8.2 Forecasting BasicsForecasting techniques are generally classified into two main types: qualitative and quantitative. Qualitative (subjective) techniques are naturally used in the absence of historical data (e.g., for new machines or products), and they are based on personal or expert judgment. On the other hand, quantitative (objective) techniques are used with existing numerical data (e.g., for old machines and products), and they are based on mathematical and statistical methods.Qualitative forecasting techniques include historical analogy, sales force composites, customer surveys, executive opinions, and the Delphi method. Quantitative techniques are classified into two types: (1) growth or time-series models that use only past values of the variable being predicted, and (2) causal or predictor-variable models that use data of other (predictor) variables.Nahmias (2005) makes the following observations about forecasts: (1) forecasts are usually not exact, (2) a forecast range is better than a single number, (3) aggregate forecasts are more accurate than single-item forecasts, (4) accuracy of forecasts is higher with shorter time horizons, and (5) forecasts should not ignore known and relevant information. To choose a forecasting technique, the main criteria include: (1) objective of the forecast, (2) time horizon for the forecast, and(3) data availability for the given technique. In order to develop a quantitative forecasting model, the steps below should be followed:1. Define the variable to be predicted, and identify possible cause-effect relationships and associated predictor variables;2. Collect and validate available data for errors and outliers;3. Plot the data over time, and look for major patterns including stationarity, trends, and seasonality;4. Propose several forecasting models, and determine the parameters and forecasts of each model;5. Use error analysis to test and validate the models and select the best one; and6. Refine the selected model and try to improve its performance. Maintenance Forecasting and Capacity Planning 159Quantitative forecasting techniques are classified into time-series and causal models. They aim to identify, from past values, the main patterns that will continue in the future. The most frequent patterns, illustrated in Figure 8.1, include the following:1. Stationary: level or constant demand;2. Growth or trend: long-term pattern of growth or decline;3. Seasonality: cyclic pattern repeating itself at fixed intervals; and4. Economic cycles: similar to seasonality, but length and magnitude of cycle may vary.
đang được dịch, vui lòng đợi..
