WO2021022977A1 - 预测资产价格走势的方法、服务器及计算机可读存储介质 - Google Patents

预测资产价格走势的方法、服务器及计算机可读存储介质 Download PDF

Info

Publication number
WO2021022977A1
WO2021022977A1 PCT/CN2020/101642 CN2020101642W WO2021022977A1 WO 2021022977 A1 WO2021022977 A1 WO 2021022977A1 CN 2020101642 W CN2020101642 W CN 2020101642W WO 2021022977 A1 WO2021022977 A1 WO 2021022977A1
Authority
WO
WIPO (PCT)
Prior art keywords
year
asset
sequence
observation
data
Prior art date
Application number
PCT/CN2020/101642
Other languages
English (en)
French (fr)
Inventor
林晓明
Original Assignee
华泰证券股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华泰证券股份有限公司 filed Critical 华泰证券股份有限公司
Publication of WO2021022977A1 publication Critical patent/WO2021022977A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0283—Price estimation or determination
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/06—Asset management; Financial planning or analysis

Definitions

  • This application relates to the field of computer technology, and in particular to a method, server, and computer-readable storage medium for predicting asset price trends.
  • the main purpose of this application is to provide a method, a server and a computer-readable storage medium for predicting asset price trends, aiming to solve the technical problem of the lack of a method for obtaining asset trend information based on the law of asset economic cycle changes in the prior art.
  • this application provides a method for predicting asset price trends, and the method for predicting asset price trends includes the following steps:
  • the step of performing frequency domain filtering on the year-on-year sequence data of each type of the asset according to a plurality of the observation periods, and obtaining the filtering sequence of the asset in each of the observation periods includes:
  • the step of performing inverse Fourier transform on the first intermediate sequence to obtain the filtering sequence of the asset in each observation period includes:
  • data points are intercepted from the second intermediate sequence to obtain a filtering sequence of the asset in each observation period.
  • the year-on-year sequence data of each type of asset includes corresponding year-on-year sequence data of multiple asset targets, and the year-on-year sequence data of each type of asset is frequency-based on the multiple observation periods.
  • Domain filtering to obtain the filtering sequence of the asset under each observation period includes:
  • the filtering subsequences corresponding to the multiple asset targets in each observation period are combined to obtain the filtering sequence of the asset in each observation period.
  • the merging processing step includes Hilbert transformation and iterative calculation of merging weights.
  • the step of predicting the trend of the asset according to the regression parameter corresponding to each type of the asset and the filtering and merging sequence includes:
  • the first data point corresponding to the observation time is determined from the filtering and merging sequence corresponding to each observation period, and the first data point in each observation period and each type of asset The corresponding regression parameters are fitted to obtain the first year-on-year data point at the observation time;
  • the first year-on-year data point is compared with the second year-on-year data point, and the price trend of the asset is predicted according to the comparison result, wherein, when the first year-on-year data point is less than the second year-on-year data point, The return on assets increases, and when the first year-on-year data point is greater than the second year-on-year data point, the return on assets falls.
  • the method before the step of determining the first data point corresponding to the observation time from the filtering and merging sequence corresponding to each observation period, the method further includes:
  • Frequency domain filtering is performed on the filtering combination sequence corresponding to each observation period.
  • the step of acquiring the observation time and acquiring the year-on-year sequence data of multiple types of assets corresponding to the observation time includes:
  • this application also provides a server, which includes a memory, a processor, and a processing program that is stored on the memory and can run on the processor for predicting asset price trends.
  • a server which includes a memory, a processor, and a processing program that is stored on the memory and can run on the processor for predicting asset price trends.
  • the processing program for predicting asset price trends is executed by the processor, the steps of the method for predicting asset price trends as described above are realized.
  • this application also proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a processing program for predicting asset price trends, and the processing program for predicting asset price trends is processed When the device is executed, the steps of the method for predicting asset price trends as described above are implemented.
  • This application first performs frequency domain filtering on the year-on-year sequence data of each type of asset according to multiple preset observation periods. The frequency domain filter sequence of the, and then the frequency domain filter sequence of multiple types of assets are combined. Based on the filter combination sequence of multiple types of assets, the linear regression method is used to fit the year-on-year sequence data of multiple types of assets to predict the trend of each type of asset. Provides a method for obtaining asset trend information based on the economic cycle change law of assets, which can accurately predict asset price trends.
  • FIG. 1 is a schematic diagram of the structure of a server involved in a solution of an embodiment of the present application
  • FIG. 2 is a schematic flowchart of the first embodiment of the method for predicting asset price trends in this application;
  • FIG. 3 is a schematic flowchart of a second embodiment of the method for predicting asset price trends in this application.
  • FIG. 4 is a schematic flowchart of a third embodiment of the method for predicting asset price trends in this application.
  • FIG. 5 is a schematic flowchart of a fourth embodiment of the method for predicting asset price trends in this application.
  • Fig. 6 is a schematic flowchart of a fifth embodiment of a method for predicting asset price trends in this application.
  • the main solution of the embodiment of the present application is to obtain the observation time and the observation period, and obtain the year-on-year sequence data of multiple types of assets corresponding to the observation time;
  • the sequence data is filtered in the frequency domain to obtain the filter sequence of the asset in each observation period; the filter sequences of multiple types of the asset in each observation period are combined to obtain the corresponding observation period Filtering and merging sequence; inputting the year-on-year sequence data of each type of asset and the filtering and merging sequence into a linear regression model to obtain the regression parameter corresponding to the asset; according to the regression parameter and the filter corresponding to each type of asset
  • the combined sequence predicts the trend of the asset price.
  • This application firstly performs frequency domain filtering on the year-on-year sequence data of each type of asset according to multiple preset observation periods to obtain the corresponding frequency domain filter sequence, and then merges the frequency domain filter sequences of multiple types of assets, based on the filtering of multiple types of assets Combine the sequence and use the linear regression method to fit the year-on-year sequence data of multiple types of assets to predict the trend of each type of asset. It provides a method for obtaining asset trend information based on the economic cycle of the asset, which can accurately predict the asset Price trend.
  • Fig. 1 is a schematic structural diagram of a server in a hardware operating environment involved in a solution of an embodiment of the present application.
  • the server may include a processor 1001, such as a CPU, a communication bus 1002, and a memory 1003.
  • the communication bus 1002 is used to implement connection and communication between these components.
  • the memory 1003 may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as a magnetic disk memory.
  • the memory 1003 may also be a storage device independent of the aforementioned processor 1001.
  • server structure shown in FIG. 1 does not constitute a limitation on the server, and may include more or fewer components than shown in the figure, or a combination of certain components, or different component arrangements.
  • the memory 1003 which is a computer storage medium, may include an operating system and a processing program for predicting asset price trends.
  • the processor 1001 may be used to call a processing program for predicting asset price trends stored in the memory 1003, and perform the following operations:
  • the processor 1001 may call a processing program for predicting asset price trends stored in the memory 1003, and also perform the following operations:
  • the processor 1001 may call a processing program for predicting asset price trends stored in the memory 1003, and also perform the following operations:
  • data points are intercepted from the second intermediate sequence to obtain a filtering sequence of the asset in each observation period.
  • the processor 1001 may call a processing program for predicting asset price trends stored in the memory 1003, and also perform the following operations:
  • the filtering subsequences corresponding to the multiple asset targets in each observation period are combined to obtain the filtering sequence of the asset in each observation period.
  • the processor 1001 may call a processing program for predicting asset price trends stored in the memory 1003, and also perform the following operations:
  • the processor 1001 may call a processing program for predicting asset price trends stored in the memory 1003, and also perform the following operations:
  • the first data point corresponding to the observation time is determined from the filtering and merging sequence corresponding to each observation period, and the first data point in each observation period and each type of asset The corresponding regression parameters are fitted to obtain the first year-on-year data point at the observation time;
  • the first year-on-year data point is compared with the second year-on-year data point, and the price trend of the asset is predicted according to the comparison result, wherein, when the first year-on-year data point is less than the second year-on-year data point, The return on assets increases, and when the first year-on-year data point is greater than the second year-on-year data point, the return on assets falls.
  • the processor 1001 may call a processing program for predicting asset price trends stored in the memory 1003, and also perform the following operations:
  • Frequency domain filtering is performed on the filtering combination sequence corresponding to each observation period.
  • the processor 1001 may call a processing program for predicting asset price trends stored in the memory 1003, and also perform the following operations:
  • the first embodiment of the present application provides a method for predicting asset price trends, the method includes:
  • Step S10 Obtain the observation time and the observation period, and acquire year-on-year sequence data of multiple types of assets corresponding to the observation time;
  • Various types of assets include stocks, bonds, and commodities.
  • For each type of asset first determine an observation time, and then obtain the year-on-year serial data of that type of asset corresponding to the observation time.
  • the observation time can be measured in months, such as January 2019, or days, such as January 31, 2019.
  • the data included in the year-on-year sequence data of a certain type of asset are all monthly year-on-year data.
  • Each month's year-on-year data is calculated in a fixed 12-month cycle.
  • the ratio of the stock closing price at the end of January 2009 to the stock closing price at the end of January 2008 or the logarithm of the ratio is used as the year-on-year data for January 2009 .
  • the year-on-year data Calculated according to the following formula:
  • the year-on-year sequence data of a certain type of asset corresponding to the observation time includes the monthly year-on-year data from the observation time and the monthly year-on-year data before the observation time (L1-1). For example, if the observation time is December 2018, and the length of the year-on-year sequence data is 120, the year-on-year sequence data of stocks corresponding to the observation time includes the year-on-year data of stocks from January 2009 to December 2018.
  • the method of calculating the monthly average price or the method of recording the month-end value can also be used. Among them, it is preferable to use The statistical method for recording the value at the end of the month.
  • Step S20 Perform frequency domain filtering on the year-on-year sequence data of each type of asset according to a plurality of observation periods to obtain a filter sequence of the asset in each observation period;
  • the asset data sequence in a continuous period of time has a time dimension.
  • stock price data is a numerical sequence in continuous time, so the resulting year-on-year sequence data can be regarded as a time sequence.
  • the time series can be analogous to the time-domain signal generated by the movement of an economic and financial asset, and the time-domain signal has the expression of Fourier series. Therefore, Fourier transform can be performed on the year-on-year sequence data to obtain the corresponding frequency. Domain data is analyzed and processed in the frequency domain.
  • the observation cycle in this embodiment Different types of assets have their common economic cycle in the unified financial and economic system, that is, the observation cycle in this embodiment, and the corresponding asset year-on-year data show regular changes in the observation cycle.
  • the observation period is 42 months, 100 months, and 200 months.
  • each observation period corresponds to a target frequency signal.
  • These target frequency signals are stable and continuous, and the remaining short-term unsustainable frequency signals can be regarded as noise. Therefore, in this step, frequency domain filtering is performed on the year-on-year sequence data of each type of asset, in order to retain these target frequency signals that contain important information about the asset's economic cycle and reduce noise interference.
  • a set of filter coefficients gauss win is determined according to each observation period period, and a set of filter coefficients gauss win and frequency domain data wave fft is obtained according to each set of filter coefficients gauss win First intermediate sequence
  • nfft is the zero-padded length of the year-on-year data
  • the gauss index is a sequence of numbers from 1 to nfft
  • center frequency represents the center frequency, which is the frequency corresponding to the period factor to be extracted
  • gauss alpha is a parameter that affects the Gaussian filter bandwidth.
  • inverse Fourier transform is performed on each set of first intermediate sequences to obtain a set of second intermediate sequences; data points are intercepted from the second intermediate sequences according to the preset sequence length LEN Obtain the filtering sequence of each type of asset under each observation period Among them, the preset sequence length LEN is equal to the sum of the corresponding sequence length L1 and the extrapolated length L2.
  • Real(Z) is the real part of Z.
  • Step S30 combining the filtering sequences of the multiple types of assets in each observation period to obtain the filtering combined sequence corresponding to the observation period;
  • the merging processing steps in this step include the iterative calculation of Hilbert transformation and merging weights. Specifically, taking the three types of assets, stocks, bonds, and commodities as examples, it is subdivided into the following steps S31 to S34:
  • Step S31 Obtain a first filter matrix according to the filter sequence of stocks, bonds and commodities
  • the filter sequences of stocks, bonds, and commodities are Regarding each filtering sequence as a vector, the three vectors are combined to form a filtering matrix M1.
  • Step S32 Hilbert transform is performed on the first filter matrix to obtain a corresponding second filter matrix
  • Step S33 Perform iterative calculation of merging weights according to the second filter matrix.
  • the combined weight vector whose observation period is period It is a vector of all 1s with length N, where N is the number of vectors in matrix M2, that is, the number of asset categories, and the value is 3 when the asset only includes stocks, bonds, and commodities.
  • weight m mean(weight) (7)
  • M*N represents the multiplication of the matrix M and N
  • (M)′ is the conjugate transpose of M
  • diag(W) contains W on the main diagonal Diagonal matrix
  • conj(M) is the complex conjugate of M
  • M.*N represents the dot product of matrix M and N
  • mean(M) is the mean value of each column of M.
  • dcnt 100 is selected in this embodiment.
  • the filter sequences of various assets can be regarded as time-domain signals, they are affected by the cycle of the entire economic and financial system and are similar to the principles of signal propagation. Most of them are affected by strong noise interference, and the signal-to-noise ratio is usually not high.
  • the iterative calculation method of merging weight given in the step reduces the phase difference estimation error, and obtains the optimal weight of the filter sequence merging of various assets, thereby effectively improving the signal-to-noise ratio and stability of the merging sequence of the filter sequence of various assets Sex.
  • Step S34 Obtain a filtered combined sequence according to the second filter matrix and the combined weight vector.
  • the filtered synthesis sequence X period whose observation period takes the value of period is a vector, namely
  • abs(W) is the complex magnitude of each element of W (each element of W is a complex number), sum(W) is the sum of the elements of W.
  • Step S40 input the year-on-year sequence data of each type of the asset and the filtered merged sequence into a linear regression model to obtain regression parameters corresponding to the asset;
  • the year-on-year serial data of the i-th asset As the explained variable (dependent variable), input the filtered combined sequence X period under each observation period as the explanatory variable (independent variable) into the linear regression model, and the observation period is 42 months, 100 months, and 200 months as For example, the following linear regression formula is obtained:
  • the least square estimation algorithm is used to obtain the estimated values of the regression parameters (b 1 , b 2 , b 3 , b 4 ) in the regression formula.
  • Step S50 Predict the trend of the asset price according to the regression parameter corresponding to each type of asset and the filtering and merging sequence.
  • Step S51 Determine the first data point corresponding to the observation time from the filtering and merging sequence corresponding to each observation period, and determine the first data point corresponding to the observation time in each observation period and each category Fitting regression parameters corresponding to the asset to obtain the first year-on-year data point at the observation time;
  • the L1 data point in the middle is the first data point corresponding to the observation time t.
  • the observation period including 42 months, 100 months, and 200 months as an example, fit according to the following formula
  • Step S52 Determine a second data point corresponding to the prediction time from the filtering and merging sequence corresponding to each observation period, and determine the second data point corresponding to the prediction time in each observation period and the second data point in each observation period and each type of observation period. Fitting the regression parameters corresponding to the asset to obtain the second year-on-year data point at the predicted moment;
  • the prediction time is later than the observation time t, which is (t+1), and taking the observation period including 42 months, 100 months, and 200 months as an example, the prediction time (t+1) is obtained by fitting according to the following formula
  • the second year-on-year data point is obtained by fitting according to the following formula
  • Step S53 comparing the first year-on-year data point with the second year-on-year data point, and predicting the trend of the asset based on the comparison result, wherein, when the first year-on-year data point is less than the second year-on-year data point When the rate of return on assets increases, when the first year-on-year data point is greater than the second year-on-year data point, the rate of return on assets falls.
  • the first year-on-year data point can be calculated according to the following formula
  • the second year-on-year data point The first difference of when Greater than zero indicates that the economic cycle factor at the predicted moment has a greater driving force on asset prices, and the asset return rate will increase; on the contrary, when Less than zero indicates that the return on assets at the predicted moment will decrease.
  • the server is a server that provides asset trend information, responsible for the collection, analysis and processing of various asset information, and management of user terminal authentication requests and information acquisition requests.
  • the server includes an acquisition module, a frequency domain filtering module, a merging module, a linear regression module, a prediction module, and an information sending module.
  • the user terminal installs and runs application software, the user registers an account with the server through the application software, and uses the registered account to log in to the server to send a request for obtaining asset information.
  • the application software includes an information generating module, an information sending module, an information receiving module, and an information display module.
  • the server and the user terminal can be in the same local area network, through a wired connection or a wireless connection, or the server and the terminal are connected to the Internet and communicate through the Internet.
  • the user sets the type of assets to be observed and the time of observation in the application software running on the user terminal.
  • the information generation module of the application software receives the user's setting information, and generates and obtains asset request information according to the setting information, and the application software information transmission module A request for acquiring asset information is sent to the server, where the asset information request includes the asset type to be observed and the observation time.
  • the server's acquisition module determines the observation time and the asset type to be observed from the received asset information request, and on the other hand acquires the observation period.
  • the observation period can be a preset observation period, or the observation period can be obtained based on the frequency domain data of the asset to be observed.
  • the specific method is: obtain the year-on-year sequence data of the asset to be observed, and perform Fourier transform on it to obtain the corresponding Frequency domain data, obtain the amplitude value of each frequency component in the frequency domain data of the asset to be observed, and determine the frequency component whose amplitude value meets the preset condition, and use the sine signal period corresponding to the frequency component whose amplitude value meets the preset condition as the observation period .
  • the preset condition may be one maximum amplitude value, or three maximum amplitude values, or the amplitude value is greater than or equal to a preset amplitude threshold.
  • the frequency domain filtering module of the server performs frequency domain filtering on the year-on-year sequence data of each type of asset according to the acquired multiple observation periods, and obtains the filtering sequence of each type of asset in each observation period.
  • the merging module of the server merges the filter sequences of multiple types of assets in each observation period to obtain the filter merge sequence corresponding to each observation period.
  • the linear regression module of the server obtains the regression parameters corresponding to each type of asset based on the year-on-year sequence data of each type of asset and the filtered and combined sequence.
  • the server's prediction module predicts the trend of each type of asset based on the regression parameters and filter merge sequence corresponding to each type of asset, and the server's information sending module sends the predicted trend information of each type of asset to be predicted to the user terminal .
  • the information receiving module of the application software running on the user terminal receives the trend information of the asset to be predicted, and the information display module of the application software analyzes the trend information of the asset to be predicted and displays it on the display screen of the user terminal, for example Display the rise and fall information of the asset to be predicted on the display in the form of text or graphs.
  • the corresponding frequency domain filter sequence is obtained by first performing frequency domain filtering on the year-on-year sequence data of each type of asset according to multiple preset observation periods, and then combining the frequency domain filter sequences of multiple types of assets, based on multiple
  • the filtering and merging sequence of various types of assets uses linear regression to fit the year-on-year sequence data of multiple types of assets to predict the trend of each type of asset. It provides a method of obtaining asset trend information based on the law of changes in the economic cycle of assets. Accurately predict asset price trends.
  • the second embodiment of the present application provides a method for predicting asset price trends based on the first embodiment.
  • the year-on-year sequence data of each type of asset includes the corresponding year-on-year sequence data of multiple asset targets.
  • the steps in step S20 include:
  • Step S21 Obtain the corresponding year-on-year sequence data of each asset target in the year-on-year sequence data of the asset;
  • the year-on-year sequence data of each type of asset includes the corresponding year-on-year sequence data of multiple asset targets.
  • it includes the year-on-year serial data of China's 10-year government bonds, the year-on-year serial data of U.S. 10-year government bonds, Year-on-year serial data on 10-year government bonds in the United Kingdom, serial data on 10-year German government bonds, and serial data on Japanese 10-year government bonds.
  • the year-on-year sequence data of multiple asset targets included in it showed extremely relevant behavior.
  • the stock markets of various countries around the world show a strong correlation, and there are also strong correlations between various financial and economic indicators of different countries. Therefore, for each type of asset, it is necessary to obtain the year-on-year sequence data of the multiple assets it contains, and extract the common economic cycle change characteristics according to the following steps S22 and S23 and merge them.
  • the combined sequence can reflect the market A unified system-level periodic movement can be used to better fit the year-on-year sequence data of various asset prices in subsequent processing.
  • Step S22 performing frequency domain filtering on the corresponding year-on-year sequence data of each asset target according to a plurality of observation periods, to obtain the filtering subsequence of the corresponding year-on-year sequence data of the asset target in each observation period;
  • the frequency domain filtering step includes zero-filling the same-year sequence data, performing Fourier transform, Gaussian frequency domain filtering, and inverse Fourier transform.
  • step S23 the corresponding filter subsequences of the multiple asset targets in each observation period are combined to obtain the filter sequence of the asset in each observation period.
  • step S40 the year-on-year sequence data of each type of asset input to the linear regression model is a sequence that has not been Gaussian filtered and merged, that is, the i-th sequence is combined with the same merging process in step S30.
  • the corresponding year-on-year sequence data of all asset targets of the class asset Perform the merger to obtain the year-on-year serial data of the combined i-th asset
  • the frequency domain filtering module of the server respectively performs frequency domain filtering on the corresponding year-on-year sequence data of multiple asset targets in each type of asset according to the acquired multiple observation periods to obtain the corresponding year-on-year sequence of each asset target. Filter sequence under each observation period.
  • the merging module of the server merges the filter sequences of multiple asset targets in each observation period to obtain the filter merge sequence corresponding to each type of asset.
  • the corresponding frequency domain filtering sequence is obtained by performing frequency domain filtering on the corresponding year-on-year sequence data of each asset object of each type of asset according to a plurality of preset observation periods, and then combining multiple asset objects corresponding Frequency domain filter sequence to obtain the combined frequency domain filter sequence corresponding to each type of asset. Since this combined frequency domain filter sequence is obtained by combining a larger number of samples with strong correlation, the combined frequency domain filter sequence It can better represent the periodic movement characteristics of each type of asset, and further improve the accuracy of the asset trend prediction results.
  • a third embodiment of the present application provides a method for predicting asset price trends based on the first embodiment. This embodiment further includes before step S51:
  • Step S60 Perform frequency domain filtering on the filtering combined sequence corresponding to each observation period.
  • the filtering and merging sequence corresponding to each observation period participating in the fitting is performed by the frequency domain filtering processing method in step S20 above.
  • the specific processing steps for each filtering combined sequence include:
  • Step S61 Perform zero padding on the filtered combined sequence, and perform Fourier transform on the filtered combined sequence after zero padding to obtain frequency domain data corresponding to the filtered combined sequence;
  • Step S62 Determine the corresponding filter coefficient according to the observation period to which the filter combined sequence belongs, and obtain a third intermediate sequence according to the frequency domain data corresponding to the filter coefficient and the filter combined sequence;
  • Step S63 Perform inverse Fourier transform on the third intermediate sequence to obtain a filtered combined filter sequence.
  • frequency-domain filtering is performed on the filtering combined sequence corresponding to each observation period to obtain a filtering combined sequence with a higher signal-to-noise ratio, which further improves The accuracy of the asset forecast trend results.
  • step S10 includes:
  • Step S11 Obtain the observation time, the preset lag period, and the preset year-on-year sequence length;
  • the research found that the original asset price lags behind its year-on-year sequence data, and there is a lag period.
  • the original asset price is about 5 months slower than the year-on-year sequence data, that is, the change in the cycle phase will be reflected in the asset return after 5 months Rate.
  • the observation time t if It is believed that the return on assets corresponding to the predicted time t+5 will rise, otherwise if It is believed that the return on assets corresponding to period t+5 will decrease.
  • a preset lag period For example, set the preset lag period to 5 months.
  • Step S12 obtaining an end time according to the observation time and the preset lag period, and obtaining a start time according to the end time and the preset year-on-year sequence length;
  • Step S13 Obtain year-on-year sequence data of multiple types of assets from the start time to the end time.
  • the required year-on-year sequence data is [t-(120+5)month, t-5month], that is, the date starts from 2000
  • the monthly sequence of stock prices starting on January 31, 2009 and ending on December 31, 2009, is used to predict the stock price trend on May 31, 2010.
  • the year-on-year sequence data of multiple types of assets is obtained according to the lag period of the original asset price relative to the same-year sequence data, which reduces the error of predicting the trend of each type of asset.
  • the present application also provides a server, which includes a memory, a processor, and a processing program for predicting asset price trends stored on the memory and running on the processor, and the processing program for predicting asset price trends When executed by the processor, the steps of the method for predicting asset price trends are realized.
  • the embodiment of the present application also proposes a computer-readable storage medium, the computer-readable storage medium stores a processing program for predicting asset price trends, and the processing program for predicting asset price trends is implemented when executed by a processor. The steps of the method for predicting asset price trends described above.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Development Economics (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Theoretical Computer Science (AREA)
  • Marketing (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Human Resources & Organizations (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Technology Law (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种预测资产价格走势的方法、服务器和计算机可读存储介质，该方法包括：获取观测时刻对应的多类资产的同比序列数据(S10)；根据多个观测周期分别对每一类资产的同比序列数据进行频域滤波，得到每一类资产在各个观测周期下的滤波序列(S20)；对每一观测周期下的多类资产的滤波序列进行合并处理，得到每一观测周期对应的滤波合并序列(S30)；将每一类资产的同比序列数据和滤波合并序列输入线性回归模型，得到每一类资产对应的回归参数(S40)；根据每一类资产对应的回归参数和滤波合并序列预测每一类资产价格的走势(S50)。

Description

预测资产价格走势的方法、服务器及计算机可读存储介质
优先权信息
本申请要求于2019年8月6日申请的、申请号为201910724548.3、名称为“预测资产价格走势的方法、服务器及计算机可读存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机技术领域，尤其涉及一种预测资产价格走势的方法、服务器及计算机可读存储介质。
背景技术
针对大类资产配置所制定的投资策略受到越来越多的关注，现阶段对大类资产周期的研究主要集中在美林时钟。通过对美林时钟的分析发现：传统的宏观择时方法发现资产轮动规律，是建立在能够经验性地对经济周期数据与资产收益率数据匹配进而研究的基础上的。这些结论一则依赖于划分经济周期的方式，二则依赖于对资产收益率的统计。不同的人依据不同标准，很有可能会划分出不同的周期阶段；即便对于周期有大致统一的认识，关于周期阶段与具体资产收益率的对应关系也难以完全确定。
这种基于经济周期阶段的资产轮动规律并不是十分稳定可靠，即目前为止缺乏基于资产的经济周期变化规律获取资产走势信息的方法。大类资产投资的相关研究集中在宏观策略研究，在量化技术领域尚未形成体系，因此相关策略可操作性差。
上述内容仅用于辅助理解本申请的技术方案，并不代表承认上述内容是现有技术。
发明内容
本申请的主要目的在于提供一种预测资产价格走势的方法、服务器和计算机可读存储介质，旨在解决现有技术中缺乏基于资产的经济周期变化规律获取资产走势信息的方法的技术问题。
为实现上述目的，本申请提供一种预测资产价格走势的方法，所述预测资产价格走势的方法包括如下步骤：
获取观测时刻和观测周期，并获取所述观测时刻对应的多类资产的同比序列数据；
根据多个所述观测周期分别对每一类所述资产的同比序列数据进行频域滤波，得到所述资产在各个所述观测周期下的滤波序列；
对每一所述观测周期下的多类所述资产的滤波序列进行合并处理，得到所述观测周期对应的滤波合并序列；
将每一类所述资产的同比序列数据和所述滤波合并序列输入线性回归模型，得到所述资产对应的回归参数；
根据每一类所述资产对应的回归参数和所述滤波合并序列预测所述资产价格的走势。
在一实施例中，所述根据多个所述观测周期分别对每一类所述资产的同比序列数据进行频域滤波，得到所述资产在各个所述观测周期下的滤波序列的步骤包括：
对所述同比序列数据进行补零，并对补零后的同比序列数据进行傅里叶变换，得到对应的频域数据；
根据每一个所述观测周期确定一组滤波器系数，并根据所述滤波器系数与所述频域数据得到第一中间序列；
对所述第一中间序列进行逆傅里叶变换，得到所述资产在各个所述观测周期下的滤波序列。
在一实施例中，所述对所述第一中间序列进行逆傅里叶变换，得到所述资产在各个所述观测周期下的滤波序列的步骤包括：
对所述第一中间序列进行逆傅里叶变换，得到第二中间序列；
根据预设序列长度从所述第二中间序列中截取数据点得到所述资产在各个所述观测周期下的滤波序列。
在一实施例中，每一类所述资产的同比序列数据包括多个资产标的对应的同比序列数据，所述根据多个所述观测周期分别对每一类所述资产的同比序列数据进行频域滤波，得到所述资产在各个所述观测周期下的滤波序列的步骤包括：
获取所述资产的同比序列数据中的各个资产标的对应的同比序列数据；
根据多个观测周期分别对每一所述资产标的对应的同比序列数据进行频域滤波，得到所述资产标的对应的同比序列数据在各个所述观测周期下的滤波子序列；
对每一所述观测周期下的多个资产标的对应的滤波子序列进行合并处理，得到所述资产在各个所述观测周期下的滤波序列。
在一实施例中，所述合并处理步骤包括希尔伯特变换和合并权重的迭代计算。
在一实施例中，所述根据每一类所述资产对应的回归参数和所述滤波合并序列预测所述资产的走势的步骤包括：
从每一所述观测周期对应的滤波合并序列中确定与所述观测时刻对应的第一数据点，并根据各个所述观测周期下的所述第一数据点以及所述每一类所述资产对应的回归参数拟合得到所述观测时刻下的第一同比数据点；
从每一所述观测周期对应的滤波合并序列中确定与预测时刻对应的第二数据点，并根据各个所述各个观测周期下的所述第二数据点以及所述每一类所述资产对应的回归参数拟合得到所述预测时刻下的第二同比数据点；
将所述第一同比数据点与所述第二同比数据点进行比较，根据比较结果预测所述资产的价格走势，其中，当所述第一同比数据点小于所述第二同比数据点时，所述资产收益率上涨，当所述第一同比数据点大于所述第二同比数据点时，所述资产收益率下跌。
在一实施例中，所述从每一所述观测周期对应的滤波合并序列中确定与所述观测时刻对应的第一数据点的步骤之前还包括：
对每一所述观测周期对应的滤波合并序列进行频域滤波。
在一实施例中，所述获取观测时刻，并获取所述观测时刻对应的多类资产的同比序列数据的步骤包括：
获取观测时刻、预设滞后期以及预设同比序列长度；
根据所述观测时刻与所述预设滞后期得到结束时刻，并根据所述结束时刻以及所述预设同比序列长度得到开始时刻；
获取从所述开始时刻到所述结束时刻的多类资产的同比序列数据。
此外，为实现上述目的，本申请还提供一种服务器，该服务器包括：存 储器、处理器及存储在所述存储器上并可在所述处理器上运行的预测资产价格走势的处理程序，所述预测资产价格走势的处理程序被所述处理器执行时实现如上所述的预测资产价格走势的方法的步骤。
此外，为实现上述目的，本申请还提出一种计算机可读存储介质，其中，所述计算机可读存储介质上存储有预测资产价格走势的处理程序，所述预测资产价格走势的处理程序被处理器执行时实现如上所述的预测资产价格走势的方法的步骤。
本申请实施例提出的一种预测资产价格走势的方法、服务器和可读计算机存储介质，本申请通过先根据多个预设观测周期分别对每一类资产的同比序列数据进行频域滤波得到对应的频域滤波序列，再合并多类资产的频域滤波序列，基于多类资产的滤波合并序列，采用线性回归的方法拟合多类资产的同比序列数据，从而预测每一类资产的走势，提供了一种基于资产的经济周期变化规律获取资产走势信息的方法，可以准确地预测资产价格走势。
附图说明
图1是本申请实施例方案涉及的服务器的结构示意图；
图2为本申请预测资产价格走势的方法第一实施例的流程示意图；
图3为本申请预测资产价格走势的方法第二实施例的流程示意图；
图4为本申请预测资产价格走势的方法第三实施例的流程示意图；
图5为本申请预测资产价格走势的方法第四实施例的流程示意图；
图6为本申请预测资产价格走势的方法第五实施例的流程示意图。
本申请目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请实施例的主要解决方案是：获取观测时刻和观测周期，并获取所述观测时刻对应的多类资产的同比序列数据；根据多个所述观测周期分别对每一类所述资产的同比序列数据进行频域滤波，得到所述资产在各个所述观测周期下的滤波序列；对每一所述观测周期下的多类所述资产的滤波序列进 行合并处理，得到所述观测周期对应的滤波合并序列；将每一类所述资产的同比序列数据和所述滤波合并序列输入线性回归模型，得到所述资产对应的回归参数；根据每一类所述资产对应的回归参数和所述滤波合并序列预测所述资产价格的走势。
本申请通过先根据多个预设观测周期分别对每一类资产的同比序列数据进行频域滤波得到对应的频域滤波序列，再合并多类资产的频域滤波序列，基于多类资产的滤波合并序列，采用线性回归的方法拟合多类资产的同比序列数据，从而预测每一类资产的走势，提供了一种基于资产的经济周期变化规律获取资产走势信息的方法，可以准确地预测资产价格走势。
如图1所示，图1是本申请实施例方案涉及的硬件运行环境的服务器的结构示意图。
如图1所示，该服务器可以包括：处理器1001，例如CPU，通信总线1002，存储器1003。其中，通信总线1002用于实现这些组件之间的连接通信。存储器1003可以是高速RAM存储器，也可以是稳定的存储器(non-volatile memory)，例如磁盘存储器。存储器1003可选的还可以是独立于前述处理器1001的存储装置。
本领域技术人员可以理解，图1中示出的服务器结构并不构成对服务器的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。
如图1所示，作为一种计算机存储介质的存储器1003中可以包括操作系统和预测资产价格走势的处理程序。
在图1所示的装置中，处理器1001可以用于调用存储器1003中存储的预测资产价格走势的处理程序，并执行以下操作：
获取观测时刻和观测周期，并获取所述观测时刻对应的多类资产的同比序列数据；
根据多个所述观测周期分别对每一类所述资产的同比序列数据进行频域滤波，得到所述资产在各个所述观测周期下的滤波序列；
对每一所述观测周期下的多类所述资产的滤波序列进行合并处理，得到所述观测周期对应的滤波合并序列；
将每一类所述资产的同比序列数据和所述滤波合并序列输入线性回归模 型，得到所述资产对应的回归参数；
根据每一类所述资产对应的回归参数和所述滤波合并序列预测所述资产价格的走势。
进一步地，处理器1001可以调用存储器1003中存储的预测资产价格走势的处理程序，还执行以下操作：
对所述同比序列数据进行补零，并对补零后的同比序列数据进行傅里叶变换，得到对应的频域数据；
根据每一个所述观测周期确定一组滤波器系数，并根据所述滤波器系数与所述频域数据得到第一中间序列；
对所述第一中间序列进行逆傅里叶变换，得到所述资产在各个所述观测周期下的滤波序列。
进一步地，处理器1001可以调用存储器1003中存储的预测资产价格走势的处理程序，还执行以下操作：
对所述第一中间序列进行逆傅里叶变换，得到第二中间序列；
根据预设序列长度从所述第二中间序列中截取数据点得到所述资产在各个所述观测周期下的滤波序列。
进一步地，处理器1001可以调用存储器1003中存储的预测资产价格走势的处理程序，还执行以下操作：
获取所述资产的同比序列数据中的各个资产标的对应的同比序列数据；
根据多个观测周期分别对每一所述资产标的对应的同比序列数据进行频域滤波，得到所述资产标的对应的同比序列数据在各个所述观测周期下的滤波子序列；
对每一所述观测周期下的多个资产标的对应的滤波子序列进行合并处理，得到所述资产在各个所述观测周期下的滤波序列。
进一步地，处理器1001可以调用存储器1003中存储的预测资产价格走势的处理程序，还执行以下操作：
进行希尔伯特变换和合并权重的迭代计算。
进一步地，处理器1001可以调用存储器1003中存储的预测资产价格走势的处理程序，还执行以下操作：
从每一所述观测周期对应的滤波合并序列中确定与所述观测时刻对应的 第一数据点，并根据各个所述观测周期下的所述第一数据点以及所述每一类所述资产对应的回归参数拟合得到所述观测时刻下的第一同比数据点；
从每一所述观测周期对应的滤波合并序列中确定与预测时刻对应的第二数据点，并根据各个所述各个观测周期下的所述第二数据点以及所述每一类所述资产对应的回归参数拟合得到所述预测时刻下的第二同比数据点；
将所述第一同比数据点与所述第二同比数据点进行比较，根据比较结果预测所述资产的价格走势，其中，当所述第一同比数据点小于所述第二同比数据点时，所述资产收益率上涨，当所述第一同比数据点大于所述第二同比数据点时，所述资产收益率下跌。
进一步地，处理器1001可以调用存储器1003中存储的预测资产价格走势的处理程序，还执行以下操作：
对每一所述观测周期对应的滤波合并序列进行频域滤波。
进一步地，处理器1001可以调用存储器1003中存储的预测资产价格走势的处理程序，还执行以下操作：
获取观测时刻、预设滞后期以及预设同比序列长度；
根据所述观测时刻与所述预设滞后期得到结束时刻，并根据所述结束时刻以及所述预设同比序列长度得到开始时刻；
获取从所述开始时刻到所述结束时刻的多类资产的同比序列数据。
参照图2，本申请第一实施例提供一种预测资产价格走势的方法，所述方法包括：
步骤S10，获取观测时刻和观测周期，并获取所述观测时刻对应的多类资产的同比序列数据；
多类资产包括股票、债券以及商品。对于每一类资产，先确定一个观测时刻，再获取观测时刻对应的该类资产的同比序列数据，其中，观测时刻可以以月为单位，例如2019年1月，也可以以天为单位，例如2019年1月31日。
在本实施例中，某一类资产的同比序列数据中所包含的数据均为月度同比数据。每一个月度同比数据以固定的12月周期进行计算，例如将2009年1月月末的股票收盘价格与2008年1月月末的股票收盘价格的比值或比值的对数值作为2009年1月的同比数据，该同比数据
Figure PCTCN2020101642-appb-000001
根据下述公式计算得到：
Figure PCTCN2020101642-appb-000002
t 0＝2008年1月,t 0+12＝2009年1月。
在同比序列数据长度为L1时，观测时刻对应的某一类资产的同比序列数据包括从观测时刻所在的月度同比数据以及观测时刻之前(L1-1)个月度同比数据。例如，观测时刻为2018年12月，同比序列数据长度为120，那么观测时刻对应的股票的同比序列数据包括从2009年1月至2018年12月的股票同比数据。
需要说明的是，在确定股票、债券、商品等多类资产在一个月的观测周期中的观测值时，可以使用计算月度平均价格的方式，也可以使用记录月末值的方式，其中，优选使用记录月末值的统计方式。
步骤S20，根据多个观测周期分别对每一类所述资产的同比序列数据进行频域滤波，得到所述资产在各个所述观测周期下的滤波序列；
在一段连续时间内的资产数据序列具有时间维度，例如股票价格数据为连续时间上的数值序列，因此由此而得的同比序列数据可看作一个时间序列。该时间序列可以类比为一项经济金融资产运动产生的时域信号，而该时域信号具有傅里叶级数的表达方式，因此可对该同比序列数据进行傅里叶变换，得到对应的频域数据，在频域对其进行分析处理。
不同类型的资产在统一的金融经济系统中有其共同的经济周期，即本实施例中的观测周期，其对应的资产同比数据在该观测周期内呈现有规律的变化，其中，优选地，在本实施例中观测周期为42个月、100个月以及200个月。在频域上，每一个观测周期对应着一个目标频率信号，这些目标频率信号是稳定且持续的，其余的短期不可持续的频率信号可视为噪声。因此，本步骤中对每一类资产的同比序列数据进行频域滤波，旨在保留这些包含着资产的经济周期重要信息的目标频率信号，降低噪声的干扰。
进一步地，由于傅里叶变换中存在栅栏效应，在求取每一类资产的滤波序列时，先对同比序列数据进行补零，并对补零后的同比序列数据进行傅里叶变换，得到对应的频域数据。以
Figure PCTCN2020101642-appb-000003
代表第i类资产的同比序列数据，以
Figure PCTCN2020101642-appb-000004
代表第i类资产的同比序列数据对应的频域数据。
如下述公式(1)、(2)、(3)所示，根据每一个观测周期period确定一组滤波器系数gauss win，并根据每一组滤波器系数gauss win与频域数据wave fft得到 一组第一中间序列
Figure PCTCN2020101642-appb-000005
Figure PCTCN2020101642-appb-000006
Figure PCTCN2020101642-appb-000007
Figure PCTCN2020101642-appb-000008
其中，nfft是同比数据补零后长度，gauss index是1至nfft的数列，center frequency代表中心频率即所要提取周期因子对应的频率，gauss alpha为影响高斯滤波带宽的参数。这些参数可优选设置为：nfft取4096，period取对应的42个月、100个月或200个月，gauss alpha取10。
需要说明的是，根据下面的公式(4)对
Figure PCTCN2020101642-appb-000009
进行共轭对称操作：
Figure PCTCN2020101642-appb-000010
接着对第i类资产中各个观测周期period的第一中间序列
Figure PCTCN2020101642-appb-000011
进行逆傅里叶变换，得到第i类资产在各个观测周期下的滤波序列
Figure PCTCN2020101642-appb-000012
具体地，如下面公式(5)所示，对每一组第一中间序列进行逆傅里叶变换，得到一组第二中间序列；根据预设序列长度LEN从第二中间序列中截取数据点得到每一类资产在各个观测周期下的滤波序列
Figure PCTCN2020101642-appb-000013
其中，预设序列长度LEN等于同比序列长度L1与外推长度L2之和。
Figure PCTCN2020101642-appb-000014
其中，Real(Z)是Z的实数部分。
步骤S30，对每一所述观测周期下的多类所述资产的滤波序列进行合并处理，得到所述观测周期对应的滤波合并序列；
在面对同一个全球经济金融环境，多类资产受到同样的经济周期的驱使而呈现出相关性极强的行为。因此，对于每一类资产，需要将同一观测周期下的多类资产的滤波序列进行合并，即将它们相似的共同经济周期变化特征并进行合并，合并后的序列能够反映市场中统一的系统级别的周期运动，以 供后续处理中较好地对各类资产的价格同比序列数据进行拟合。
本步骤中的合并处理步骤包括希尔伯特变换和合并权重的迭代计算，具体地，以股票、债券以及商品这三大类资产为例，细分为下述步骤S31～S34：
步骤S31，根据股票、债券以及商品的滤波序列得到第一滤波矩阵；
股票、债券以及商品的滤波序列分别为
Figure PCTCN2020101642-appb-000015
Figure PCTCN2020101642-appb-000016
将每一个滤波序列作为一个向量，由三个向量组合形成一个滤波矩阵M1。
步骤S32，对所述第一滤波矩阵进行希尔伯特变换，得到对应的第二滤波矩阵；
在本步骤中，如下式所示，调用软件平台库函数hibert对M1进行希尔伯特变换：
M2＝hilbert  (M1)
步骤S33，根据第二滤波矩阵进行合并权重的迭代计算。
具体地，先初始化观测周期取值为period的合并权重向量
Figure PCTCN2020101642-appb-000017
为长度为N的全1的向量，其中，N为矩阵M2中向量数目，即资产的类别数，当资产只包含股票、债券以及商品时取值为3。
根据下述公式(6)、(7)、(8)进行合并权重向量
Figure PCTCN2020101642-appb-000018
的迭代计算：
Figure PCTCN2020101642-appb-000019
weight m＝mean(weight)  (7)
Figure PCTCN2020101642-appb-000020
其中,
Figure PCTCN2020101642-appb-000021
代表经过第k次迭代计算后的合并权重向量,M*N代表矩阵M和N相乘，(M)′是M的共轭转置，diag(W)是包含W在主对角线上的对角矩阵，conj(M)是M的复共轭，M.*N代表矩阵M和N点乘,mean(M)是M的每列均值。
当迭代计算次数到达预设迭代次数阈值dcnt，合并权重收敛，得到最终的合并权重向量
Figure PCTCN2020101642-appb-000022
优选地，本实施例中选取dcnt＝100。
由于各类资产的滤波序列可视为时域信号，其受到整个经济金融系统的周期的影响与信号传播的原理有相似之处，大多受到强烈的噪音干扰，信噪比通常不高，因此本步骤所给出的合并权重迭代计算方法通过减少相位差估计误差，得出各类资产的滤波序列合并的最优权重，以此有效提高各类资产的滤波序列的合并序列的信噪比和稳定性。
步骤S34，根据第二滤波矩阵和合并权重向量得到滤波合并序列。
根据下述公式(9)得到的观测周期取值为period的滤波合成序列X period是一个向量，即
Figure PCTCN2020101642-appb-000023
abs(W)是W的每个元素的复数幅值(W的每个元素是复数)，sum(W)是W的元素总和。
步骤S40，将每一类所述资产的同比序列数据和所述滤波合并序列输入线性回归模型，得到所述资产对应的回归参数；
将第i类资产的同比序列数据
Figure PCTCN2020101642-appb-000024
作为被解释变量(因变量)，将各个观测周期下的滤波合并序列X period作为解释变量(自变量)输入线性回归模型，以观测周期取值为42个月、100个月、200个月为例，得到下述线性回归公式：
Figure PCTCN2020101642-appb-000025
线性回归模型中基于上述线性回归公式，采用最小二乘估计算法得到回归公式中的回归参数(b 1、b 2、b 3、b 4)的估计值。
步骤S50，根据每一类所述资产对应的回归参数和所述滤波合并序列预测所述资产价格的走势。
由于多个观测周期下的滤波合并序列能够反映经济金融市场中统一的系统级别的周期运动，因此能较好地对各类资产的价格走势进行拟合。
在本步骤中，具体的，如图3所示，依据下述步骤S51～S53预测每一类资产的走势：
步骤S51，从每一所述观测周期对应的滤波合并序列中确定与所述观测时 刻对应的第一数据点，并根据各个所述观测周期下的所述第一数据点以及所述每一类所述资产对应的回归参数拟合得到所述观测时刻下的第一同比数据点；
Figure PCTCN2020101642-appb-000026
中第L1个数据点为与观测时刻t对应的的第一数据点，对于第i类资产，以观测周期包括42个月、100个月、200个月为例，根据下述公式拟合得到观测时刻t下的第一同比数据点：
Figure PCTCN2020101642-appb-000027
步骤S52，从每一所述观测周期对应的滤波合并序列中确定与预测时刻对应的第二数据点，并根据各个所述各个观测周期下的所述第二数据点以及所述每一类所述资产对应的回归参数拟合得到所述预测时刻下的第二同比数据点；
预测时刻晚于观测时刻t，为(t+1)，并以观测周期包括42个月、100个月、200个月为例，根据下述公式拟合得到预测时刻(t+1)下的第二同比数据点：
Figure PCTCN2020101642-appb-000028
步骤S53，将所述第一同比数据点与所述第二同比数据点进行比较，根据比较结果预测所述资产的走势，其中，当所述第一同比数据点小于所述第二同比数据点时，所述资产收益率上涨，当所述第一同比数据点大于所述第二同比数据点时，所述资产收益率下跌。
具体地，可以根据下述公式计算第一同比数据点
Figure PCTCN2020101642-appb-000029
与第二同比数据点
Figure PCTCN2020101642-appb-000030
的一阶差分值
Figure PCTCN2020101642-appb-000031
当
Figure PCTCN2020101642-appb-000032
大于零说明预测时刻时的经济周期因子对资产价格有着更大的推动力，资产收益率将会提高；反之，当
Figure PCTCN2020101642-appb-000033
小于零则说明预测时刻的资产收益率将会降低。
Figure PCTCN2020101642-appb-000034
为了更好的理解本实施例的实现过程，对本实施例适用的资产走势信息处理系统中的硬件设备以及硬件设备之间的通信流程进行介绍。
本实施例适用的资产走势信息处理系统包括：
服务器，服务器为提供资产走势信息的服务器，负责各类资产信息的搜集与分析处理、管理用户终端的认证请求和获取信息请求。其中，服务器包括获取模块、频域滤波模块、合并模块、线性回归模块、预测模块以及信息 发送模块。
用户终端，用户终端安装并运行着应用软件，用户通过该应用软件向服务器注册账号，并使用注册账号登录服务器发出获取资产信息请求。其中，应用软件包括信息生成模块、信息发送模块、信息接收模块以及信息显示模块。
服务器和用户终端之间可处于同一个局域网中，通过有线连接或无线连接，或者服务器和终端均接入互联网，通过互联网进行通信。
下面给出用户终端与服务器之间通信流程的一种示例：
1、用户在用户终端上运行的应用软件中设置待观测的资产类型和观测时刻，应用软件的信息生成模块接收用户的设置信息，并根据设置信息生成获取资产请求信息，应用软件的信息发送模块给服务器发送获取资产信息请求，其中，所述资产信息请求包括待观测的资产类型和观测时刻。
2、服务器的获取模块一方面从接收到的获取资产信息请求中确定观测时刻和待观测的资产类型，另一方面获取观测周期。
其中，观测周期可以是预设好的观测周期，也可以根据待观测资产的频域数据获取观测周期，具体方法为：获取待观测资产的同比序列数据，对其进行傅里叶变换得到对应的频域数据，获取待观测资产的频域数据中各个频率分量的幅度值，并确定幅度值符合预设条件的频率分量，将幅度值符合预设条件的频率分量对应的正弦信号周期作为观测周期。预设条件可以为一个最大的幅度值，也可以为三个最大的幅度值，或者为幅度值大于或等于一个预设幅度阈值。
3、服务器的频域滤波模块根据获取的多个观测周期分别对每一类资产的同比序列数据进行频域滤波，得到每一类资产在各个观测周期下的滤波序列。
4、服务器的合并模块对每一个观测周期下的多类资产的滤波序列进行合并处理，得到每一个观测周期对应的滤波合并序列。
5、服务器的线性回归模块根据每一类资产的同比序列数据和滤波合并序列得到每一类资产对应的回归参数。
6、服务器的预测模块根据每一类资产对应的回归参数和滤波合并序列预测每一类资产的走势，并且服务器的信息发送模块将预测得到的每一类待预测资产的走势信息发送给用户终端。
7、用户终端上运行的应用软件的信息接收模块接收待预测资产的走势信息，并且应用软件的信息显示模块在解析完待预测资产的走势信息后将其显示在用户终端的显示屏上，例如以文字或图表的形式在显示屏上显示待预测资产的涨跌信息。
在本实施例中，通过先根据多个预设观测周期分别对每一类资产的同比序列数据进行频域滤波得到对应的频域滤波序列，再合并多类资产的频域滤波序列，基于多类资产的滤波合并序列，采用线性回归的方法拟合多类资产的同比序列数据，从而预测每一类资产的走势，提供了一种基于资产的经济周期变化规律获取资产走势信息的方法，可以准确地预测资产价格走势。
进一步的，参照图4，本申请第二实施例基于第一实施例提供一种预测资产价格走势的方法，每一类所述资产的同比序列数据包括多个资产标的对应的同比序列数据，本实施例在步骤S20中步骤包括：
步骤S21，获取所述资产的同比序列数据中的各个资产标的对应的同比序列数据；
在本实施例中，每一类资产同比序列数据中包括多个资产标的对应的同比序列数据，例如对于债券类资产，包括中国10年期国债同比序列数据、美国10年期国债同比序列数据、英国10年期国债同比序列数据、德国10年期国债同比序列数据以及日本10年期国债同比序列数据等。
在面对同一个全球经济金融环境，对于每一类资产，其所包括的多个资产标的的同比序列数据表现出了相关性极强的行为。例如在股票市场，全球各个国家的股票市场表现出极强的相关性，以及不同国家各金融经济指标之间也存在较强的相关性。因此，对于每一类资产，需要获取其所包含的多个资产标的的同比序列数据，依据下述步骤S22与步骤S23提取其共同经济周期变化特征并进行合并，合并后的序列能够反映市场中统一的系统级别的周期运动，以供后续处理中较好地对各类资产的价格同比序列数据进行拟合。
步骤S22，根据多个观测周期分别对每一所述资产标的对应的同比序列数据进行频域滤波，得到所述资产标的对应的同比序列数据在各个所述观测周期下的滤波子序列；
以
Figure PCTCN2020101642-appb-000035
代表第i类资产的第j个资产标的对应的同比序列数据，以与上述步骤S20中相同的方法对
Figure PCTCN2020101642-appb-000036
进行频域滤波，得到代表第i类资产 的第j个资产标的对应的同比序列数据对应的在各个观测周期下的滤波子序列
Figure PCTCN2020101642-appb-000037
其中，频域滤波步骤包括对同比序列数据补零、做傅里叶变换、高斯频域滤波以及反傅里叶变换。
步骤S23，对每一所述观测周期下的多个资产标的对应的滤波子序列进行合并处理，得到所述资产在各个所述观测周期下的滤波序列。
以与上述步骤S30中相同的合并处理步骤将第i类资产的所有资产标的对应的滤波子序列进行合并，得到第i类资产在各个观测周期下的滤波序列
Figure PCTCN2020101642-appb-000038
需要说明的是，在执行步骤S40时，输入线性回归模型的每一类资产的同比序列数据均为未进行高斯滤波且合并后的序列，即以上述步骤S30中相同的合并处理步骤将第i类资产的所有资产标的对应的同比序列数据
Figure PCTCN2020101642-appb-000039
进行合并，得到合并后的第i类资产的同比序列数据
Figure PCTCN2020101642-appb-000040
在本实施例中，服务器的频域滤波模块根据获取的多个观测周期分别对每一类资产中的多个资产标的对应的同比序列数据进行频域滤波，得到每一个资产标的对应的同比序列在各个观测周期下的滤波序列。
对于每一类资产，服务器的合并模块对每一个观测周期下的多个资产标的滤波序列进行合并处理，得到每一类资产对应的滤波合并序列。
在本实施例中，通过根据多个预设观测周期分别对每一类资产的每一个资产标的对应的同比序列数据进行频域滤波得到对应的频域滤波序列，再合并多个资产标的对应的频域滤波序列，得到每一类资产对应的合并后的频域滤波序列，由于这个合并后的频域滤波序列由数目更多的相关性强的样本合并而得，合并后的频域滤波序列更能代表每一类资产的周期运动特征，进一步提高了资产走势预测结果的准确性。
进一步的，参照图5，本申请第三实施例基于第一实施例提供一种预测资产价格走势的方法，本实施例在步骤S51之前还包括：
步骤S60，对每一所述观测周期对应的滤波合并序列进行频域滤波。
在本实施例中，在拟合观测时刻下的第一同比数据点拟合之前，对参与 拟合的每一个观测周期对应的滤波合并序列，以上述步骤S20的频域滤波处理方法对其进行频域滤波，对于每个滤波合并序列的具体处理步骤包括：
步骤S61，对滤波合并序列进行补零，并对补零后的滤波合并序列进行傅里叶变换，得到滤波合并序列对应的频域数据；
步骤S62，根据滤波合并序列所属的观测周期确定对应的滤波器系数，并根据所述滤波器系数与所述滤波合并序列对应的频域数据得到第三中间序列；
步骤S63，对所述第三中间序列进行逆傅里叶变换，得到滤波后的滤波合并序列。
在本实施例中，在拟合每一类资产的同比数据点之前，对每一所述观测周期对应的滤波合并序列进行频域滤波，得到了信噪比更高的滤波合并序列，进一步提高了资产预测走势结果的准确性。
进一步的，参照图6，本申请第四实施例基于上述第一至第三实施例中的任一实施例提供一种预测资产价格走势的方法，本实施例在步骤S10中包括：
步骤S11，获取观测时刻、预设滞后期以及预设同比序列长度；
经研究发现，资产原始价格滞后于其同比序列数据，存在一个滞后期，例如资产原始价格比起同比序列数据慢5个月左右，即周期相位上的变化会在5个月后反映在资产收益率上。换言之，对于观测时刻t，若
Figure PCTCN2020101642-appb-000041
则认为预测时刻t+5对应资产收益率会上升，反之若
Figure PCTCN2020101642-appb-000042
则认为t+5期对应资产收益率会下降。
因此，在本实施例中，在获取了观测时刻和预设同比序列长度后，还需要获取一个预设滞后期。例如，将预设滞后期设置为5个月。
步骤S12，根据所述观测时刻与所述预设滞后期得到结束时刻，并根据所述结束时刻以及所述预设同比序列长度得到开始时刻；
步骤S13，获取从所述开始时刻到所述结束时刻的多类资产的同比序列数据。
例如，若预设同比序列长度为120，对于观测时刻t＝2010年5月31日,需要的同比序列数据是[t-(120+5)月，t-5月]，即以日期从2000年1月31日开始到2009年12月31日结束的股票价格月度序列，用来预测2010年5月31日时股票的价格走势。
在本实施例中，在对资产走势进行预测时，根据资产原始价格相对于其 同比序列数据的滞后期获取多类资产的同比序列数据，降低了对每一类资产走势预测的误差。
本申请还提供一种服务器，该服务器包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的预测资产价格走势的处理程序，所述预测资产价格走势的处理程序被所述处理器执行时实现所述的预测资产价格走势的方法的步骤。
此外，本申请实施例还提出一种计算机可读存储介质，所述计算机可读存储介质上存储有预测资产价格走势的处理程序，所述预测资产价格走势的处理程序被处理器执行时实现所述的预测资产价格走势的方法的步骤。
需要说明的是，在本文中，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者系统不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者系统所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、方法、物品或者系统中还存在另外的相同要素。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在如上所述的一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，空调器，或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (10)

  1. 一种预测资产价格走势的方法，其中，所述预测资产价格走势的方法包括以下步骤：
    获取观测时刻和观测周期，并获取所述观测时刻对应的多类资产的同比序列数据；
    根据多个所述观测周期分别对每一类所述资产的同比序列数据进行频域滤波，得到所述资产在各个所述观测周期下的滤波序列；
    对每一所述观测周期下的多类所述资产的滤波序列进行合并处理，得到所述观测周期对应的滤波合并序列；
    将每一类所述资产的同比序列数据和所述滤波合并序列输入线性回归模型，得到所述资产对应的回归参数；
    根据每一类所述资产对应的回归参数和所述滤波合并序列预测所述资产价格的走势。
  2. 如权利要求1所述的预测资产价格走势的方法，其中，所述根据多个所述观测周期分别对每一类所述资产的同比序列数据进行频域滤波，得到所述资产在各个所述观测周期下的滤波序列的步骤包括：
    对所述同比序列数据进行补零，并对补零后的同比序列数据进行傅里叶变换，得到对应的频域数据；
    根据每一个所述观测周期确定一组滤波器系数，并根据所述滤波器系数与所述频域数据得到第一中间序列；
    对所述第一中间序列进行逆傅里叶变换，得到所述资产在各个所述观测周期下的滤波序列。
  3. 如权利要求2所述的预测资产价格走势的方法，其中，所述对所述第一中间序列进行逆傅里叶变换，得到所述资产在各个所述观测周期下的滤波序列的步骤包括：
    对所述第一中间序列进行逆傅里叶变换，得到第二中间序列；
    根据预设序列长度从所述第二中间序列中截取数据点得到所述资产在各个所述观测周期下的滤波序列。
  4. 如权利要求1所述的预测资产价格走势的方法，其中，每一类所述资 产的同比序列数据包括多个资产标的对应的同比序列数据，所述根据多个所述观测周期分别对每一类所述资产的同比序列数据进行频域滤波，得到所述资产在各个所述观测周期下的滤波序列的步骤包括：
    获取所述资产的同比序列数据中的各个资产标的对应的同比序列数据；
    根据多个观测周期分别对每一所述资产标的对应的同比序列数据进行频域滤波，得到所述资产标的对应的同比序列数据在各个所述观测周期下的滤波子序列；
    对每一所述观测周期下的多个资产标的对应的滤波子序列进行合并处理，得到所述资产在各个所述观测周期下的滤波序列。
  5. 如权利要求1至4任一项所述的预测资产价格走势的方法，其中，所述合并处理步骤包括希尔伯特变换和合并权重的迭代计算。
  6. 如权利要求1至4任一项所述的预测资产价格走势的方法，其中，所述根据每一类所述资产对应的回归参数和所述滤波合并序列预测所述资产价格的走势的步骤包括：
    从每一所述观测周期对应的滤波合并序列中确定与所述观测时刻对应的第一数据点，并根据各个所述观测周期下的所述第一数据点以及所述每一类所述资产对应的回归参数拟合得到所述观测时刻下的第一同比数据点；
    从每一所述观测周期对应的滤波合并序列中确定与预测时刻对应的第二数据点，并根据各个所述各个观测周期下的所述第二数据点以及所述每一类所述资产对应的回归参数拟合得到所述预测时刻下的第二同比数据点；
    将所述第一同比数据点与所述第二同比数据点进行比较，根据比较结果预测所述资产的价格走势，其中，当所述第一同比数据点小于所述第二同比数据点时，所述资产收益率上涨，当所述第一同比数据点大于所述第二同比数据点时，所述资产收益率下跌。
  7. 如权利要求6所述的预测资产价格走势的方法，其中，所述从每一所述观测周期对应的滤波合并序列中确定与所述观测时刻对应的第一数据点的步骤之前还包括：
    对每一所述观测周期对应的滤波合并序列进行频域滤波。
  8. 如权利要求1至4任一项所述的预测资产价格走势的方法，其中，所述获取观测时刻，并获取所述观测时刻对应的多类资产的同比序列数据的步 骤包括：
    获取观测时刻、预设滞后期以及预设同比序列长度；
    根据所述观测时刻与所述预设滞后期得到结束时刻，并根据所述结束时刻以及所述预设同比序列长度得到开始时刻；
    获取从所述开始时刻到所述结束时刻的多类资产的同比序列数据。
  9. 一种预测资产价格走势的服务器，其中，所述预测资产价格走势的服务器包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的预测资产价格走势的处理程序，所述预测资产价格走势的处理程序被所述处理器执行时实现如权利要求1至8中任一项所述的预测资产价格走势的方法的步骤。
  10. 一种计算机可读存储介质，其中，所述计算机可读存储介质上存储有预测资产价格走势的处理程序，所述预测资产价格走势的处理程序被处理器执行时实现如权利要求1至8中任一项所述的预测资产价格走势的方法的步骤。
PCT/CN2020/101642 2019-08-06 2020-07-13 预测资产价格走势的方法、服务器及计算机可读存储介质 WO2021022977A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910724548.3A CN110458352A (zh) 2019-08-06 2019-08-06 预测资产价格走势的方法、服务器及计算机可读存储介质
CN201910724548.3 2019-08-06

Publications (1)

Publication Number Publication Date
WO2021022977A1 true WO2021022977A1 (zh) 2021-02-11

Family

ID=68485076

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/101642 WO2021022977A1 (zh) 2019-08-06 2020-07-13 预测资产价格走势的方法、服务器及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN110458352A (zh)
WO (1) WO2021022977A1 (zh)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110458352A (zh) * 2019-08-06 2019-11-15 华泰证券股份有限公司 预测资产价格走势的方法、服务器及计算机可读存储介质
CN111091466A (zh) * 2019-11-25 2020-05-01 华泰证券股份有限公司 资产投资建议信息的生成方法、装置和可读存储介质
CN111126666A (zh) * 2019-11-25 2020-05-08 华泰证券股份有限公司 资产投资建议信息的生成方法、装置和可读存储介质
CN114462679A (zh) * 2022-01-04 2022-05-10 广州杰赛科技股份有限公司 基于深度学习的网络流量预测方法、装置、设备及介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130198110A1 (en) * 2012-01-27 2013-08-01 Robert M. Sellers, Jr. Method for buying and selling stocks and securities
CN106126910A (zh) * 2016-06-22 2016-11-16 上海垒土资产管理有限公司 基于马尔科夫状态转移模型的状态转换预测方法及系统
CN110458352A (zh) * 2019-08-06 2019-11-15 华泰证券股份有限公司 预测资产价格走势的方法、服务器及计算机可读存储介质
CN111091466A (zh) * 2019-11-25 2020-05-01 华泰证券股份有限公司 资产投资建议信息的生成方法、装置和可读存储介质
CN111126666A (zh) * 2019-11-25 2020-05-08 华泰证券股份有限公司 资产投资建议信息的生成方法、装置和可读存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130198110A1 (en) * 2012-01-27 2013-08-01 Robert M. Sellers, Jr. Method for buying and selling stocks and securities
CN106126910A (zh) * 2016-06-22 2016-11-16 上海垒土资产管理有限公司 基于马尔科夫状态转移模型的状态转换预测方法及系统
CN110458352A (zh) * 2019-08-06 2019-11-15 华泰证券股份有限公司 预测资产价格走势的方法、服务器及计算机可读存储介质
CN111091466A (zh) * 2019-11-25 2020-05-01 华泰证券股份有限公司 资产投资建议信息的生成方法、装置和可读存储介质
CN111126666A (zh) * 2019-11-25 2020-05-08 华泰证券股份有限公司 资产投资建议信息的生成方法、装置和可读存储介质

Also Published As

Publication number Publication date
CN110458352A (zh) 2019-11-15

Similar Documents

Publication Publication Date Title
WO2021022977A1 (zh) 预测资产价格走势的方法、服务器及计算机可读存储介质
US11314823B2 (en) Method and apparatus for expanding query
Andreou et al. Detecting multiple breaks in financial market volatility dynamics
Chen et al. Performance of information criteria for selection of Hawkes process models of financial data
WO2021103572A1 (zh) 资产投资建议信息的生成方法、装置和可读存储介质
Nason et al. Should we sample a time series more frequently?: decision support via multirate spectrum estimation
CN116362823A (zh) 用于行为稀疏场景的推荐模型训练方法、推荐方法及装置
Tone et al. DEA SCORE CONFIDENCE INTERVALS WITH PRESENT–FUTURE-BASED RESAMPLING1
CN102402552B (zh) 一种基于分析结果提供推广账户优化建议的方法与设备
WO2023050649A1 (zh) 基于数据补全的esg指数确定方法及相关产品
WO2021103571A1 (zh) 资产投资建议信息的生成方法、装置和可读存储介质
CN110569190A (zh) 交易压力测试方法和装置、以及电子设备和可读存储介质
US11544281B2 (en) Query-oriented approximate query processing based on machine learning techniques
EP2454686A2 (en) Evaluation of website visitor based on value grade
CN110472137B (zh) 识别模型的负样本构建方法、装置和系统
WO2021051976A1 (zh) 金融时间序列的合成方法、装置和存储介质
Xiaohui et al. An algorithm of generating random number by wavelet denoising method and its application
Liu et al. Jump-detection and curve estimation methods for discontinuous regression functions based on the piecewise B-spline function
CN111930604B (zh) 联机交易性能分析方法及装置、电子设备和可读存储介质
Zhang et al. Uncertainty of financial time series based on discrete fractional cumulative residual entropy
Boubaker A generalized arfima model with smooth transition fractional integration parameter
Ben et al. Chirp signal denoising based on convolution neural network
CN110827070B (zh) 基于动态扩展因子的用户成长计算方法及装置
US11521274B2 (en) Cost allocation estimation using direct cost vectors and machine learning
Shareef et al. Separation of overlapping non-stationary signals and compressive sensing-based reconstruction using instantaneous frequency estimation

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20850539

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20850539

Country of ref document: EP

Kind code of ref document: A1