Spoiler: Tax avoidance/evasion
Work with Valeria Secchini (PhD candidate at COPTAX, Charles University)
First order
Second order
PH
PL
...
Xu, Jian, Thanuka L. Wickramarathne, and Nitesh V. Chawla. "Representing higher-order dependencies in networks." Science advances 2.5 (2016): e1600028.
Q: Is PH a better representation of the data than PL?
PH
PL
Test: are the two distributions different?
Kullback-Leibler divergence > δ
Problem: overfitting
(high number of false positives)
On realistic synthetic data:
Precision: 7%
Sensitivity: 62%
Our solution: Train/Validation approach
Find pattern in train dataset, keep if it holds in validation
On realistic synthetic data:
Precision: 90%
Sensitivity: 60%
Validation
Training
60%
50%
40%
50%
100%
40%
60%
We want to find if PH is different from PL:
* Divergence between PH and PL in training > δ1
PH
PL
...
...
...
We want that distance to be reliable:
* Divergence between PH in training and validation > δ2
* Divergence between PL in training and validation > δ3
Which thresholds to use?
1. Create synthetic data at the path level
2. Fit a model on that data
Open questions:
How to best create synthetic data?
How to handle heterogeneity? Higher-order dependencies may be "owner dependent" (the first node in the chain)
How to handle time series? (e.g. detect changes in structures, measure effect of policy)
(lots on results: community, ranking, simulation)
Results:
Applied to multinational corporations
70 patterns, 75% containing a tax haven
p.
p.
p.
8150 companies
9.3/window
Garcia-Bernardo, Javier, Joost Witteman, and Marilou Vlaanderen. "Uncovering the size of the illegal corporate service provider industry in the Netherlands: a network approach." EPJ Data Science 11.1 (2022): 23.
Strategy:
Open questions:
Better way to find illegal directors? (GNNs? MSc of Andrea Longobucco)
Motifs mining: What are some motifs linked to providing trust services
(lots on applications: e.g., suspicious transactions)
Results:
Illegal trust service providers:
Garcia-Bernardo, Javier, and Petr Janský. "Profit shifting of multinational corporations worldwide." arXiv preprint arXiv:2201.08444 (2022).
Garcia-Bernardo, J., Haberly, D., Janský, P., Palanský, M., & Secchini, V. (2022). The indirect costs of corporate tax avoidance exacerbate cross-country inequality (No. 2022/33). WIDER Working Paper.
Garcia-Bernardo, J., Haberly, D., Janský, P., Palanský, M., & Secchini, V. (2022). The indirect costs of corporate tax avoidance exacerbate cross-country inequality (No. 2022/33). WIDER Working Paper.
Garcia-Bernardo, J., Haberly, D., Janský, P., Palanský, M., & Secchini, V. (2022). The indirect costs of corporate tax avoidance exacerbate cross-country inequality (No. 2022/33). WIDER Working Paper.
Garcia-Bernardo, J., Haberly, D., Janský, P., Palanský, M., & Secchini, V. (2022). The indirect costs of corporate tax avoidance exacerbate cross-country inequality (No. 2022/33). WIDER Working Paper.
Garcia-Bernardo, J., Haberly, D., Janský, P., Palanský, M., & Secchini, V. (2022). The indirect costs of corporate tax avoidance exacerbate cross-country inequality (No. 2022/33). WIDER Working Paper.
Open questions:
How to implement policy interventions in the model?
How to calibrate the model to find time trends?
Javier García-Bernardo
javier.gbe@pm.me