In economic research, and in particular in development economics, it is important to have access to precise income data. Yet, getting data on income has always been a problem, and even more so before the invention of modern national accounting in the 1930s. One of the particular problems that persisted for a long time was how to deal with price differences across countries. That problem was alleviated – although not entirely solved – by the advent of the International Comparison Project (read the story here og Tim Taylors account of the project here).

Researchers have for the last couple of decades either used various versions of the national account data in the Penn World Tables (PWT) or from the World Bank. No one claims that they are perfect, but both sources tend to yield very similar results. They have also both been regularly updated, with the PWT ending at version 7.3. However, many people have wanted better and even more precise data.

PWT 8.0 was therefore introduced to great fanfare in 2013, when a new team of researchers changed the basic methodology behind the data. These changes were outlined by Robert Feenstra, Robert Inklaar and Marcel Timmer in a highly profiled article in no less than the American Economic Review. Many researchers nonetheless began to get doubts because the new Penn data turned out to look rather different from the old data, and several known economic events were nowhere to be found in the new version. Maxim Pinkovskiy and Xavier Sala-I-Martin for example dcoumented (ungated version here) that both the 2011 version of the World Bank dataset and PWT 7.1 provide a much better fit with development, when one compare the data to long-run changes in light intensity at night, which cannot be manipulated or fiddled with. Something was obviously amiss with PWT 8.0 and 8.1.

This year saw the publication of PWT 9.0 with the implicit claim that the beginners’ problems of the new methodology had been sorted out. The question then is which data you can trust and what you use? Interested readers can of course make up their own minds, for example by looking at the Danish data. All three series from the PWT start in 1950, 7.1 ends in 2010 and 9.0 in 2014. Penn provides two series in version 9.0, as the new methodology implies that there are GDP series calculated from expenditures (the ’e’-series in the figures) and other series calculated from output (the ’o’-series). Beginning in 1966, these series can be compared to the purchasing power adjusted series from Statistics Denmark (DS). The clue here is that Denmark has some of the most precise national accounting data, not least because of the tradition of registering everything at the source, which ought to provide a ”most precise” benchmark; a sort of gold standard.  And as is obvious in the figures, the four series match up rather nicely in the very long run. The lowest correlation is 0.96 (between  PWT 9.0 ’o’ and DS). However, as is particularly obvious in the second figure, where I focus on the shorter period between 1970 and 1995, there are substantial and worrying differences in the short-run dynamics. Looking at the annual growth rates, it turns out that PWT 7.1 and DS are almost identical (the correlation is 0.98), while PWT 9.0 is different.

The correlation between growth rates in DS and PWT 9.0 ’e’ is a mere 0.08! The output-based series looks better, but correlations of 0.76 with DS and 0.84 with PWT 7.1 are far from impressive, when the data ought to be calculated from the same base information. The figure also reveals that the new data exhibit a much larger decline in the last years of the Anker Jørgensen governments, and a visibly slower recovery after the conservative government of Poul Schlüter took over in 1982. A few years later, the PWT 7.1 and DS series also exhbit rather different growth rates for the Fogh/Løkke governments between 2001 and 11. PWT 7.1 has an annual average growth rate of 0.25 % and DS one of 0.43 %, while the new PWT data show rates of 1.52 and 2.23 %.

The conclusion is that something is obviously wrong in the new data. New is not always better and I feel obliged to recommend that colleagues, students and anyone else interested in the topic do not use the new PWT series, although they have been sold to the academic community as improved measures.