Do forecasts improve over time?

Purpose - Accounting and decision making rely heavily on forecasts. For several reasons, we should expect ongoing increases in forecasting accuracy. The purpose of this paper is to test the hypothesis of improved forecasts over time. Design/methodology/approach - The paper analyzes original monthly sales plans and current data for three different car models in six different countries over 15 years and over several product life cycles (PLCs). Forecasting accuracy is calculated as one minus forecasting error. Forecasting error is measured with MAD/MEAN for periods of years or relative deviations per month. The hypothesis of decreasing forecasting errors is tested with the non‐parametric Mann/Kendall trend test. Additional interviews with managers were conducted to elicit details of internal forecasting organization and instruments. Findings - The paper finds no evidence of increased forecasting accuracy in general over 15 years or over subsequent PLCs. This seems surprising, given improved statistical methods and software in general, and experience and learning effects of the organization itself. However, there is evidence from the case, that the reason lies in environmental uncertainty and volatility and not in internal factors within the control of the company. Research limitations/implications - Evidence from one case study is limited in its external validity. Future studies should analyze the forecasts of more companies, more industries and different forecasting objects, the latter including consumer, industrial goods and services. In the absence of further research, the results seem to negate the common assumption, that companies are generally able to make accurate forecasts, including those for accounting purposes. This hypothesis is clearly confuted. Practical implications - The paper describes a methodology for companies to analyze their own forecasting accuracy and to identify possible reasons for a lack of accuracy, or basic approaches to increasing it. Originality/value - Most studies on forecasting accuracy rely on interviews and questionnaires, entailing bias that is difficult to control. Few studies analyze archival data in order to measure forecasting accuracy; so that our study avoids much of the bias mentioned above. Despite the inevitable limitations of case studies, a study such as the present one at least allows us to dispute a common hypothesis about forecasting accuracy in practice.