Rule-based forecasting: development and validation of an expert systems approach to combining time series extrapolations
Article Abstract:
Rule-based forecasting is a method of generating forecasts based on the features of the data using forecasting expertise and domain knowledge. To test the feasibility of this procedure, a rule base for extrapolations of forecasts for annual economic and demographic time series was developed and applied. The rule base integrated forecasts generated by the extrapolation methods of random walk, regression, Brown's linear exponential smoothing, and Holt's exponential smoothing based on rules applying 18 characteristics of time series. Results showed that rule-based forecasting does allow the application of forecasting skills and domain knowledge suited to the features of time series. It was also found that rule-based forecasting was more effective than random walk and equal-weights combining.
Publication Name: Management Science
Subject: Business, general
ISSN: 0025-1909
Year: 1992
User Contributions:
Comment about this article or add new information about this topic:
Sliding simulation: a new approach to time series forecasting
Article Abstract:
An empirical test which examined 111 time series subsamples was conducted to test a new approach to time series forecasting. The novel approach indicates that a model/method is used not because it fits historical data well, but because it can accurately forecast out-of-sample specific data. A technique is picked out from among various methods run in parallel and utilizing out-of-sample data. Additionally, model/methods are optimized for each prediction horizon separately, which makes it possible to have individual models to forecast for different horizons. Research results indicate that the new method has a better performance than other techniques by a large margin.
Publication Name: Management Science
Subject: Business, general
ISSN: 0025-1909
Year: 1990
User Contributions:
Comment about this article or add new information about this topic:
Forecasting when pattern changes occur beyond the historical data
Article Abstract:
The article investigates the use of a new forecasting technique that is based on both a long-term and a short-term model. The models are used to produce the final forecasts by establishing specific parameters that are related to the extent, number, and duration of recent pattern changes. The technique that is advocated was applied to the 111 series used for M-competition.
Publication Name: Management Science
Subject: Business, general
ISSN: 0025-1909
Year: 1986
User Contributions:
Comment about this article or add new information about this topic:
- Abstracts: Redundant, overlapping organization: a Japanese approach to managing the innovation process. Business alliances and the strategy of the Japanese firm
- Abstracts: Continuous process improvement at Brooktree. GAPV: a new approach to joint venture investments. Are learning curves still relevant?
- Abstracts: Merger strategies and capital market risk. Stockholder reactions to CEO changes in large corporations. Diversification, systematic risk, and shareholder return: a capital market extension of Rumelt's 1974 study
- Abstracts: Individuals' perceptions and misperceptions of time series properties of quarterly earnings. Did firms undertake debt-equity swaps for an accounting paper profit or true financial gain?
- Abstracts: The lease vs. purchase decision. Managing risk with derivatives. Teaching activity-based costing