NYU Tandon Co-Sponsors a Conference for Financial Forecasters
Analysts skilled at accurate forecasting can reduce costs, increase opportunities, and improve outcomes for their clients, and gaining the knowledge and experience needed to do so is vital. To aid in that process, in early December, NYU Tandon co-sponsored the M4 Conference, which was keynoted by Distinguished Professor of Risk Engineering Nassim Nicholas Taleb, who discussed uncertainty in financial forecasting.
Master’s degree candidate Thomas Haversang was in attendance and said, “Professor Taleb’s talk was excellent; he focused a lot on what he called ‘the expert problem,’ which is especially pertinent to forecasting — and to academia as a whole. He explained fat tails, the limitations that they present in modeling, and the new statistical tools we need to use. He made a fascinating point that even though we cannot model the fat-tailed real world, we can still structure our models to be robust to such random and detrimental phenomenon.”
Another Tandon student, Rupal Jain, concurred, saying, “The M4 conference was a great learning experience for me as it demonstrated a blend of Machine Learning and Forecasting and their importance in the financial industry. It provided a chance to learn from pioneers of the field such as Professor Taleb and Spyros Makridakis [the director of the Institute for the Future at the University of Nicosia, co-sponsor of the event], who shared their own experiences and presented a different perspective on risk and forecasting it.”
The Conference was being held following the completion of the M4 Competition, the fourth iteration of a challenge that had been launched in 1982 with the goal of enabling participants to learn to improve forecasting accuracy and advance the field as much as possible.
M4 drew numerous academicians and practitioners, who competed using a variety of techniques to predict time-series data. The focus was practical, rather than theoretical, and the event was aimed at providing an empirical look at which major time-series methods objectively performed best. Organizers pointed out that accurate predictions and the correct assessment of the uncertainty surrounding them are indispensable for all types of future-oriented decisions: from determining appropriate inventory and sales levels to buying and selling financial instruments. “The findings of the M4 Competition that just ended have provided a wealth of practical information for improving the accuracy of forecasts and the correct assessment of uncertainty based on a huge database of 100,000 series covering six application domains (macro, micro, demographic, industry, financial and others) and six time frequencies (yearly, quarterly, monthly, weekly, daily and hourly),” they wrote.
“We had a wide-ranging discussion of when machine learning was more relevant than statistical models and came to the general conclusion that a hybrid approach was most effective,” he said. “It was a pleasure to be part of such a well-organized and well-attended event, and I forecast that Tandon will be involved in M5 when it is held.”