We use cookies to improve your experience on our website. By accepting you are agreeing to our use of cookies

Accept Close

Winton last week hosted a data science workshop at its San Francisco office with some of the leading thinkers in the field. The workshop considered the topic: have official measures of economic activity kept pace with technology and globalisation, and how might data science tools improve the status quo?

The workshop's attendees included some of the biggest names in the Bay Area data science and tech community; representatives from financial institutions and large corporations; and two dozen Winton employees. Presentations were made by Winton CEO and founder David Harding, Bloomberg Beta founding member Shivon Zilis, Google Chief Economist Hal Varian and ex-U.S. Chief Data Scientist DJ Patil.

David kicked off the workshop by asking whether ‘fundamental' economic data - as measured by official statistics bureaux - are truly fundamental to financial markets. To answer this, David drew on work from Winton Research that included looking at the relative market impact of 12 key measures of U.S. economic activity. Announcement days produce bigger swings in volatility-adjusted returns, but the picture is not stationary; market sentiment towards different indicators changes significantly through time.

David concluded that fundamental statistics present fundamental problems for investors. "Most trading around official economic data is sound and fury, a zero-sum game," David said. Even if a trader were to know in advance exactly what number would print for the non-farm payrolls data - a monthly indicator of U.S. employment levels - the potential returns on offer would only be significantly attractive for the single trading day preceding the release. Knowing the announcement a week, a month or a year in advance appears to be of no value. Fertile ground, therefore, for research into alternative economic indicators.

The proliferation and increasing sophistication of satellite-generated data could provide some answers, argued venture capitalist Shivon Zilis. Shivon was sceptical of the accuracy of economists' traditional tools for tracking economic activity. One of her areas of focus at Bloomberg Beta is machine intelligence, and she feels the pace of change in this exciting domain is instructive. "The switch was flipped last year," said Shivon. Just a couple of years ago, it was a niche area of research dominated by academics. "Now, Fortune 500 company executives want to apply it.”

Google's Hal Varian showed that unusual and seemingly esoteric data points can prove to be potent predictors of economic behaviour. In an entertaining speech that ranged from hangovers and gift-shopping to the Greek referendum and predictors of morbidity, Hal demonstrated that picking out signals from available real-time data can help society produce up-to-date measures of economic activity.  Official indicators, by contrast, are often produced with a lag of a month or more. The new, unconventional signals might also improve accuracy, given many indicators are frequently revised at later dates. The positive implications for public policy include greater efficiency and responsiveness.

That was a theme developed further by DJ Patil, who served as the United States' first chief data scientist in the Obama White House. The analysis of big data can contribute a huge amount to public sector governance, DJ explained. The priorities that guided DJ and his team's work were high-level: societal benefits worth $1 trillion or more; problems that affected over half the country's population; and challenges that lacked alternative solutions. Yet DJ’s team's initiatives also produced visible and meaningful improvements at an individual level, such as in healthcare outcomes.

DJ stressed the need for the responsible and rigorous application of quantitative techniques. Gathering, cleaning and combining multiple large data sets produce the best inferences, DJ said. That chimes with Winton's experience of large-scale data gathering, within its Systematic Data Collection business.

In the question-and-answer session, Hal drew on demographic trends to make the positive case for future 21st century employment prospects. Automation may be rendering some jobs obsolete, but a coming wave of retiring baby boomers may offset the pressure to a degree, Hal said. And David, in response to a question about Winton's investment strategies, pointed out that machine learning may be a phrase that is in vogue, but it's a technique Winton has been applying for many years.

As the formal part of the evening drew to a close, the audience had a chance to mingle and pose further teasers to the distinguished panel over sushi, cocktails and canapés. Just a year after Winton opened its San Francisco office, it is supporting the data science community's push to debate and develop the newest thinking. Stay tuned to catch news of the next of these exciting workshops.