Chapter 5 (The forecaster’s toolbox) Flashcards
Mean method
Forecasts of all future values are equal to the average of the historical data
- often not preferred
Naïve method
All forecasts to be the value of the last observation
- use when no trend, no seasonality
Seasonal naïve method
Each forecast to be equal to the last observed value from the same season
- use when seasonality is strong
Drift method
amount of change over time (called the drift) is set to be the average change
drawing a line between first and last observations, and extrapolating it
- use when there is trend
Why have the 4 simple forecasting methods?
Any forecasting methods that are developed will be compared to these simple methods to ensure new method is better than them
if not, new method is not worth considering
Fitted values
Predicted values for the outcome variable
forecasts that uses previous observations
Residuals
difference between the observations and the corresponding fitted/predicted values
Residual diagnostics
uz-cn
good forecasting method will yield innovation residuals with the following properties:
1. innovation residuals are uncorrelated
2. innovation residuals have zero mean
useful (but not necessary) for the residuals to also have the following two properties:
3. innovation residuals have constant variance
4. innovation residuals are normally distributed
- these impact prediction intervals
Portmanteau tests (Ljung-Box test) for autocorrelation
If p-value greater than 0.05 –> accept hypothesis of white noise
- white noise present, ideal
If p-value less than 0.05 –> reject hypothesis of white noise
- autocorrelation present, not ideal
Prediction intervals
an interval within which we expect yt to lie with a specified probability
1.96 for 95%
1.28 for 80%
accuracy of forecasts
determined by considering how well a model performs on new data that were not used when fitting the model
Forecast errors
difference between an observed value and its forecast
- residuals are calculated on the training set while forecast errors are calculated on the test set
- residuals are based on one-step forecasts while forecast errors can involve multi-step forecasts
RMSE
Root Mean Square Error
Minimising the RMSE will lead to forecasts of the mean
Time series cross-validation
uses several training and testing sets at different points to achieve a possible forecast
training and testing uses known data, thus will perform slightly better