Big data. Correlation vs. causation. Most people think that it’s imperative that we harness big data to find any correlations that exist. Without once determining what the theory is that could substantiate the correlation. (If you check back to my 19 April post, you will find the references to determine what is wrong with this concept.)

One of the persons who commented on that on that blog, a man I respect for his sci/tech writings, Gustavo Silva, left a comment that provides a great lead-in for the discussion today. He recalled a lecturer announcing that a high rate of births in Paris correlates well with excessively high stork populations- ergo, storks must bring babies. OK, now, let’s discuss something that’s a little less obviously wrong (only if you don’t pay attention).

Officials from the European Commission, the European Central Bank and certain politicians here in the US are professing that we must cut spending and taxes to improve the economy. As proof, they cite a 2010 study by Carmen Reinhart and Kenneth Rogoff (Harvard University, published by the National Bureau of Economic Research [NBER]). Reinhart and Rogoff averred that high government debt levels (they arbitrarily chose debt levels over 90% of the gross domestic product [GDP]) hinders economic growth.

Olli Rehn (Commissioner of European Economic and Monetary Affairs) actually referred to this study often in his speeches and news conferences explaining why they act the way they do.  It was the proof for the claim that “[h]igh debt levels can crowd out economic activity and entrepreneurial dynamism, and thus hamper growth. This conclusion is particularly relevant at a time when debt levels in Europe are now approaching the 90% threshold, which the U.S. has already passed.”

Without spending time discussing the conservative ties of the NBER, which has its own agenda, the facts behind the study prove otherwise to the stated conclusions.  As Reinhart-Rogoff state, there is a correlation between debt and GDP- but that link does not prove that increased debt increases lower GDP, as opposed to a lower GDP’s that causes a rise in the country’s debt. (You know, the chicken and egg problem).

As a matter of fact, it turns out that it’s the opposite causation; weak economic conditions depress tax revenues, simultaneously placing a demand for social support systems, which, therefore, cause a country’s debt to rise.  And, by using the full complement of the data that was officially part (see the next paragraph as to why I phrased it thusly) of the Reinhart-Rogoff article, one can see that economic growth in one year can predict the debt load over the next few (two or three years) more than the debt load can predict whether growth does or does not occur.

Thomas Herndon, Michael Ash, and Robert Pollin (economists from the University of Massachusetts) have fully analyzed the Reinhart-Rogoff findings and published their analysis in PERI (Political Economy Research Institute). They disclose that this oft-quoted paper contain basic errors, including a grievous one that their spreadsheets omitted five countries (Australia, Austria, Belgium, Canada, and Denmark) from inclusion of their analysis and results. That last error causes the stated Reinhart-Rogoff conclusions to disappear, i.e., to be totally false.

(And, this has nothing to do with the quality of peer-review that the paper may have undergone. Most reviewers don’t have the time or the ability- or even the actual data themselves- to review each calculation and discern if an error were made in the calculations. Moreover, when dealing with economic data from long ago, one must recognize that the data itself can be highly suspect- because the world has gotten better at collecting said data in real time and with better precision.)

Top that omission of 5 countries with these additional errors in the paper: the Reinhart-Rogoff study excludes several years of high debt that was associated with high growth in various advanced economies. And, another of it’s prime examples, the high debt of the New Zealand economy from 1946 to 1949, with resulting 1951 negative growth (-7.6% was quoted) was presumed to be a perfect example of their findings. But, Statistics New Zealand reports that the number was based on academic estimates; it was not until 1955 that New Zealand produced any official GDP estimates or figures. Moreover, the world now employs margins of errors (and restatements of the actual data at a later date) for GDP figures- that did not occur decades ago.

Correcting for those errors and few other problems brings the average growth rate of the advanced countries in years when debt exceed 90% of gross domestic product up to 2.2%. The Reinhart-Rogoff study found the annual growth rate for countries with high debt was -0.1%. Moreover, the growth (2.2%, not -0.1%) is only about a point lower than those for debt-to-GDP ratios of 60 to 90%. Obviously, the findings of the paper are totally invalid, as a result.

Then, we get to the bigger issues.

The proper way to propose solutions is to start with an underlying theory, one that is tested against reality. That’s the ‘big data’ problem- it believes that correlation does not need a theory, does not need an understanding of why things are. After all, one could argue that the 90% threshold is pretty arbitrary- except that the deficit hawks already believed this was a magic number and so seized on the flawed study to bolster their beliefs- in spite of the facts.

The next problem is somewhat similar to the one above. One that I, as a chemical engineer, medical researcher, environmental system designer or product developer (depends upon the day of the week), do not confront on a daily basis. I get to build prototypes, I get to test scale models. These can then be used to see if my theory of why things “are” really “flies”. But, economists can’t tinker with a nation’s economies with their macroeconomic theories, because the consequences of a mistake are dire. (Oh, wait- they are- and look how Europe is tanking!)

Then, there’s the issue of sample size. The universe of economic data that one wants to compare are typically small. And, gets smaller still, when one recognizes that the financial systems have evolved and changed over the years. So, there are insufficient data points with which one can prove or disprove a theory. Add to that the fact that large countries have currencies that drive economies while smaller countries’ currencies are buffeted by the external changes to their abodes, and you have a true ‘hornets nest’ to evaluate.

Finally, the concept that there IS a debt-to-GDP threshold that obtains for each country is an issue. It’s a problem because the causation is clearly backwards to this theory, as stated above. The concept that a universal relationship exists between incurring debt and developing negative growth, regardless of time or country, regardless of whether an economy is strong or weak has clearly been proven chimeric. Instead, it’s negative growth that leads to the stacking up of debt.

Now, if only Europe and a certain group of Americans would recognize they had their facts backward…