STATISTICS COLLOQUIUM
Thursday, June 3rd, 2010 at 4:10pm, MSB 1147 (Colloquium Room)
Speaker: Candace N. Metoyer (Intel Corporation)
Title: Reacting to New Information: The use of monthly data to inform a quarterly forecast
Abstract: When producing short-term sales forecasts during volatile economic times, the ability to react quickly to new information is essential (and sometimes, "no reaction" is the appropriate reaction). Typical econometric forecasting models require that the time series used as model inputs be of the same frequency as the forecasting output. For example, most quarterly forecasting models are designed to accept only quarterly series as inputs. As a result, information from higher-frequency data, such as monthly data, is often not fully utilized. In this talk, a method that uses monthly data to inform a quarterly forecast is described.
This method is applied to real computer product sales data and a current-quarter forecast is generated. It is found that incorporation of the monthly sales data reduced the absolute relative forecast error from 18% to 3%.
