Statistics & Data Science Seminar
— 5:00pm
Location:
In Person
-
Gates Hillman 4307
Speaker:
GLENN SHAFER
,
University Professor and Broad of Governor's Professor, Department of Accounting and Information Systems, Rutgers University
https://www.business.rutgers.edu/faculty/glenn-shafer
Modernizing Cournot’s principle
In everyday English, a forecast is something less than a prediction. It is more like an estimate. When an economist forecasts 3.5% inflation in the United States next year, or my weather app forecasts 0.55 inches of rain, these are not exactly predictions. When the forecaster gives rain a 30% probability, this too is not a prediction. A prediction is more definite about what is predicted and about predicting it.
We might say that a probability is a prediction when it is very close to one. But this formulation has a difficulty: there are too many high probabilities. There is a high probability against every ticket in a lottery, but we cannot predict that no ticket will win.
Game-theoretic statistics resolves this problem by showing how some high probabilities are simpler than others. The simpler ones qualify as predictions. This story has roles for Cournot’s principle, Kolmogorov’s algorithmic complexity, and de Finetti’s previsione. Visit this and my two books on the topic with Vladimir Vovk.
—
In the 1970s, Glenn Shafer launched the “Dempster-Shafer” theory. The Belief Functions and Applications Society, devoted to this theory, has been holding international conferences since 2010. During the past 25 years, Glenn and Vladimir Vovk launched game-theoretic probability and statistics. Their two books on the topic appeared in 2001 and 2019.
Glenn has published more than 20 papers on the history of probability and statistics. His most recent book, The Splendors and Miseries of Martingales: Their History from the Casino to Mathematics, co-edited with Laurent Mazliak, was published by Birkhäuser in 2022.
Glenn served in the Peace Corps in Afghanistan. At the Rutgers Business School, he served as director of the doctoral program for ten years and as dean for four years.
Event Website:
https://www.cmu.edu/dietrich/statistics-datascience/events/index.html