Method and apparatus for predicting and reporting a real estate value based on a weighted average of predicted values

• US 6,876,955 B1
• Filed: 12/20/2002
• Issued: 04/05/2005
• Est. Priority Date: 12/28/2001
• Status: Active Grant

• Current Assignee: Fannie Mae
• Original Assignee: Fannie Mae
• Inventors: Kuo, Chionglong, Fleming, Mark
• Primary Examiner(s): Tsai, Carol S W
• Application Number:
  US10/324,075
• Time in Patent Office: 837 Days
• Field of Search: 702/181, 702/182, 702/194, 705/351, 705/38, 706/151, 706/161, 706/21
• US Class Current: 702/194
• CPC Class Codes: G06Q 30/06 Buying, selling or leasing transactions
• CPC Class Codes: G06Q 40/00 Finance; Insurance; Tax strategies; Processing of corporate or income taxes

Abstract

A system and method of accurately predicting and reporting a value of a property based on a weighted average of values predicted by at least two prediction models. The system and method include the steps of accessing predicted values, determined by the prediction models, for the property; determining property-specific proportional prediction error distribution information for each predicted value determined by each prediction model; assigning a weight to the predicted value determined by each prediction model by using the property-specific proportional prediction error distribution information; and generating a property-specific weighted average value based on combination of the weight and the predicted value determined by each prediction model and reporting the property-specific weighted average value to minimize prediction error during prediction of the property value.

Claims

1. A method of predicting a value of a property based on a weighted average of values predicted by at least two prediction models, the method comprising the steps of:

accessing predicted values, determined by the prediction models, for the property;

determining property-specific proportional prediction error information for the predicted values determined by each of the prediction models, the property-specific proportional prediction error information including a proportional prediction error that is proportional to a value of the property;

assigning a weight to the predicted value determined by each prediction model by using the property-specific proportional prediction error information; and

generating a property-specific weighted average value based on combination of the weight and the predicted value determined by each prediction model and reporting the property-specific weighted average value to minimize prediction error during prediction of the property value.

2. The method of claim 1, wherein the property-specific proportional prediction error information comprises a correlation coefficient and a standard deviation, the method further comprising:

identifying the correlation coefficient by correlating the proportional prediction errors determined for a pair of prediction models; and

determining the standard deviation for each prediction model by determining a standard deviation for a set of proportional prediction errors that exist for the model.

3. The method of claim 2, further comprising:

using the standard deviations to determine the weight for each prediction model.

4. The method of claim 2, further comprising:

using the correlation coefficients to assign the weight for each prediction model.

13. The method of claim 1, wherein the property-specific proportional prediction error information includes a proportional prediction error distribution parameter that is correlated to at least one statistically significant variable.

14. The method of claim 13, wherein the at least one statistically significant variable is a time interval since the last sale of the property.

15. The method of claim 13, wherein the at least one statistically significant variable is a time interval since the last property assessment.

16. The method of claim 13, wherein the at least one statistically significant variable is a magnitude of the property value.

17. The method of claim 13, wherein the proportional prediction error distribution parameter is correlated to a first statistically significant variable for a first of the prediction models having a first prediction model type, and the proportional prediction error variance is correlated to a second statistically significant variable for a second of the prediction models having a second prediction model type.

18. The method of claim 17, wherein the prediction models of the first prediction model type are a repeat sales model and/or tax assessment model, and the first statistically significant variable is a time interval.

19. The method of claim 17, wherein the prediction models of the second prediction model type are a physical characteristic model and/or a neural network model, and the second statistically significant variable is a magnitude of the property value.

20. The method of claim 13, wherein the proportional prediction error distribution parameter is a standard deviation.

21. The method of claim 13, wherein the proportional prediction error distribution parameter is a variance.

5. A method of accurately predicting and reporting a value of a property based on a weighted average of values predicted by at least two prediction models, the method comprising the steps of:

accessing predicted values, determined by the prediction models, for the property;

determining property-specific proportional prediction error information for each predicted value determined by each prediction model, wherein the property-specific proportional prediction error information comprises a proportional prediction error, a correlation coefficient and a standard deviation, wherein the proportional prediction error is proportional to a value of the property, the correlation coefficient is identified by performing a correlation of the proportional prediction error determined for a pair of prediction models and the standard deviation is determined for each prediction model by determining a standard deviation for a set of proportional prediction errors that exist for the model;

assigning a weight to the predicted value determined by each prediction model by using the property-specific proportional prediction error information; and

generating a property-specific weighted average value based on combination of the weight and the predicted value determined by each prediction model and reporting the property-specific weighted average value to minimize prediction error associated with the predicted property value.

6. The method of claim 5, further comprising the step of determining a predicted standard deviation based on a correlation between a proportional prediction error variance and at least one other predefined variable.

7. The method of claim 6, wherein the at least one other predefined variable includes a time interval since the last sale of the property.

8. The method of claim 6, wherein the at least one other predefined variable includes a time interval since the last property assessment.

9. The method of claim 6, wherein the at least one other predefined variable includes a magnitude of the property value.

10. The method of claim 9, further comprising the step of applying an exclusion filter to at least one prediction model.

11. The method of claim 6, wherein the at least one predefined variable is a statistically significant variable for determining the weight assigned to each prediction model.

34. The method of claim 5, wherein the correlation coefficient is one of a plurality of correlation coefficients, such that for any given one of the prediction models, there is a correlation coefficient corresponding to each of the remaining prediction models.

35. The method of claim 34, wherein the correlation coefficients are determined based upon sample correlation coefficients at a geographic level with adequate sample size.

36. The method of claim 35, wherein the geographic level is a county level.

12. A system for predicting the value of a property based on a weighted average of values predicted by at least two prediction models, the system comprises:

means for accessing predicted values for the property determined by the prediction models and for assigning a weight to the predicted value determined by each prediction model;

means for determining property-specific proportional prediction error information for each predicted value determined by each prediction model, wherein the property-specific proportional prediction error information comprises a proportional prediction error, a correlation coefficient and a standard deviation, wherein the proportional predication error is proportional to a value for the property, the correlation coefficient is identified by performing a correlation of the proportional prediction error determined for a pair of prediction models, and the standard deviation is determined for each prediction model by determining a standard deviation for a set of proportional prediction errors that exist for the model;

means for assigning a weight to the predicted value determined by each prediction model by using the property-specific proportional prediction error information; and

means for generating a property-specific weighted average value based on combination of the weight and the predicted value determined by each prediction model and reporting the property-specific weighted average value to minimize prediction error associated with the predicted property value.

22. The system of claim 12, wherein the property-specific proportional prediction error information includes a proportional prediction error distribution parameter that is correlated to at least one statistically significant variable.

23. The system of claim 22, wherein the at least one statistically significant variable is a time interval since the last sale of the property.

24. The system of claim 22, wherein the at least one statistically significant variable is a time interval since the last property assessment.

25. The system of claim 22, wherein the at least one statistically significant variable is a magnitude of the property value.

26. The system of claim 22, wherein the proportional prediction error distribution parameter is correlated to a first statistically significant variable for a first of the prediction models having a first prediction model type, and the proportional prediction error variance is correlated to a second statistically significant variable for a second of the prediction models having a second prediction model type.

27. The system of claim 26, wherein the prediction models of the first prediction model type are a repeat sales model and/or tax assessment model, and the first statistically significant variable is a time interval.

28. The system of claim 26, wherein the prediction models of the second prediction model type are a physical characteristic model and/or a neural network model, and the second statistically significant variable is a magnitude of the property value.

29. The system of claim 22, wherein the proportional prediction error distribution parameter is the standard deviation.

30. The system of claim 22, wherein the proportional prediction error distribution parameter is a variance.

31. The system of claim 12, wherein the correlation coefficient is one of a plurality of correlation coefficients, such that for any given one of the prediction models, there is a correlation coefficient corresponding to each of the remaining prediction models.

32. The system of claim 31, wherein the correlation coefficients are determined based upon sample correlation coefficients at a geographic level with adequate sample size.

33. The system of claim 32, wherein the geographic level is a county level.