130 likes | 304 Views
Forecasting. Time Series Analysis - Projecting future values from historical trends. Approaches. Moving Average Weighted Moving Average Exponential Smoothing. Data Collection. More Data is Better GIGO (garbage in garbage out) Timely Relevant. Three Year Moving Average.
E N D
Forecasting Time Series Analysis - Projecting future values from historical trends.
Approaches • Moving Average • Weighted Moving Average • Exponential Smoothing
Data Collection • More Data is Better • GIGO (garbage in garbage out) • Timely • Relevant
Three Year Moving Average • Forecastt = (Salest-1 +Salest-2 + Salest-3)/3
Three Year Weighted Moving Average • Forecast = W1 Salest-1 + W2 Salest-2 + W3 Salest-3 • W1+W2+W3 = 1 • The most recent period is generally the best indicator and therefore gets the greatest weight.
Number of Periods and Weighting • A shorter time period reacts more quickly and is more appropriate when the series is less stable. • Greater weights should be applied to more recent periods when the series is less stable.
Exponential Smoothing • Forecastt = Forecastt-1 + (Salest-1-Forecastt-1) • A large Alpha () emphasizes recent demand therefore it is more responsive to changes in the underlying average. • The starting Forecast is generally the Actual for that period.
Evaluating Forecast Models • Cumulative Forecast Error - CFE • Mean Absolute Deviation - MAD • Root Mean Square Error -RMSE
Cumulative Forecast Error(CFE) • Errort = Actualt - Forecastt • CFE = Errors • This measures bias, or whether the forecast is consistently above or below the actual. The smaller the CFE the less bias.
Mean Absolute Deviation (MAD) MAD = Absolute Value(Errors) n This is the average absolute error. The lower the MAD, the better the model.
Root Mean Square Error (RMSE) RMSE = Errors2 n This is the standard deviation of the errors around the actual values. The lower the RMSE the better the model.
Excel Example • Develop a forecast for 1998 using: • 3 year moving average • 3 year weighted moving average (.7, .2, .1) • Exponential Smoothing ( = .15)