Forecasting Methods for Managers

Multi-Variate Modelling including Lagged variables and Dummy Variables

2

Topics for Today
• Multi-variate relationships • Correlation matrices • Doing a multiple regression in Excel • Multi-collinearity • Lagged variables • Dummy variables
▫ For modelling qualitative data ▫ For modelling seasonality

3

Multi-Variate Relationships
• So far we have only looked at Time Series. These are where:
. . . . one dependent variable, eg: sales, temperature . . . . varies with time

• We have identified no underlying drivers of the relationship • We just made forecasts one or more periods ahead • These are commonly used business models . . . . but the business world is not that simple:
▫ The variables we need to forecast do not just depend on time ▫ Multi-variate models are required ▫ We can then identify the ‘levers’ to pull to ‘drive’ our variable

4

An Example
In previous years this was a double module
• Attendance at tutorials varied as the year progressed • Time is one factor but other factors could be:
▫ ▫ ▫ ▫ ▫ ▫ Students’ perception that the tutorial will help them pass Weather conditions: eg temperature on morning of tutorial Time of day for the tutorial (9am tutorials are not popular) Students dropping out of the module or the university Volume of background reading in the recommended texts Assignment marks achieved – low marks produce attendance

5

An Example
How suitable are these other factors?
Suitable
• • • • Tutorial week: this is equivalent to the passage of time Students’ perception that the tutorial will help them pass Weather conditions: seems likely to have an impact Amount of reading required . . . . . . . . large amounts may increase or decrease attendance

Not Suitable
• Time of day: tutorials are the same time each week • Students dropping out is