Forecasting
Forecasting is the process of predicting future events based on historical data and observations. It encompasses a range of applications, from weather and traffic predictions to economic and business forecasts. While the term "prediction" can sometimes refer to past events, forecasting specifically pertains to potential future occurrences, often with varying degrees of probability and uncertainty. The practice of forecasting employs both qualitative and quantitative methods, tapping into mathematical and statistical analyses to create models that help reduce uncertainty.
Various forecasting techniques exist, including autoregressive models and simulation methods, each suited to different types of data and contexts. The accuracy of forecasts is typically evaluated against randomness and error magnitude, leading to ongoing debates about the effectiveness of expert judgment versus mathematical modeling. The complexity of forecasting increases with the interdependence of variables, making accurate predictions particularly challenging in dynamic environments, such as fashion or technology. Ethical considerations also arise, particularly regarding the impact of forecasts on societal behaviors and decisions, prompting discussions about integrating cultural values into forecasting practices. As a multifaceted discipline, forecasting remains a critical tool across numerous fields, shaping our understanding of potential futures.
Forecasting
Summary: The science of prediction is grounded in statistics, data analysis, and modeling, applied to such areas as traffic, sales, and the stock market.
Forecasting essentially means predicting. Prediction of various phenomena has been of interest to mankind ever since humans have inhabited the planet. One prediction of early human beings may have been that the sun would rise the next day, along with where animals or other food might appear. These predictions would be based on observation and experience. Once human beings began to investigate natural laws, certain predictions like the rising of the sun came to be regarded as certainties by many scientists. Generally, all predictions are based on experience but may be formulated with varying degrees of mathematical rigor that involve different levels of probability or uncertainty. “Prediction” may refer to guessing about the past but “forecasting” is always used to mean guessing events that may or may not happen in the future. Forecasting may be qualitative or quantitative, and events may not occur at all or may only occur after a very long period of time.
Mathematicians and statisticians have explored forecasting in a variety of fields, such as traffic flow, ocean waves, and asset price forecasts. Many mathematicians have contributed toweather forecasting, such as Johann Werner, James Glaisher, Lewis Richardson, Vilhelm Bjerknes, and Edward Lorenz. With regard to business, economics, and marketing, in the eleventh century, Shen Kua explored price forecasting and the theory of supply and demand. Statisticians George Box and Gwilym Jenkins in 1970 published a book on time series analysis for forecasting. In 2001, George Tiao won the Samuel Wilks Award of the American Statistical Association, in part for his work in forecasting, and in 2003, David Wallace won the same award, in part for his research on forecasting elections.
Forecasting Models
Forecasting models are created using a wide variety of analytical and computational methods from mathematics and statistics. In general, the quantification and reduction of uncertainty are required to make forecasting models accurate enough to help businesses make sound decisions.
Several issues arise while forecasting, including the time range of the forecast (the time until which the forecast may be applicable) and the availability and reliability of the data. Some traditional data analytic methods must be modified to account for the serial correlation common in data resulting from processes observed repeatedly over time. A large class of mathematical forecasting models involves applying weighted smoothing methods to fit functions or trends to historical data. Smoothing constants and other parameters may depend on choices made by the forecaster, so different models based on exactly the same data might produce varying forecasts.
Autoregressive moving average (ARMA) models, sometimes called “Box–Jenkins models” because they are estimated using a methodology developed by Box and Jenkins, along with integrated moving average (ARIMA) models, are widely applied to what are known as observable, nonstationary processes with serially correlated data. Financial data commonly falls into this process category. They may also use adaptive filtering, widely found in other applications such as signal processing, to remove noise.
Decomposition forecasting models mathematically separate overall trend, seasonal, and random components in data. Scatterplots, simple linear regression, and curve fitting may be useful for explorations and some modeling. Simulation methods facilitate dynamic models and exploration of “what-if” scenarios. The cross-impact matrix method explicitly takes into account the fact that the occurrence of one event can impact the likelihood of other events, so probabilities can be assigned to produce an intercorrelational structure to examine relationships between system components. Multiple regression is also used to examine multifactor influences. In general, the greater the interdependence of components, the more difficult it becomes to make a prediction about any single component. Decision trees, game theory, and chaos theory are other mathematical areas that have been used to explore systems to make forecasts.
Forecasting Validity
Ultimately, forecasts are usually judged by their accuracy, often in a subjective manner, and there are many theories regarding how to measure the utility of forecasts. One criterion is to assess whether the forecast differs from pure randomness. Another is to quantify the magnitude of error. Decision scientist Spyros Makridakis has stated that in many situations, judgmental forecasting by human experts has been shown to be superior to mathematical models. However, in terms of optimization, he also noted that forecasting many complex problems is unfeasible without computer modeling. For example, simultaneously forecasting inventory levels for thousands of items for sale at a major retailer or needed by a manufacturing company is likely beyond the scope of subjective judgmental forecasting. Computer technology also allows for the creation of complex decision algorithms with subsystems and feedback loops.
Stability in the system being modeled is also an important factor in determining whether model extrapolation will be valid and reliable for forecasting. Developmental inertia is the idea that some systems are less variable and therefore more easily predictable than others. For example, the rapidly changing fashion industry is a low-inertia or unstable system and new trends are difficult to predict mathematically. Decisions also need not be dichotomies but rather probabilities along multiple paths. Mathematical concepts from decision theory and utility theory, such as expected value, have also been incorporated into forecasting modeling and decisions.
Forecasting Ethics
An ethical consideration raised by forecasting is whether probabilistic inferences actually create the future, since decisions made today by individuals, businesses, and policy makers undoubtedly affect actions taken later. In his 1970 novel Future Shock, sociologist Alvin Toffler discussed the impact of evolving technology on humans and asserted the need for value impact forecasting, which is the idea that social forecasting must incorporate cultural and societal values. Mathematicians and others continue to study and debate these theories and problems and to seek ways to quantify psychological and qualitative variables considered essential by many forecasters.
Bibliography
Chase, Charles. Demand-Driven Forecasting: A Structured Approach to Forecasting. Hoboken, NJ: Wiley, 2009.
Hanke, John. Business Forecasting. 9th ed. Upper Saddle River, NJ: Prentice Hall, 2008.
Howe, Leo, and Alan Wain. Predicting the Future. New York: Cambridge University Press, 2005.
Morlidge, S., and S. Player. Future Ready: How to Master Business Forecasting. Hoboken, NJ: Wiley, 2010.