BELGIUM: Big data analysis of the perception of the social security contribution on company cars

In Belgium, employers who provide company cars to their employees pay lower social security contributions on this benefit than they would on regular wages. Specifically, they pay a flat-rate contribution based on the vehicle’s CO₂ emissions—referred to as the CO₂ contribution. The Belgian National Social Security Office (NSSO) is responsible for collecting these contributions. It receives information on contribution amounts and vehicle license plates through the quarterly social security declaration (DMFA). Legislation also grants the NSSO access to other data sources, notably from the Federal Public Service (FPS) Mobility, which records technical vehicle data such as CO₂ emission levels, to verify the declared contributions.

During its audit of the social security contributions on company vehicles, the Belgian Court of Audit assessed whether the NSSO was collecting these contributions correctly—specifically, whether it was conducting verifications based on reliable and usable data. The audit revealed that the NSSO does not utilize the control mechanisms provided by law. There is no systematic verification of the accuracy of declared amounts, which poses risks to the proper collection of contributions. In this context, the Court of Audit conducted a quantitative analysis to determine whether the declared amounts aligned with the legally due amounts and to quantify any discrepancies.

Quantitative Analysis Method

The comparison was conducted for all vehicles declared to the NSSO in the fourth quarter of 2022—over 550,000 vehicles. According to the legal formula, the contribution is calculated on a flat-rate monthly basis according the vehicle’s CO₂ emission rate, fuel type, and engine type. In all cases, the monthly amount cannot be less than 20.83 euros (indexed).

Although the CO₂ contribution is calculated monthly per vehicle, it is declared quarterly in aggregate (for all vehicles provided by an employer) via the DMFA. The monthly amount must be multiplied by the number of months the vehicle was used during the quarter (1, 2, or 3). To determine the total quarterly amount due per employer, the contributions for each vehicle must be summed.

To perform this analysis, the Court identified the necessary fields in the NSSO and FPS Mobility datasets required to recalculate the contribution formula. A detailed review of the technical glossaries of both administrations was a crucial step. This review provided added value for the audit by revealing technical issues that led to findings—such as missing fields, non-blocking controls, and the risk of outliers. Based on the list of data strictly necessary for the audit—considered personal data—the Court obtained authorization to process it. From data receipt to processing, all operations were conducted securely (encrypted transfers, secure storage, restricted access, and deletion of data six months after publication).

The analysis was conducted using the R programming language on CSV files. A series of operations were performed on the result of a data crossing of previously cleaned NSSO/FPS Mobility data (harmonization of field names and value units, management of missing data). The datasets were linked using license plates as a unique key. However, the match between the two datasets was not perfect. In some cases, the Court had to substitute missing CO₂ emission values with the average CO₂ emissions for company vehicles in the same year.

Figure 1 – Quantitative analysis method

On the basis of this data crossing, the Court recalculated the contributions using the legal formula. These calculated amounts were then compared with the declared amounts (C), and the audit team quantified the proportion of declarations that deviated by more than 20% from the expected value.

The analysis revealed that the NSSO lacks the basic data needed to verify declared amounts and that there are concerns about the quality of DMFA data. For instance, the ‘license plate’ field—one of the dataset’s most critical fields—accepts any value or symbol. Nearly 8% of license plates for the quarter were unusable (47,000 plates were either unknown to FPS Mobility or deregistered).

Even after allowing a 20% margin of error between the amounts of contributions declared and the amount due and substituting missing CO₂ values with averages, 17% of employers showed a risk of incorrect declarations. This represents a potential revenue loss of €15 million per year. Notably, more than half of this amount is attributable to a small number of employers.

Conclusion

Through this analysis of raw and big data from the administrations, the Belgian Court of Audit was able to make significant observations regarding data quality and issued recommendations to improve the collection of social security contributions.

Specifically, the Court recommends that the NSSO:

• Review the procedures for declaring the CO₂ contribution
• Strengthen data quality controls
• Conduct targeted controls based on identified risk factors

By Hélène Mastrodicasa and Pierre Vandoorne, performance auditors at the Belgian Court of Audit