Markets are slaves to ‘fake’ data-a statistic is not a fact!
If you think forecasting is a humbling experience have a thought for the poor statistician trying to collect accurate economic data. The truth is forecasters struggle to forecast sunrise and the statisticians can only provide guestimates. Both forecasts and data are subject to regular revisions. In layman’s terms they are invariably wrong.
So what has gone wrong with the data? Economists and the media have presented statistics as fact. They are not, but their religious attention to the monthly data releases creates a false sense of confidence in their accuracy. They counter this by stating they are merely satisfying demand. GDP, Inflation, Current account, Unemployment are artistic interpretations framed by international convention, but their outcomes can topple a government.
The major issue is that the world has changed. The rapid structural changes, led by technology, is poorly captured in practices designed over 60 years ago when the UK and the West was largely manufacturing based, as opposed to the current dominant service sector.
Charlie Bean (ex BoE, MPC) concluded that the current statistics are antiquated (Independent Review of UK economic statistic: final report, 2016). The digital revolution has given rise to new ways of exchanging and providing services and made it far harder to measure accurately economic output. In the past many of these services would be paid for and captured in GDP but now are free, albeit bundled with advertising. Ironically, GDP may actually fall while the quantity and quality of services are increasing. Moreover, companies such as Amazon, Skype and Google operate across national boundaries which make it difficult to allocate value added to particular countries. The versatility of modern technology has also spawned the rise of the self employed- about 15% of the UK workforce. This has confused the issue on many fronts not least measuring hours worked and labour productivity. Capital spending may in certain cases be underreported as consumer spending e.g. laptop, tablet and these businesses make better use of non-business assets. Traditional employment and income data will no longer capture the reality of the labour market. Income, traditionally paid by salary, has been understated as the self employed incorporate and receive proprietors’ income and/or dividends. Recognising this in September the UK Office for National Statistics (ONS) announced one of its largest revisions to the national accounts. As a result, real household disposable income grew 5.3% in 2015, not the 3.5% currently recorded and the savings ratio rose from 6.1% to 9.2%! What this ultimately means for the alarming Q1 2017 savings ratio, which was the lowest for 50 years at 1.75%, is problematic.
Bean suggests that if the digital economy was fully captured by official statistics, it could add between 0.3-0.75% to the growth rate of the UK economy.
This may not trouble you but another ONS study found that foreigners own more UK equities and corporate bonds than previously thought and an error in recording of dividends shows that there are more income outflows than previously thought. The effect of these changes will be for the 2015 current account to deteriorate from a deficit of £80bn to £98bn, nearly 1% of GDP.
Does it really matter? Yes. While the household sector might be in a slightly better position than thought the UK is increasingly dependent on foreign flows and on ‘the kindness of strangers’.
We have just received another data bombshell from the Office of Budget Responsibility (OBR, Forecast Evaluation Report, Oct. 2016). They are reducing their assumption for potential productivity growth over the next five years which would weaken the medium term outlook for public finances. The OBR describe the arithmetic divergence between their central forecasts and the subsequent outturns as ‘differences’ rather than ‘errors’ because when the forecast was made they were not in possession of sufficient information!
Given the litany of data ‘failures’ you may well be wondering then how the UK Treasury came to the conclusion that ‘brexit’ would make us worse off in fifteen years’ time.
The problems with data accuracy are actually getting worse. It is not just a UK issue. US GDP is a case in point. Since 2010 75% of initial releases have been revised higher. A major part of the problem is that most economic data is survey based, at least for the initial releases.
Meyer, Mok and Sullivan (Household Surveys in Crisis,2015) examine the decline in quality of US household surveys. The overall message was that households are overburdened by surveys, leading to a decline in many measures of survey cooperation and quality. Indeed, they have become increasingly less likely to answer surveys at all. In one example they note that replies for the consumer expenditure survey dropped from over 85% in the early 1980s to only 65% in 2103. This is an important survey as it helps generate the weights used in consumer price inflation.
Households are reluctant to answer certain questions, especially details on income. To overcome missing data most surveys typically impute a response. When households do provide answers, they are less likely to be accurate. One way to test for measurement error is to link survey data for instance on welfare payments that households say they have received with administrative micro data on the amounts actually provided to each household. They show that survey measures of whether an individual receives income and of how much income is received from major transfer programs are both sharply biased downward, and this bias has risen over time.
Survey data becomes more problematical when surveys are voluntary. Selection bias and loss aversion can apply. Selection bias arises because there is an opportunity cost to participating in a survey, and so voluntary surveys will tend to attract only the more passionate participants in that particular field; the apathetic majority tend not to participate. With loss aversion people tend to apply a disproportionately larger weight to a loss than they apply to a gain. As a result people are likely to be more passionate about negative than positive opinions again skewing voluntary polls.
Even when a variable is seemingly well defined, such as the total number of people working at a particular time, in most cases it still has to be estimated, usually by sampling a fraction of the total population. Ensuring that this sample is representative is challenging in itself.
Markets like surveys as ostensibly forward looking indicators. The problem with surveys is that there has been a tendency for survey results to overreact to underlying economic data. This media obsession with incomplete or simply inaccurate data is a concern.
The brexit debate was framed by bogus forecasts and ‘antiquated’ data. This problem is not set to diminish as we proceed along the ‘fourth industrial revolution’. We don’t even know the margin of error. The Bank of England assume there is about 3-4% range of possible growth numbers around the official GDP data!
Commentators, politicians and economists need to educate the public of these data deficiencies and be cautious on policy response to numbers we do not understand or unable to compute with any certainty.
James Sharpe
October 15th 2017