Calibration scripts for the BeforeIT.jl ABM for the 27 EU member countries

This repository contains scripts that can be used to generate initial conditions and parameters for the BeforeIT.jl agent-based model.

The scripts are based on the Matlab and PostgreSQL scripts used in Poledna et al. (already extended to work for all EU27 countries) and were kindly shared with us. To ease the accessability of this work, we translated the scripts to Julia. The only prerequisite for this repository is Julia, there are no outside (technical) dependencies.

Installation

To be able to use the package, you can activate a new Julia environment in any folder from the terminal by typing

julia --project=.

Then, whithin the Julia environment, you can install CalibrateBeforeIT.jl as

using Pkg
Pkg.add(url = "https://github.com/ViennaInstitute/CalibrateBeforeIT.jl")

You can ensure to have installed all dependencies via

Pkg.instantiate()

Now you should be able to run the example code scripts described just below.

Aside: ] enters the "pkg" mode where you can make changes to the needed package dependencies by this project. Press backspace to return to normal "julia" mode. Equivalently, ? is for help mode, ; for shell mode.

Example Usage

Step 1: download Eurostat tables as .parquet files

Run script 01_download_raw_eurostat_tables.jl. Reserve at least 3GB of disk space and allow the script to run one or two hours.

The following steps are carried out:

1. Download of the necessary Eurostat tables, convert to long format and save as .parquet files in the data/ directory.

After that, the data/ directory is populated with the .parquet files that we then need for the next steps.

Step 2: preprocess Eurostat tables

Run script 02_preprocess_raw_eurostat_tables.jl.

The following steps are carried out:

1. Input-output coefficients are separated into three tables (2010-2014, 2015-2019, 2020 onwards). Because we need a time-series, we append the three tables into one.
2. Firm counts are not available through just one table: We have to extract the necessary data items and write them to disk for later easier querying.

The data are preprocessed to help in the generation of calibration inputs.

Step 3: generate the calibration input data for a given country

Run script 03_create_calibration_data.jl. In the script, one can specify the country, year ranges, etc that are parameters to the calibration process.

The following steps are carried out:

1. Import data for the EA19 country aggregate. This is used as an imputation device for countries that lack certain time series.
2. Then, for a given country:
   1. Import Figaro (input-output) data
   2. Import GDP, GVA, imports, exports time series
   3. Import data series for initial stocks and flows
3. From the three data sources and the EA19 data, initial conditions and parameters for a given point in time are generated. These initial conditions and parameters serve as input data for the model.

References

Glielmo, A., Devetak, M., Meligrana, A., & Poledna, S. (2025). BeforeIT. jl: High-Performance Agent-Based Macroeconomics Made Easy. arXiv preprint arXiv:2502.13267. https://doi.org/10.48550/arXiv.2502.13267

Poledna, S., Miess, M. G., Hommes, C., & Rabitsch, K. (2023). Economic forecasting with an agent-based model. European Economic Review, 151, 104306. https://doi.org/10.1016/j.euroecorev.2022.104306