Time series models are a highly useful forecasting method. but arcdeficient in the sense that they merely extrapolate past patterns inthe data without taking into account the expected irregular futureevents. To overcome this limitation. forecasting experts in practicejudgmentally adjust the statistical forecasts. Typical judgmentalfactors may be treated as outliers in statistical analysis. Toautomate the judgmental adjustment process. neural network models aredeveloped in this study. To collect the data for judgmental events,judgmental effects are filtered out of raw data. The main trend iscaptured by a neural network model using the filtered data. whilejudgmental effects are modeled by another neural network. Then thejudgmental effects are additively adjusted. Performance of thisarchitecture is tested in comparison with five other architectures.According to the experiments. the architecture of neural networkbused additive judgmental adjustment significantly improves theforecasting performance.
展开▼