Economic phenomenon predicting and analyzing system using neural network

• US 5,444,819 A
• Filed: 04/27/1993
• Issued: 08/22/1995
• Est. Priority Date: 06/08/1992
• Status: Expired due to Fees

• Current Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Inventors: Negishi, Michiro
• Primary Examiner(s): Downs, Robert W.
• Assistant Examiner(s): Dorvil, Richemond
• Application Number:
  US08/053,143
• Time in Patent Office: 847 Days
• Field of Search: 395/22, 395/23, 395/24, 364/401, 364/402
• US Class Current: 706/21
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
• CPC Class Codes: G06Q 40/02 Banking, e.g. interest calculation or account maintenance credit or loans G06Q40/03
• CPC Class Codes: Y10S 706/925 Business

Abstract

An economic phenomenon predicting and/or analyzing system using a neural network. In the disclosed system, time series data indicating economic phenomena are input to preparation modules, and moving-average values and their differences are generated. One of the preparation modules performs a predetermined process over the time series data indicating an economic phenomenon, i.e. the change of TOPIX, to remove trends. A pattern sorter sorts the trend-free data into a certain number of groups. Average values of various time series data, their differences and the result of pattern sorting are input to input layer neurons of the network. The network is provided in advance with learning information of the change of TOPIX in the past. The output of the output layer neurons will be a value of prediction of the change of TOPIX. For the output of hidden layer neurons, principal components are obtained by principal analysis modules. A correlation analysis module obtains a distribution of frequency of principal component rankings and analyzes the correlation between the explanation variants and the output of the neural network based on the obtained distribution of frequency.

Claims

1. An economic phenomenon predicting system comprising:

(a) a predicting neural network comprising (a1) an input layer having a plurality of input layer neurons for inputting signals indicating respective economic phenomena, (a2) a predetermined number of hidden layers comprising a plurality of hidden layer neurons, respectively, each of said input neurons making synaptic combinations with arbitrary ones of said hidden layer neurons, each synaptic combination having a weight, and (a3) an output layer comprising a predetermined number of output layer neurons, each of said hidden layer neurons making synaptic combinations with arbitrary ones of said output layer neurons, each synaptic combination having a weight, and each of said output layer neurons outputting an output signal, and (a4) wherein the weights of the synaptic combinations between said input layer neurons and hidden layer neurons and the weights of the synaptic combinations between said hidden layer neurons and said output layer neurons are organized by learning so that when data indicating variations of various principal economic indices including an economic phenomenon to be predicted and data indicating a variation pattern of the economic phenomenon to be predicted are input to said input layer neurons, said signals output from said output layer neurons represent a result of prediction of the economic phenomena;

(b) moving-average-value arithmetic means for inputting time series data indicating various principal economic indices including the economic phenomenon to be predicted and for obtaining moving-average values for a plurality of recent predetermined periods, said moving-average-value arithmetic means supplying said obtained moving-average values, as part of said data indicating the variations of the various principal economic indices including the economic phenomenon to be predicted, to said input layer neurons;

(c) difference arithmetic means for obtaining difference at least a first to an n-th order between said moving-average values, said difference arithmetic means supplying the obtained differences, as part of said data indicating the variations of said various principal economic indices including the economic phenomenon to be predicted;

(d) trend removing means for removing trends from the time series data indicating the economic phenomenon to be predicted by subtracting from said time series data indicting the economic phenomenon to be predicted the individual moving-average value of the economic phenomenon to be predicted for any of the plurality of predetermined periods; and

(e) pattern-sorting means for sorting said time series data indicating the economic phenomenon to be predicted after removing the trends into patterns, said pattern-sorting means outputting patterns, said patterns being obtained from the sorting, as data indicating a variation pattern of the economic phenomenon to be predicted to said input layer neurons.

2. An economic phenomenon predicting system to claim 1, wherein the number of said hidden layers is one.

3. An economic phenomenon predicting system according to claim 1, wherein the number of output layer neurons is one.

4. An economic phenomenon predicting system according to claim 1, wherein said moving-average-value arithmetic means performs a moving-average-value arithmetic operation that is weighted so that a sum of the moving average values is 1.

5. An economic phenomenon predicting system according to claim 1, wherein said hierarchical-difference arithmetic means operates second order differences of said moving-average values.

6. An economic phenomenon predicting system according to claim 1, wherein each of the output signals of said output layer neurons is a signal representing an index of the economic phenomenon to be predicted.

7. An economic phenomenon predicting system according to claim 1, wherein the economic phenomenon to be predicted is a stock index representing a price of stocks on the market.

8. An economic phenomenon predicting system according to claim 1, wherein said various principal economic indices other than the economic phenomenon to be predicted include any of a turnover, various interest rates, prices of staple commodities, a financial index, a trade index and a price index.

9. An economic phenomenon predicting system according to claim 1, wherein said pattern-sorting means is a neural network provided by learning pattern that have frequently occurred in past periods in such a manner that when the time series data indicating the economic phenomenon to be predicted are input, the neural network outputs corresponding patterns obtained by said sorting.

10. An economic phenomenon predicting system according to claim 9, wherein said pattern-sorting means is a self-organizing map.

11. An economic phenomenon predicting system according to claim 1, further comprising a post-process means for converting the output signals from said output layer neurons of said neural network into a rate of change of indices of the economic phenomenon to be predicted.

12. An economic phenomenon predicting and analyzing system comprising:

(a) a neural network for predicting economic phenomenon, comprising (a1) an input layer having a plurality of input layer neurons for inputting signals indicating respective economic phenomena, (a2) a predetermined number of hidden layers having respective hidden layer neurons, respectively, each of said input neurons making synaptic combinations with arbitrary ones of said hidden layer neurons, each synaptic combination having a weight, and (a3) an output layer having a predetermined number of output layer neurons, each of said hidden layer neurons making a synaptic combinations with arbitrary ones of said output layer neurons, each synaptic combination having a weight, and each of said output layer neurons providing an output signal, wherein the weights of the synoptic combinations between said input layer neurons and hidden layer neurons and the weights of the synoptic combinations between said hidden layer neurons and said output layer neurons are organized by learning so that when data indicating variations of various principal economic indices including an economic phenomenon to be predicted and data indicating a variation pattern of the economic phenomenon to be predicted are input to said input layer neurons, said signals output from said output layer neurons represent a result of prediction of the economic phenomenon;

(b) moving-average-value arithmetic means for inputting time series data indicating various principal economic indices and for obtaining moving-average values for a plurality of recent predetermined periods, said moving-average-value arithmetic means supplying said obtained moving-average values, as part of said data indicating the variations of the various principal economic indices including the economic phenomenon to be predicted, to said input layer neurons;

(c) difference arithmetic means for obtaining a first to an n-th order difference between said moving-average values, said difference arithmetic means outputting the obtained differences, as part of said data indicating the variations of said various principal economic indices including the economic phenomenon to be predicted;

(d) principal component arithmetic means for obtaining principal components of the output of said hidden layer neurons by principal analysis; and

(e) correlation analyzing means for analyzing a correlation between variation of the economic phenomenon to be predicted and variation of the output of said output layer neurons by analyzing the obtained principal components.

13. An economic phenomenon predicting and analyzing system according to claim 12, wherein the number of said hidden layers is one.

14. An economic phenomenon predicting and analyzing system according to claim 12, wherein the number of output layer neurons is one.

15. An economic phenomenon predicting and analyzing system according to claim 12, wherein said moving-average-value arithmetic means performs a moving-average-value arithmetic operation using the weights determined so that a sum of said moving average values is 1.

16. An economic phenomenon predicting and analyzing system according to claim 12, wherein said hierarchical-difference arithmetic means operates a second order differences of said moving-average values.

17. An economic phenomenon predicting and analyzing system according to claim 12, wherein each of the output signals of said output layer neurons are signals representing an index of the economic phenomenon to be predicted.

18. An economic phenomenon predicting and analyzing system according to claim 12, wherein the economic phenomenon to be predicted is a stock index representing the price of stock on the market.

19. An economic phenomenon predicting and analyzing system according to claim 12, wherein said various principal economic indices other than the economic phenomenon to be predicted include a turnover, any kind of interest rates, prices of staple marchandize, a financial index, a trade index or a price index.

20. An economic phenomenon predicting and analyzing system according to claim 12, further comprising a post-process means for converting each of the output signals from said output layer neurons of said neural network into a rate of change of indices of the economic phenomenon to be predicted.

21. An economic phenomenon predicting and analyzing system according to claim 12, further comprising:

trend removing means for removing trends from inputting time series data indicating the economic phenomena to be predicted by subtracting from said time series data the individual moving-average value of the economic phenomenon to be predicted for any of the predetermined periods; and

pattern-sorting means for sorting said time series data indicating the economic phenomenon to be predicted after removing trends from patterns, said pattern-sorting means generating output patterns, said output patterns being obtained from the sorting, as data indicating a variation pattern of the economic phenomenon to be predicted to said input layer neurons.

22. An economic phenomenon predicting and analyzing system according to claim 21, wherein said pattern-sorting means is a neural network that has been trained with learning patterns that have frequently occurred in past periods so that when the time series data indicating the economic phenomenon to be predicted are input, the neural network generates output patterns corresponding to patterns obtained by the sorting.

23. An economic phenomenon predicting and analyzing system according to claim 12, wherein said pattern-sorting means is a self-organizing map.

24. An economic phenomenon predicting and analyzing system according to claim 22, wherein said principal component arithmetic means obtains principal components by operating eigen vectors of covariance matrix of the output of said hidden layer neurons.

25. An economic phenomenon predicting and analyzing system according to claim 24, wherein said principal component arithmetic means obtains the principal components, for the output of said hidden layer neurons when said learning data are input to said input layer neurons, by operating said eigen vectors according to a Jacobi method.

26. An economic phenomenon predicting and analyzing system according to claim 25, wherein said principal component arithmetic means inputs the inputs of said neural network in the form of outputs of said hidden layer neurons and calculates the individual explanation variants based on a plurality of last-named outputs, and

wherein said correlation analyzing means obtains a distribution of frequency of principal component rankings for said principal components obtained, for every learning data, by said principal component arithmetic means, obtains from said distribution of frequency a degree of correlation between said principal component rankings and the explanation variants, and obtains from said correlation an influence of said principal component rankings upon the economic phenomenon to be predicted.

27. An economic phenomenon predicting and analyzing system according to claim 26, wherein said correlation analyzing means obtains the degree of correlation between said principal component rankings and the explanation variants by:

obtaining from said distribution of frequency a first section including only a maximal value, a second section that includes a predetermined proportion of upper learning data except the maximal value, a third section that includes a predetermined proportion of lower learning data except a minimal value, and a fourth section including only the minimal value;

obtaining average values and standard deviations for the explanation variants in each of said second and third sections; and

judging the explanation variants, wherein a difference between the average values of said second and third sections is relatively large and wherein a ratio of a standard deviation to an absolute value of said difference is small, to indicate that the correlation to said principal component rankings is relatively strong.

28. An economic phenomenon predicting and analyzing system according to claim 26, wherein said correlation analyzing means obtains the influence of said principal component rankings upon the economic phenomenon to be predicted by:

obtaining from said distribution of frequency a first section including only a maximal value, a second section to which a predetermined proportion of upper learning data except the maximal value belong, a third section to which a predetermined proportion of lower learning data except a minimal value belong, and a fourth section including only the minimal value;

summing a plurality of products of said principal component rankings and the weights of synaptic combinations of the hidden layer neurons and output layer neurons in respective sections to generate a ranking sum; and

multiplying the ranking sum by the range of the individual section.

29. An economic phenomenon predicting and analyzing system according to claim 12, wherein said correlation analyzing means inputs the inputs of said neural network in the form of outputs of said hidden layer neurons and calculates an explanation variants based on a last-named of said outputs.

30. An economic phenomenon predicting and analyzing system according to claim 29, wherein said explanation variants are at least any of the following:

the inputs of said neural network, differences between said moving-average values in different periods for the time series data representing the same principal economic indices and differences between said moving-average values in the same period for the time series data representing different principal economic indices.

31. An economic phenomenon predicting and analyzing method comprising the steps of:

(a) obtaining moving-average values of time series data indicating various principal economic indices for a plurality of recent predetermined periods, each of said various principal economic indices including an economic phenomenon to be predicted;

(b) obtaining differences between said moving-average values, which are obtained in an individual common period, for at least a first to an n-th order difference;

(c) removing trends from time series data indicating the economic phenomenon to be predicted by subtracting from time series data indicating the economic phenomenon to be predicted an individual moving-average value of the economic phenomenon for any of the predetermined periods;

(d) sorting said time series data indicating the economic phenomenon to be predicted, after removing trends, into patterns; and

(e) predicting the economic phenomenon, based on said moving-average values, said differences and said sorting, using a neural network including (e1) an input layer comprising a plurality of input layer neurons for inputting said moving-average values and differences, and for inputting data indicating the variations of various principal economic indices including the economic phenomenon to be predicted and said patterns as data indicating a variation patterns of the economic phenomenon, (e2) a predetermined number of hidden layers comprising a plurality hidden layer neurons, respectively, each of said input neurons making synaptic combinations with arbitrary ones of said hidden layer neurons, each synaptic combination having a weight, and (e3) an output layer comprising a predetermined number of output layer neurons, each of said hidden layer neurons making synaptic combinations with arbitrary ones of said output layer neurons, each synaptic combination having a weight, and each of said output layer neurons generating an output signal representing a result of prediction of the economic phenomenon, (e4) wherein weights of the synaptic combinations between said input layer neurons and hidden layer neurons and weights of the synaptic combinations between said hidden layer neurons and said output layer neurons are organized by learning in such a manner that when a plurality of data indicating variations of the various principal economic indices and a plurality of data indicating the variation patterns of the economic phenomenon are input to said input layer neurons, said signals output from said output layer neurons represent the result of prediction of the economic phenomenon.

32. An economic phenomenon predicting and analyzing method comprising the steps of:

(a) obtaining moving-average values of time series data indicating various principal economic indices for a plurality of recent predetermined periods, each of said various principal economic indices including an economic phenomenon to be predicted;

(b) obtaining differences between said moving-average values, which are obtained in the individual common period, for at least a first to an n-th order difference;

(c) predicting the economic phenomenon, based on said moving-average values and differences, using a neural network including (c1) an input layer having a plurality of input layer neurons for inputting said moving-average values and differences, data indicating the variations of various principal economic indices including the economic phenomenon to be predicted, (c2) a predetermined number of hidden layers having a plurality of hidden layer neurons, respectively, each of said input neurons making synaptic combination with arbitrary ones of said hidden layer neurons, each synaptic combination having a weight, and (c3) an output layer having a predetermined number of output layer neurons, each of said hidden layer neurons making a synaptic combinations with arbitrary one of said output layer neurons, each of said output layer neurons generating an output signal representing the result of prediction of the economic phenomenon, and each synaptic combination having a weight, (c4) wherein weights of the synaptic combinations between said input layer neurons and hidden layer neurons and weights of the synaptic combinations between said hidden layer neurons and said output layer neurons are organized by learning in such a manner that when a plurality of data indicating the variations of the various principal economic indices and a plurality of data indicating the variation pattern of the economic phenomenon are input to said input layer neurons, said signals output from said output layer neurons represent a result of prediction of the economic phenomenon;

(d) obtaining a plurality of principal components by a principal component analysis for the output of said hidden layer neurons; and

(e) analyzing a correlation between the variation pattern of economic phenomenon and the variation of output of said output layer neurons by analyzing the obtained principal components.

33. An economic phenomenon predicting system comprising:

(a) a predetermined number of neural networks organized to output a result of prediction of an economic phenomenon when data indicating variations of various principal economic indices including the economic phenomenon and data indicating a variation pattern of the economic phenomenon are provided as input;

(b) moving-average-value arithmetic means for obtaining moving-average values of time series data indicating various principal economic indices for a plurality of recent predetermined periods, said moving-average-value arithmetic means supplying said obtained moving-average values, as part of said data indicating the variations of the various principal economic indices including the economic phenomenon to be predicted, to said networks;

(c) difference arithmetic means for obtaining a plurality of differences between said moving-average values obtained in an individual common period, for at least from a first to an n-th difference, said difference arithmetic means supplying the obtained differences, as part of said data indicating the variations of said various principal economic indices including the economic phenomenon to be predicted, to said networks;

(d) trend removing means for removing a plurality of trends from the time series data indicating the economic phenomenon to be predicted by subtracting from said time series data an individual moving-average value of the economic phenomenon for any of the predetermined periods; and

(e) pattern-sorting means for sorting said time series data indicating the economic phenomenon to be predicted into patterns, said pattern-sorting means generating output patterns, which are obtained from the sorting, as data indicating variation patterns of the economic phenomenon to said networks.

34. An economic phenomenon predicting and analyzing system comprising:

(a) a predetermined numbers of neural networks that generates a number of hidden layer outputs according to both data indicating variations of various principal indices, including an economic phenomenon to be predicted, and data indicating a variation pattern of the economic phenomenon, and so as to output signals indicating the economic phenomenon by combining said hidden layer outputs;

(b) moving-average-value arithmetic means for obtaining moving-average values of inputted time series data indicating various principal economic indices for a plurality of recent predetermined periods, said moving-average-value arithmetic means supplying said obtained moving-average values, as part of said data indicating the variations of the various principal economic indices, including the economic phenomenon to be predicted, to said networks;

(c) difference arithmetic means for obtaining a difference between said moving-average values, Which are obtained in an individual common period, for at least from a first to an n-th order difference, said difference arithmetic means including means for supplying the obtained differences, as part of said data indicating the variations of said various principal economic indices including the economic phenomenon to be predicted, to said networks;

(d) principal component arithmetic means for obtaining principal components of the hidden layer outputs by principal analysis; and

(e) correlation analyzing means for analyzing a correlation between variation of the economic phenomenon to be predicted and variation of the output of said output layer neurons by analyzing the obtained principal components.