ch 4. part two

Question 1

forecast performance evaluation: purpose?

Answer 1

1. to monitor the performance if any given model (over time)

2. to select the most suitable model for a given forecasting situation

Question 2

commonly used criteria for forecast performance evaluation?

Answer 2

1. bias
2. mad (mean absolute deviation)
3. mape (mean absolute percent error)
4. mse (mean squared error)

Question 3

bias

Answer 3

1. average forecast error (could be positive or negative)
2. reveals direction of error
   
   • negative bias indicates overcast
     - positive bias indicates under forecast
3. positive and negative errors may cancel each other, which could suppress the magnitude of error

Question 4

mad (mean absolute deviation)

Answer 4

1. average of absolute errors (always positive)
2. reveals magnitude of errors
3. does not reveal direction of errors
4. unlike bias, does not suppress the magnitude of errors

Question 5

mape (mean absolute percent error)

Answer 5

1. percent error (always positive)

2. reveals proportional error (i.e. the size of error with respect tot he size of actual demand)

Question 6

mse (mean squared error)

Answer 6

1. always positive
2. reveals variation among errors
3. large errors are weighted more, small errors are weighted less

Question 7

ex post forecast

Answer 7

1. The procedure requires to split the entire past data into two halves.
2. The first half of the data is used to identify one or more models that fit the data best.
3. These models are then used to forecast the other half of the data (as if it was the future)
4. The model that predicts the “assumed” future best is chosen.
   Ex-post means after the fact

Question 8

tracking signal (TS)

Answer 8

1. TS is used to monitor the performance of the model
2. TS is computed at the end of each period
3. TS of up to plus or minus 3 is considered okay
4. Whenever TS goes outside the limit, model needs to be revised
5. TS should not show any discernible pattern, such as a steady upward or downward movement.
6. The numerator is also called RSFE (Running Sum of Forecast Errors). It is not the same as bias.

Question 9

simple linear regression for forecasting?

Answer 9

y=a + bX

a=intercept
b=slope
y=dependent variable (forecast demand)
x=independent variables (variables that influence demand)

Question 10

forecasting by multiple regression?

Answer 10

• includes more than on independent variable, but one dependent variable
• applies in situations where demand depends on more than one factor
• in addition to the unusual f tests and t tests, model needs to be checked for multicollinearity, autocorrelation, etc.