Accurate predictive models are the heart and soul of portfolio management, particularly in today's rapid-fire trading environment. Investors are constantly seeking any advantage — no matter how small — when choosing instruments to add to, retain in, or remove from portfolios.
No stone is thus left unturned in the pursuit of good modeling and forecasting: if factors such as the daily news, changing opinions and tastes, or even the social interactions between corporate directors and analysts can potentially tip off investors to price moves minutes, hours or days before new earnings information becomes public, that information becomes incredibly valuable.
But do these factors accurately predict market movements? A Stevens researcher decided to find out — and learned something surprising.
Dr. Germán Creamer, a professor of quantitative finance and business analytics, performed pathbreaking research demonstrating that link mining, relationship mining, and news and sentiment analysis can all be used to help anticipate and predict earnings surprises and other asset-price moves.
To test his theories about the 'corporate interlock' of financial social networks, Creamer created a model of relationships between directors of publicly traded S&P 500 companies and the analysts who provide coverage of those companies. He then used specially designed algorithms to determine whether analysts' surprise predictions for companies with whose leadership they were socially 'close' were more likely to be accurate.
And they were. Earnings surprise forecasts correlated much more strongly to actual surprises when analysts were socially close to directors. Interestingly, Creamer found this effect diminished somewhat after 2001, when the federal Regulation Fair Disclosure brought far greater transparency to company financial information and effectively ended the process of selective corporate disclosure of operations to small groups of analysts.
Calling Creamer's technique a 'different angle,' Deutsche Bank's influential newsletter Market Research singled out the Stevens research for special interest.
"This approach could help to disentangle the conflicts of interests between company management, investors and analysts, and hence providing better forecasts for earnings surprises and abnormal returns," noted the newsletter's authors.
Working with a team from Columbia University, Creamer is also creating new algorithms that parse published news and predict the future directions of European stocks.
"The correlation is between 10 percent and 30 percent," Creamer explains.
"Again, it is another signal, not strong to use alone but something that adds to your equities research and refines it."
Creamer's work is just part of a comprehensive program of financial research and education at Stevens, much of it enabled by the Hanlon Financial Systems Lab — a leading-edge facility featuring Bloomberg terminals streaming real-time market data and the capability to analyze historical data with state-of-the-art software packages.
Stevens' financial offerings also include the nation's first undergraduate major in quantitative finance; leadership of a respected annual HFT (high-frequency trading) conference; pathbreaking business intelligence and analytics (BI&A) graduate coursework; and comprehensive financial engineering programming and research.
To learn more about Stevens financial and management research, visit stevens.edu/howe.