Nicht lieferbar
The Combination of Forecasts Using Rank-Based Techniques
Versandkostenfrei!
Nicht lieferbar
The combination of forecasts is getting more and more important since it is easier to obtain multiple forecasts for the same event with the advance of computing. There is also increasing competition between forecasters leading to a variety of forecasts, e.g. Meteo Consult and Deutscher Wetterdienst for weather forecasts in Germany.The typical situation is the following: A decision maker, say a politician, a company executive, or a farmer is presented with the dilemma: There are several forecasts for the same event say the GDP growth rate, sales of a certain item, or the daily temperature. Thes...
The combination of forecasts is getting more and more important since it is easier to obtain multiple forecasts for the same event with the advance of computing. There is also increasing competition between forecasters leading to a variety of forecasts, e.g. Meteo Consult and Deutscher Wetterdienst for weather forecasts in Germany.
The typical situation is the following: A decision maker, say a politician, a company executive, or a farmer is presented with the dilemma: There are several forecasts for the same event say the GDP growth rate, sales of a certain item, or the daily temperature. These are just examples for many more cases where we are confronted with this situation. Here, the combination of forecasts sets in. Even if the decision maker says: "model x was best in the past, let's stick with forecast x" he is already using a simple but justified combining technique. He may also say: "well, let's average those forecasts (to be on the safe side)". This is an even simpler combining method since it does not consider any information on the prior performance of the forecaster or forecast models, respectively.
This book is going to present new non-parametric forecast combining methods based on ranks of past forecast errors. These techniques are compared to existing combining methods by using an extensive data analysis, simulations, and a selection procedure. The main goal is to find out when or why a combining technique is best to come up with an universal algorithm, in this case the selection process.
Contents:
1. Introduction
2. Literature review of combining techniques
2.1 Simple combining methods
2.2 Variance-covariance based methods
2.3 Regression-based methods
2.4 Bayesian approach
2.5 Rank-based methods
2.6 Nonlinear methods
2.7 Other methods
3. Rank-based techniques
3.1 The univariate case
3.2 The multivariate case
4. The data
4.1 Combined forecast data in literature
4.2 German macro economic forecast data
4.3 Micro economic data - drug store sales
4.4 Weather data
5. Combination of forecasts - empirical results
5.1 The forecast errors
5.2 Analysis framework
5.3 Forecast quality measures
5.4 Combination of forecasts
5.5 Analysis of coefficients
5.6 Multivariate combination of forecasts
6. Simulation study
6.1 Analysis of the variance-covariance structure
6.2 Simulation setup
6.3 Simulation results
7. Model selection
7.1 Description of the selection process
7.2 Conducting the selection process
7.3 Inferences from the selected methods
8. Summary
The typical situation is the following: A decision maker, say a politician, a company executive, or a farmer is presented with the dilemma: There are several forecasts for the same event say the GDP growth rate, sales of a certain item, or the daily temperature. These are just examples for many more cases where we are confronted with this situation. Here, the combination of forecasts sets in. Even if the decision maker says: "model x was best in the past, let's stick with forecast x" he is already using a simple but justified combining technique. He may also say: "well, let's average those forecasts (to be on the safe side)". This is an even simpler combining method since it does not consider any information on the prior performance of the forecaster or forecast models, respectively.
This book is going to present new non-parametric forecast combining methods based on ranks of past forecast errors. These techniques are compared to existing combining methods by using an extensive data analysis, simulations, and a selection procedure. The main goal is to find out when or why a combining technique is best to come up with an universal algorithm, in this case the selection process.
Contents:
1. Introduction
2. Literature review of combining techniques
2.1 Simple combining methods
2.2 Variance-covariance based methods
2.3 Regression-based methods
2.4 Bayesian approach
2.5 Rank-based methods
2.6 Nonlinear methods
2.7 Other methods
3. Rank-based techniques
3.1 The univariate case
3.2 The multivariate case
4. The data
4.1 Combined forecast data in literature
4.2 German macro economic forecast data
4.3 Micro economic data - drug store sales
4.4 Weather data
5. Combination of forecasts - empirical results
5.1 The forecast errors
5.2 Analysis framework
5.3 Forecast quality measures
5.4 Combination of forecasts
5.5 Analysis of coefficients
5.6 Multivariate combination of forecasts
6. Simulation study
6.1 Analysis of the variance-covariance structure
6.2 Simulation setup
6.3 Simulation results
7. Model selection
7.1 Description of the selection process
7.2 Conducting the selection process
7.3 Inferences from the selected methods
8. Summary