Stock-Flow Consistent Agent-Based Model engine for studying phase transitions in AI-driven labor market automation. Powers all five papers in the complexity-econ series (40,000+ Monte Carlo simulations).
Simulates 10,000 heterogeneous firms across 6 sectors (GUS 2024 calibration) interacting on a configurable network topology over 120 months. Firms decide whether to adopt AI, go hybrid, or stay traditional based on profitability thresholds, network mimetic pressure, and sector-specific CES elasticity of substitution (σ). A BDP (basic disposable payment) shock at month 30 triggers the automation transition.
Every monetary flow is tracked bilaterally: government taxes and spends, banks lend and absorb losses, the central bank sets rates via Taylor rule, and the exchange rate floats via balance-of-payments (PLN) or stays fixed under SGP fiscal constraint (EUR).
src/main/scala/sfc/
├── Main.scala Entry point + Monte Carlo orchestrator
├── config/
│ └── SimConfig.scala 40+ parameters, 2 regimes, 4 topologies, env var overrides
├── agents/
│ ├── Firm.scala TechState (Traditional→Hybrid→Automated→Bankrupt), decision logic
│ ├── Household.scala Aggregate household state
│ └── CentralBank.scala NBP/ECB rate wrapper
├── engine/
│ ├── Simulation.scala Monthly step: labor → demand → firms → bank → inflation → forex → fiscal
│ └── World.scala Immutable world state (macro + per-sector σ vector)
├── dynamics/
│ ├── SigmaDynamics.scala Endogenous σ evolution (Arthur-style learning-by-doing)
│ └── DynamicNetwork.scala Death-birth rewiring with preferential attachment
├── networks/
│ └── WattsStrogatz.scala WS, Erdos-Renyi, Barabasi-Albert, ring lattice generators
└── sfc/
└── BalanceSheet.scala GovState, BankState, ForexState (SFC accounting)
- PLN: Floating exchange rate, endogenous NBP Taylor rule (α=1.5, β=0.8, inertia=0.70)
- EUR: Fixed rate at 4.33 PLN/EUR, exogenous ECB Taylor rule, SGP fiscal constraint (3% deficit, 60% debt ceiling, austerity κ=2.0)
- Watts-Strogatz small-world (k=6, p=0.10) — default
- Erdos-Renyi random graph
- Barabasi-Albert scale-free (preferential attachment)
- Ring lattice (no rewiring)
- 4-state technology transition: Traditional → Hybrid → Automated (+ Bankrupt)
- Profitability threshold via CES elasticity:
θ = min(1.0, 0.88 + 0.075·log₁₀(σ)) - Network mimetic pressure: 40% weight on neighbors' adoption + 40% global panic
- Uncertainty discount reduced by demonstration effect (neighbors already automated)
- Sector-specific digital readiness, failure rates, and hybrid worker retention
- σ evolution:
σ(t+1) = σ(t) + λ·σ(t)·adoption(t)·(1 - σ(t)/cap)with ratchet + hard cap - Network rewiring: bankrupt firms replaced with probability ρ, new entrants wire via preferential attachment
- Both disabled by default (λ=0, ρ=0) — backward compatible with Papers 01–04
- Phillips curve inflation (demand-pull + cost-push + import + tech-deflation)
- Soft deflation floor at -1.5%/mo (Bewley 1999 wage rigidity)
- Banking sector with NPL tracking, CAR constraint, endogenous lending rates
- Balance-of-payments forex with IRP arbitrage (PLN) or fixed rate (EUR)
- Government fiscal balance with SGP constraints (EUR only)
| Sector | Share | σ (CES) | Digital readiness | Hybrid retain |
|---|---|---|---|---|
| BPO/SSC | 3% | 50.0 | 0.50 | 50% |
| Manufacturing | 16% | 10.0 | 0.45 | 60% |
| Retail/Services | 45% | 5.0 | 0.40 | 65% |
| Healthcare | 6% | 2.0 | 0.25 | 75% |
| Public | 22% | 1.0 | 0.08 | 90% |
| Agriculture | 8% | 3.0 | 0.12 | 85% |
- JDK 17+
- sbt
# BDP = 2000 PLN, 30 seeds, PLN regime
sbt "run 2000 30 baseline pln"# EUR regime with SGP
sbt "run 2000 30 baseline eur"# Fat JAR (faster, no sbt startup overhead)
sbt assembly
java -jar target/scala-3.5.2/sfc-abm.jar 2000 30 baseline pln# Network topology
TOPOLOGY=er java -jar sfc-abm.jar 2000 30 test pln # Erdos-Renyi
TOPOLOGY=ba java -jar sfc-abm.jar 2000 30 test pln # Barabasi-Albert
TOPOLOGY=lattice java -jar sfc-abm.jar 2000 30 test pln # Ring lattice# Custom σ per sector
SIGMAS="8.97,4.88,4.95,1.0,0.8,1.44" java -jar sfc-abm.jar 2000 30 test pln
# σ multiplier (all sectors)
SIGMA_MULT=2.0 java -jar sfc-abm.jar 2000 30 test pln
# System size
FIRMS_COUNT=1000 java -jar sfc-abm.jar 2000 30 test pln
# Decision rule tuning
DEMO_THRESH=0.25 java -jar sfc-abm.jar 2000 30 test pln
# Endogenous σ (Paper-05)
SIGMA_LAMBDA=0.02 java -jar sfc-abm.jar 2000 30 test pln
# Dynamic network (Paper-05)
REWIRE_RHO=0.1 java -jar sfc-abm.jar 2000 30 test pln
# Full endogenous (both mechanisms)
SIGMA_LAMBDA=0.02 REWIRE_RHO=0.1 java -jar sfc-abm.jar 2000 30 test plnCSV files written to mc/ (European format: semicolon separator, comma decimals):
mc/<prefix>_terminal.csv— per-seed terminal values at month 120 (26 columns)mc/<prefix>_timeseries.csv— aggregated time series (mean, std, p05, p95)
26 output columns: Seed, Inflation, Unemployment, TotalAdoption, ExRate, MarketWage, GovDebt, NPL, RefRate, PriceLevel, AutoRatio, HybridRatio, 6× sector adoption, EffectiveBDP, 6× sector σ, MeanDegree
93 tests across 8 suites, run on every push via GitHub Actions:
sbt test| Suite | Tests | Covers |
|---|---|---|
| FirmSpec | 14 | Tech states, worker counts, capacity, σ threshold |
| SimConfigSpec | 12 | Sector calibration, parameter bounds, regime detection |
| NetworkSpec | 10 | WS/ER/BA/lattice degree, symmetry, connectivity |
| BalanceSheetSpec | 10 | NPL ratio, CAR, lending rate, credit constraint |
| SimulationSpec | 12 | Labor market, inflation channels, Taylor rules |
| SigmaDynamicsSpec | 6 | Learning-by-doing, ratchet, cap, backward compat |
| DynamicNetworkSpec | 6 | Death-birth, preferential attachment, backward compat |
| IntegrationSpec | 23 | End-to-end: 120×26 output, reproducibility, bounds |
- Scala 3.5.2 / sbt
- Zero runtime dependencies (stdlib only)
- ScalaTest 3.2.19 for unit tests
- sbt-assembly for fat JAR packaging
See CITATION.cff
MIT