Behavioural Macro and Finance

Agent-based models of the macroeconomy, taught with BeforeIT.jl — the Julia implementation of the Poledna et al. macro ABM, calibrated on Austrian and Italian national accounts.

The course combines two lectures with five tutorials and an optional Julia primer. The notebooks move from inspecting a calibrated economy to comparing interventions, extending the model and estimating its parameters. Each lesson asks you to predict an outcome, run a controlled experiment and explain what the result establishes.

Start here

• Lecture slides — both decks as PDF.
• Running the notebooks — install Julia and Pluto, or just read online.
• Core route: tutorials 01–03 — understand the model, assess forecasts and compare policy responses.
• Advanced route: tutorials 04–05 — build an extension and investigate likelihood-free inference.

Read the saved notebook outputs online, or run the source notebooks locally to change parameters. Tutorial 05 is rendered in reading mode; its inference experiments run locally.

What the tutorial covers

00 — Julia + Pluto primer. Optional; skip it if you already write Julia.

01 — Meet the model. Run the Austrian economy forward and read off GDP, then open the box: the agent types and their counts, the input–output network, what happens inside a quarter, and whether the stock-flow accounting closes.

02 — Calibration, no-burn-in and forecasting. Identify data-based estimates and policy assumptions, inspect initialization without a burn-in, and compare indexed model paths with a simple benchmark. Interpret simulation bands and distinguish dispersion across paths from evidence of forecasting accuracy.

03 — Shocks, expectations and policy. Compare shocks with matched baselines using the same random seeds. Interpret output, consumption, wages and interest rates, then investigate how expectations and the euro-area monetary policy rule affect the response.

04 — Extending the model: CANVAS. Implement a CANVAS-inspired pricing extension, apply a persistent wage-norm intervention and vary pass-through channels. Compare each configuration with its own baseline and interpret price levels and cumulative inflation.

05 — Likelihood-free calibration. Estimate the extension’s parameters from simulated observations with ABC, NPE and optional NRE. Work through parameter recovery, posterior checks and the limits of these experiments for empirical and policy validation. Reading mode is lightweight; inference requires a local Python environment, and runtime depends on the simulation budget and hardware.

The tutorials connect to the lecture material and include guided exercises with hints and diagnostic checks. Record the seed, horizon, units and baseline for each experiment.

Acknowledgement
The design of this website is based on Computational Thinking, a live online Julia/Pluto textbook. (computationalthinking.mit.edu)