3. Forecasting Flashcards
In sample period
The sample period over which the models are estimated
Out of sample period
The data segment we hold out in order to evaluate the estimated models for predictive power.
Forecast origin
The exact time period at which the forecast is being made
Forecast horizon
The time between the forecast origin and event being predicted.
One step ahead forecasts
When the forecast horizon is one period
Point forecasts
These estimate a particular value of the variable being forecast
Interval forecasts
These give intervals within which the forecasted value should be found for a particular % of the time
Forecast error
The difference between actual (observed) value of the variable and its predicted value
Standard error of the prediction
The square root of the forecast error variance.
What are the different types of information criterion?
Akaike’s information criterion (AIC)
Schwarz’s / Bayesian information criterion (BIC)
How do we choose the best model based on information criterion?
- Estimate serveral ARDL (p,q) models and for each model check for serial correlation
- Eliminate all models with serial correlation
- Select remaining models which fit the in sample data best using AIC and BIC. Smallest value is best
Forecast error û
The difference between the actual value of the variable and its predicted value
How do we make an interval forecast?
We take our predicted value GÛ and add or minus 1.96 times the standard error times the forecast error to give a 95% confidence interval
What are the two ways of comparing the forecasting ability of models
•root mean squared error RMSE
•mean absolute error (MAE)
the smaller the value the better the model
What is the mean absolute error
The average of the absolute forecast errors
