Presentation: Prediction, Coupling and Causation.
Biographical Information: George Sugihara is the McQuown Chair and Distinguished Professor of Natural History, at Scripps Institution of Oceanography. He holds degrees from the University of Michigan and Princeton University. He is a theoretical biologist who has worked across a wide variety of fields, including landscape ecology, algebraic topology, algal physiology and paleoecology, neurobiology, atmospheric science, fisheries science, and quantitative finance. He is the inaugural holder of the McQuown Chair in Natural Science at the Scripps Institution of Oceanography. Most of his early work was motivated exclusively by pure science, and the later work more by pragmatic utility and environmental concerns. Nearly all of it is based on extracting information from observational data (turning data into information). His initial work on fisheries as complex, chaotic systems led to work on financial networks and prediction of chaotic systems. He is one of 18 members of the National Academies Board on Mathematical Sciences and their Applications, and was a Managing Director at Deutsche Bank. He helped found Prediction Company (sold to UBS) and Quantitative Advisors LLC. He has been a consultant to the Bank of England, the Federal Reserve Bank of New York, and to The Federal Reserve System on questions of international security: systemic risk in the financial sector. Other notable research relates some of his early work on topology and assembly in ecological systems to recent work on social systems and work on generic early warning signs of critical transitions that apply across many apparently different classes of systems.
Although correlation is neither necessary nor sufficient to establish causation, it remains deeply ingrained in our heuristic thinking. With increasing recognition that nonlinear dynamics are ubiquitous, and that relationships among variables will depend on system state, the use of correlation to infer causation becomes more difficult. Here we examine a criterion that identifies time series variables as causally related if they interact as part of the same dynamic system. Rather than using diet overlap as a proxy for the network of interactions, we directly deduce the operative network of realized dynamic linkages from information embedded in time series. Our approach, based on nonlinear state space reconstruction, addresses Berkeley’s 301-year correlation vs. causation dilemma and identifies basic problems when the current solution, Granger causality, is applied to nonlinear ecosystems. This criterion applies even in highly nonlinear cases and provides a conceptual framework for studying coupling and catastrophic change in nature. As a speaker in SS23: Ecosystem Change and Predictability of Aquatic Ecosystems on Wednesday, 11 July, Dr. Sugihara will discuss further details of the method.