Page 9 - IOTA Good Practice Guide
P. 9
Risk indicators – Hierarchical clustering on declaration data
Level: intermediate
Contact person: Jean-Luc Wichoud, Switzerland
Problem:
We always try to find indicator covering a large variety of
taxpayer for our predictive model to tend to cover all the risks
and find other ways to enrich it.
Solution:
A risk indicator could be more complex, it could be an output
of an advanced analytics model. In the attachment on the GPG
website, you find an example from Switzerland, where a risk
indicator is calculated using an unsupervised method (hierarchical cluster). It gives a more advanced
example of a risk indicator used as a component to select taxpayers at risk.
Risk indicators – Network analysis with Panorama
Level: advanced
Contact person: Ádám Fajkusz, Hungary
Problem:
Quite often and in particular in the case of VAT-fraud,
risk is linked to a network of taxpayers. How could we
extract information from a network to define a risk
indicator?
Solution:
Other advanced techniques could be used for predictive
modelling. For instance, network analysis is very useful to detect fraudulent organizations. The
Hungarian Tax Administration uses the tool Panorama to create networks that help detecting VAT
fraud.
Combine and optimize – VAT risk model
Level: intermediate
Contact person: Lisanne van der Breggen, The Netherlands
Problem:
How to build a predictive model to integrate risk
indicators and optimize the selection of VAT-declaration
with refund for check?
Solution:
The predictive model presented is focused on VAT
declaration with refund, it is a risky area in many
countries. The Netherlands has implemented an iterative
approach to create the model. Risk indicators are the input, a training sample is extracted, the model
is optimized on the sample and compared with another one, so iteratively the model is improved. The
advantage of this iterative approach is a real optimization of the output of your model.
7
