[No Chan-do’s ‘Prospective Models’] How You Use Them Is What Matters

[ 8/19/2008 ]

The word “forecasting” is always around us. From weather forecasts in daily life to major decisions by governments and businesses, forecasting is inescapable. It matters because decisions are made on the basis of expectations about the future. To my knowledge, the first forecasting model began with Einstein’s 1905 mathematical description of “Brownian motion,” which represents irregular movement. Describing Einstein, famous for the theory of relativity, as a pioneer of forecasting models may sound unexpected. Every forecast tries to express random movement as a formula and identify a pattern in it. The central task of a forecasting model is therefore to give structure to randomness. In 1827, a British botanist became interested in the way tiny particles moved irregularly yet showed a pattern. In 1905, Einstein gave this random movement a mathematical description. Before Einstein, however, Louis Bachelier of the University of Paris submitted a doctoral thesis in 1900 titled “Theory of Speculation,” treating stock-market movements as a form of random Brownian motion. His adviser, the renowned mathematician and physicist Henri Poincaré, reportedly regarded the thesis as difficult to understand and ahead of its time. As late as 1950, Bachelier’s work on stock-market speculation had largely been forgotten, and virtually no scholars knew of it. MIT economics professor Paul Samuelson then came across the work by chance in the University of Paris library. He praised it as an extraordinary piece of work and tried to revise and complete what he saw as Bachelier’s unfinished theory, but ultimately did not succeed. Bachelier’s 1900 stock-market formula has since been assessed as remarkably close to—and, in some respects, reaching beyond—the Nobel Prize-winning work on option pricing developed in 1973. Tracing the origins and development of forecasting models helps us understand both their limits and their potential. In an Oxford University classroom, a professor teaching financial mathematics asked students: “Is financial mathematics a science or a social science?” His answer was: “Science studies the work of the Creator, while social science studies the work of human beings. And human creations are much harder to predict.” In practice, the stock market has proved one of the hardest subjects for any forecasting model to predict. Financial institutions therefore revise their stock-market models frequently. A model is considered good if one of its factors can explain more than 10% of market movements. Forecasting expert Diebold has stressed that a model is only a model; what matters more is the process of putting it to work in the real world. The crucial question is how we use forecasting models.