In addition to using historic trends when attempting to forecast the future, we can use another method called ‘regression forecasting’.This method of forecasting calculates the strength of the dependency between two variables with a result of 0 representing no dependency relationship at all (totally random) and 1 (direct relationship). This calculation is called a ‘correlation co-efficient’ (drop me an email if you want to know the maths behind this).
If the dependency between the two variables is strong (e.g. if one goes up so does the other one roughly in proportion) and we know the value of one of them we can use this to drive the value of our forecasted variable.
Great, but how is this useful?
Some example of some things we might want to calculate
• Size of winning margin vs. if a team is playing at home vs. away (i.e. is there a home ground advantage) used for calculating the odds when betting?
• Interest rates and the general economy (Reserve Bank)?
• The selling price you plan on setting for your product and the forecasted number of items sold (Marketing Departments)?
• This quarters Sales vs. Next quarters consulting revenue (consulting executives)?
• The movement in the current price of gold today vs. the price of a gold mining stock on the stock exchange tomorrow (stock brokers, hedge funds)?
• Smoking and lung disease?
• The current temperature in the pacific and next month weather (El-Nino effect?)
It’s often easier to start with items that you already think have a strong relationship, however if you have a large data set (for example the sales history for all items for the last two years) you can use ‘data mining’ techniques to sift through data to pick up correlations that you may not have thought previously existed.
If the predictor value is known (or can be controlled) and there is a delay between the two variables this method of forecasting is especially useful. In this circumstance the forecasting can be done with no (or little) history, especially useful if the history is irrelevant, inaccurate, there is not trend or if the data is simply not available.
(btw Shell Stadium a.k.a Kardina Park is Geelong Football teams home ground, a strong relationship exists between winning and playing at this ground)
It’s often easier to start with items that you already think have a strong relationship, however if you have a large data set (for example the sales history for all items for the last two years) you can use ‘data mining’ techniques to sift through data to pick up correlations that you may not have thought previously existed.
If the predictor value is known (or can be controlled) and there is a delay between the two variables this method of forecasting is especially useful. In this circumstance the forecasting can be done with no (or little) history, especially useful if the history is irrelevant, inaccurate, there is not trend or if the data is simply not available.
(btw Shell Stadium a.k.a Kardina Park is Geelong Football teams home ground, a strong relationship exists between winning and playing at this ground)
