Model creation tool for econometric models
View Patent ↗A model creation tool is provided that automatically evaluates potential econometric models and, given a set of historical data on which the model is intended to operate and other constraints, automatically identifies an optimum model. The optimum model may be an autoregressive model. The optimum model is defined by a set of optimum model parameters and optimum model coefficients. The optimum model parameters may include a model order such as an autoregressive model order, a seasonality function parameter, a trend function parameter, and a parameter that sets a tolerance that is used when performing principal components dimension reduction operations on the historical data.
1. A method for using an econometric model creation tool implemented on computing equipment to automatically create an optimum econometric model that generates an economic forecast over a forecast horizon based on a given set of historical data, wherein the model is defined by model parameters and model coefficients, the method comprising:
selecting a plurality of trial sets of model parameters for the model using the model creation tool implemented on the computing equipment;
estimating model coefficients for the model defined by each set of trial model parameters using the model creation tool implemented on the computing equipment;
with the model creation tool implemented on computing equipment, using the estimated model coefficients and the given set of historical data in the model that is defined by each set of trial parameters to generate a corresponding error metric value that is indicative of how well that model generates forecasts;
with the model creation tool implemented on computing equipment, using the error metric values to identify which of the sets of trial parameters and estimated model coefficients defines the optimum model; and
using the model creation tool implemented on computing equipment to perform an automated principal components analysis on the given set of historical data.
2. The method defined in claim 1
wherein using the model creation tool to perform the automated principal components analysis and estimating the model coefficients comprises constructing a singular value decomposition of a regression matrix.
3. The method defined in claim 1 wherein estimating the model coefficients comprises constructing a limited dimensional pseudo-inverse matrix.
4. The method defined in claim 1 wherein using the estimated model coefficients and the given set of historical data in the model that is defined by each set of trial parameters to generate the corresponding error metric value comprises creating forecast history matrices from the given set of historical data.
5. The method defined in claim 1 wherein using the estimated model coefficients and the given set of historical data in the model that is defined by each set of trial parameters to generate the corresponding error metric value comprises creating forecast history matrices from the given set of historical data and computing variances between elements in different rows in the forecast history matrices.
6. The method defined in claim 1 wherein the model is an autoregressive (AR) model having an AR order, wherein the order of the AR model is one of the model parameters in each set of trial model parameters, and wherein using the error metric values to identify which of the sets of trial parameters and estimated model coefficients defines the optimum model comprises identifying an optimum AR order.
7. The method defined in claim 1 wherein each set of trial model parameters includes a seasonality function parameter and wherein using the error metric values to identify which of the sets of trial parameters and estimated model coefficients defines the optimum model comprises identifying an optimum seasonality function parameter.
8. The method defined in claim 1 wherein each set of trial model parameters includes a trend function parameter and wherein using the error metric values to identify which of the sets of trial parameters and estimated model coefficients defines the optimum model comprises identifying an optimum trend function parameter.
9. The method defined in claim 1 wherein each set of trial model parameters includes a principal components dimensional reduction tolerance parameter and wherein using the error metric values to identify which of the sets of trial parameters and estimated model coefficients defines the optimum model comprises identifying an optimum principal components dimensional reduction tolerance parameter.
10. The method defined in claim 1 wherein each set of trial model parameters includes an autoregressive model order parameter, a seasonality function parameter, a trend function parameter and wherein using the error metric values to identify which of the sets of trial parameters and estimated model coefficients defines the optimum model comprises identifying an optimum autoregressive model order parameter, an optimum seasonality function parameter, and an optimum trend function parameter.
11. The method defined in claim 1 wherein estimating the model coefficients comprises constructing a pseudo-inverse matrix.
12. The method defined in claim 1
wherein using the model creation tool to perform the automated principal components analysis and estimating the model coefficients comprises constructing an approximate singular value decomposition of a regression matrix, wherein constructing the singular value decomposition comprises constructing a symmetric eigenvalue decomposition of a power of the regression matrix.