The Expert Notes

Welcome to the latest edition of "The Expert Notes”. This time, we are focussing on the link between innovation and flexibility in tax practice.

Sven Sistig provides an impressive account of how technological innovation and agility, through e-invoicing and cashless payments, have a transformative impact on tax compliance and enforcement.

To the latest VAT GAP report 2023 from yesterday, there is a VAT GAP drop of 1/3 from EUR 99bn (2020) to EUR 61bn (2021). Potential reasons, according to EU COM: “Another cause of the observed evolution of the VAT compliance gap might be related to changes in consumption patterns towards categories and channels where compliance is generally higher (e.g., online shopping) and the increased share of cashless payments.”

Furthermore, E-Invoicing and Reporting obligations took a role, as the drop in the gap in Italy and Poland shows! It is interesting to see the results in one year for 2022, keeping in mind the macroeconomic situation due to the Ukraine War, inflation, return from COVID-19, and quick fixes effective as of July 2021. States could be on a good way to consider local and cross-border E-Invoicing, which will reduce the VAT gap, most probably significantly. Question should be allowed: Will extension of deemed supplier ruling for marketplaces be needed any more (pillar of ViDA)?

Felix Lamp shows that the technological revolution goes beyond compliance. Machine learning and flexible modelling increase the precision and efficiency of our tax systems, paving the way for smarter tax administration.

In my career, I have often helped decision-makers understand what benefits the introduction of machine learning models could bring. To be able to make technology decisions that use machine learning, you must understand what a parameterised model is. ?

1. A parameterised model is a point of view: Say you develop an application for tax filing, and you want to remind the user that a filing deadline is approaching. You could reason that the user needs about two weeks to prepare the filing and formulate the following rule: ? If deadline - today <= 2 weeks => warn user else … ? But you could also take the view that time (t) and action (a) are parameters that you can choose: ? If deadline - today <= t => take action a else … ?

2. A parameterised model allows you to have data determine the parameters: In the above example, if we measured the cost associated with the filing (e.g. fines, the cost to employees …), the actual filing date and the actions we took for past filings, we could derive the cost-minimising combination of t and a. Point 2 has two important implications: ?

a. If data determines the parameters, we do not have to set them. If we have to set parameters, our model necessarily needs to be small (i.e. have few parameters) as the mental cost for setting parameters is high and is incurred for each change of the model (think about setting the billions of parameters that you encounter in a Large Language Model yourself).

Therefore, a parameterised model allows us to consider many parameters and, therefore, has a better chance to provide more accurate results (it can incorporate a larger context in its decision-making). ?

b. Our decision-making becomes more empirical. Instead of sitting in an armchair at the fireplace contemplating what the parameters could be, we are confronted by data. We can even quantify the cost of making different decisions.

Do we always need a lot of data to work with a machine-learning model? ? => No, but generally, the fewer the data, the more parameters we have to set. This means that we have to invest in modelling expertise and rely on domain knowledge. Working with small samples has a long tradition in statistics. ? Machine learning shines where domain knowledge is poorly formalised, and we need a large context to make accurate decisions. ?

There is a caveat: We must be careful about what claims we derive from our model. In general, the more control we have about how the data is collected and the more thought we put into the model, the stronger the claims are that we can make. With this caveat in mind, there are a lot of low-hanging machine learning fruits in tax made, and adopting the parameterised model mindset is a good start to grab them.

Meanwhile, Monika Hagmann reports that France is intensifying the use of digital technologies to increase efficiency in tax monitoring, which is modernising the tax administration and, at the same time, strengthening its integrity and efficiency.

???Another Year, Another French Finance Bill – Part 4 ???

For 2024, the Draft French Finance Bill, released on September 27, 2023, provides some changes to the VAT legislation. ?Enhanced Investigative Powers: Further to the described changes, Article 19 of the Draft Bill proposes to offer Tax administration agents the possibility to conduct active investigations under a pseudonym on websites, social networks and messaging applications. These investigations are, of course, not limited to VAT! ?

???Scope of Application: Who??Tax agents with at least the rank of public finance controller. What Actions Can They Perform? Access publicly available information. Engage in electronic exchanges. Extract and store data and potential evidence. Ensure actions do not induce any legal breaches.

Applicable Platforms As per Regulation 2022/2065, this includes those listed in lit. i) and m) of Article 3 of Regulation 2022/2065, i.e. i) “online platforms” defined as a ”hosting” service that, at the request of a recipient of the service, stores and disseminates information to the public, unless that activity is a minor and purely ancillary feature of another service or a minor functionality of the principal service and, for objective and technical reasons, cannot be used without that other service, and the integration of the feature or functionality into the other service is not a means to circumvent the applicability of this Regulation “online interfaces” defined as ”any software”, including a website or a part thereof, and applications, including mobile applications”

???♀??Investigation Focus for the purposes of investigating or ascertaining the breaches mentioned in c of 1 of Article 1728, Article 1729, I of Article 1729-0 A and the last paragraph of Article 1758 of the French General Tax Code, i.e. Hidden activities Inaccuracies or omissions in a tax return, as well as the restitution of a tax claim whose payment has been improperly obtained Taxes imposed on financial assets hidden in foreign accounts, etc.

???Upcoming Detailed Regulations:?An application decree will soon clarify the specifics, particularly concerning the data retention period.

And Ole Z. emphasises that the dynamic nature of global trade requires continuous adaptability, which is crucial to ensure trade efficiency and regulatory compliance.

#EU proposes to postpone parts of the new preferential rules and #duties for electric cars and batteries. Trade news that even made it to the #BBC and other UK news sites, as this was one of the hard realities of #Brexit for manufacturing, resulting in extremely high costs. https://lnkd.in/eSyYUd75 ?? In my view, this shows that nothing is set in stone, and parties can always agree to re-negotiate or change the terms of a contract or agreement. https://lnkd.in/eZ4fG_aa

This wraps up our latest edition. As we move forward, expect to dive deeper into these subjects, offering thorough perspectives and dialogues on prospective changes. Keep an eye out for our comprehensive analyses and expert opinions on these crucial issues. Don't forget to subscribe for more!