Deep learning-based quantitative transaction strategy dynamic optimization system and method thereof

Abstract

The invention discloses a quantitative transaction strategy dynamic optimization system and method based on deep learning, and belongs to the technical field of quantitative transactions, and the system comprises a data collection module which is used for collecting data of a target market invested by an investor from a data source; the data source comprises a security exchange and financial media; the acquisition frequency setting module is used for setting the frequency of data acquisition from the data source according to the transaction demand of the investor, and the acquisition frequency setting module comprises a distinguishing unit and a data setting unit; the distinguishing unit is used for distinguishing the transaction demands into high-frequency transaction demands and intermediate-frequency transaction demands according to the transaction demands. Through dynamic data adaptation, quantitative risk grading, multi-feature prediction and hierarchical fund management, accurate optimization and risk controllability of a quantitative transaction strategy are realized, and adaptability and income stability of the strategy in a complex market environment are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of quantitative trading technology, in particular to a deep learning-based quantitative trading strategy dynamic optimization system and method thereof. BACKGROUND

[0002] Deep reinforcement learning combines the perception ability of deep learning and the decision-making ability of reinforcement learning, which can find patterns in high-dimensional data and optimize decision-making strategies through interaction between agents and the environment. In the field of quantitative trading, this technology can process large amounts of market data and optimize the execution of trading strategies through intelligent decision-making systems.

[0003] For this research, the application number CN202211693912.2 provides a deep learning-based quantitative trading strategy prediction method, device and storage medium, which includes: obtaining the quotation data of the target market; inputting the quotation data into a pre-set index model to obtain the index value output by the index model; wherein the index model is trained using a reinforcement learning algorithm, and the index model includes a target model, a timing model, and a risk control model. This technical solution disassembles the quantitative trading process, making it easier to train and run in parallel mode for index generation and decision-making processes, effectively shortening the index generation time and strategy decision-making time.

[0004] Another application number CN202311033708.2 provides a deep learning-based high-frequency quantitative trading method and system, which includes: step 1: obtaining target transaction records; step 2: training target transaction records based on deep learning technology to obtain a high-frequency quantitative trading model; step 3: obtaining current transaction data; step 4: performing corresponding high-frequency quantitative trading based on current transaction data and the high-frequency quantitative trading model; step 5: determining delay attribution based on the transaction delay events in the high-frequency quantitative trading process, and processing the transaction delay events based on the delay attribution. This technical solution trains a high-frequency quantitative trading model based on target transaction records, enabling short-term fast trading with low buy-in and high sell-out, making the judgment of trading points faster.

[0005] The optimization of quantitative trading strategies aims to improve the overall performance of the trading system, with specific goals including increasing returns, reducing risks, enhancing stability, and improving adaptability. However, the quantitative strategies of the above technical solutions still have deficiencies. When investors invest in the stock market, the yield will quickly decline when faced with market mechanism switching (such as policy impact, unexpected events), requiring manual readjustment, which in turn increases the investment risk of investors. SUMMARY

[0006] In view of the above problems existing in the prior art of quantification transaction technology, the present application is proposed.

[0007] Therefore, one of the purposes of the present application is to provide a deep learning-based quantification transaction strategy dynamic optimization system and method, which realizes the precise optimization and risk control of quantification transaction strategy through dynamic data adaptation, quantification risk grading, multi-feature prediction and hierarchical fund management, and improves the adaptability and yield stability of the strategy in complex market environment.

[0008] To solve the above technical problems, the present application provides the following technical solutions:

[0009] On the one hand, the present application provides a deep learning-based quantification transaction strategy dynamic optimization system, which comprises:

[0010] A data acquisition module is used to acquire data of a target market in which an investor invests from a data source; the data source includes a stock exchange and a financial media;

[0011] A collection frequency setting module is used to set the frequency of collecting data from the data source according to the transaction demand of the investor; the collection frequency setting module comprises a distinguishing unit and a data setting unit;

[0012] The distinguishing unit is used to distinguish the transaction demand into high-frequency transaction demand and medium-frequency transaction demand according to the transaction demand; in the transaction demand corresponding to the high-frequency transaction demand, the frequency of collecting data from the data source is set to real-time; in the transaction demand corresponding to the medium-frequency transaction demand, the frequency of collecting data from the data source is set to non-real-time; the non-real-time includes collecting data from the data source once every 5-10 minutes;

[0013] The data setting unit is used to set the risk of the collected data in the transaction demand in response to the transaction demand corresponding to the high-frequency transaction demand; including setting low risk, medium risk and high risk;

[0014] A data fusion processing module, the data fusion processing module comprises a transaction volume division unit, a feature extraction unit, a model construction unit and a calculation unit;

[0015] The transaction volume division unit is used to obtain the transaction volume of a preset period according to the set risk, the transaction volume includes the transaction amount, and the preset period is the period when the transaction price of the target market is in a falling state;

[0016] The feature extraction unit is used to extract the state feature of the transaction price in the falling state in response to the preset period; the state feature includes the proportion of the price dimension and the volume dimension to the amount.

[0017] As a preferred scheme of the present application, wherein: in the data setting unit, the risk setting step includes:

[0018] Collecting relevant data for the investors to obtain profits in the transaction demand corresponding to the high-frequency transaction demand, the relevant data including the highest price and the lowest price of the target market;

[0019] Setting a floating interval between the highest price and the lowest price, the floating interval including an upward floating interval and a downward floating interval;

[0020] Setting a floating value for the highest price in the downward floating interval, and setting a calculation of the floating value every 20-30 seconds;

[0021] Pre-setting a risk threshold for the floating value, the risk threshold including 3% lower than the highest price, the 3% corresponding to the low risk.

[0022] As a preferred scheme of the present application, wherein: the model construction unit is used to construct a dynamic prediction model, in the dynamic prediction model, including market volume and price before the transaction price is in a falling state, and state characteristics and external information of the falling state as input features, including the set risk level and the falling probability as output features;

[0023] The calculation unit is used to calculate the associated influence of the transaction volume on the transaction price under different risk levels in response to the output features.

[0024] As a preferred scheme of the present application, wherein: in the calculation unit, the associated influence of the transaction volume on the transaction price under different risk levels is calculated according to the following formula:

[0025] Wherein, Q represents the transaction volume, the transaction volume representing the transaction volume under the specified risk level; the transaction volume including the number of shares and the number of contracts;

[0026] Wherein, R represents the risk level, C represents the total amount of the investor's account, P entry represents the entry price, P stop represents the stop loss price, and θ represents the leverage multiple, which is 1 if no leverage is used.

[0027] As a preferred scheme of the present application, wherein: in the calculation unit, the associated influence of the transaction volume on the transaction price under different risk levels also includes calculation according to the following formula:

[0028] Wherein, ΔP represents the price impact, the price impact represents the price impact in the case of a specified risk level;

[0029] In the formula, k represents the market impact coefficient, sigma represents the asset volatility, Q represents the trading volume, and V represents the daily average trading volume.

[0030] As a preferred scheme of the present application, wherein: based on the calculated correlation impact, the trading volume corresponding to the transaction price is divided into n represents the nth divided trading volume, and scenario training is performed on each divided trading volume, which includes setting different loss amounts according to the divided trading volume, the loss amount being the trading volume of the last investor causing monetary loss in the target market before reaching the specified risk level, marking the trading volume as a reference trading volume, obtaining the median of the reference trading volume, if the investor trades in the target market at a future time, if the trading volume corresponding to the transaction is the same as the median, then obtaining the difference between the current transaction price and the highest price, if it is not lower than the highest price, the system does not perform risk prompting; otherwise, risk prompting is performed.

[0031] As a preferred scheme of the present application, wherein: if risk prompting is not performed, an amount corresponding to the difference is obtained, and at least three critical values are preset based on the amount, and the first critical value accounts for half of the amount; if the difference between the transaction price and the highest price at a future time is higher than the first critical value, the system performs risk prompting.

[0032] As a preferred scheme of the present application, wherein: after the system performs risk prompting, the remaining amount is obtained, the amount is divided according to the proportions of 20%, 50% and 30%, the amount corresponding to the 20% is set as a first amount, if the investor trades using the first amount after the system performs risk prompting, the difference between the transaction price after the transaction and the highest price is obtained, if the difference is an expanding change, the system performs risk prompting; otherwise, risk prompting is not performed.

[0033] On the other hand, the present application provides a method applied to the deep learning-based quantitative trading strategy dynamic optimization system as described above, comprising the following steps:

[0034] Collecting data of the target market in which the investor invests from a data source; the data source includes stock exchanges and financial media;

[0035] According to the transaction demand of the investor, the frequency of collecting data from the data source is set, including according to the transaction demand, the transaction demand is divided into high-frequency transaction demand and medium-frequency transaction demand, in the transaction demand corresponding to the high-frequency transaction demand, the frequency of collecting data from the data source is set to real-time, in the transaction demand corresponding to the medium-frequency transaction demand, the frequency of collecting data from the data source is set to non-real-time, and the non-real-time includes collecting data from the data source once every 5-10 minutes;

[0036] In the transaction demand, the change of the collected data is risk set; including setting low risk, medium risk and high risk;

[0037] According to the set risk, the transaction amount of a preset period is obtained, the transaction amount includes a transaction amount, and the preset period is a period when the transaction price of the target market is in a falling state;

[0038] The state characteristics of the transaction price in the falling state are extracted; the state characteristics include the price dimension and the proportion of the trading volume in the amount dimension;

[0039] A dynamic prediction model is constructed, in the dynamic prediction model, the market quantity and price before the transaction price is in a falling state and the state characteristics of the falling state and external information are included as input characteristics, and the set risk level and the falling probability are included as output characteristics;

[0040] According to the output characteristics, the correlation influence of the transaction amount on the transaction price under different risk levels is calculated.

[0041] Beneficial effects:

[0042] 1、The application distinguishes high-frequency and medium-frequency transaction demand through the collection frequency setting module, and dynamically adjusts the data collection frequency (real-time or every 5-10 minutes), so that the strategy can quickly respond to market changes, avoid decision deviation caused by data lag, and improve the adaptability of the strategy to multi-level market rhythm;

[0043] 2、In high-frequency trading, the application sets the floating interval of the highest price and the lowest price and the risk threshold (such as a 3% price drop corresponding to low risk), and calculates the floating value every 20-30 seconds to realize real-time quantitative classification of transaction risk. This mechanism can accurately identify extreme volatility scenarios and prevent excessive trading or stop loss errors caused by violent price fluctuations;

[0044] 3、The data fusion processing module extracts the characteristics of the price falling state (such as the price dimension and the trading volume proportion), and constructs a dynamic prediction model combined with market quantity and price data and external information. The model can integrate multi-source data, capture the implicit rules of price trends, and improve prediction accuracy;

[0045] 4、The present application calculates price impact through market impact coefficient, asset volatility rate and other parameters, and sets multiple critical values (such as 50% and 20% stratification of the amount difference), when the transaction price approaches the critical point, the system automatically triggers a risk prompt to help investors avoid liquidity risk or market manipulation in advance. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor. Among them:

[0047] Fig. 1 The modular structure diagram of the deep learning-based quantitative trading strategy dynamic optimization system of the embodiments of the present application;

[0048] Fig. 2 The method flowchart of the embodiments of the present application;

[0049] The figure label: 110-data acquisition module; 120-acquisition frequency setting module; 1201-distinguishing unit; 1202-data setting unit; 130-data fusion processing module; 1301-transaction volume division unit; 1302-feature extraction unit; 1303-model construction unit; 1304-computing unit. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme of the embodiments of the present application will be described clearly and completely in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the present application.

[0051] Because of the deficiencies of the existing quantitative strategy, when investors invest in the stock market, the yield will rapidly decay when facing market mechanism switching (such as policy impact, unexpected events), which needs to be manually adjusted, thereby leading to the increase of investment risk of investors.

[0052] Based on this, the present application proposes a deep learning-based quantitative trading strategy dynamic optimization system and method, which realizes the precise optimization and risk control of quantitative trading strategy through dynamic data adaptation, quantitative risk grading, multi-feature prediction and hierarchical fund management, and improves the adaptability and yield stability of the strategy in complex market environment.

[0053] The scheme will be further specifically explained below by means of embodiments and in conjunction with the drawings.

[0054] With reference to Figs. 1-2 For an embodiment of the present application, the embodiment provides a deep learning-based quantitative trading strategy dynamic optimization system, comprising:

[0055] The data acquisition module 110 is configured to acquire data of a target market in which an investor invests from a data source; the data source includes a stock exchange and a financial media;

[0056] In the embodiment, before data acquisition, the required data source needs to be determined first, including a stock exchange, a financial media, a professional data provider, etc., and the determination of the data source helps to determine the accuracy, real-time performance and completeness of the data;

[0057] According to the characteristics of the data source, a suitable acquisition tool is selected, for example, for real-time transaction data, a professional data interface or a subscription service may be used; for historical data, an API or a data download service provided by a data provider can be selected;

[0058] The target market includes a stock market;

[0059] It should be noted that the data of the target market obtained from the data source such as the stock exchange and the financial media can cover core indicators such as price and trading volume;

[0060] The acquisition frequency setting module 120 is configured to set the frequency of acquiring data from the data source according to the trading demand of the investor; the acquisition frequency setting module 120 includes a distinguishing unit 1201 and a data setting unit 1202;

[0061] The distinguishing unit 1201 is configured to distinguish the trading demand into high-frequency trading demand and medium-frequency trading demand according to the trading demand; in the trading demand corresponding to the high-frequency trading demand, the frequency of acquiring data from the data source is set to real-time; in the trading demand corresponding to the medium-frequency trading demand, the frequency of acquiring data from the data source is set to non-real-time; the non-real-time includes acquiring data from the data source once every 5-10 minutes;

[0062] The data setting unit 1202 is configured to set the risk of the acquired data in the trading demand in response to the trading demand corresponding to the high-frequency trading demand; including setting low risk, medium risk and high risk;

[0063] The steps of risk setting include:

[0064] The relevant data for the investor to obtain income in the trading demand corresponding to the high-frequency trading demand includes the highest price and the lowest price of the target market;

[0065] a floating interval between the highest price and the lowest price is set, the floating interval includes an upward floating interval and a downward floating interval;

[0066] a floating value about the highest price is set in the downward floating interval, and a calculation of the floating value is performed every 20-30 seconds;

[0067] a risk threshold about the floating value is preset, the risk threshold includes a 3% drop below the highest price, and the 3% corresponds to a low risk;

[0068] In this embodiment, high-frequency transactions collect data in real time to capture transient opportunities, and medium-frequency transactions reduce data redundancy to balance timeliness and computing cost;

[0069] And in high-frequency transactions, an extreme fluctuation scenario is identified in advance through risk level setting (such as a 3% drop threshold), and loss stoppage errors caused by data delay are avoided;

[0070] And the risk setting mechanism of the data setting unit is that in high-frequency transactions, the system collects the highest price and the lowest price of the target market, sets a two-way floating interval (upward / downward), and calculates the floating value of the highest price every 20-30 seconds in the downward interval, and presets a risk threshold (such as a 3% drop below the highest price for a low risk);

[0071] This dynamic calculation of the floating value and comparison with the threshold convert the risk into an operable graded signal (low / medium / high), reducing subjective judgment errors;

[0072] And the risk state is updated every 20-30 seconds to meet the stringent requirements of high-frequency transactions on timeliness and reduce non-rational operations caused by short-term price shocks;

[0073] The data fusion processing module 130 includes a transaction volume division unit 1301, a feature extraction unit 1302, a model construction unit 1303, and a calculation unit 1304.

[0074] The transaction volume division unit 1301 is configured to obtain transaction volume in a preset period according to the set risk, the transaction volume including transaction amount, and the preset period being a period when the transaction price of the target market is in a falling state;

[0075] The feature extraction unit 1302 is configured to extract state features of the transaction price in the falling state in response to the preset period; the state features including a price dimension and a proportion of the trading volume to the amount dimension;

[0076] The model construction unit 1303 is configured to construct a dynamic prediction model, in which the dynamic prediction model includes market quantity and price before the transaction price is in a falling state, and state features and external information in the falling state as input features, and includes the set risk level and the falling probability as output features.

[0077] The computing unit 1304 is responsive to the output features for calculating the associated impact of the trading volume on the trading price in the case of different risk levels;

[0078] In this embodiment, the external information includes policy shocks and emergencies;

[0079] It should be noted that the price dimension includes the continuous falling duration, the falling amplitude and the falling rate;

[0080] The proportion of the trading volume to the amount dimension includes the cumulative trading volume in the falling period, the proportion of the falling trading volume to the total trading volume, and the proportion of the falling trading volume to the total trading volume;

[0081] It should be noted that the trading volume (such as the trading amount) in the price falling period is obtained according to the risk level;

[0082] The features of the price falling state are extracted, including the price dimension (such as the falling amplitude slope) and the proportion of the trading volume to the amount dimension (the liquidity indicator);

[0083] The market volume and price, the falling features and the external information (such as macroeconomic data) are input, and the risk level and the falling probability are output;

[0084] The associated impact of the trading volume on the price is quantified by the formula, and the price impact (such as the market impact coefficient x asset volatility x trading volume / day average trading volume) is calculated;

[0085] The price trend, liquidity and external information can be combined to improve the explanatory power of the model for complex market environment and avoid single indicator prediction bias;

[0086] The model outputs the risk level and the falling probability, which provides probabilistic basis for transaction decision and enhances the adaptability of the strategy;

[0087] And through the formulaic association of the trading volume and the price, the influence degree of the trading size on the market is clear, and the trading volume distribution is optimized;

[0088] In the computing unit, the associated impact of the trading volume on the trading price in the case of different risk levels is calculated according to the following formula:

[0089] Wherein, Q represents the trading volume, and the trading volume represents the trading volume in the case of a specified risk level; the trading volume includes the number of shares and the number of contracts;

[0090] In the formula, R represents the risk level, C represents the total amount of the account of the investor, P entry represents the entry price, P stopθ represents the leverage ratio, 1 if no leverage is used;

[0091] In this embodiment, an example of this formula is as follows:

[0092] If the total account funds are 100,000, the risk level is 0.01, the entry price is 100, the stop loss price is 95, and no leverage is used, then:

[0093]

[0094] It should be noted that this formula dynamically calculates the optimal transaction volume under the current risk level by the risk level, account fund size, entry / stop loss price, and leverage ratio. The core logic is to convert risk control into transaction size constraints to ensure that the potential loss of a single transaction does not exceed the preset proportion of the total account amount (linked to the risk level);

[0095] In the calculation unit, the correlation between transaction volume and transaction price under different risk levels is also calculated, which is calculated according to the following formula:

[0096] Where ΔP represents the price impact, and the price impact represents the price impact under the specified risk level.

[0097] In the formula, k represents the market impact coefficient, σ represents the asset volatility, Q represents the transaction volume, and V represents the daily trading volume.

[0098] In this embodiment, an example of this formula is as follows:

[0099] If k = 0.3, σ = 0.02 (2% daily volatility), Q = 200 shares, and V = 10,000 shares, then:

[0100] That is, a price change of 0.027%);

[0101] It should be noted that this formula quantifies the instantaneous impact of transactions on prices through market microstructure parameters (impact coefficient, volatility, and daily trading volume). The core logic is that the larger the transaction volume, the more significant the impact on market prices, especially in markets with low liquidity, where large transactions may cause prices to deviate from the reasonable range.

[0102] The first formula calculates the maximum allowed transaction volume from the investor's risk tolerance, ensuring that the loss of a single transaction is controllable.

[0103] The second formula calculates the impact of transaction volume on prices from the perspective of market liquidity, avoiding the destruction of market stability due to excessively large transaction sizes.

[0104] Both form a closed loop through transaction volume - investors set transaction volume according to their risk preferences, while meeting market liquidity constraints (i.e. price impact within an acceptable range), if the transaction volume calculated by the first formula results in too high price impact, the transaction volume or risk level needs to be adjusted to balance risk and liquidity;

[0105] As can be seen, the first formula provides an upper limit constraint for transaction volume (based on account risk), which is a "top-down" capital management rule;

[0106] And the second formula provides a lower limit constraint for transaction volume (based on market impact), which is a "bottom-up" market adaptation rule;

[0107] The first formula focuses on the risk tolerance of investors, ensuring that transaction volume matches account funds;

[0108] The second formula focuses on the impact of market liquidity, ensuring that transaction volume matches market depth;

[0109] Based on the calculated correlation impact, the transaction volume corresponding to the transaction price is divided into △1, △2,..., △ n n represents the nth transaction volume, and each transaction volume is trained in a scenario, which includes setting different loss amounts according to the divided transaction volume, the loss amount is the transaction volume of the last investor causing monetary loss in the target market before reaching the specified risk level, the transaction volume is marked as a reference transaction volume, the median of the reference transaction volume is obtained, if the investor trades in the target market at a future time, if the transaction volume corresponding to the transaction is the same as the median, the difference between the current transaction price and the highest price is obtained, if it is not lower than the highest price, the system does not perform risk prompt; otherwise, it performs risk prompt;

[0110] If no risk prompt is performed, the amount corresponding to the difference is obtained, and at least three critical values are preset based on the amount, and the first critical value accounts for half of the amount; if the difference between the transaction price and the highest price at a future time is higher than the first critical value, the system performs risk prompt;

[0111] After the system performs risk prompt, the remaining amount is obtained, the amount is divided according to the proportions of 20%, 50% and 30%, the amount corresponding to 20% is set as the first amount, if the investor uses the first amount to trade after the system performs risk prompt, the difference between the transaction price after the transaction and the highest price is obtained, if the difference is an expanding change, the system performs risk prompt; otherwise, it does not perform risk prompt;

[0112] The transaction volume is divided into multiple levels, and the corresponding loss amount (reference transaction volume median) of each level is set. If the actual transaction volume is consistent with the median, it is determined whether to prompt the risk according to the difference between the price and the highest price;

[0113] If the risk prompt is not triggered, the system presets multiple critical values (such as 50% and 20% layers) based on the difference amount. If the difference exceeds the first critical value (50% of the amount), the prompt is triggered;

[0114] After the risk prompt, the system allocates the remaining amount by 20%, 50%, and 30%. If the investor uses the first amount (20%) and the price difference expands, the risk is prompted again;

[0115] In this embodiment, by referring to the median of the transaction volume and the critical value layering, the maximum potential loss of a single transaction is limited, and the large fund withdrawal under extreme market conditions is avoided;

[0116] And the secondary verification mechanism forms a "trigger-verification-adjustment" closed loop to prevent misoperation caused by a single signal and improve the accuracy of risk control;

[0117] And the 20%-50%-30% layered fund management balances offensiveness and defensiveness, and optimizes the allocation efficiency of funds in a volatile market.

[0118] Based on the above, the application realizes the precise optimization and risk control of the quantitative trading strategy through dynamic data adaptation, quantitative risk grading, multi-feature prediction, and layered fund management, and improves the adaptability and income stability of the strategy in complex market environment.

[0119] The embodiment combines the above-mentioned deep learning-based quantitative trading strategy dynamic optimization system, and further proposes a working method of the system, as follows:

[0120] S10: Collect data of a target market invested by an investor from a data source; the data source includes a stock exchange and a financial media;

[0121] S20: Set the frequency of collecting data from the data source according to the trading demand of the investor, including dividing the trading demand into high-frequency trading demand and medium-frequency trading demand according to the trading demand; in the trading demand corresponding to the high-frequency trading demand, the frequency of collecting data from the data source is set to real-time; in the trading demand corresponding to the medium-frequency trading demand, the frequency of collecting data from the data source is set to non-real-time; non-real-time includes collecting data from the data source once every 5-10 minutes;

[0122] S30: Risk setting is performed on the changes of the collected data in the trading demand; including setting low risk, medium risk, and high risk;

[0123] S40: Obtain the transaction volume of the preset period according to the set risk, the transaction volume including the transaction amount, and the preset period being a period when the transaction price of the target market is in a falling state;

[0124] S50: Extract state features of the transaction price in the falling state; the state features including a price dimension and a proportion of the trading volume to the amount dimension;

[0125] S60: Construct a dynamic prediction model, in which market volume and price before the transaction price is in the falling state and state features of the falling state and external information are included as input features, and the set risk level and the falling probability are included as output features;

[0126] S70: According to the output features, calculate the associated influence of the transaction volume on the transaction price under different risk levels.

[0127] In summary, the present application realizes the guarantee of the quantitative trading strategy in risk control, income optimization and fund management through five core mechanisms of dynamic data adaptation, quantitative risk grading, multi-feature fusion prediction, scenario-based loss control and price impact quantification.

[0128] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A dynamic optimization system for quantitative trading strategies based on deep learning, characterized in that, include: The data acquisition module is used to collect data from data sources about the target markets that investors invest in; The data sources include stock exchanges and financial media; The data collection frequency setting module is used to set the frequency of data collection from the data source according to the investor's trading needs. The data collection frequency setting module includes a differentiation unit and a data setting unit. The differentiation unit is used to differentiate the transaction demand into high-frequency transaction demand and medium-frequency transaction demand according to the transaction demand; in the transaction demand corresponding to the high-frequency transaction demand, the frequency of data collection from the data source is set to real-time; in the transaction demand corresponding to the medium-frequency transaction demand, the frequency of data collection from the data source is set to non-real-time; the non-real-time includes collecting data from the data source once every 5 to 10 minutes. The data setting unit responds to the transaction requirements corresponding to the high-frequency trading requirements and is used to set risk for changes in the collected data in the transaction requirements. This includes setting low risk, medium risk, and high risk; The data fusion processing module includes a transaction volume segmentation unit, a feature extraction unit, a model building unit, and a calculation unit. The trading volume segmentation unit is used to obtain the trading volume for a preset period based on the set risk. The trading volume includes the trading amount, and the preset period is the period when the trading price of the target market is in a downward trend. The feature extraction unit responds to the preset time period and is used to extract the state features of the transaction price in a falling state; the state features include the price dimension and the proportion of transaction volume to amount dimension.

2. The deep learning-based dynamic optimization system for quantitative trading strategies as described in claim 1, characterized in that, In the data setting unit, the risk setting step includes: The relevant data that enables the investor to obtain profits is collected from the trading demand corresponding to the high-frequency trading demand, and the relevant data includes the highest and lowest prices of the target market; A floating range is set between the highest price and the lowest price, allowing the highest price to fluctuate towards the lowest price. This floating range includes an upward floating range and a downward floating range. Within the downward floating range, a floating value is set for the highest price, and the floating value is calculated every 20 to 30 seconds. A preset risk threshold is set for the floating value, which includes being 3% below the maximum price, with the 3% corresponding to the low risk.

3. The deep learning-based dynamic optimization system for quantitative trading strategies as described in claim 1, characterized in that, The model building unit is used to build a dynamic prediction model. The dynamic prediction model includes market volume and price before the transaction price is in a downward state, the state characteristics of the downward state, and external information as input features, and includes the set risk level and the probability of decline as output features. The computing unit responds to the output feature to calculate the correlation between trading volume and trading price under different risk levels.

4. The deep learning-based dynamic optimization system for quantitative trading strategies as described in claim 3, characterized in that, In the calculation unit, the correlation between trading volume and trading price under different risk levels is calculated according to the following formula: Where Q represents the trading volume, which is the trading volume under a specified risk level; the trading volume includes the number of shares and the number of contracts. In the formula, R represents the risk level, C represents the total amount in the investor's account, and P... entry P represents the entry price. stop The stop-loss price is represented by θ, which represents the leverage ratio. If no leverage is used, the value is 1.

5. The deep learning-based dynamic optimization system for quantitative trading strategies as described in claim 4, characterized in that, The calculation unit calculates the correlation between trading volume and trading price under different risk levels, and also includes calculating it according to the following formula: Wherein, ΔP represents a price shock, which is a price shock under a specified risk level; In the formula, k represents the market shock coefficient, σ represents asset volatility, Q represents trading volume, and V represents average daily trading volume.

6. The deep learning-based dynamic optimization system for quantitative trading strategies as described in any one of claims 4 to 5, characterized in that, Based on the calculated associated impact, the trading volume corresponding to the trading price is categorized within the specified risk level. n represents the nth distinct trading volume, and scenario training is performed on each distinct trading volume. The scenario training includes setting different loss amounts based on the distinct trading volume. The loss amount is the trading volume of the last investor who caused a loss in the target market before reaching the specified risk level. The trading volume is marked as a reference trading volume, and the median of the reference trading volume is obtained. If the investor trades in the target market in the future, and the trading volume corresponding to the trade is the same as the median, the difference between the current trading price and the highest price is obtained. If it is not lower than the highest price, the system does not issue a risk warning. Conversely, a risk warning will be issued.

7. The deep learning-based dynamic optimization system for quantitative trading strategies as described in claim 6, characterized in that, If no risk warning is issued, the system obtains the amount corresponding to the difference and presets at least three threshold values ​​based on the amount, with the first threshold value accounting for half of the amount; if the difference between the transaction price and the highest price in the future is higher than the first threshold value, the system issues a risk warning.

8. The deep learning-based dynamic optimization system for quantitative trading strategies as described in claim 7, characterized in that, After the system issues a risk warning, the remaining amount is obtained and divided into 20%, 50% and 30% proportions. The amount corresponding to the 20% is set as the first amount. If the investor uses the first amount to trade after the system issues a risk warning, the difference between the transaction price after the transaction and the highest price is obtained. If the difference increases, the system issues a risk warning. Conversely, no risk warning will be given.

9. A method applied to the dynamic optimization system for quantitative trading strategies based on deep learning as described in claim 1, characterized in that, Includes the following steps: Data is collected from data sources to identify the target markets in which investors invest. The data sources include stock exchanges and financial media; The frequency of data collection from the data source is set according to the investor's trading needs, including classifying the trading needs into high-frequency trading needs and medium-frequency trading needs; in the trading needs corresponding to the high-frequency trading needs, the frequency of data collection from the data source is set to real-time; in the trading needs corresponding to the medium-frequency trading needs, the frequency of data collection from the data source is set to non-real-time; the non-real-time setting includes collecting data from the data source once every 5 to 10 minutes. Risk settings are applied to changes in the collected data within the aforementioned transaction requirements; This includes setting low risk, medium risk, and high risk; The trading volume for a preset period is obtained based on the set risk level. The trading volume includes the trading amount. The preset period is the period when the trading price of the target market is in a downward trend. Extract the state characteristics of the transaction price in a downward trend; the state characteristics include the price dimension and the proportion of transaction volume to the amount dimension; A dynamic prediction model is constructed, which includes taking the market volume and price before the transaction price is in a downward state, the state characteristics of the downward state, and external information as input features, and taking the set risk level and the probability of decline as output features. Based on the output characteristics, the correlation between trading volume and trading price under different risk levels is calculated.

Images