 100 | 500 | Illustrative agent count, not a study sample | | Banks | 3–5 | 4 | Illustrative sector | | CBDC holding limit | 0–100, step 5 | 20 | Model units, not an announced CBRT limit | | No limit | off/on | off | Removes the CBDC cap | | CBDC available | off/on | on | Enables cash-only comparison | | Crisis intensity | 0–1 | 0.50 | Signal from tick 10 | | Social influence | 0–1 | 0.30 | Previous neighbour-transfer weight | | Initial reserve ratio | 0.05–0.50 | 0.20 | Illustrative liquidity, not a regulatory ratio | | Privacy scenario | 0–1 | 0.80 | Illustrative design perception | | Interest appeal | off/on | off | Adds 0.04 to appeal; no interest accrual | | Effective redress | off/on | on | Changes the sign of the redress appeal weight | | Cash capacity | 1–20 | 4 | Successful cash transfers per bank and step | | Seed | 1–100 | 1 | Reproducible initialisation | ## 6 Submodels Appeal combines redress confidence weighted by +0.08 or -0.08, payment autonomy weighted by 0.04, and privacy expectation multiplied by the privacy scenario and 0.04. Interest appeal, when enabled, adds 0.04. Surveillance concern multiplied by one minus the privacy scenario and 0.20 is the privacy penalty. Base withdrawal tendency sums crisis news, social influence times the previous neighbour-transfer share, and 0.25 if the household’s bank was stressed. If CBDC is available, the household has room under the holding limit and net appeal is at least -0.03, the target is CBDC and net appeal is added to withdrawal tendency. Otherwise the target is cash. A transfer is requested when tendency exceeds the individual panic threshold. CBDC requests are 8% of remaining deposits, capped at 10 units per step. Cash requests are 2%, capped at three units, and no more than the bank’s cash-service capacity can succeed in a step. The actual amount is bounded by the request, available bank reserve and, for CBDC, the unused individual limit. Every successful transfer reduces deposits and bank reserves by the same amount and increases cash or CBDC accordingly. Household wealth, bank balance identity and total central-bank liabilities are checked to a tolerance of 0.000001 units. Because aggregate withdrawal equals reserve depletion, initial reserves mechanically cap total withdrawals. Reaching a 20% ceiling when the reserve ratio is 0.20 is not an independently discovered behavioural law. ## 7 Experiments and verification The original smoke experiment has 12 runs; the limit/crisis experiment has 150 runs across five limits, three crisis intensities and ten seeds; the interface-extreme checks have 26 runs. The downloadable CSV contains the original limit/crisis outputs, not the later 320-run mechanism audit. Compilation is checked in both the self-hosted web compiler and native NetLogo 7.0.4. Accounting verification is not empirical validation. A separate instrumented research model compares six regimes, two crisis intensities, 120/480-step horizons and, where applicable, 10/50-unit caps with eight paired seeds per cell. Its 320 native runs isolate appeal coupling and equalise cash transfer conditions. This is a separate experiment model; the interactive page still runs original v0.1. ## 8 Limitations and calibration No credit contraction, bank capital losses, asset sales, liquidity support, deposit insurance, interbank transfers, interest accrual or daily retail purchases are modelled. Reserve depletion is not bank failure. Very low initial reserve ratios can trigger stress without a crisis. Network homophily and bank assignment need independent controls. Behavioural scores and weights are illustrative; future measurement or conjoint scores are not causal transfer coefficients without further estimation. Holding limits, offline tools and interface illustrations are not official product specifications. ## 9 Sources Grimm et al. (2020). The ODD Protocol for Describing Agent-Based and Other Simulation Models. JASSS 23(2), 7. https://doi.org/10.18564/jasss.4259 Wilensky, U. (1999). NetLogo. Northwestern University. https://docs.netlogo.org/ Barucci, Gurgone, Iori and Azzone (2025). Central Bank Digital Currency, Flight-to-Quality, and Bank-Runs in an Agent-Based Model. https://arxiv.org/abs/2510.21071 Model motivation is informed by the literature, but behavioural coefficients were not estimated from these sources.