Financial transaction strategy determination method and device and related equipment

Abstract

The invention provides a financial transaction strategy determination method and device and related equipment, and the method comprises the steps: obtaining a plurality of pieces of first data related to a financial transaction, and the plurality of pieces of first data comprise unstructured text data, image data and market data; generating a transaction popularity score according to the image data, wherein the transaction popularity score is used for representing the public opinion emotion tendency of the financial transaction; constructing a composite investment factor according to the unstructured text data, the transaction popularity score and the market data, wherein the composite investment factor is used for correcting a financial transaction strategy; and correcting a preset financial transaction strategy according to the composite investment factor. Therefore, the plurality of first data are acquired, and the financial transaction strategy can be corrected according to the plurality of first data, so that the probability that the accuracy of the financial transaction strategy is relatively low due to partial data distortion is reduced, and the accuracy of the finally generated financial transaction strategy is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus and related equipment for determining financial trading strategies. Background Technology

[0002] With the rapid development of artificial intelligence technology, mainstream quantitative strategies in the field of quantitative financial trading mainly combine historical market data with algorithms such as linear regression and random forests to build predictive models, and then output quantitative strategies based on these predictive models. However, if historical market data is distorted, the accuracy of the financial trading strategies determined by the constructed predictive models can easily be low. Therefore, the accuracy of currently determined financial trading strategies is relatively low. Summary of the Invention

[0003] This application provides a method, apparatus, and related equipment for determining financial trading strategies to address the problem of low accuracy in currently determined financial trading strategies.

[0004] To solve the above problems, this application is implemented as follows:

[0005] In a first aspect, embodiments of this application provide a method for determining a financial trading strategy, including:

[0006] Acquire multiple primary data related to financial transactions, including unstructured text data, image data, and market data;

[0007] A transaction popularity score is generated based on the image data, and the transaction popularity score is used to represent the public sentiment tendency of the financial transaction;

[0008] A composite investment factor is constructed based on the unstructured text data, the transaction popularity score, and the market data. The composite investment factor is used to modify financial trading strategies.

[0009] The pre-set financial trading strategy is modified based on the composite investment factors.

[0010] Secondly, embodiments of this application provide a financial transaction strategy determination apparatus, comprising:

[0011] The acquisition module is used to acquire multiple primary data related to financial transactions, including unstructured text data, image data, and market data.

[0012] The generation module is used to generate a transaction popularity score based on the image data, and the transaction popularity score is used to represent the public sentiment tendency of the financial transaction;

[0013] A construction module is used to construct a composite investment factor based on the unstructured text data, the transaction popularity score, and the market data. The composite investment factor is used to modify financial trading strategies.

[0014] The correction module is used to correct the pre-set financial trading strategy based on the composite investment factors.

[0015] Thirdly, embodiments of this application also provide an electronic device, including: a memory, a processor, and a program stored in the memory and executable on the processor; the processor is configured to read the program in the memory to implement the steps in the method described in the first aspect above.

[0016] Fourthly, embodiments of this application also provide a readable storage medium for storing a program, which, when executed by a processor, implements the steps of the method described in the first aspect above.

[0017] Fifthly, embodiments of this application also provide a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the method described in the first aspect above.

[0018] In this embodiment, multiple sets of first data related to financial transactions can be acquired, including unstructured text data, image data, and market data. A transaction popularity score is then generated based on the image data, and a composite investment factor is constructed based on the unstructured text data, the transaction popularity score, and the market data. A pre-set financial trading strategy is then revised based on the composite investment factor. Thus, by acquiring multiple sets of first data and revising the financial trading strategy based on them, the probability of low accuracy in the financial trading strategy due to partial data distortion is reduced, and the accuracy of the final generated financial trading strategy is improved. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is one of the flowcharts illustrating the financial transaction strategy determination method provided in the embodiments of this application;

[0021] Figure 2 This is an example diagram illustrating an application scenario of the financial transaction strategy determination method provided in the embodiments of this application;

[0022] Figure 3This is a second flowchart illustrating the financial transaction strategy determination method provided in the embodiments of this application;

[0023] Figure 4 This is the third flowchart illustrating the financial transaction strategy determination method provided in the embodiments of this application;

[0024] Figure 5 This is a schematic diagram of the structure of the financial transaction strategy determination device provided in the embodiments of this application;

[0025] Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] The terms "first," "second," etc., used in the embodiments of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices. Additionally, the use of "and / or" in this application indicates at least one of the connected objects, such as A and / or B and / or C, representing seven possibilities: including A alone, B alone, C alone, both A and B present, both B and C present, both A and C present, and A, B, and C present.

[0028] Please see Figure 1 , Figure 1 This is a flowchart illustrating the financial transaction strategy determination method provided in the embodiments of this application. Figure 1 The method for determining the financial trading strategy shown can be executed by electronic devices.

[0029] like Figure 1 As shown, the method for determining a financial trading strategy may include the following steps:

[0030] Step 101: Obtain multiple primary data related to financial transactions, including unstructured text data, image data, and market data.

[0031] Among these multiple first data points, the sources corresponding to different first data points can be different. Therefore, the aforementioned multiple first data points can be referred to as multimodal data, and the sources corresponding to the aforementioned multiple first data points can be referred to as multimodal data sources. Optionally, the storage databases corresponding to different first data points can be different. For example, unstructured text data can come from documents such as corporate financial reports or press releases, while image data can come from screenshots of public opinion on social media, and market data can include information such as stock prices and trading volumes.

[0032] Alternatively, the image data may also include knowledge graph data, which can be associated with the upstream and downstream of the financial transaction industry chain and can be stored in a graph database. The specific type of graph database is not limited here. For example, the graph database may include the Neo4j graph database.

[0033] It should be noted that the specific methods for obtaining the aforementioned primary data are not limited here. Optionally, unstructured text can be obtained from documents such as corporate financial reports or press releases through technologies such as Optical Character Recognition (OCR) and Natural Language Processing (NLP). Image data can be extracted through Residual Network-50 (ResNet-50). Market data can be obtained from the network through Application Programming Interface (API). It should be noted that the aforementioned market data can also be referred to as real-time market data.

[0034] It should be noted that automatically parsing unstructured text using technologies such as OCR and NLP can reduce processing time to less than or equal to 5 minutes, greatly improving the parsing efficiency of unstructured text.

[0035] Step 102: Generate a transaction popularity score based on the image data. The transaction popularity score is used to represent the public sentiment tendency of the financial transaction.

[0036] Among them, the transaction popularity score can be used to represent the public sentiment tendency of financial transactions. For example, the larger the transaction popularity score, the more positive the public sentiment tendency of financial transactions is, while the smaller the transaction popularity score, the more negative the public sentiment tendency of financial transactions is. The above positive can also be understood as positive, and the above negative can also be called the opposite.

[0037] It should be noted that the specific method for generating transaction popularity scores based on image data is not limited here. Optionally, an image classification model can be used to determine and generate transaction popularity scores, and the aforementioned image classification model can include ResNet-50.

[0038] Alternatively, the transaction popularity score can be calculated using the following formula:

[0039] H t = Among them, H t The number of positive images on a given day indicates the number of images with a positive sentiment, while the total number of images indicates the total number of images.

[0040] It should be noted that, optionally, the above-mentioned calculation of transaction popularity scores, extraction of text using OCR (which can employ the Tesseract engine), obtaining unstructured text data, identification and numericalization of key fields such as "R&D investment" and "debt-to-asset ratio" using a Bidirectional Encoder Representations from Transformers (BERT) model, and the subsequent mapping and alignment of unstructured text data, transaction popularity scores, and market data according to time sequence can all be considered preprocessing operations performed on unstructured text data. These preprocessing operations can be performed by a preprocessing module in an electronic device; see [link to relevant documentation] for details. Figure 2 .

[0041] Step 103: Construct a composite investment factor based on the unstructured text data, the transaction popularity score, and the market data. The composite investment factor is used to modify financial trading strategies.

[0042] The specific method for constructing composite investment factors based on unstructured text data, trading popularity scores, and market data is not limited here. Optionally, the unstructured text data, trading popularity scores, and market data can be numerically processed, and then the corresponding values ​​of the unstructured text data, trading popularity scores, and market data can be respectively calculated to obtain composite investment factors. The above-mentioned target calculation can include at least one of the following: addition, subtraction, weighted addition, weighted subtraction, averaging, and calculating weighted average.

[0043] Alternatively, feature vectors corresponding to the unstructured text data, trading popularity scores, and market data can be extracted separately, and then the feature vectors corresponding to the unstructured text data, trading popularity scores, and market data can be fused to obtain a composite investment factor.

[0044] In other words, the aforementioned composite investment factors can be numerical values ​​or eigenvectors. Optionally, these composite investment factors can also be weight parameters of the model, which can be understood as model parameters. For specific details, please refer to the relevant descriptions in the section on student models later, which will not be repeated here. Therefore, the specific types of the aforementioned composite investment factors are not limited here.

[0045] Step 104: Modify the pre-set financial trading strategy according to the composite investment factors.

[0046] It should be noted that the pre-set financial trading strategy can be understood as the initial financial trading strategy. Since the composite investment factor is generated based on multiple primary data, it has good timeliness. By modifying the pre-set financial trading strategy through the composite investment factor, the timeliness of the modified financial trading strategy is also good, and thus the accuracy of the modified financial trading strategy is higher.

[0047] The specific method of modifying the pre-set financial trading strategy based on the composite investment factor is not limited here. Optionally, when the composite investment factor is a numerical value, the composite investment factor can be used as a weighting coefficient and multiplied with at least a part of the pre-set financial trading strategy to obtain the modified financial trading strategy. Alternatively, when the composite investment factor is a feature vector, the feature vector corresponding to the composite investment factor can be fused with the feature vector corresponding to the pre-set financial trading strategy to obtain the modified financial trading strategy.

[0048] It should be noted that the specific content of the financial trading strategy is not limited here. For example, a financial trading strategy may include multiple executable trading instructions, and the trading instructions may include "buy the CSI 300 ETF with a weight of 50%", etc.

[0049] In this embodiment, steps 101 to 104 acquire multiple sets of first data related to financial transactions, including unstructured text data, image data, and market data. Then, a transaction popularity score is generated based on the image data, and a composite investment factor is constructed based on the unstructured text data, the transaction popularity score, and the market data. Finally, a pre-set financial trading strategy is revised based on the composite investment factor. By acquiring multiple sets of first data and revising the financial trading strategy accordingly, the probability of low accuracy due to data distortion is reduced, and the accuracy of the final generated financial trading strategy is improved. Furthermore, because the multiple sets of first data are relatively timely, revising the financial trading strategy based on these multiple sets of first data allows the financial trading strategy to better adapt to the high-frequency and volatile financial market environment.

[0050] As an optional implementation, the step of constructing a composite investment factor based on the unstructured text data, the transaction popularity score, and the market data includes:

[0051] The unstructured text data, the transaction popularity score, and the market data are mapped and aligned according to time sequence to obtain the mapped and aligned unstructured text data, the mapped and aligned transaction popularity score, and the mapped and aligned market data.

[0052] The composite investment factor is constructed based on the mapped and aligned unstructured text data, the mapped and aligned transaction popularity score, and the mapped and aligned market data.

[0053] It should be noted that mapping and aligning unstructured text data, trading popularity scores, and market data according to time sequence can be understood as unifying all unstructured text data, trading popularity scores, and market data to data within the same time period. For example, unifying all unstructured text data, trading popularity scores, and market data to data within the first time period. Optionally, since trading popularity scores and market data correspond to the second and third time periods respectively, the change curves of trading popularity scores and market data can be obtained. Then, the trading popularity score for the first time period can be calculated based on the trading popularity score and its change curve for the second time period. Similarly, the market data for the first time period can be calculated based on the market data and its change curve for the third time period.

[0054] In this embodiment, since the sources of unstructured text data, trading popularity scores, and market data are different, there may be temporal differences between them. For example, unstructured text data may be data collected in the first time period, trading popularity scores may refer to trading popularity scores in the second time period, and market data may be market data collected in the third time period. If a composite investment factor is directly constructed based on unstructured text data, trading popularity scores, and market data, the accuracy of the constructed composite investment factor is likely to be low. Therefore, it is necessary to map and align the unstructured text data, trading popularity scores, and market data according to their temporal order, and then construct the composite investment factor based on the mapped and aligned unstructured text data, mapped and aligned trading popularity scores, and mapped and aligned market data. This can improve the accuracy of the constructed composite investment factor.

[0055] As an optional implementation, constructing the composite investment factor based on the mapped and aligned unstructured text data, the mapped and aligned transaction popularity score, and the mapped and aligned market data includes:

[0056] Spline interpolation is used to complete the mapping-aligned unstructured text data, the mapping-aligned transaction popularity score, and the mapping-aligned market data, respectively.

[0057] The composite investment factor is constructed based on the data-completed unstructured text data, the data-completed transaction popularity score, and the data-completed market data.

[0058] It should be noted that spline interpolation is a mathematical method that constructs smooth curves passing through given points using variable splines. Its core characteristic is the use of piecewise polynomials to define the curve, where each polynomial is determined by adjacent data points, ensuring the continuity of derivatives at connection points between adjacent polynomials. This achieves a smooth transition between curve segments. Therefore, spline interpolation can effectively complete the data in mapping and aligning unstructured text data, mapping and aligning transaction popularity scores, and mapping and aligning market data, with high accuracy and comprehensiveness.

[0059] In this embodiment, since some data may be missing after mapping and alignment, spline interpolation is first used to complete the unstructured text data, the trading popularity score, and the market data after mapping and alignment. Then, a composite investment factor is constructed based on the unstructured text data, the trading popularity score, and the market data after data completion. This can further improve the accuracy of the constructed composite investment factor.

[0060] As an optional implementation, the step of constructing a composite investment factor based on the unstructured text data, the transaction popularity score, and the market data includes:

[0061] A sentiment score is generated based on the unstructured text data, and price change data within a target time period is calculated based on the market data. The target time period is the time period before the current moment, or the target time period is the time period including the current moment, and the current moment is the end moment of the target time period.

[0062] Calculate the first product of the sentiment score and the first weighting coefficient, calculate the second product of the price change data within the target time period and the second weighting coefficient, and calculate the third product of the transaction popularity score and the third weighting coefficient;

[0063] The sum of the first product, the second product, and the third product is determined as the composite investment factor;

[0064] The first weighting coefficient, the second weighting coefficient, and the third weighting coefficient are preset values.

[0065] In this embodiment, the sum of the first product, the second product, and the third product is determined as the composite investment factor. This increases the diversity and flexibility of the composite investment factor generation method and further improves the accuracy of the generated composite investment factor.

[0066] For example, the calculation method of the composite investment factor in the embodiments of this application can be expressed by the following calculation formula:

[0067] Ft=α t ·S NLP +β t ·M 动量 +γ t ·H t ;

[0068] Where Ft is used to represent the composite investment factor, S NLP Used to represent sentiment score, M 动量 Used to represent price change data within a target time period, H t Used to represent the popularity score of transactions, α t β t γ t These are used to represent the first weighting coefficient, the second weighting coefficient, and the third weighting coefficient, respectively, while α t ·S NLP Used to represent the first product, β t ·M 动量 Used to represent the second product, γ t ·H t Used to represent the third product.

[0069] Optionally, the sentiment score can be obtained by analyzing the financial statement text using a Bidirectional Encoder Representations from Transformers (BERT) model. The range of the sentiment score can include [-1, 1]. The specific length of the target time period is not limited here; optionally, the target time period can be 10 days, therefore M 动量 =(P t -P t-10 ) / (P t-10 The above P t P is used to represent the price on day t. t-10 Used to indicate the price 10 days ago.

[0070] As an optional implementation, before calculating the first product of the sentiment score and the first weighting coefficient, the second product of the price change data within the target time period and the second weighting coefficient, and the third product of the transaction popularity score and the third weighting coefficient, the method further includes:

[0071] The first weight coefficient, the second weight coefficient, and the third weight coefficient are determined according to the reinforcement learning algorithm and the objective function;

[0072] The objective function is the function that maximizes the Sharpe ratio.

[0073] The specific type of reinforcement learning algorithm is not limited here; optionally, it may include proximal policy optimization (PPO) algorithms. The Sharpe ratio, also known as the Sharpe index, is a standardized indicator for evaluating fund performance.

[0074] In this embodiment, the importance of sentiment scores, price change data, and transaction popularity scores changes over time and with the continuous evolution of application scenarios. Therefore, the first weight coefficient, the second weight coefficient, and the third weight coefficient can be determined based on the reinforcement learning algorithm and the objective function. This allows for flexible determination of the first weight coefficient, the second weight coefficient, and the third weight coefficient according to changes in time and application scenarios, increasing the flexibility and diversity of the determination methods. At the same time, it also makes the determined first weight coefficient, the second weight coefficient, and the third weight coefficient more closely matched with the current time and application scenario, thereby improving the accuracy of determining the first weight coefficient, the second weight coefficient, and the third weight coefficient.

[0075] It should be noted that constructing a composite investment factor by combining reinforcement learning algorithms can optimize the first, second, and third weight coefficients, resulting in higher accuracy of these coefficients. This, in turn, leads to higher accuracy of the composite investment factor, which in turn leads to higher accuracy of the modified financial trading strategy, and consequently, higher accuracy of the decisions made based on the modified financial trading strategy.

[0076] As an optional implementation, the step of constructing a composite investment factor based on the unstructured text data, the transaction popularity score, and the market data includes:

[0077] The trained teacher model is obtained by training the teacher model based on the unstructured text data, the transaction popularity score, and the market data.

[0078] The student model, which is controlled to acquire data, learns the output data of the trained teacher model through a distillation loss function, and updates the model parameters of the student model based on the output data. The model parameters include the composite investment factor.

[0079] The coefficients of the distillation loss function are determined based on the market volatility of the financial transaction.

[0080] It should be noted that the aforementioned unstructured text data, transaction popularity scores, and market data training can be referred to as new training data (including market style labels), which can be represented by Dnew, as shown in the example below. Figure 3 As shown, the new training data can be updated weekly. The specific type of teacher model is not limited here; optionally, the teacher model can be a Transformer model, and the output data of the teacher model can also be referred to as the predicted soft label P_teacher.

[0081] It should be noted that, optionally, both the teacher model and the student model mentioned above can be artificial intelligence (AI) models, and the model parameters of the student model include composite investment factors, which can be understood as at least one of the model parameters of the student model.

[0082] It should be noted that the implementation method of this application can be referred to as the distillation process of the student model or the parameter update process, and the distillation loss function mentioned above can be found in the following description:

[0083] L distill =α·KL divergence (P teacher ,P student )+β·MSE(y true ,y pred );

[0084] Here, α and β can be understood as the coefficients of the distillation loss function, and α and β are determined based on the market volatility of the financial transaction. For example, when market volatility > 5%, α = 0.7 and β = 0.3. distill The calculated numerical value used to represent the distillation loss function, KL divergence (P teacher ,P student ) represents the KL divergence between the teacher model and the student model, MSE(y) true ,y pred The mean square error (MSE) is used to represent the difference between the output data of the teacher model and the student model.

[0085] In this embodiment, a teacher model can be trained based on unstructured text data, transaction popularity scores, and market data to obtain a trained teacher model. Then, a student model is obtained by controlling the acquisition of the teacher model to learn the output data of the trained teacher model through a distillation loss function, and the model parameters of the student model are updated according to the output data. In this way, the student model is distilled by the teacher model, so that more than 2,000 feature parameters can be automatically updated every week, enabling the student model to achieve dynamic adaptive updates and improving the accuracy of the student model.

[0086] As an optional implementation, updating the model parameters of the student model based on the output data includes:

[0087] The prediction error of the student model after the model parameters were updated was measured over N consecutive days, and the backtest Sharpe ratio of the student model after the model parameters were updated was measured.

[0088] If the prediction error of the student model after model parameter update is less than or equal to a first preset value for N consecutive days, and the backtest Sharpe ratio of the student model after model parameter update is less than or equal to a second preset value, the model parameters of the student model are updated according to the output data, and the student model after model parameter update is deployed.

[0089] Where N is an integer greater than 1.

[0090] In this embodiment, if the prediction error of the student model after model parameter update is less than or equal to a first preset value for N consecutive days, and the backtest Sharpe ratio of the student model after model parameter update is less than or equal to a second preset value, it indicates that the accuracy of the student model is high. Therefore, the model parameters of the student model can be updated according to the output data, and the student model after model parameter update can be deployed, thereby making the accuracy of the deployed student model high.

[0091] As an optional implementation, after detecting the N-day prediction error of the student model after the model parameter update and the backtest Sharpe ratio of the student model after the model parameter update, the method further includes:

[0092] If the prediction error of the student model after the model parameters are updated is greater than the first preset value for N consecutive days, the student model will be incrementally trained.

[0093] If the prediction error of the student model after model parameter update is less than or equal to the first preset value for N consecutive days, and the backtest Sharpe ratio of the student model after model parameter update is greater than the second preset value, the student model and the teacher model are retrained.

[0094] Incremental training can be referred to as updating some of the model parameters of the student model. For example, when the prediction error of the student model after the model parameter update is greater than the first preset value (such as 2%) for three consecutive days, incremental training can be performed on the student model.

[0095] Retraining the student and teacher models can be understood as updating all the model parameters of the student and teacher models.

[0096] In this embodiment, if the prediction error of the student model after model parameter update is less than or equal to the first preset value for N consecutive days, and the backtest Sharpe ratio of the student model after model parameter update is greater than the second preset value, it indicates that the accuracy of both the student model and the teacher model is low at this time, and the student model and the teacher model need to be retrained to improve the accuracy of the student model and the teacher model.

[0097] It should be noted that the detailed flowchart of the implementation method of this application can be found in [reference needed]. Figure 3 As shown, N can be Figure 3 The first preset value for the 3-day period can be... Figure 3 The second preset value can be 2%. Figure 3 10% of the total.

[0098] Additionally, you can also see Figure 4 , Figure 4 A flowchart illustrating the process of updating composite investment factors, as shown below. Figure 4 As shown, data after the market closes each day can be obtained to update the composite investment factors. This ensures a high degree of matching between the composite investment factors and the actual situation of the daily financial trading market, thereby enabling more accurate determination of financial trading strategies based on the composite investment factors.

[0099] It should be noted that network parameters can be represented using θ, and the updated network parameters can be represented using θ0. k+1 It means that θ k+1 The calculation formula can be found in the following description:

[0100] =arg .

[0101] Here, At can be called the dominance function, and argmax is a function that modulates the function E. t A function to find the parameters, and π θ (α t |s t ), π θold (α t |s t ), π θ πθold ε represents the probability distribution of network parameters under different states, min means finding the minimum value, clip is used to represent multimodal model fusion (Constrastive Language-Image Pre-training), and ϵ is used to represent the exploration rate.

[0102] It should be noted that, in order to more fully illustrate the above embodiments, a specific embodiment will be used as an example below. For detailed steps, please refer to [link / reference needed]. Figure 2 As shown in the embodiments of this application, the electronic device may include the aforementioned data layer, model layer, and strategy layer.

[0103] It should be noted that, optionally, the process of generating a pre-set financial trading strategy can refer to the following steps:

[0104] By inputting the user's risk preference (low / medium / high risk) and investment horizon (short-term / medium-term), a financial trading strategy is generated through the following steps:

[0105] 1) Target selection: Based on momentum ranking (M) 动量 The pool of initial targets is composed of the top 20% and moving average signals (20-day > 60-day).

[0106] 2) Weight allocation: The targets are grouped according to the feature space using a clustering algorithm (DBSCAN), and the optimal weight combination is selected within each group through Monte Carlo simulation;

[0107] 3) Strategy output: Generate executable trading instructions (such as "Buy CSI 300 ETF, weight 50%)".

[0108] Alternatively, after the financial trading strategy is modified, the strategy layer can also make decisions based on the aforementioned financial trading strategy, for example:

[0109] 1) Stress test: Generate 1,000 market paths based on Monte Carlo simulation and calculate the portfolio value at risk (VAR).

[0110] 2) Dynamic hedging: If VAR exceeds the threshold (e.g., 5%), hedging recommendations are automatically generated (e.g., "Buy VIX futures, weighted X%)".

[0111] 3) Circuit breaker mechanism: When the real-time drawdown is greater than 8%, the position will be forcibly liquidated and switched to cash assets.

[0112] Specifically, when the detected VAR exceeds the threshold, the strategy layer can generate a hedging recommendation. When a real-time drawdown is detected to be greater than 8%, the strategy layer can make a forced liquidation decision to ensure the safety of financial transactions. Additionally, when market volatility is detected to be less than 3%, the strategy layer can restart the aforementioned financial trading strategy.

[0113] See Figure 5 , Figure 5 This is a structural diagram of the financial trading strategy determination device 500 provided in the embodiments of this application. The financial trading strategy determination device 500 includes:

[0114] The acquisition module 501 is used to acquire multiple primary data related to financial transactions, including unstructured text data, image data, and market data.

[0115] The generation module 502 is used to generate a transaction popularity score based on the image data, wherein the transaction popularity score is used to represent the public opinion sentiment tendency of the financial transaction;

[0116] The construction module 503 is used to construct a composite investment factor based on the unstructured text data, the transaction popularity score and the market data, and the composite investment factor is used to modify the financial trading strategy;

[0117] The correction module 504 is used to correct the pre-set financial trading strategy according to the composite investment factor.

[0118] As an optional implementation, construction module 503 includes:

[0119] The mapping and alignment submodule is used to map and align the unstructured text data, the transaction popularity score, and the market data according to time sequence to obtain the mapped and aligned unstructured text data, the mapped and aligned transaction popularity score, and the mapped and aligned market data.

[0120] A submodule is constructed to build the composite investment factor based on the mapped and aligned unstructured text data, the mapped and aligned transaction popularity score, and the mapped and aligned market data.

[0121] As an optional implementation, a submodule is constructed, including:

[0122] The data replenishment unit is used to complete the mapped and aligned unstructured text data, the mapped and aligned transaction popularity score, and the mapped and aligned market data using spline interpolation.

[0123] The construction unit is used to construct the composite investment factor based on the data-completed unstructured text data, the data-completed transaction popularity score, and the data-completed market data.

[0124] As an optional implementation, construction module 503 includes:

[0125] The generation submodule is used to generate a sentiment score based on the unstructured text data, and to calculate price change data within a target time period based on the market data. The target time period is the time period before the current time, or the target time period is the time period including the current time, and the current time is the end time of the target time period.

[0126] The calculation submodule is used to calculate the first product of the sentiment score and the first weight coefficient, the second product of the price change data within the target time period and the second weight coefficient, and the third product of the transaction popularity score and the third weight coefficient.

[0127] The first determining submodule is used to determine the sum of the first product, the second product, and the third product as the composite investment factor;

[0128] The first weighting coefficient, the second weighting coefficient, and the third weighting coefficient are preset values.

[0129] As an optional implementation, construction module 503 further includes:

[0130] The second determining submodule is used to determine the first weight coefficient, the second weight coefficient, and the third weight coefficient based on the reinforcement learning algorithm and the objective function.

[0131] The objective function is the function that maximizes the Sharpe ratio.

[0132] As an optional implementation, construction module 503 includes:

[0133] The acquisition submodule is used to obtain the trained teacher model based on the unstructured text data, the transaction popularity score, and the market data.

[0134] The update submodule is used to control the acquired student model to learn the output data of the trained teacher model through the distillation loss function, and update the model parameters of the student model according to the output data, wherein the model parameters include the composite investment factor;

[0135] The coefficients of the distillation loss function are determined based on the market volatility of the financial transaction.

[0136] As an optional implementation, the update submodule includes:

[0137] The detection unit is used to detect the prediction error of the student model over N consecutive days after the model parameters are updated, and to detect the backtest Sharpe ratio of the student model after the model parameters are updated.

[0138] The update unit is used to update the model parameters of the student model according to the output data and deploy the student model with updated model parameters when the prediction error of the student model after the model parameter update is less than or equal to a first preset value and the backtest Sharpe ratio of the student model after the model parameter update is less than or equal to a second preset value.

[0139] Where N is an integer greater than 1.

[0140] As an optional implementation, the update submodule also includes:

[0141] An enhanced training unit is used to perform incremental training on the student model when the prediction error of the student model after model parameter updates is greater than a first preset value over N consecutive days.

[0142] The retraining unit is used to retrain the student model and the teacher model when the prediction error of the student model after model parameter update is less than or equal to a first preset value over N consecutive days, and the backtest Sharpe ratio of the student model after model parameter update is greater than a second preset value.

[0143] The financial trading strategy determination device 500 can achieve the implementation described in the embodiments of this application. Figure 1 The various processes in the method embodiments, and the ways to achieve the same beneficial effects, will not be repeated here to avoid repetition.

[0144] This application also provides an electronic device. Please refer to [link to relevant documentation]. Figure 6 The electronic device may include a processor 601, a memory 602, and a program 6021 stored in the memory 602 and executable on the processor 601. When the program 6021 is executed by the processor 601, it can achieve... Figure 1 Any steps in the corresponding method embodiments and the achievement of the same beneficial effects will not be repeated here.

[0145] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by hardware related to program instructions, and the program can be stored in a readable medium. This application also provides a readable storage medium storing a computer program, which, when executed by a processor, can implement the above-described methods. Figure 1 Any step in the corresponding method embodiment can achieve the same technical effect, and will not be repeated here to avoid repetition.

[0146] The storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0147] This application also provides a computer program product, including computer instructions, which, when executed by a processor, can perform the above-described functions. Figure 1 Any step in the corresponding method embodiment can achieve the same technical effect, and will not be repeated here to avoid repetition.

[0148] The above description represents the preferred embodiments of this application. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for determining a financial trading strategy, characterized in that, include: Acquire multiple primary data related to financial transactions, including unstructured text data, image data, and market data; A transaction popularity score is generated based on the image data, and the transaction popularity score is used to represent the public sentiment tendency of the financial transaction; A composite investment factor is constructed based on the unstructured text data, the transaction popularity score, and the market data. The composite investment factor is used to modify financial trading strategies. The pre-set financial trading strategy is modified based on the composite investment factors.

2. The method according to claim 1, characterized in that, The construction of a composite investment factor based on the unstructured text data, the transaction popularity score, and the market data includes: The unstructured text data, the transaction popularity score, and the market data are mapped and aligned according to time sequence to obtain the mapped and aligned unstructured text data, the mapped and aligned transaction popularity score, and the mapped and aligned market data. The composite investment factor is constructed based on the mapped and aligned unstructured text data, the mapped and aligned transaction popularity score, and the mapped and aligned market data.

3. The method according to claim 2, characterized in that, The construction of the composite investment factor based on the mapped and aligned unstructured text data, the mapped and aligned transaction popularity score, and the mapped and aligned market data includes: Spline interpolation is used to complete the mapping-aligned unstructured text data, the mapping-aligned transaction popularity score, and the mapping-aligned market data, respectively. The composite investment factor is constructed based on the data-completed unstructured text data, the data-completed transaction popularity score, and the data-completed market data.

4. The method according to claim 1, characterized in that, The construction of a composite investment factor based on the unstructured text data, the transaction popularity score, and the market data includes: A sentiment score is generated based on the unstructured text data, and price change data within a target time period is calculated based on the market data. The target time period is the time period before the current moment, or the target time period is the time period including the current moment, and the current moment is the end moment of the target time period. Calculate the first product of the sentiment score and the first weighting coefficient, calculate the second product of the price change data within the target time period and the second weighting coefficient, and calculate the third product of the transaction popularity score and the third weighting coefficient; The sum of the first product, the second product, and the third product is determined as the composite investment factor; The first weighting coefficient, the second weighting coefficient, and the third weighting coefficient are preset values.

5. The method according to claim 4, characterized in that, Before calculating the first product of the sentiment score and the first weighting coefficient, the second product of the price change data within the target time period and the second weighting coefficient, and the third product of the transaction popularity score and the third weighting coefficient, the method further includes: The first weight coefficient, the second weight coefficient, and the third weight coefficient are determined according to the reinforcement learning algorithm and the objective function; The objective function is the function that maximizes the Sharpe ratio.

6. The method according to claim 1, characterized in that, The construction of a composite investment factor based on the unstructured text data, the transaction popularity score, and the market data includes: The trained teacher model is obtained by training the teacher model based on the unstructured text data, the transaction popularity score, and the market data. The student model, which is controlled to acquire data, learns the output data of the trained teacher model through a distillation loss function, and updates the model parameters of the student model based on the output data. The model parameters include the composite investment factor. The coefficients of the distillation loss function are determined based on the market volatility of the financial transaction.

7. The method according to claim 6, characterized in that, The step of updating the model parameters of the student model based on the output data includes: The prediction error of the student model after the model parameters were updated was measured over N consecutive days, and the backtest Sharpe ratio of the student model after the model parameters were updated was measured. If the prediction error of the student model after model parameter update is less than or equal to a first preset value for N consecutive days, and the backtest Sharpe ratio of the student model after model parameter update is less than or equal to a second preset value, the model parameters of the student model are updated according to the output data, and the student model after model parameter update is deployed. Where N is an integer greater than 1.

8. The method according to claim 7, characterized in that, After detecting the continuous N-day prediction error of the student model after the model parameter update and the backtest Sharpe ratio of the student model after the model parameter update, the method further includes: If the prediction error of the student model after the model parameters are updated is greater than the first preset value for N consecutive days, the student model will be incrementally trained. If the prediction error of the student model after model parameter update is less than or equal to the first preset value for N consecutive days, and the backtest Sharpe ratio of the student model after model parameter update is greater than the second preset value, the student model and the teacher model are retrained.

9. A financial trading strategy determination device, characterized in that, include: The acquisition module is used to acquire multiple primary data related to financial transactions, including unstructured text data, image data, and market data. The generation module is used to generate a transaction popularity score based on the image data, and the transaction popularity score is used to represent the public sentiment tendency of the financial transaction; A construction module is used to construct a composite investment factor based on the unstructured text data, the transaction popularity score, and the market data. The composite investment factor is used to modify financial trading strategies. The correction module is used to correct the pre-set financial trading strategy based on the composite investment factors.

10. An electronic device, comprising: A memory, a processor, and a program stored in the memory and executable on the processor; characterized in that the processor is configured to read the program in the memory to implement the steps in the financial trading strategy determination method as described in any one of claims 1 to 8.

11. A readable storage medium for storing a program, characterized in that, When the program is executed by the processor, it implements the steps in the financial trading strategy determination method as described in any one of claims 1 to 8.

12. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps in the financial trading strategy determination method as described in any one of claims 1 to 8.

Images