Forecasting/ Causal Model

1
Forecasting/ Causal Model

• MGS 8150

2
Forecasting
3
Quantitative Forecasting
--Forecasting based on data and models

• Causal Models

Price Population Advertising
Causal Model
Year 2000 Sales

• Time Series Models

Sales1999 Sales1998 Sales1997
Time Series Model
Year 2000 Sales
4
Causal versus Correlation

• There is some confusion between causality and
  correlation.
• All causality has some correlation but all
  correlations do not indicate causality.

5
Causal forecasting

• Regression
• Find a straight line that fits the data best.
• y Intercept slope x
• Slope change in y / change in x

Best line!
Intercept
6
Causal Forecasting Models

• Curve Fitting Simple Linear Regression
• One Independent Variable (X) is used to predict
  one Dependent Variable (Y).
• Prediction line is written as Y a b X,
  where a is the Intercept of the line and b is the
  Coefficient.
• a and b are estimated by software (say, Excel).
  No need to learn formulas for them. Find the
  regression line with Excel
• Use Excels Data Data Analysis Regression
• (You may need a plug-in Analysis Tool Pack)
• Curve Fitting Multiple Regression
• Two or more independent variables are used to
  predict the dependent variable
• Y b0 b1X1 b2X2 bpXp

7
Using of the Model

• Make a forecast or prediction
• Interpretation of the coefficients of X (aka,
  independent variables)
• Interpretation of the intercept (optional)

8
Evaluating Goodness of the Model

• Check the following
• Check R-squared (for the whole model).
• Indicates how much of the total sum of squared
  (SS column in the Excel output) is explained away
  or removed by the Regression model. R-squared
  SS Regression/ SS Total.
• Higher the better absolute acceptable values
  depend on the knowledge level of the field
• Check F-value and its significance (for the whole
  model).
• F-value indicates overall goodness of the model.
  Higher the better.
• Check its significance value, simply put, is the
  probability that the model is not a good fit.
  Lower the better.
• Check p-values (for the individual variables)
• P-value of a variable, simply put, is the
  probability that the variable is not a
  significant player in the model. Lower the better.