13 November 2014
Dr Andreas Hoepner on whether big data analytics will revolutionise the way we think about the economy
Dr Andreas Hoepner
Will big data analytics revolutionise the way we think about the economy?
Having seen many chief economists of large financial institutions present at industry conferences, I would argue that a pattern of thinking has become visible that is also prevalent among many academic economists. This pattern is: theoretical assumption first, evidence from actual data second.
In other words, chief economists tend to present simplistic charts of basic statistics of macroeconomic variables to back up what appears to be a preformed personal view about the current state of the economy. These patterns become particularly evident when two economists present back to back and interpret very similar statistics in very different directions.
So is this science? Well, in my view it is much better characterised as economic arts, since the researcher (i.e. economist) views their own personal views as substantially more relevant than the insights that could be gained from much more advanced statistical analysis. Such an emphasis on personal assumption is common in many arts, certainly in the absence of better analytical methods.
So how is financial data science different, and what is it anyway?
Financial data science combines three realms of expertise: classic computational data science, advanced statistical analysis, and in-depth content knowledge about the analysed segment of financial markets. Given this complexity, it is nearly always practiced in teams, as one rarely finds an individual with expertise in all three realms.
The financial data science teams do not start their work by conceptualising their personal assumptions but by defining and deeply exploring the relevant data. As a result, every measurement common in current economics is questioned according to its actual practical suitability. For instance, financial economics use standard deviation, beta or tracking error frequently as measures of investment risk these days. These measures, however, are driven as much by upside deviations from expected values as by downside deviations, with virtually every investor appreciating upside deviations. Hence, financial data scientist would use the purely downside-focused version of these risk measures (i.e. semi standard deviation, downside beta and trailing error).
Even more importantly, financial data scientists have an actual measure for the quality of their work, which is more than the subjective approval of their colleagues. They simply use the explanatory power of their models, especially out of sample, as a clear statistical indicator of the quality of their work.
For instance, if a financial data science model explains over 95% of the variation in the outcome variable, it lacks understanding of less than 5% of the real-world relationships the scientist studied. Consider in contrast that many financial economists studying banking confirm their theoretical assumptions with empirical models that often understand less than 20% of the variation in the outcome variable. At these statistical levels, where often there is over an 80% lack of understanding, it is not surprising that financial economists were not excessively hailed as providing warning signals before, or solutions after, the banking crisis.
So will financial data science revolutionise the way we think about the economy? Well, only time will tell but it will certainly provide those who use it much more accurate microscopes to study economic relationships.
Dr Andreas Hoepner
Associate Professor of Finance
Director of Enterprise
ICMA Centre, Henley Business School
Watch a video of Dr Andreas Hoepner interviewed by the Swedish Economist (video in English): http://bit.ly/1pMHRGj