SIMPLE LINEAR REGRESSION

 

Assumed Linear Model:

 

               

 

 

Theoretical basis for selection of dependent and independent variables

           

            1.  Dependent variable is the variable that we desire to forecast (predict)

 

            2.  Independent or predictor variable(s) are assumed to have a one-way causal effect on the dependent variable.

 

Criteria for selection of independent variables:

 

          1.  There is a sound theoretical basis for including the predictor variable in the model.

 

            2.  There is an adequate data for all of the variables of interest.  Rule of thumb:  there should be at least 5 observations for every independent variable used in the model to guard against “overfitting.”

 

            3.  There should be future projections of the independent variables or a lead-lag relationship should be established, so that the dependent variable is predicted for “known” independent variables.

 

            4.  The historical relationship is likely to continue into the future.

 

Steps in Estimating the Relationship

 

            1.  Always begin with a scatter diagram with the dependent variable on the y-axis and the independent variable on the x-axis. 

            2.   Examine the correlation coefficient matrix among the dependent and independent variable(s). 

3.      Estimate the least squares regression line.

 

Linear Regression Model 

 

1.      Statistical formulation (what determines the values of the coefficients?)

 

2.      Statistical estimation (how do we estimate the regression coefficients?)

 

3.      Properties of the least-squares model (assumptions that model should satisfy)

a.      No specification error (linear, relevant independent variables included)

b.      No measurement error

c.       Error term (random, constant variance, no autocorrelation, uncorrelated with independent variables, normal distribution.)

 

`           4.  Estimated regression equation versus true regression equation.

 

a.      Parameter estimation

b.      Statistical validation (hypothesis testing)

 

5.  Model performance

c.       Standard error of the estimate

d.      Analysis of variance

 

Forecasting with Linear Regression Models

 

  1. Point forecasts

 

  1. Interval forecasts
    1. Selection of level of confidence
    2. The standard error of the forecast with small samples
    3. The standard error with large samples

 

  1. Evaluation the forecast performance
    1. MAD, MSE, MAPE, etc. 
    2. Turning point accuracy