Texas A&M University
David Bessler received undergraduate and master’s degrees in economics and agricultural economics from the University of Arizona and his PhD in Agricultural Economics at the University of California at Davis. He moved to Texas A&M in 1982; earlier he held an academic appointment as an Assistant Professor at Purdue University. He was a Simon Fellow in Agricultural Economics and Econometrics at Manchester University in the United Kingdom in 1991. He was formerly co-editor and editorial council member of the Western Journal of Agricultural Economics and associate editor of the Journal of Forecasting.
Bessler studies causation, machine learning, and forecasting, all with respect to agriculturally-based economic systems. Most of his work is with time-ordered observational data. His work on causation follows recent work in philosophy and computer science in exploring a counterfactual or manipulation view of causation, using graph theory for representations. Fundamental to this work is the distinction between passively observing a relationship in data and actively intervening on that relationship (as suggested to us by Haavelmo more than fifty years ago). The distinction shows up in interpretations possible from inference methods used for experimental versus observational inquiries of nature (the economy). This work is embedded in the modern machine learning literature and its associated algorithms. Applications by Bessler and his colleagues have been to price discovery in markets separated by space, time or form; dynamic relationships among prices, quantities consumed and expenditures; the nexus among literacy, birth rates, malnutrition and potential latent variables; and equity market pricing.
His forecasting work focuses on motivation and evaluation of probability forecasts using calibration metrics and scoring rules (as developed in meteorology and statistics). The idea here is that it is generally not enough to offer a probability forecast (whether subjective or objective) to guide decision-making. One might do well to demonstrate its goodness in terms of both calibration (do the issued probabilities agree, ex post, with relative frequencies?) and the ability to reliably sort (distinguish) between events that occur and events that do not occur. Applications by Bessler and his colleagues have been to agricultural commodity prices, macro-economic systems, consumer purchases, and equity market prices. Probabilities emanating from both econometric models and human beings show considerable room for improvement, especially with respect to the sorting criterion; suggesting potential for contributions by future generations of scientists trying to understand the economy.
This post is part of an ongoing series of profiles of AAEA members. Have a suggestion for a future profile? Send them to Info@aaea.org.