Computer-implemented regression systems and methods for time series data analysis

• US 7,171,340 B2
• Filed: 05/02/2005
• Issued: 01/30/2007
• Est. Priority Date: 05/02/2005
• Status: Active Grant

• Current Assignee: SAS Institute Incorporated
• Original Assignee: SAS Institute Incorporated
• Inventors: Brocklebank, John C.
• Primary Examiner(s): Hoff; Marc S.
• Assistant Examiner(s): SUAREZ, FELIX E
• Application Number:
  US11/119,670
• Publication Number:
  US 20060247900A1
• Time in Patent Office: 638 Days
• Field of Search: 702/84, 702/104, 702/181, 702/183, 702/190, 702/191, 702/189, 700/29, 704/201, 704/228, 706/22
• US Class Current: 702/189
• CPC Class Codes: G06F 2111/08 Probabilistic or stochastic CAD
• CPC Class Codes: G06F 30/20 Design optimisation, verification or simulation optimisation, verification or simulation of circuit designs G06F30/30

Abstract

Computer-implemented systems and methods for analyzing time series data. Statistical techniques are performed upon candidate autoregressive components and regressor components using the time series data. Autoregressive and regressor components are included in a predictive model based upon the autoregressive and regressor components'"'"' significance levels as determined by the statistical techniques.

Claims

1. A computer-implemented method for determining a parametric model to process time series data with respect to regressor components and autoregressive components, comprising:

receiving time series data;

performing a first stepwise statistical technique upon autoregressive components using the time series data and determining significance level of an autoregressive component while performing the first stepwise statistical technique;

performing a second stepwise statistical technique upon regressor components using the time series data and determining significance level of a regressor component while performing the second stepwise statistical technique;

including, in a parametric model, autoregressive and regressor components based upon the autoregressive and regressor components'"'"' significance levels;

providing as output the parametric model that contains the autoregressive and regressor components that were included based upon the autoregressive and regressor components'"'"' significance levels;

wherein the parametric model is a predictive model of the time series data and adjusts for autocorrelation in the time series data.

2. The method of claim 1, wherein the first and second stepwise statistical techniques include stepwise regression techniques;

wherein the stepwise regression techniques include a technique selected from the group comprising;

a forward selection regression technique and a backward elimination regression technique.

3. The method of claim 1, wherein the first and second stepwise statistical techniques include stepwise regression techniques;

wherein the stepwise regression techniques include a technique selected from the group comprising;

a forward selection regression technique, a backward elimination regression technique, and a forward-selection with changes regression technique.

4. The method of claim 3, wherein multiple parametric models are automatically generated by using different permutations of the group'"'"'s stepwise regression techniques upon the autoregressive components and the regressor components;

wherein the generated parametric models are predictive models of the time series data and adjust for autocorrelation in the time series data.

5. The method of claim 1, wherein autocorrelation is measured during the first and second stepwise statistical techniques;

wherein because of the measurement of the autocorrelation and the adjustment for autocorrelation, the parametric model improves accuracy with respect to R-squares, t-tests, and p-values associated with use of the parametric model.

6. The method of claim 1, wherein the first and second stepwise statistical techniques permit combining individual factor coefficients at different lags;

wherein coefficients are combined for different lags of the same variable type by using integrated elasticity measurement;

wherein the integrated elasticity measurement includes assessing effect of percentage change in a dependent variable relative to an independent variable.

7. The method of claim 6, wherein a first integrated elasticity measurement is larger than a second integrated elasticity measurement;

wherein the larger first integrated elasticity measurement indicates the dependent variable of the first integrated elasticity measurement is more responsive to changes in the independent variable of the first integrated elasticity measurement than is the dependent variable of the second integrated elasticity measurement relative to changes in the independent variable of the second integrated elasticity measurement.

8. The method of claim 1, wherein a list of candidate autoregressive components and regressor components are provided;

wherein the first stepwise statistical technique is performed upon the candidate autoregressive components using the time series data and determining significance level of an autoregressive component while performing the first stepwise statistical technique;

performing a second stepwise statistical technique upon the candidate regressor components using the time series data and determining significance level of a regressor component while performing the second stepwise statistical technique.

9. The method of claim 1, wherein the regressor and autoregressive components include components selected from the group comprising variables and parameters.

10. The method of claim 1, wherein forecasting is performed upon the included autoregressive and regressor components in order to generate the parametric model.

11. The method of claim 1, wherein source of the time series data is a data warehouse;

wherein a data mining module performs the steps of claim 1.

12. The method of claim 1, wherein the autoregressive components used for the first stepwise statistical technique is from a candidate set of autoregressive components;

wherein the regressor components used for the second stepwise statistical technique is from a candidate set of regressor components.

13. The method of claim 1, wherein the autoregressive and regressor components included in the parametric model are for use in analyzing time lags associated with the time series data.

14. The method of claim 13, wherein the parametric model includes an independent variable Y_t;

wherein Y_t is expressed as;

Y_t=β

₀+β

_1,1X_1,t+β

_1,2X_1,t-1+β

_1,3X_1,t-2+ . . . +β

_2,1X_2,t+β

_2,2X_2,t-1+β

_2,3X_2,t-2+ . . . +β

_v,nX_v,t-k+(e_t/(1−

ρ

₁B−

ρ

₂B²−

. . . −

ρ

_sB^s));

wherein BY_t=Y_t-1;

wherein B^sY_t=Y_t-s;

wherein Y_t represents a target or independent variable;

wherein β

_v,n represents independent variables coefficients with “

n”

number of them for “

v”

number of independent variables;

wherein X_v,t-k represent independent variables with “

v”

number of them at different times “

t”

with “

k”

time lags;

wherein e_t represents an error term at different times “

t”

;

wherein ρ

_s represents autoregressive parameters with “

s”

of them;

wherein B represents backshift factors.

15. The method of claim 14, wherein at least one of the X_v,t-k terms is not used to express the independent variable because the coefficient of said X_v,t-k term has a value of approximately zero.

16. A data signal that is transmitted across a network, wherein the data signal includes the parametric model of claim 1.

17. Computer-readable medium capable of causing a computing device to perform the method of claim 1.

18. A memory that stores a parametric model to process time series data with respect to regressor components and autoregressive components, wherein the parametric model was constructed based upon a method comprising:

receiving time series data;

performing a first stepwise statistical technique upon autoregressive components using the time series data and determining significance level of an autoregressive component while performing the first stepwise statistical technique;

performing a second stepwise statistical technique upon regressor components using the time series data and determining significance level of a regressor component while performing the second stepwise statistical technique;

including, in a parametric model, autocorrelation and regressor components based upon the autocorrelation and regressor components'"'"' significance levels;

providing as output the parametric model that contains the autoregressive and regressor components that were included based upon the autoregressive and regressor components'"'"' significance levels;

wherein the parametric model is a predictive model of the time series data and adjusts for autocorrelation in the time series data.

19. A computer-implemented system for determining a parametric model to process time series data with respect to regressor components and autoregressive components, comprising:

first computer instructions configured to perform a first statistical technique upon autoregressive components using time series data and to determine significance level of an autoregressive component while performing the first statistical technique;

second computer instructions configured to perform a second statistical technique upon regressor components using the time series data and to determine significance level of a regressor component while performing the second statistical technique;

wherein autoregressive and regressor components are included in the parametric model based upon the autoregressive and regressor components'"'"' significance levels;

third computer instructions configured to provide as output the parametric model that contains the autoregressive and regressor components that were included based upon the autoregressive and regressor components'"'"' significance levels;

wherein the parametric model is a predictive model of the time series data and adjusts for autocorrelation in the time series data.

20. The system of claim 19, wherein the time series data contains a sequence of observations related to an advertising activity;

wherein the parametric model is a predictive model for use in analyzing the advertising activity and adjusts for autocorrelation in the time series data which is related to the advertising activity.