Bad data

Among the top catchphrases used by market players, two at least give a prominent place to statistics: the timeless refrain, “The figures are better than expected” and its negative counterpart: “The figures are below expectations“. If you find yourself in a trading room in France on a Friday afternoon around 2:30 PM when US job statistics are released, it is not possible to avoid at least one of these two phrases, sometimes punctuated by a curse or an emphatic “Yes!“. And if by chance, another figure is released by China in the day you will certainly have an opportunity of hearing the inevitable “You cannot trust Chinese statistics” from one of the trading floor’s many English-speaking speakers.

And yet while it is generally accepted that “One cannot trust Chinese statistics“, can one really trust the others as well? Should one really give such prominence to macroeconomic data, and how should such data be integrated by the markets? Difficult questions that were addressed in a solid study published by the bank UBS [1].

This problem of markets adjusting to publications is, above all, of emotional nature: published data generates a sentiment among consumers or investors that pushes them to react. This sentiment has been modelled by UBS through its “blasé barometer”. The idea: compare the “volatility of sentiment” with the actual volatility of the pertinent underlying data. The results speak for themselves: the blasé barometer becomes so volatile that it might be renamed the “stressed out barometer”. For the last two years, we have seen markets increasingly overreact to macroeconomic stimulus. The recent months during which bonds, equities and foreign exchange have registered extremely large swings, have offered a vivid illustration.

If markets overreact to macroeconomic data, it is no doubt because stock market investors give them too much weight. In fact, one might think that in this age of big data, the data we possess is increasingly reliable. Regretfully, reality is not so simple.

Let us take the case of consumer data: whereas studies on the subject boasted response rates of 85% in the 1980s, this figure has since fallen down to 65% and the responses are frequently skewed. The statisticians try to get around this weakness by seeking out data directly on the web but, in so doing, they introduce a bias that underweights non-Internet consumer spending. From big data to “bad data”, the line is quickly crossed and the paradox is obvious: if the granularity of data is today much stronger, the figures’ reliability in the end is not necessarily greater.

In the same spirit, let us take a look at what is going on across the channel: always very pragmatic, the Bank of England is now publishing data in the form of intervals rather than exact figures. To say that in 2015, English growth was between 2% and 3% seems less ambitious than asserting that it was 2.5%, though no doubt much more intellectually honest.

In light of these considerations, keeping statistics into perspective should help investors regain a little serenity. In the same way, considering less popular statistics is sometimes more meaningful. An example? Today, 50% of the inhabitants of Shanghai order their meals out rather than making it themselves. An observation that is ultimately more meaningful than uncertain GDP figures and also reassuring, both in terms of the transformation of the Chinese economy towards a service economy and the growth prospects of the world’s second-largest economy.

Didier Le Menestrel

[1]  “Is sentiment data sentimental nonsense? ” Paul Donovan, Economist, UBS, 24 February 2015