Forecasting

The history of forecasting goes back at least as far as the Oracle at Delphi in Greece. Stripped of its mystique, this was what we now refer to as “unaided judgment,” the only forecasting method available for centuries. As a formal area of study, the earliest examples are from the weather forecasters of the nineteenth century. The early years of the twentieth century saw increasing interest in business and economic forecasting, which is the focus of this article. Novel methods were applied to agricultural yields and prices as well as analysis of the business cycle. Researchers became interested in both methods and in producers of forecasts who may or may not use formal methods. Construction of macroeconomic models occurred in the 1940s associated with the work of the Cowles Commission. As the first computers became available, these models were estimated and used for forecasting. The 1960s and 1970s marked the era of univariate forecasting; ARIMA modeling and exponential smoothing both date from this time and are widely used today in business forecasting. The 1980s saw a further institutionalization of the subject and increasing exchange between the different groups of researchers, economists, statisticians, engineers, and, later, psychologists and data scientists. The Journal of Forecasting and subsequently the International Journal of Forecasting were founded; both aimed to integrate the disparate aspects of forecasting. It was also the era of some major forecasting developments: unit-root testing, vector autoregression, cointegration, state-space modeling, and ARCH modeling. Questions about the best forecasting method were tested in “competitions” between methods, but clear answers were not forthcoming. As computing power has continued to increase, more sophisticated and complex forecasting methods have emerged, based on neural networks and decision trees. The 2020s has seen computing power used to handle more data, more nonlinear methods, more emphasis on forecast distributions, and perhaps more stress on the limitations of business and economic forecasting and the strategies that should be followed. After introducing books and papers that examine the breadth of forecasting, this bibliography’s structure recognizes that the fundamental methods of business and economic forecasting—judgment, extrapolative time series methods, and econometrics—still have distinct development paths. The newer area of computationally intensive methods, artificial intelligence and machine learning, was initially applied to predicting individual behavior and events, such as bankruptcy but now used extensively in a wide range of applications, is included separately. Forecasting software is also surveyed briefly, as its dissemination has been critical both to research innovation and to changes in forecasting practice. Finally, the bibliography covers various important areas of application.