In October 2012 the Property Services Regulatory Authority (PSRA) in Ireland launched the Residential Property Price Register (RPPR). Up until this point, there had been no official government dataset of individual residential property sales prices across the country. The lack of such data was a significant gap in housing information and the launch was welcome, even if it was happening several years too late. In 2012, Ireland was nearing the bottom of a massive housing crash. House prices were down over 50 per cent, and apartment prices by over 60 per cent, from their height in 2007. House building had dropped from 88,000 new units in 2006 to a few thousand. The RPPR tracked all residential property sales from 2010 onwards, giving address, date of sale and price, though missing other useful information such as property type, floor size or number of bedrooms.

Along with one of my colleagues, Eoghan McCarthy, I started to examine the dataset keen to see the geographic pattern of prices. It quickly became clear that there were some issues with the data. Within half an hour of its release, Eoghan had already found over 200 suspected errors in the data. Most were identified by simply sorting the data by price value. For example, a semidetached house in Limerick City was recorded as selling for €18m in 2010. Our assumption given the property type and location was that two additional noughts had been added to €180,000, or a decimal point had been omitted. The other end of the sorted spreadsheet showed that some houses in very desirable parts of Dublin had sold for just a few thousand euro rather than hundreds of thousands. In other cases, multiple transactions had been lumped together. For example, one transaction of €4.5m was for 30 apartments. It would have been more appropriate to record 30 entries of €150k given that all the other transactions were sales of individual properties. Ronan Lyons, a real-estate economist working with a property company (Daft.ie) spotted the same errors and when he and his colleagues compared the dataset to their nationwide sales dataset they ended up removing 7.7 per cent of properties where there was a significant difference between the sale price as advertised and recorded.