Asset price prediction method and device, computer storage medium and processor

Abstract

The invention discloses an asset price prediction method and device, a computer storage medium and a processor. According to the scheme, historical asset price sequence data of a to-be-predicted subject in a first time period is acquired; inputting the historical asset price sequence data into a target physical information neural network model, and fitting a volatility parameter of the asset price of the subject in a first time period by the target physical information neural network model based on the historical asset price sequence data; the volatility parameters comprise an initial volatility, a mean regression speed of the volatility, a long-term volatility mean value, volatility of the volatility and a correlation coefficient between the asset price and the volatility; and inputting the volatility parameter into the Heston model, and predicting target asset price sequence data of the subject in the second time period by the Heston model based on the volatility parameter. According to the technical scheme, the volatility parameter can be determined in a relatively simple manner, and the accuracy of asset price prediction is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an asset price prediction method, device, computer storage medium and processor. Background Art

[0002] In the field of risk management technology, the analysis and prediction of asset price data are particularly important. The analysis and prediction of asset price data are often affected by a variety of complex factors, especially some technical factors. At present, the existing asset price data prediction models (such as the Heston model) are mainly based on multiple key parameters such as the initial volatility of asset prices, the long-term volatility mean, and the mean reversion speed of volatility to analyze and predict future asset prices. However, obtaining and determining multiple key parameters such as the initial volatility, the long-term volatility mean, and the mean reversion speed of volatility often requires complex calculations and processing of historical asset price data. For example, the above parameters are calculated through more than 20 first-order derivative approximate formulas and hundreds of fourth-order derivative inference formulas, which leads to deviations between the calculated multiple key parameters and the actual parameters, which in turn leads to the low accuracy of the trend of future asset price data predicted by the asset price data prediction model.

[0003] Based on this, in the field of risk management technology, it is necessary to provide a less complex prediction solution to accurately predict asset price data. Summary of the invention

[0004] Based on the above problems, the present application provides an asset price prediction method, device, computer storage medium and processor, the purpose of which is to determine volatility parameters in a relatively concise manner, improve the accuracy of asset price prediction, and thereby enhance the application prospects of the predicted asset prices in the field of risk management technology.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] In a first aspect, the present application provides an asset price prediction method, the method comprising:

[0007] Obtain the historical asset price series data of the subject to be predicted in the first period;

[0008] Inputting historical asset price series data into the target physical information neural network model, the target physical information neural network model fits the volatility parameters of the subject's asset price in the first period based on the historical asset price series data; the volatility parameters include initial volatility, mean reversion speed of volatility, long-term volatility mean, volatility of volatility, and correlation coefficient between asset price and volatility;

[0009] The volatility parameters are input into the Heston model, and the Heston model predicts the target asset price series data of the subject in the second period based on the volatility parameters; the second period is later than the first period.

[0010] In an optional implementation, the historical asset price series data is input into the target physical information neural network model, and the target physical information neural network model fits the volatility parameters of the subject's asset price in the first period based on the historical asset price series data, including:

[0011] The historical asset price series data is input into the partial differential neural network of the target physical information neural network model, and the partial differential neural network performs differential processing on the historical asset price series data to obtain the asset price volatility at each moment in the first period, and determines the statistical characteristics based on the historical asset price series data and the asset price volatility at each moment; the statistical characteristics are used to describe the volatility of the historical asset price series data;

[0012] The statistical characteristics are input into the parameter neural network of the target physical information neural network model, and the parameter neural network generates volatility parameters based on the statistical characteristics.

[0013] In an optional implementation, the target physical information neural network model is obtained through the following process:

[0014] Obtain a sample data set; the sample data set includes M random asset price series data and the random volatility parameters corresponding to each random asset price series data, the random volatility parameters include the initial volatility within a preset parameter range, the mean reversion speed of the volatility, the long-term volatility mean, the volatility of the volatility, and the correlation coefficient between the asset price and the volatility; M is a positive integer;

[0015] Input each random asset price series data into the initial physical information neural network model to obtain the output result of the initial physical information neural network model; the output result includes the predicted volatility parameter corresponding to each random asset price series data;

[0016] Determine a first loss value based on a predicted volatility parameter corresponding to each piece of random asset price series data and a random volatility parameter corresponding to the piece of random asset price series data;

[0017] The parameters of the initial physical information neural network model are iteratively updated based on the first loss value until the parameters meet the training cutoff condition, and the model training is stopped to obtain the target physical information neural network model.

[0018] In an optional implementation, each random asset price sequence data is input into the initial physical information neural network model to obtain the output result of the initial physical information neural network model, including:

[0019] Input each random asset price series data into the partial differential neural network respectively, and use the partial differential neural network to perform differential processing on each random asset price series data to obtain the volatility corresponding to each random asset price series data, and determine the target statistical feature corresponding to each random asset price series data based on each random asset price series data and the volatility of the random asset price series data;

[0020] The target statistical features corresponding to each random asset price series data are input into the parameter neural network, and the parameter neural network generates the predicted volatility parameters corresponding to each random asset price series data.

[0021] In an optional implementation, after determining the target statistical feature corresponding to each random asset price series data, the asset price prediction method further includes:

[0022] Obtain a target training sample set; the target training sample set includes the volatility corresponding to N random asset price series data and the actual volatility corresponding to each random asset price series data; N is a positive integer, and N is less than M;

[0023] Determine a second loss value based on the volatility corresponding to the N random asset price series data and the actual volatility corresponding to each random asset price series data;

[0024] The parameters of the partial differential neural network are updated based on the second loss value.

[0025] In an optional implementation, before obtaining the sample data set, the asset price prediction method further includes:

[0026] Obtain preset parameter ranges and parameter distribution types; preset parameter ranges include initial volatility generation range, volatility mean reversion speed generation range, long-term volatility mean generation range, volatility volatility generation range, and correlation coefficient generation range; parameter distribution types include normal distribution;

[0027] Generate multiple sets of random volatility parameters based on preset parameter ranges and parameter distribution types;

[0028] Input each set of random volatility parameters into the Heston model respectively, and obtain the random asset price series data corresponding to each set of random volatility parameters output by the Heston model;

[0029] A sample data set is determined based on multiple sets of stochastic volatility parameters and random asset price series data corresponding to each set of stochastic volatility parameters.

[0030] In a second aspect of the present application, an asset price prediction device is provided, the device comprising:

[0031] An acquisition module, used to acquire historical asset price series data of the subject to be predicted in the first period;

[0032] A determination module is used to input the historical asset price sequence data into the target physical information neural network model, and the target physical information neural network model fits the volatility parameters of the subject's asset price in the first period based on the historical asset price sequence data; the volatility parameters include initial volatility, mean reversion speed of volatility, long-term volatility mean, volatility of volatility, and correlation coefficient between asset price and volatility;

[0033] The prediction module is used to input the volatility parameter into the Heston model, and the Heston model predicts the target asset price series data of the subject in the second period based on the volatility parameter; the second period is later than the first period.

[0034] Optionally, the determination module includes:

[0035] A first determination unit is used to input the historical asset price series data into the partial differential neural network of the target physical information neural network model, and the partial differential neural network performs differential processing on the historical asset price series data to obtain the asset price volatility at each moment in the first period, and determine statistical features based on the historical asset price series data and the asset price volatility at each moment; the statistical features are used to describe the volatility of the historical asset price series data;

[0036] The generating unit is used to input the statistical features into the parameter neural network of the target physical information neural network model, and the parameter neural network generates the volatility parameters based on the statistical features.

[0037] In a third aspect of the present application, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the above-mentioned asset price prediction method is implemented.

[0038] In a fourth aspect of the present application, a processor is provided for running a computer program, and the computer program executes the above-mentioned asset price prediction method when it is running.

[0039] Compared with the prior art, this application has the following beneficial effects:

[0040] In the technical solution of the present application, firstly, the historical asset price sequence data of the subject to be predicted in the first time period is obtained; then the historical asset price sequence data is input into the pre-trained target physical information neural network model, and the volatility parameters of the subject's asset price in the first time period can be fitted based on the asset price sequence data through the target physical information neural network model, thereby achieving accurate initial asset price volatility, mean reversion speed of volatility, long-term volatility mean, volatility, and correlation coefficient between asset price and volatility (i.e., volatility parameters) in a relatively concise manner, without the need to calculate the above parameters through dozens or even hundreds of derivative reasoning formulas, thereby avoiding the problem of low accuracy of the calculated volatility parameters due to the complex calculation process of the volatility parameters, and improving the accuracy of the determined volatility parameters; thereby, the accurate volatility parameters are input into the Heston model, and the Heston model can accurately predict the target asset price sequence data of the subject in the future time period (i.e., the second time period) based on the volatility parameters, thereby improving the accuracy of asset price prediction, and further improving the application prospects of the predicted asset prices in the field of risk control management technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0042] Figure 1 A flowchart of an asset price prediction method provided in an embodiment of the present application;

[0043] Figure 2 A schematic diagram of a target physical information neural network model provided in an embodiment of the present application;

[0044] Figure 3 A flowchart of a process for generating a volatility parameter provided in an embodiment of the present application;

[0045] Figure 4 A flowchart of a training process of a target physical information neural network model provided in an embodiment of the present application;

[0046] Figure 5 A flowchart of another training process of a target physical information neural network model provided in an embodiment of the present application;

[0047] Figure 6 A schematic diagram of the structure of an asset price prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] As described above, currently, existing asset price data prediction models (such as the Heston model) are mainly based on multiple key parameters such as the initial volatility of asset prices, the long-term volatility mean, and the mean reversion speed of volatility to analyze and predict future asset prices. However, obtaining and determining multiple key parameters such as the initial volatility, the long-term volatility mean, and the mean reversion speed of volatility often requires complex calculations and processing of historical asset price data, such as calculating the above parameters through more than 20 first-order derivative approximate formulas and hundreds of fourth-order derivative inference formulas, which leads to deviations between the calculated multiple key parameters and the actual parameters, and thus leads to a low accuracy rate in the trend of future asset price data predicted by the asset price data prediction model.

[0049] After research, the inventors proposed an asset price prediction method, which obtains the historical asset price sequence data of the subject to be predicted in the first time period; then inputs the historical asset price sequence data into a pre-trained target physical information neural network model, and can fit the volatility parameters of the subject's asset price in the first time period based on the asset price sequence data through the target physical information neural network model, thereby achieving accurate initial asset price volatility, mean reversion speed of volatility, long-term volatility mean, volatility, and correlation coefficient between asset price and volatility (i.e., volatility parameter) in a relatively concise manner, without the need to calculate the above parameters through dozens or even hundreds of derivative reasoning formulas, thereby avoiding the problem of low accuracy of the calculated volatility parameter due to the complex calculation process of the volatility parameter, and improving the accuracy of the determined volatility parameter; thereby, the accurate volatility parameter is input into the Heston model, and the Heston model can accurately predict the target asset price sequence data of the subject in the future time period (i.e., the second time period) based on the volatility parameter, thereby improving the accuracy of asset price prediction, and further improving the application prospects of the predicted asset price in the field of risk control management technology.

[0050] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0051] Keyword definitions:

[0052] Physics-informed neural network model: Physics-informed neural networks, referred to as PINNs, are used to migrate from partial differential equations (PDE, full name Partial Differential Equations) to stochastic partial differential equations (SDE, full name Stochastic Differential Equations), and can handle inverse problems such as parameter solving. The constraints of partial differential equations are solved through the PINNs network to fit the curves expressed by the partial differential equations. In parameter solving, the PINNs network can achieve parameter fitting based on the error between the target data and the fitting curve through error back propagation.

[0053] Heston model: is a stochastic volatility model used to describe the dynamics of asset prices and volatility. The Heston model captures the volatility characteristics of asset prices by introducing stochastic volatility, thereby providing more accurate asset price predictions.

[0054] Method Embodiment

[0055] The embodiment of the present application provides an embodiment of an asset price prediction method. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0056] See also Figure 1 , which is a flow chart of an asset price prediction method provided by an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0057] Step S101, obtaining historical asset price series data of the subject to be predicted in the first time period.

[0058] In an optional embodiment, an asset price prediction system can be used as the execution subject of the asset price prediction method of the embodiment of the present application. For the convenience of description, the asset price prediction system is referred to as the system below.

[0059] In this embodiment, the subject to be predicted may be an organization such as an enterprise, an institution, etc. The historical asset price series data is the asset price data recorded at fixed time intervals by the subject to be predicted in the first period.

[0060] Step S102, inputting the historical asset price series data into the target physical information neural network model, and the target physical information neural network model fits the volatility parameters of the subject's asset price in the first period based on the historical asset price series data.

[0061] In step S102, the volatility parameters include initial volatility, mean reversion speed of volatility, long-term volatility mean, volatility of volatility, and correlation coefficient between asset price and volatility.

[0062] In this embodiment, the target physical information neural network model is a pre-trained neural network model, such as Figure 2 As shown in the figure, the neural network model consists of a partial differential neural network and a parameter neural network. The system can fit the volatility parameters required by the Heston model to predict asset price series data based on historical asset price series data through the target physical information neural network model: initial volatility v0, mean reversion speed k of volatility, long-term volatility mean θ, volatility ξ of volatility, and correlation coefficient ρ between asset price and volatility. Among them, the partial differential neural network has 4 layers of neural networks, each with 400 neurons, and its output layer activation function is Liner, and the activation functions of other layers except the output layer are LeakyReLu. The parameter neural network has 4 layers of neural networks, each with 200 neurons, and its output layer activation function is Sigmoid, and the activation functions of other layers except the output layer are LeakyReLu.

[0063] In order to improve the accuracy of the determined volatility parameters and avoid the problem of low accuracy of the calculated volatility parameters due to the complex calculation process of the volatility parameters, in this embodiment, the system can input the historical asset price series data into the target physical information neural network model, and the partial differential neural network and parameter neural network in the target physical information neural network model can fit the volatility parameters of the subject's asset price in the first period based on the historical asset price series data, so as to obtain accurate initial asset price volatility, mean reversion speed of volatility, long-term volatility mean, volatility, and correlation coefficient between asset price and volatility in a relatively concise manner, without the need to calculate the above parameters through dozens or even hundreds of derivative reasoning formulas. Specifically, see Figure 3 , which is a flow chart of a process for generating a volatility parameter provided in an embodiment of the present application, and the process includes the following steps:

[0064] Step S1021, input the historical asset price series data into the partial differential neural network of the target physical information neural network model, and use the partial differential neural network to perform differential processing on the historical asset price series data to obtain the asset price volatility at each moment in the first time period, and determine the statistical characteristics based on the historical asset price series data and the asset price volatility at each moment.

[0065] In step S1021, statistical features are used to describe the volatility of historical asset price series data. In this embodiment, the statistical features may include the mean, variance, standard deviation, distribution characteristics, etc. of the historical asset price.

[0066] In this embodiment, if Figure 2 As shown, the system can input the historical asset price series data into the partial differential neural network of the target physical information neural network model, and obtain the asset price volatility x at each moment in the first time period by differential processing of the historical asset price series data y at time t through the differential neural network; then perform feature extraction on the asset price volatility x and the historical asset price series data y to obtain statistical features.

[0067] Step S1022, inputting the statistical features into the parameter neural network of the target physical information neural network model, and the parameter neural network generates volatility parameters based on the statistical features.

[0068] In this embodiment, if Figure 2 As shown, the system inputs the extracted statistical features into the parametric neural network, and the parametric neural network generates volatility parameters based on the statistical features: initial volatility v0, mean reversion speed k of volatility, long-term volatility mean θ, volatility ξ, and the correlation coefficient ρ between asset price and volatility.

[0069] Step S103, inputting the volatility parameter into the Heston model, and using the Heston model to predict the target asset price series data of the subject in the second period based on the volatility parameter.

[0070] In step S103, the second time period is later than the first time period. For example, the first time period is from 00:00 on January 1, 2023 to 00:00 on December 31, 2023, and the second time period is from 00:00 on January 1, 2024 to 00:00 on December 31, 2024.

[0071] In this embodiment, the Heston model is specifically shown in formula (1).

[0072]

[0073] Where μ is the expected return rate of the asset; S t is the asset price at time t; v t is the volatility at time t; k is the mean reversion rate of volatility, θ is the long-term volatility mean, ξ is the volatility of volatility, and are two standard Brownian motions with a correlation coefficient ρ between them.

[0074] In this embodiment, the target asset price series data includes data of multiple asset price series paths. Through the target asset price series data, trend graphs of multiple asset prices corresponding to the subject can be generated.

[0075] Based on the scheme introduced in the above steps S101 to S103, it can be known that in the present application, by obtaining the historical asset price series data of the subject to be predicted in the first time period; and then inputting the historical asset price series data into the pre-trained target physical information neural network model, the volatility parameters of the subject's asset price in the first time period can be fitted based on the asset price series data through the target physical information neural network model, thereby achieving accurate initial asset price volatility, mean reversion speed of volatility, long-term volatility mean, volatility of volatility, and the relationship between asset price and volatility in a relatively concise manner. There is no need to calculate the above parameters through dozens or even hundreds of derivative reasoning formulas, thus avoiding the problem of low accuracy of the calculated volatility parameters due to the complicated calculation process of the volatility parameters, and improving the accuracy of the determined volatility parameters; thereby, the accurate volatility parameters are input into the Heston model, and the Heston model can accurately predict the target asset price series data of the subject in the future period (i.e., the second period) based on the volatility parameters, thereby improving the accuracy of asset price prediction, and further enhancing the application prospects of the predicted asset prices in the field of risk control management technology.

[0076] In order to improve the fitting accuracy of the target physical information neural network model, in this embodiment, the system can perform model training based on M random asset price series data and the random volatility parameters corresponding to each random asset price series data. The model training process includes the following steps:

[0077] Step S1, obtaining a sample data set; the sample data set includes M random asset price series data and a random volatility parameter corresponding to each random asset price series data.

[0078] In step S1, the stochastic volatility parameters include the initial volatility within the preset parameter range, the mean reversion speed of the volatility, the long-term volatility mean, the volatility of the volatility, and the correlation coefficient between the asset price and the volatility; M is a positive integer.

[0079] In order to enhance the training effect of the model, in this embodiment, Figure 4 As shown, the system can perform dimensionality reduction processing on the data in the sample data set and normalize the data after dimensionality reduction processing. Among them, the dimensionality reduction processing can be Figure 4 The PCA principal component analysis method is shown.

[0080] In this embodiment, if Figure 4 As shown, the system can obtain a sample data set by setting a parameter distribution (i.e., a parameter distribution type) and a range (i.e., a preset parameter range). The process of obtaining the sample data set includes the following steps:

[0081] Step S11, obtaining a preset parameter range and a parameter distribution type; the preset parameter range includes an initial volatility generation range, a volatility mean reversion speed generation range, a long-term volatility mean generation range, a volatility volatility generation range, and a correlation coefficient generation range; the parameter distribution type includes a normal distribution.

[0082] In this embodiment, the parameter distribution type can be a t distribution (Student's t-Distribution) or other distribution types used to improve the accuracy and robustness of the model in addition to the normal distribution. The preset parameter range and parameter distribution type can be selected according to actual needs, so no specific limitation is made.

[0083] Step S12, generating multiple groups of random volatility parameters based on preset parameter ranges and parameter distribution types.

[0084] In this embodiment, if Figure 4 As shown, the system can generate multiple sets of random volatility parameters (v0, k, θ, ξ, ρ) based on preset parameter ranges and parameter distribution types.

[0085] Step S13, inputting each set of random volatility parameters into the Heston model respectively, and obtaining random asset price series data corresponding to each set of random volatility parameters output by the Heston model.

[0086] Taking 5000 sets of random volatility parameters as an example, the 5000 sets of random volatility parameters are input into the Heston model to obtain 5000 random asset price series data whose output length is the same as the predicted asset price series data.

[0087] Step S14, determining a sample data set based on multiple groups of random volatility parameters and random asset price series data corresponding to each group of random volatility parameters.

[0088] In this embodiment, if Figure 4 As shown, the system can combine a single parameter with the corresponding random asset price series data to obtain sample data sets: v0 data set, k data set, θ data set, ξ data set and ρ data set.

[0089] It should be noted that the above steps S11 to S14 are not shown in the figure.

[0090] Step S2, input each random asset price series data into the initial physical information neural network model respectively to obtain the output result of the initial physical information neural network model; the output result includes the predicted volatility parameter corresponding to each random asset price series data.

[0091] In this embodiment, if Figure 5 As shown, the system can input each random asset price series data into the initial physical information neural network model respectively, and the initial physical information neural network model will perform Heston model constraints and asset price constraints on the random asset price series data to obtain the predicted volatility parameters corresponding to each random asset price series data.

[0092] Specifically, the system inputs each random asset price series data into the partial differential neural network respectively, and the partial differential neural network performs differentiation processing on each random asset price series data to obtain the volatility corresponding to each random asset price series data, and determines the target statistical features corresponding to each random asset price series data based on each random asset price series data and the volatility of the random asset price series data; then the target statistical features corresponding to each random asset price series data are input into the parameter neural network, and the parameter neural network generates the predicted volatility parameters corresponding to each random asset price series data.

[0093] Optionally, in this embodiment, the system may use stochastic gradient descent method (Adam optimizer) for training, and the learning rate of the partial differential neural network is For parameterized neural networks, the learning rate is The purpose is to continuously optimize the parameters θ of the partial differential neural network g and the parameters θ of the neural network d . In each training, the number of times the discriminator is trained is: k = 10; batch size: m = 200; the dimension of the latent space is 200; the partial differential neural network and the parameter neural network are initialized using a multivariate normal distribution with a mean of 0 and a standard deviation of 0.02.

[0094] Specifically, after determining the target statistical features corresponding to each random asset price series data, the system can obtain a target training sample set, wherein the target training sample set includes the volatilities corresponding to N random asset price series data and the actual volatility corresponding to each random asset price series data; N is a positive integer, and N is less than M; then the system can determine a second loss value based on the volatilities corresponding to the N random asset price series data and the actual volatility corresponding to each random asset price series data, and update the parameters of the partial differential neural network based on the second loss value.

[0095] Optionally, when updating the parameters of the parameter neural network, the system may update the parameters of the parameter neural network based on the random volatility parameters and predicted volatility parameters corresponding to the N random asset price series data.

[0096] Step S3, determining a first loss value based on the predicted volatility parameter corresponding to each piece of random asset price series data and the random volatility parameter corresponding to the piece of random asset price series data.

[0097] In this embodiment, the system can be based on Figure 5 The target loss function shown, the predicted volatility parameter corresponding to each random asset price series data, and the random volatility parameter corresponding to the random asset price series data determine the first loss value.

[0098] In this embodiment, the system can Figure 2 The SED equation loss calculation, initialization loss calculation, data loss calculation and second loss value are shown to ensure that the total loss (ie, the first loss value) is minimized.

[0099] Step S4, iteratively update the parameters of the initial physical information neural network model based on the first loss value until the parameters meet the training cutoff condition, stop model training, and obtain the target physical information neural network model.

[0100] It should be noted that the above steps S1 to S4 are not shown in the figure.

[0101] Optionally, after obtaining the target physical information neural network model, the system can input the data in the test set into the target physical information neural network model to obtain the output results of the target physical information neural network model, and compare the differences between the predicted volatility parameters and the random volatility parameters in the output results to further adjust the target physical information neural network model and improve the accuracy of the fitted volatility parameters of the target physical information neural network model.

[0102] Alternatively, if Figure 5As shown, the system can also compare the target physical information neural network model with an existing model (such as a Monte Carlo model) to evaluate the model effect. By inputting the data in the test set into each model, the volatility parameters are fitted respectively, and the future asset price series data is generated based on the volatility parameters, and then the mean square absolute error (MSAE) and mean square relative error (MSRE) between the future asset price series data and the actual asset price series data are calculated, and the yield path is calculated. The mean, variance, skewness and kurtosis of the yield path are calculated based on the yield path, and the MSAE, MSRE and the mean, variance, skewness and kurtosis of the yield path of the target physical information neural network model are compared with the mean, variance, skewness and kurtosis of the MSAE, MSRE and the yield path corresponding to the existing model, and the fitting effect of the model is evaluated by comparing the results.

[0103] The asset price prediction method provided in the embodiment of the present application can fit the volatility parameters of the subject's asset price in the first time period based on the asset price series data through the target physical information neural network model, thereby achieving accurate initial asset price volatility, volatility mean reversion speed, long-term volatility mean, volatility, and correlation coefficient between asset price and volatility (i.e., volatility parameters) in a relatively concise manner, without the need to calculate the above parameters through dozens or even hundreds of derivative reasoning formulas, thereby avoiding the problem of low accuracy of the calculated volatility parameters due to the complex calculation process of the volatility parameters, and improving the accuracy of the determined volatility parameters; thereby, the accurate volatility parameters are input into the Heston model, and the Heston model can accurately predict the subject's target asset price series data in the future period (i.e., the second period) based on the volatility parameters, thereby improving the accuracy of asset price prediction, and further improving the application prospects of the predicted asset prices in the field of risk control management technology.

[0104] Device Embodiment

[0105] The embodiment of the present application provides an asset price prediction device, wherein: Figure 6 A schematic diagram of the structure of an asset price prediction device provided in an embodiment of the present application is shown in FIG. Figure 6 As shown, the device includes: an acquisition module 11, a determination module 12 and a prediction module 13. Figure 6 You can see the connection relationship between several modules.

[0106] The acquisition module 11 is used to acquire the historical asset price series data of the subject to be predicted in the first period;

[0107] A determination module 12 is used to input the historical asset price sequence data into the target physical information neural network model, and the target physical information neural network model fits the volatility parameters of the subject's asset price in the first period based on the historical asset price sequence data; the volatility parameters include initial volatility, mean reversion speed of volatility, long-term volatility mean, volatility of volatility, and correlation coefficient between asset price and volatility;

[0108] The prediction module 13 is used to input the volatility parameter into the Heston model, and the Heston model predicts the target asset price series data of the subject in the second period based on the volatility parameter; the second period is later than the first period.

[0109] Optionally, the determination module includes: a first determination unit and a generation unit.

[0110] A first determination unit is used to input the historical asset price series data into the partial differential neural network of the target physical information neural network model, and the partial differential neural network performs differential processing on the historical asset price series data to obtain the asset price volatility at each moment in the first period, and determine statistical features based on the historical asset price series data and the asset price volatility at each moment; the statistical features are used to describe the volatility of the historical asset price series data;

[0111] The generating unit is used to input the statistical features into the parameter neural network of the target physical information neural network model, and the parameter neural network generates the volatility parameters based on the statistical features.

[0112] Optionally, the asset price prediction device further includes: a sample acquisition module, an output module, a first determination module and an iterative update module.

[0113] The sample acquisition module is used to acquire a sample data set; the sample data set includes M random asset price series data and the random volatility parameters corresponding to each random asset price series data, and the random volatility parameters include the initial volatility within the preset parameter range, the mean reversion speed of the volatility, the long-term volatility mean, the volatility of the volatility, and the correlation coefficient between the asset price and the volatility; M is a positive integer;

[0114] An output module, used to input each random asset price series data into the initial physical information neural network model, and obtain the output result of the initial physical information neural network model; the output result includes the predicted volatility parameter corresponding to each random asset price series data;

[0115] A first determination module, configured to determine a first loss value based on a predicted volatility parameter corresponding to each piece of random asset price series data and a random volatility parameter corresponding to the piece of random asset price series data;

[0116] The iterative update module is used to iteratively update the parameters of the initial physical information neural network model based on the first loss value until the parameters meet the training cutoff condition, then stop the model training and obtain the target physical information neural network model.

[0117] Optionally, the output module includes: a processing unit and a parameter generating unit.

[0118] The processing unit is used to input each random asset price series data into the partial differential neural network respectively, and the partial differential neural network performs differential processing on each random asset price series data to obtain the volatility corresponding to each random asset price series data, and determine the target statistical feature corresponding to each random asset price series data based on each random asset price series data and the volatility of the random asset price series data;

[0119] The parameter generation unit is used to input the target statistical features corresponding to each random asset price series data into the parameter neural network, and the parameter neural network generates the predicted volatility parameters corresponding to each random asset price series data.

[0120] Optionally, the asset price prediction device further includes: a sample acquisition unit, a second determination unit and an update unit.

[0121] The sample acquisition unit is used to acquire a target training sample set after determining the target statistical feature corresponding to each random asset price series data; the target training sample set includes the volatility corresponding to N random asset price series data and the actual volatility corresponding to each random asset price series data; N is a positive integer, and N is less than M;

[0122] A second determining unit, configured to determine a second loss value based on the volatility corresponding to the N pieces of random asset price series data and the actual volatility corresponding to each piece of random asset price series data;

[0123] An updating unit is used to update the parameters of the partial differential neural network based on the second loss value.

[0124] Optionally, the asset price prediction device further includes: a parameter acquisition unit, a parameter generation unit, an asset price sequence generation unit and a third determination unit.

[0125] The parameter acquisition unit is used to acquire a preset parameter range and a parameter distribution type before acquiring a sample data set; the preset parameter range includes an initial volatility generation range, a volatility mean reversion speed generation range, a long-term volatility mean generation range, a volatility volatility generation range, and a correlation coefficient generation range; the parameter distribution type includes a normal distribution;

[0126] A parameter generation unit, used for generating multiple sets of random volatility parameters based on a preset parameter range and parameter distribution type;

[0127] The asset price sequence generating unit is used to input each set of random volatility parameters into the Heston model respectively, and obtain the random asset price sequence data corresponding to each set of random volatility parameters output by the Heston model;

[0128] The third determining unit is used to determine a sample data set based on multiple groups of random volatility parameters and random asset price series data corresponding to each group of random volatility parameters.

[0129] Storage Medium Embodiments

[0130] The embodiment of the present application provides a computer-readable storage medium, on which a program is stored, wherein when the program is executed by a processor, some or all of the steps in the asset price prediction method described in the aforementioned method embodiment of the present application are implemented. The storage medium can be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, and other media that can store program codes.

[0131] Processor Embodiments

[0132] An embodiment of the present application provides a processor for running a program, wherein part or all of the steps in the asset price prediction method introduced in the aforementioned method embodiment are executed when the program is running.

[0133] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely schematic, in which the unit described as a separate component may or may not be physically separated, and the component prompted as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0134] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An asset price prediction method, characterized in that: include: Obtain the historical asset price series data of the subject to be predicted in the first period; Inputting the historical asset price sequence data into the target physical information neural network model, the target physical information neural network model fits the volatility parameters of the subject's asset price in the first period based on the historical asset price sequence data; the volatility parameters include initial volatility, mean reversion speed of volatility, long-term volatility mean, volatility of the volatility, and correlation coefficient between the asset price and the volatility; The volatility parameter is input into the Heston model, and the Heston model predicts the target asset price series data of the subject in a second time period based on the volatility parameter; the second time period is later than the first time period.

2. The method according to claim 1, characterized in that The step of inputting the historical asset price sequence data into the target physical information neural network model, and fitting the volatility parameter of the subject's asset price in the first period based on the historical asset price sequence data by the target physical information neural network model, comprises: Inputting the historical asset price series data into the partial differential neural network of the target physical information neural network model, performing differential processing on the historical asset price series data by the partial differential neural network to obtain the asset price volatility at each moment in the first period, and determining statistical features based on the historical asset price series data and the asset price volatility at each moment; the statistical features are used to describe the volatility of the historical asset price series data; The statistical features are input into the parameter neural network of the target physical information neural network model, and the parameter neural network generates the volatility parameters based on the statistical features.

3. The method according to claim 2, characterized in that The target physical information neural network model is obtained through the following process: Acquire a sample data set; the sample data set includes M random asset price series data and a random volatility parameter corresponding to each random asset price series data, wherein the random volatility parameter includes an initial volatility within a preset parameter range, a mean reversion speed of the volatility, a long-term volatility mean, a volatility of the volatility, and a correlation coefficient between the asset price and the volatility; M is a positive integer; Input each piece of the random asset price series data into the initial physical information neural network model to obtain an output result of the initial physical information neural network model; the output result includes a predicted volatility parameter corresponding to each piece of the random asset price series data; Determine a first loss value based on the predicted volatility parameter corresponding to each piece of the random asset price series data and the random volatility parameter corresponding to the piece of the random asset price series data; The parameters of the initial physical information neural network model are iteratively updated based on the first loss value, and the model training is stopped when the parameters meet the training cutoff condition to obtain the target physical information neural network model.

4. The method according to claim 3, characterized in that The step of inputting each piece of the random asset price sequence data into the initial physical information neural network model to obtain the output result of the initial physical information neural network model includes: Input each piece of the random asset price series data into the partial differential neural network respectively, and use the partial differential neural network to perform differential processing on each piece of the random asset price series data to obtain the volatility corresponding to each piece of the random asset price series data, and determine the target statistical feature corresponding to each piece of the random asset price series data based on each piece of the random asset price series data and the volatility of the random asset price series data; The target statistical feature corresponding to each piece of the random asset price series data is input into the parameter neural network, and the parameter neural network generates the predicted volatility parameter corresponding to each piece of the random asset price series data.

5. The method according to claim 4, characterized in that After determining the target statistical feature corresponding to each piece of the random asset price series data, the method further includes: Obtain a target training sample set; the target training sample set includes the volatility corresponding to N pieces of the random asset price series data, and the actual volatility corresponding to each piece of the random asset price series data; N is a positive integer, and N is less than M; Determine a second loss value based on the volatility corresponding to the N pieces of random asset price series data and the actual volatility corresponding to each piece of random asset price series data; Update parameters of the partial differential neural network based on the second loss value.

6. The method according to claim 3, characterized in that: Before acquiring the sample data set, the method further includes: Obtaining the preset parameter range and parameter distribution type; the preset parameter range includes an initial volatility generation range, a volatility mean reversion speed generation range, a long-term volatility mean generation range, a volatility volatility generation range, and a correlation coefficient generation range; the parameter distribution type includes a normal distribution; Generate multiple groups of the stochastic volatility parameters based on the preset parameter range and parameter distribution type; Input each group of the stochastic volatility parameters into the Heston model respectively, and obtain the random asset price series data corresponding to each group of the stochastic volatility parameters output by the Heston model; The sample data set is determined based on a plurality of groups of the stochastic volatility parameters and the random asset price series data corresponding to each group of the stochastic volatility parameters.

7. An asset price prediction device, characterized in that: include: An acquisition module, used to acquire historical asset price series data of the subject to be predicted in the first period; a determination module, configured to input the historical asset price sequence data into a target physical information neural network model, and the target physical information neural network model fits the volatility parameters of the subject's asset price in the first period based on the historical asset price sequence data; the volatility parameters include initial volatility, mean reversion speed of volatility, long-term volatility mean, volatility of the volatility, and correlation coefficient between the asset price and the volatility; A prediction module is used to input the volatility parameter into a Heston model, and the Heston model predicts the target asset price series data of the subject in a second period based on the volatility parameter; the second period is later than the first period.

8. The device according to claim 7, characterized in that The determination module comprises: a first determining unit, configured to input the historical asset price series data into a partial differential neural network of the target physical information neural network model, and to perform differential processing on the historical asset price series data by the partial differential neural network to obtain asset price volatility at each moment in the first period, and to determine statistical features based on the historical asset price series data and the asset price volatility at each moment; the statistical features are used to describe the volatility of the historical asset price series data; A generating unit is used to input the statistical features into a parameter neural network of the target physical information neural network model, and the parameter neural network generates the volatility parameter based on the statistical features.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the asset price prediction method according to any one of claims 1 to 6 is implemented.

10. A processor, characterized in that: Used to run a computer program, which, when running, executes the asset price prediction method according to any one of claims 1 to 6.

Images