An increasingly bitter dispute among economists is raging over the accuracy of the preliminary official figures showing a 0.4 per cent economic contraction in the third quarter. In one camp sits Goldman Sachs, led by Kevin Daly, who says the figures are “unbelievable, literally”. Now Danny Gabay of Fathom Financial Consulting accuses the Goldman team of “some seriously flawed analysis” in a note published today.
I like and respect the people on both sides of this argument and I’ll try to summarise their cases fairly. A reasonable conclusion, I believe, is that though Gabay scores some technical points, the weight of argument still rests with Daly. Little weight should still be given to the current GDP figures unless you believe, as Gabay seems to, that people in 2019 and beyond will have a worse idea of the underlying 2009 economy than people in 2012. His view is that the benefit of hindsight is strictly time-limited.
The case against the initial GDP figures showing a contraction
Using data for the initial release of quarterly GDP figures going back to 1975 and their most recent vintage, Goldman Sachs show the correlation coefficient for the most recent 10 year period is only 0.1. The initial figures are in effect random noise and have no predictive power in explaining the final figures. Moreover, the Goldman analysis shows there are amusing cross-country stereotypes. As in the table below, Europeans are super cautious about initial estimates of growth, while the US statisticians are too gung-ho.
The initial figures are also biased, particularly when the first estimate is negative. The average change in these circumstances is to revise higher the initial estimate by 1 percentage point. And finally a bunch of other contemporary indicators – PMI data, CBI surveys, BCC surveys, Employment growth and even eurozone flash GDP estimates – are better predictors of the final UK GDP quarterly figure than the initial published estimate. I’ve looked at the Goldman spreadsheet and there is no doubt all of this is accurate.
The case in favour of the initial GDP figures
Fathom counters with the argument that the Goldman Sachs analysis, while numerically accurate, is conceptually flawed. So bad in fact, Gabay says the Goldman’s work is “baloney”. Goldman Sachs, “have based their analysis on the wrong data,” he says because the most recent vintage of data includes both new information and methodological changes. It would be impossible for the initial data to predict methodological changes in calculating GDP, so “counting these improvements as errors or bias in the preliminary estimates is clearly wrong”.
Fathom argues that the only fair comparison is between the initial data and that vintage of data three years later when all the reconciliation is complete, not the version of the data that is printed now. On this comparison, the initial GDP data has a correlation of 0.6 with the data produced three years later (not at all bad) and other sources of data are worse predictors than the initial data.
This dispute reduces to an argument over whether the best description of the economy is the current version of history or the version produced three years after the initial estimate.
Gabay has produced a clever note showing how this change makes a big difference to a plausible assessment of the initial GDP figures. But let’s take his logic a little further.
- If the initial GDP figures are the best predictor of the GDP data 3 years on and those numbers are the relevant comparison, then we should throw away the current view of economic history and all the subsequent methodological improvements because these do not add to our knowledge about what was happening then.
- The current set of official figures on economic history are therefore, to use Danny’s terminology, “baloney”, and he is making a serious criticism of the Office for National Statistics.
Of course, the past initial estimates cannot hope to take account of methodological changes such as a difference in the measurement of financial services, which was introduced recently. On this Gabay is quite correct.
But, if you take the view, as I do, that the view of history using current GDP methodology is the best guide to past economic conditions, that is the data to use. After all, policy makers at the time would have wanted to know what demand conditions were on the best available GDP methodology, even if it was yet to be invented.
I am sure the Bank of England would love to see now the final GDP figures for Q3 2009 as they will be published in 2019 and not just the ones published in 2012. The 2019 vintage would incorporate both the full set of data and future methodological improvements to capture better what was happening this year.