| Occasionally, demand data are not issued in equal time periods and even some demand histories are missing. This dissertation presents forecasting techniques which enable the generation of forecasts when demand data are not issued in the uniform time space associated with missing demand history. The estimate of total demand for each observation is derived by the integration of the demand function. Then, the method of ordinary least squares (OLS) is used to solve for the coefficients needed to forecast the demand. Four models are proposed as follows: Horizontal Model, Trend Model, Three-Term Model (Seasonal Model), and Four-Term Model (Trend Seasonal Model). The coefficient of variation of the forecast errors (COV) is used to measure the performance of forecasting. In this study, 20,000 cases of simulations are generated to test each model. The results from four models show that the forecasting performs better with a higher number of observations and a lower percentage of missing data. |
