The volume of Dutch online consumer purchases from foreign EU webshops is calculated on the basis of VAT declarations. Filing VAT returns in the Netherlands is compulsory for companies selling products to Dutch consumers out of other countries, i.e. with a threshold amount of 100 thousand euros in annual turnover. It is not compulsory for those companies with an annual turnover in the Netherlands which falls below this amount. These companies fall outside the scope of this survey. With the VAT returns as a source, it is possible to compile an overview of the foreign EU webshops where Dutch consumers order their purchases.
The first step in this method is to identify the foreign companies in the VAT declarations. Foreign companies are identifiable because they use a tax representative and/or are handled by a specific tax office.
Subsequently, the potential turnover from online sales to Dutch consumers is calculated by adding up the turnover from the VAT declarations of the selected companies. VAT returns that have been passed on to the Dutch customer are not taken into account. In these cases, the customer is a business and by definition there is no consumer turnover.
The next step is to identify the webshops within the selection of companies. For the businesses with the highest VAT declarations, this is done manually and on the basis of the information available on the internet.
The selected companies with lower reported turnover are classified automatically. First, a legal company name and a web crawler are used to locate company websites. On the web pages that have been located, characteristic terms such as ‘shopping cart’ are counted. Finally, a Machine Learning (ML) algorithm is used to determine whether this is in fact a webshop.
In the final step of this method, the publication figures are calculated by adding together the results of the manual and automatic webshop classifications. As the automatic classification method is not error-free, corrections take place to adjust for the effect of this error on the publication totals (bias) and a margin of error is derived.