Can you give all the criteria to evaluate the forecasting performance of the regression estimators?
It depends on how many series (objects) your data consists of.
If you have many series, the following information may be interesting for you.
Our review of available measures have suggested that popular measures (such as MAPE, MASE, and others) have serious limitations.
We recommend a scheme based on average relative MAE (AvgRelMAE):
https://www.researchgate.net/publication/282136084_Measuring_Forecasting_Accuracy_Problems_and_Recommendations_%28by_the_Example_of_SKU-Level_Judgmental_Adjustments%29
https://www.researchgate.net/publication/257026708_Measuring_Forecasting_Accuracy_The_Case_Of_Judgmental_Adjustments_To_Sku-Level_Demand_Forecasts
The approach is based on finding geometric mean of relatives MAEs. You need to 1) select a banchmark, 2) calculate relative MAEs according to your benchmark, 3) find geometric mean of relative MAEs.
In the above papers we describe the advantages of this approach and the reasons why alternative approaches may lead to false conclusions.
Article Measuring Forecasting Accuracy: The Case Of Judgmental Adjus...
Chapter Measuring Forecasting Accuracy: Problems and Recommendations...
It would be useful to know whether you are predicting or forecasting in a time or spatial series
Regression estimators are based on past data, and good forecast depends on future data. So, the forecasting performance is linked with the stability of the past and the future data more than in the regression estimator itself alone.
The regression model is estimated from the past data. Best fitted model always reduces the forecast error. The model fitness can be determined through estimating the R-square value. If the R-square value is more than 90% then the model estimated is treated as best fitted model for prediction. Again through estimating forecast error (in terms of MAPE, MASE etc) you can check the accuracy of the regression model.
Hi dear Espejo!
could please guide me how can I get future for the variable(s) which I want to forecast. Appreciated if you could give illustrations either in E-Views or Grtl or SPSS.
Looking forward for your early and positive response.
Let for example, I want to forecast health care expenditure ad GDP to the period 2023 and 2025. I have annual data from 1981 to 2015 on the variables.
There are different forecasting techniques are available for prediction. One popular model is time series ARIMA model. Others are artificial intelligence techniques such as ANN, ANFIS, LSSVM and genetic programming. You can apply any of these to predict the future value of a variable, if you want to predict the changing variance of a variable then GARCH is a suitable model.
Thanks dear Sanjita!
I am stuck to the point that how can I get the future values of the variables. I am using ARIMA. I mean that for value for the period 2016, 2017 etc etc.
I am using E-Views and Grtl if u know the procedure then plz help.
Thnx.
Habib
Sorry, I have no idea on E-Views and Grtl. I have used STATISTICA and Minitab software for time series modelling.
ok Thanks , it will better if you could help me in Minitab, with codes and illustrations by using hypothetical data or ur own work of time series forecasting.
Habib
To check how good your regression model is, you can use the following metrics:
- Badges
- Science topic