Preprocessing the Data Missing Date Values Many forecasting techniques will not tolerate missing values. Things that can be done: Use forecasting technique to forecast value forward or back-cast Use average or ?eyeball? If seasonal, use average of previous seasonal values If multiple (causal) regression- drop observations or use estimates Do sensitivity analysis Work-Day Adjustments Adjustment for variation in actual # of work days in a time period Procedure Collect data on the number of working days or business days occurring each month for the several years covered by the time-series Use this data to adjust actually monthly data so that each month will reflect what the results would have been had that month contained the average number of working days Find the average number of working days for each of the 12 months. Then adjust the raw data by the appropriate percentage to reflect what actual values would have been if the month had contained the average number of working days E.g., if the average number of working days = 20 and you find a month that had 22 working days then Xadj=(20/22)Xt Where Xadj = adjusted value of output or data Xt = raw value of output or data Special (unusual) events Replace values with amore typical value Forecast or back-cast Make sure that these fluctuations will not occur in the future. If not sure, then leave the special/unusual event in Standardizing Places different time-series on a common basis (takes out level) Helps in comparing different time-series on the basis of forecasting model parameters Eliminating Trend (There are methods that cannot handle trend pg. 40 which can and which can?t) To achieve stationary Some methods perform better when only single pattern (horizontal, trend, seasonal, cycle) is modeled Trend is eliminated or reduced using a technique called differencing Pg. 38 De-seasonalizing the Data Forecast Adjustments Raw Data (Xt) Transform Workday, Deflate Trend, Seasonality Adjusted Data Forecast Adjusted Data Forecast (adjusted) Transform Workday, Deflate, Trend, Seasonality Forecast (Ft)
Want to see the other 2 page(s) in Lecture 9-Preprocessing the Data.docx?JOIN TODAY FOR FREE!