Serbia – a tax haven and a shelter haven for debtors

Category: Best data-driven reporting (small and large newsrooms)

Country/area: Serbia

Organisation: BIRN Serbia

Organisation size: Small

Publication date: 31/05/2019

Credit: Slobodan Georgiev, Milorad Ivanovic , Jelena Veljkovic, Aleksandar Djordjevic, Miodrag Markovic, Natalija Jovanovic

Project description:

Agency called World Business Solutions buys indebted companies for profit giving their old owners a fresh start. Our investigation revealed a scheme of 10 people taking over almost 2000 troubled companies from early 2014 by May 2019. All of them combined were 53 million EUR in debt. In most cases, the state was the biggest creditor meaning companies were not paying taxes timely. According to the Serbian law, this scheme was partially legal. Limited Liability Company (LLC) is not the responsibility of the owner or director, but the legal entity, that is the company itself. Prosecution and Police are investigating

Impact reached:

For the organisation with our scale, the story had an impressive public impact with 26 republications, including some of the major outlets in Serbia such as Al Jazeera, Blic, Danas, N1, RTS, etc. Investigative reporting portal Insajder did a follow-up https://insajder.net/sr/sajt/tema/15051/.  

The video about the residential building with more than 400 companies registred had more than 58.000 views at it was shared via FB and TW (YT – 418 views:TW – 36,540 views, FB – 21, 053 views).

 Data visualization sets were praised as Top 10 by GIJN for May 2019. See here https://gijn.org/2019/06/06/gijns-data-journalism-top-10-european-electiondata-via-audio-tax-fraud-parserator/ . Google analytics at javno.rs shows that the story had 13,513 pageviews (Unique Pageviews: 12,507). On top of that, the story was shared 12,262 times and the average time spent on page is 5 min and 37 seconds.

Techniques/technologies used:

BIRN team combined investigative reporting and data journalism to tell the story about tex heaven in the making. It all started as an accident when my colleague noticed one owner had more than 20 companies registered on his name. Most of them had blocked bank accounts.  

Why would someone buy a blocked company, he wondered. After some digging, he came across another case – another owner of hundreds of blocked companies. In just a few months, he would learn from the policeman investigating these cases how this problem is “the cancer of the Serbian economy.”

Investigative reporters obtained a list of 10 owners with the most companies through FOI request from Serbian Business Registry. Since the number of companies was changing, the data team has been tracking them daily. 

The scraper BIRN team designed was pulling company ID numbers and addresses from the Public Registry. Using that list of company ID numbers, we ran another scraper through the National Bank’s Single Register of Accounts Search to gather data on the amount of debt for each company. We compared the same list of companies with the Tax Administration’s List of the biggest debtors determining how much some of them owe for tax. We used Python for analysis and Flourish for visualizations. 

While the data team was wrangling the sets, investigative reporters went on the field. They met previously mentioned sources from the police who revealed how “some Belgrade lawyer” was behind it and how the police couldn’t find information about his identity. Reporters visited Jasonova 11 St, a residential building in the family neighborhood with more than 420 companies registered on its address. As they were trying to enter, the public bailiff officer was leaving the notice in the lobby failing to find the company’s office door.

What was the hardest part of this project?

Exposing an organized debt evasion scheme had its challenges both on the investigative and data side of the production. Data journalists required a lot of precision and nerves.  

The biggest challenge was creating an initial dataset in the daily changing environment. To overcome this problem, we conducted two analyses at different points in time. During the first one in February, we developed the code and methodology. In May, a couple of days before the publishing we repeated the process to have fresh data.

New dataset was created by scraping and merging together  different entities from tree publicly accessible databases – Serbian Business Registry, National Bank database, and Tax Administration’s List of the biggest debtors. 

We created a list of company ID numbers by the owner we obtained through FOI. Extracting ID numbers and Addresses for each company from the Serbian Business Registry was challenging since the website uses reCAPTCHA. Data on the amount of blockage for each company was deeply buried within the National Bank database,  so the scraping process took longer. Tax Administration’s List is in PDF format, which required additional cleaning.

We combined all the elements mentioned to determine the total amount of blockage, tax debt, several addresses with a couple of hundreds of companies registered on it. All our findings are available as visualizations in the link section.

What can others learn from this project?

This project is a great example of investigative and data journalists working together on revealing different angles of a knotted story. As numbers were setting the direction, the investigative part provided context and human perspective. When data journalists calculated 420 companies registered on one address, investigative reporters visited it to learn neighbors are very concerned about this problem. We believe this project sets a good example for other newsrooms interested in covering complex topics through data and investigative reporting. Findings our data team provided served as leads for investigative reporters or as facts they faced officials with. Not only that we managed to mine the data from different sources offering our readership more information than the official institutions, but we shed light on struggles police investigation was facing regarding this case. Another great resource for other organizations interested in data journalism could be the code we produced for scrapers and analysis.

Project links: