How to Construct Absurd Data - the UK Experience

For the last few months I have been preparing the Coldwater Economics Good Stats Guide, which has involved checking the methodology, provenance and track record of the hundreds of data-series which I follow in my Shocks & Surprises work. It has been a constant reminder of how much we take ‘the economic data’ for granted, trusting that these eyes and ears on the world economy are reasonably accurate accounts of what’s actually going on.  

This illusion sometimes cannot survive a close examination of what is involved in producing them. Sometimes the problems are merely empirical (trouble doing the counting), but very often they are conceptually profound and possibly intractable (for example, estimating services activity, counting ‘real’ investment spending). Sometimes, the problems are simply perverse, with seemingly reasonable methods producing predictably absurd results. 

Take, for example, Britain’s construction output index, which for five out of the last six months has reported outright contraction, which has resulted in yoy output falling consistently for the last four months. June’s numbers showed output falling 0.9% mom sa and falling 2.4% yoy. 

Can this possibly be right? On the face of it, it seems unlikely in an economy in which during the last four quarters has been growing at a rate of 0.5% qoq on average, during which time gross fixed capital formation outpaced GDP growth.  It also seems unlikely given that employment in the construction industry grew by 5.4% yoy in 1Q16 (latest data available), and that wages in the construction industry were growing at 9% yoy by April 2016 (latest data), far outstripping wage growth in any other sector of the economy.  How many industries do you know that compete for labour like this when they are in recession?  If all the data is correct, management is presiding over an epic collapse in productivity while paying handsomely for the privilege.

This is the point at which it pays to know how the construction output index is constructed. On the face of it, construction seems a fairly tangible thing to count; naively, one might count the number of houses, factories, offices, floor space completed in a month. But the difficulties are fairly obvious: one square metre of nuclear power plant probably isn’t comparable with one square metre of warehouse space. 

Rather, what happens is the Office of National Statistics (ONS) surveys about 8,000 construction companies for their monthly billings, of which typically 60% to 70% reply. This allows them to form an estimate of the value of output.   For the record, this grew 4.6% mom nsa in June, and rose 1.2% yoy, and the 6m movement against trend is modestly positive.  To produce the output volume result, however, this billings estimate needs to be deflated by a cost index. And that’s where the problems really surface. The methodology chosen seems reasonable: measure changes in the prices of inputs via the appropriate PPI subindexes, and apply the appropriate weightings (determined by EU standards).  

But here you run into the problem directly: for most construction projects, labour is roughly half the cost. When the construction industry is busy, competition for labour pushes up prices sharply (like now). But in turn, that pushes up the deflator, with the result that ‘real output’ registers a decline.  By contrast, when business slackens, wages will fall, and consequently so will the output deflator, with the result that ‘real output’ will be measured as rising. Whenever there is a greater upward and downward volatility in construction wages than there is in underlying billings, this resulting ‘construction output index’ will consistently produce perverse results.  At every step, the ONS methodology seems reasonable, but the result is predictably absurd. 

I want to be fair to the ONS. The problem was dropped in its lap when in December 2014, the Department of Business Innovation & Skills gave up producing its own estimates of changes in construction sector prices and costs. In the aftermath ONS scrambled to construct their own price series by April 2015. The solution they lit upon is acknowledged as only an interim solution, and the construction output series has been de-registered as National Statistics. The ONS knows there are problems with the series, but neither it, nor the market which greets this data as an interesting indicator, has really acknowledged the absurdity at its core.