Intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filter

Abstract

This intelligent investment advisory asset allocation adaptive system, based on multi-source fusion Kalman filtering, relates to the field of asset allocation technology. It generates behavioral labels from individual behavioral data, inputs this data into a preset first filter to generate a first behavioral state vector, calculates a behavioral resonance index based on the first behavioral state vector and the market state vector, inputs the data into a preset second filter to generate a second behavioral state vector, updates the baseline allocation ratio for preset asset classes and the floating allocation ratio around the baseline ratio, and generates an asset allocation ratio. The system also updates the calculation parameters of the behavioral resonance index based on the statistical characteristics of the behavioral resonance intensity over multiple consecutive preset monitoring periods. This system can significantly improve the personalization level of asset allocation and the effectiveness of risk control.

Description

Technical Field

[0001] This application relates to the field of asset allocation technology, specifically to an intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering. Background Technology

[0002] Traditional asset allocation typically employs static risk assessment models, which struggle to respond in real-time to rapid market changes. This results in investor risk tolerance assessments lagging behind actual market conditions. In particular, current technology lacks the ability to precisely identify investor motivations, failing to effectively distinguish between planned cash flows reflecting long-term financial planning and emotional trading behaviors influenced by short-term market sentiment.

[0003] This makes asset allocation benchmarks frequently subject to irrational market fluctuations. Market practice shows that when investor sentiment is highly synchronized with macro market fluctuations, existing systems cannot identify this resonance phenomenon in a timely manner and implement forward-looking interventions. This often leads to the failure of asset allocation strategies during periods of sharp market volatility, exacerbating the potential risk exposure of portfolios. To address these issues, existing technologies urgently need improvement. Summary of the Invention

[0004] The purpose of this application is to provide an intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering, which can respond to market changes in real time, accurately distinguish investors' behavioral motivations, and establish a dynamic correlation mechanism between micro-behavior and macro market, thereby significantly improving the personalization level of asset allocation and the effectiveness of risk control.

[0005] The objective of this application can be achieved through the following technical solution: Firstly, an intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering includes the following modules:

[0006] The data acquisition module is used to simultaneously acquire personal behavior data and macro market data within a preset monitoring period;

[0007] The first filtering module is used to generate corresponding behavior labels for the personal behavior data, input the personal behavior data carrying the behavior labels into a preset first filter, output a first feature parameter representing the short-term behavior state, and generate a first behavior state vector.

[0008] The behavioral resonance module is used to generate a market state vector representing the market state based on the macro market data, and to calculate a behavioral resonance index representing the degree of matching between the first behavioral state vector and the market state vector.

[0009] The second filtering module is used to input personal behavior data carrying behavior tags into a preset second filter to output a second feature parameter representing the long-term behavior state and generate a second behavior state vector.

[0010] The configuration adjustment module is used to update the configuration benchmark ratio for a preset asset category according to the second behavior state vector, update the configuration floating ratio around the configuration benchmark ratio according to the behavior resonance index and the first behavior state vector, and generate the corresponding asset configuration ratio.

[0011] The feedback optimization module is used to obtain the behavioral resonance intensity within a preset monitoring period based on the behavioral resonance index, and to update the calculation parameters of the behavioral resonance index based on the statistical characteristics of the behavioral resonance intensity within multiple consecutive preset monitoring periods.

[0012] Secondly, the adaptive asset allocation method for intelligent investment advisors based on multi-source fusion Kalman filtering includes the following steps:

[0013] Simultaneously acquire personal behavior data and macro market data within a preset monitoring period;

[0014] Generate corresponding behavior labels for the personal behavior data, input the personal behavior data carrying the behavior labels into a preset first filter, output a first feature parameter representing the short-term behavior state, and generate a first behavior state vector.

[0015] A market state vector representing the market state is generated based on the macro market data, and a behavioral resonance index representing the degree of matching between the first behavioral state vector and the market state vector is calculated based on the first behavioral state vector and the market state vector.

[0016] Personal behavioral data carrying behavioral tags is input into a preset second filter to output a second feature parameter representing the long-term behavioral state and generate a second behavioral state vector.

[0017] The configuration benchmark ratio for the preset asset category is updated according to the second behavior state vector, the configuration floating ratio around the configuration benchmark ratio is updated according to the behavior resonance index and the first behavior state vector, and the corresponding asset configuration ratio is generated.

[0018] The behavioral resonance intensity within a preset monitoring period is obtained based on the behavioral resonance index, and the calculation parameters of the behavioral resonance index are updated based on the statistical characteristics of the behavioral resonance intensity within multiple consecutive preset monitoring periods.

[0019] Thirdly, a computer storage medium stores computer-executable instructions, which, when executed, implement the intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering described in the first aspect.

[0020] Compared with the prior art, the beneficial effects of this application are:

[0021] This application integrates data acquisition, filtering, behavioral resonance analysis, asset allocation adjustment, and optimization feedback modules to achieve real-time monitoring and dynamic matching of investor behavior and market conditions. It can effectively distinguish between planned and emotional trading, establish a correlation mechanism between micro and macro data, and optimize configuration parameters in a closed loop based on historical data. It has the advantages of real-time response to market changes, accurate identification of behavioral motivations, and dynamic adjustment of asset allocation. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the module of the intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering in this application;

[0023] Figure 2 This is a schematic diagram illustrating the steps of the intelligent investment advisory asset allocation adaptive method based on multi-source fusion Kalman filtering in this application. Detailed Implementation

[0024] The technical solutions 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, and not all embodiments. The components described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but only to illustrate selected embodiments of this application.

[0025] Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item has been defined in one figure, it does not need to be further defined and explained in subsequent figures. The terms "first", "second", etc. are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0026] Traditional robo-advisory systems face multiple technical challenges in asset allocation. Specifically, the system cannot dynamically capture the real-time evolution of investors' risk tolerance as market conditions change, driven by external market fluctuations. Simultaneously, the system lacks a precise mechanism to distinguish investor motivations, leading to the mixing of planned and emotional behaviors in data processing, thus allowing irrational trading to interfere with the allocation benchmark. Furthermore, the failure to establish a closed-loop integration path between micro-level behavioral data and macro-level market conditions prevents personalized and adaptive optimization of allocation strategies. These issues directly result in decreased benchmark stability and insufficient accuracy in state-based responses during asset allocation decisions.

[0027] During periods of sudden market volatility, investors may execute multiple outflows of funds due to short-term sentiment. The system may misidentify these emotional outflows as a shift in long-term risk appetite, leading to erroneous adjustments to asset allocation ratios. In this scenario, the lack of behavioral labeling prevents the effective identification of trading motives, and the absence of a dual-state vector estimation mechanism causes short-term sentiment and long-term discipline to become intertwined during data processing, ultimately causing the allocation strategy to deviate from the investor's true behavioral patterns. Specifically, this problem manifests as the system continuously outputting allocation plans that do not match the investor's actual needs, causing unnecessary fluctuations in asset allocation ratios. If these issues are not resolved, the system will continue to generate allocation benchmarks that deviate from the investor's true behavioral characteristics, resulting in unreasonable deviations in asset allocation ratios during market volatility. This will weaken the system's risk control capabilities, potentially leading to unexpected losses in asset value, and reducing the reliability and applicability of robo-advisory services in dynamic market environments.

[0028] Therefore, this application provides an adaptive asset allocation system for intelligent investment advisory services based on multi-source fusion Kalman filtering, such as... Figure 1 As shown, it includes the following modules:

[0029] The data acquisition module is used to simultaneously acquire personal behavior data and macro market data within a preset monitoring period;

[0030] The first filtering module is used to generate corresponding behavior labels for the personal behavior data, input the personal behavior data carrying the behavior labels into a preset first filter, output a first feature parameter representing the short-term behavior state, and generate a first behavior state vector.

[0031] The behavioral resonance module is used to generate a market state vector representing the market state based on the macro market data, and to calculate a behavioral resonance index representing the degree of matching between the first behavioral state vector and the market state vector.

[0032] The second filtering module is used to input personal behavior data carrying behavior tags into a preset second filter to output a second feature parameter representing the long-term behavior state and generate a second behavior state vector.

[0033] The configuration adjustment module is used to update the configuration benchmark ratio for a preset asset category according to the second behavior state vector, update the configuration floating ratio around the configuration benchmark ratio according to the behavior resonance index and the first behavior state vector, and generate the corresponding asset configuration ratio.

[0034] The feedback optimization module is used to obtain the behavioral resonance intensity within a preset monitoring period based on the behavioral resonance index, and to update the calculation parameters of the behavioral resonance index based on the statistical characteristics of the behavioral resonance intensity within multiple consecutive preset monitoring periods.

[0035] The aforementioned technical solution, by introducing a multi-source fusion Kalman filter mechanism, achieves deep insights into investor behavior and adaptive adjustments to asset allocation, representing a significant technological advancement compared to existing technologies. For example, existing robo-advisory systems may rely solely on user-completed risk questionnaires to determine their risk tolerance; this static assessment method cannot dynamically capture an investor's true risk tolerance as market conditions change. However, this system can accurately identify and distinguish between investors' short-term emotional fluctuations and long-term financial discipline, thereby constructing a unique two-tiered allocation framework for each investor that reflects their true behavioral patterns.

[0036] Furthermore, existing systems often struggle to distinguish between "planned behavior" representing long-term financial discipline and "emotional behavior" driven by short-term market sentiment when allocating assets, making the allocation benchmark susceptible to irrational trading. This system effectively differentiates between these two types of behavior by generating behavioral labels for individual behavioral data and inputting them into preset first and second filters. This results in more stable updates to the allocation benchmark and more flexible and targeted adjustments to the floating ratio.

[0037] Furthermore, this system innovatively introduces a behavioral resonance module, which quantifies the coupling degree between individual sentiment and overall market sentiment by calculating a behavioral resonance index. For example, when the behavioral resonance index shows a high degree of resonance between user sentiment and market sentiment, the allocation adjustment module can automatically apply reverse constraints, such as tightening the upper limit of stock fluctuation ratios, thereby achieving proactive risk intervention. The feedback optimization module adaptively updates the calculation parameters of the behavioral resonance index based on the statistical characteristics of the behavioral resonance intensity, forming a closed-loop optimization of the allocation strategy, significantly improving the personalization and adaptability of the strategy. As a result, this system can more accurately identify risks, more effectively allocate assets, and provide investors with more valuable intelligent investment advisory services.

[0038] It should be further explained that, in the specific implementation process, the process of simultaneously acquiring personal behavior data and macro market data within the preset monitoring period includes:

[0039] The personal behavior data refers to an ordered set of data that reflects the trading behavior, fund flow and other related operations of investors or account holders, obtained at continuous time points or within a preset monitoring period (such as daily, weekly or monthly); the macro market data refers to multi-dimensional indicator data that reflects the overall operation of the financial market, the macroeconomic environment and related market sentiment and liquidity.

[0040] The personal behavior data includes: transaction timestamps, transaction amounts, transaction frequency, fund inflow / outflow direction, trading instruments, changes in holdings, login frequency, query behavior, etc.; data sources typically include user trading terminal logs, account transaction records provided by brokerages or custodians, fund flow files, and user interaction behavior tracking data within the investment advisory platform; the data collection frequency is consistent with the preset monitoring period to ensure coverage of all key behavioral events within a complete monitoring period, and the data is arranged and stored according to a unified timeline for subsequent behavioral tag generation and state vector construction;

[0041] For example, the system periodically retrieves the latest transaction data from brokerage trading systems or fund custody platforms via pre-defined data interfaces, analyzing information such as transaction time, amount, direction, and underlying asset code. Simultaneously, it collects behavioral data from user terminals or service platforms, recording user clicks on specific information, the frequency and duration of position inquiries, and interactions related to risk assessments. All personal behavioral data undergoes preliminary cleaning and formatting after collection, and is then segmented and archived according to a pre-defined monitoring cycle.

[0042] The macroeconomic market data includes: daily or cyclical returns and volatility of market indices (such as stock indices and bond indices), sector performance, market liquidity indicators (such as trading volume and turnover rate), market sentiment indicators (such as the VIX volatility index and changes in margin trading balance), and macroeconomic indicators (such as interest rates, inflation rates, and PMI). Data sources typically include publicly available financial market databases, market data released by stock exchanges, indicator data released by macroeconomic statistics agencies, and sentiment indices calculated based on text data such as news and social media.

[0043] For example, basic market data such as opening price, closing price, highest price, lowest price, and trading volume of major market indices can be periodically acquired via financial data APIs during the monitoring period, and then their returns and volatility can be calculated. Simultaneously, cross-asset information such as changes in the bond market yield curve, exchange rate fluctuations in the foreign exchange market, and commodity price indices can be collected. Furthermore, a text sentiment index calculated using natural language processing technology can be introduced to reflect the overall sentiment of market participants. All macroeconomic market data must undergo time alignment and normalization after collection, and be unified to the same monitoring period timestamp as individual behavioral data to ensure time consistency when subsequently calculating market state vectors.

[0044] In other embodiments, this application further proposes specific implementation methods for generating behavioral labels from personal behavior data, as well as the specific type and processing range of the first filter in the first filtering module. Specifically, the generation of behavioral labels is achieved through a supervised learning classifier. The transaction time and amount extracted from personal behavior data, and the market volatility extracted from macroeconomic market data, are input into the supervised learning classifier to output behavioral labels corresponding to the personal behavior data, including planned inflows, event-driven inflows, and emotional outflows. The first filter is a Kalman filter with a first process noise covariance. Personal behavior data carrying event-driven inflows and emotional outflows are input into the first filter to output first feature parameters containing emotional sensitivity and impulsive tendency values. The first feature parameters are then sequentially combined into a first behavioral feature vector.

[0045] Supervised learning classifiers are machine learning models that learn from training data with known labels to predict the category label of new data, thereby classifying data points. These classifiers can employ various models, such as support vector machines, decision trees, or random forests. They can be fed with personal behavioral data features like transaction time and amount, as well as macroeconomic market data features like market volatility, to help the classifier accurately identify the intrinsic motivations and nature of user behavior.

[0046] The first filter is defined as a Kalman filter with a first process noise covariance. A Kalman filter is a highly efficient recursive filter used to optimally estimate the state of a dynamic system in the presence of noise, combining a system model and measurement data. The process noise covariance is a key parameter in the Kalman filter, reflecting the degree of influence of uncertainties in the system model or unmodeled external disturbances on the system state. It can be initially set based on historical data analysis or expert experience, or an adaptive Kalman filtering method can be used to dynamically adjust it based on actual observation data during system operation.

[0047] Personal behavioral data carrying both event-driven inflows and emotional outflows is input into the first filter, clarifying the data types processed by the first filter. Event-driven inflows and emotional outflows typically represent irrational or short-term behaviors of users driven by specific events or influenced by emotions. After acquiring personal behavioral data, behavioral labels are generated through a supervised learning classifier. The system can then filter the data based on these labels, directing only data streams labeled "event-driven inflows" and "emotional outflows" to the first filter for processing.

[0048] The emotional sensitivity and impulsive tendency values ​​output by the first filter are key indicators characterizing the user's short-term behavioral state. Emotional sensitivity reflects the degree to which a user is affected by market sentiment fluctuations, while impulsive tendency measures the user's tendency to react irrationally and immediately during decision-making. The Kalman filter can be designed to estimate a state vector whose elements are the emotional sensitivity and impulsive tendency values, and these estimates are dynamically updated by fusing observation data. Combining emotional sensitivity and impulsive tendency values ​​into the first behavioral feature vector is to form a unified, structured data representation, facilitating processing and analysis by subsequent modules.

[0049] For a single preset monitoring cycle Extract all transaction records marked as emotional outflows, for each transaction Record its transaction time and transaction amount Simultaneously, obtain sentiment indicators from macro market data within the same monitoring period. For each emotional outflow trade, find the standard deviation of the sentiment indicator around the time of the trade (e.g., the previous 30 minutes). The emotional impact value of a transaction is calculated by multiplying the transaction amount by its standard deviation. ; Obtain the sum of the emotional impact values ​​of all emotional outflow transactions within the monitoring period. ,Will Divide by the total assets of the user during the monitoring period The emotional sensitivity during the monitoring period was obtained by normalization. ;

[0050] During the monitoring period Within this scope, identify all event-driven inflows and sentiment-driven outflows. Calculate the cumulative number of these transactions. and cumulative amount Calculate the cumulative amount of planned inflow transactions within the same monitoring period. The impulsive tendency value ,in This refers to the total number of transactions during the monitoring period. All are preset weighting coefficients.

[0051] In some other implementations, this application further proposes to extract multi-dimensional market characteristic indicators based on macro market data, including sentiment indicators representing market sentiment state, volatility indicators representing market volatility state, and liquidity indicators representing market liquidity state. After normalization, each market characteristic indicator is combined in sequence into a corresponding market state vector.

[0052] Macroeconomic market data refers to economic, policy, and social data that influences the overall performance of the financial market. This data can be obtained through publicly available financial data interfaces or by using web scraping techniques to retrieve relevant reports and data from official statistical agencies' websites and then processing them in a structured manner. The extracted multi-dimensional market characteristic indicators are derived from the processed and analyzed macroeconomic market data. They are used to quantify the numerical values ​​describing different market attributes, aiming to capture the complexity of market behavior.

[0053] Sentiment indicators are quantitative measures of market investor optimism or pessimism, reflecting market participants' psychological expectations and risk appetite. They are calculated or obtained directly from market trading data (such as the put / call ratio and the VIX fear index). Volatility indicators measure the degree of price fluctuations over a period of time, reflecting market uncertainty and risk levels. They are obtained by calculating the standard deviation of returns of assets such as stock indices, commodity prices, or exchange rates, as well as historical volatility or implied volatility. Liquidity indicators measure the ease with which an asset can be quickly bought or sold without significantly affecting its price, reflecting market activity and depth. They are measured by calculating market volume, bid-ask spreads, impact costs, and market depth.

[0054] The sentiment index Monitoring cycle The normalized value of the put / call ratio (CPR) within the period, where CPR = (total call option volume / total put option volume), and the volatility indicator uses the monitoring period. The annualized realized volatility of the intraday return of the benchmark index (such as the CSI 300) is the normalized value. The liquidity indicator adopts the monitoring period. The normalized value of the average daily turnover rate of all stocks in the domestic market.

[0055] In other embodiments, this application further proposes weighted summation of each element in the aforementioned first-beginning state vector to obtain a sentiment index for a corresponding preset monitoring period, and sequentially combining the sentiment indices from multiple consecutive preset monitoring periods into a sentiment index sequence. Simultaneously, weighted summation of each element in the aforementioned market state vector to obtain a market index for a corresponding preset monitoring period, and sequentially combining the market indices from multiple consecutive preset monitoring periods into a market index sequence. Based on this, the Pearson correlation coefficient between the sentiment index sequence and the market index sequence is obtained. Combined with preset adjustment parameters Obtaining the Behavioral Resonance Index And it is used as the behavioral resonance index of the latest preset monitoring cycle in the multiple consecutive preset monitoring cycles, and the value range of the preset adjustment parameter is [0.5, 1.5].

[0056] Specifically, a weighted summation of each element within the first behavioral state vector is performed to obtain the corresponding sentiment index for a preset monitoring period. This aims to condense multi-dimensional short-term behavioral state information into a single, quantifiable sentiment indicator. This is achieved by assigning preset weights to each element (emotional sensitivity, impulsive tendency value) in the first behavioral state vector and performing a linear combination. The sentiment indices from multiple consecutive preset monitoring periods are sequentially combined into a sentiment index sequence to capture the dynamic trend of individual emotional states over time, providing a foundation for subsequent time series analysis. This sequence is constructed using a fixed-length sliding window method, meaning that the latest sentiment index is added and the oldest index is removed with each update.

[0057] Similarly, weighted summation of the elements within the market state vector to obtain the market index for the corresponding preset monitoring period integrates complex macroeconomic market data (sentiment indicators, volatility indicators, liquidity indicators) into a comprehensive market state indicator. This can also be achieved through weighted summation with preset weights or through statistical methods such as principal component analysis. Combining market indices from multiple consecutive preset monitoring periods into a market index sequence aims to construct a time series of market states for comparative analysis with the sentiment index sequence. Obtaining the Pearson correlation coefficient between the sentiment index sequence and the market index sequence is a standard statistical method for quantifying the degree of linear correlation between the two. Combining preset adjustment parameters to obtain the behavioral resonance index maps the Pearson correlation coefficient to a more interpretable indicator within the range [0, 1]. The introduction of preset adjustment parameters allows the system to adjust the sensitivity of the behavioral resonance index to the correlation coefficient according to actual needs.

[0058] In some other embodiments, this application further proposes that the second filter is a Kalman filter with a second process noise covariance, the second process noise covariance being smaller than the first process noise covariance. Personal behavior data carrying planned inflows are input into the second filter to output a second feature parameter containing discipline intensity and long-term risk preference value, and the second feature parameter is sequentially combined into a second behavioral feature vector.

[0059] The second filter also employs a Kalman filter. The second process noise covariance reflects the uncertainty of the system model or the degree of random fluctuation in the system state over time. A smaller process noise covariance indicates that the system state changes are relatively stable or predictable; the filter will place more trust in its own predictions and respond more slowly to new measurement data, thus achieving a smoother estimate. Conversely, a larger process noise covariance indicates that the system state changes drastically or is unpredictable; the filter will place more trust in new measurement data and respond more quickly to state estimation. The second process noise covariance is smaller than the first process noise covariance, indicating that the second filter, when processing its corresponding data, has a lower expectation of inherent changes or uncertainties in the system state. This means that the second filter is designed to provide a smoother and more stable state estimate, reducing sensitivity to short-term fluctuations.

[0060] The planned inflow of personal behavioral data refers to users' clearly planned and purposeful fund inflows. This type of behavior typically exhibits high stability and consistency, reflecting the user's deep-seated investment philosophy and financial planning. This data is input into the second filter as the measurement input for the second Kalman filter, used to update and estimate the user's long-term behavioral status. The discipline strength reflects the user's ability to adhere to a predetermined plan and resist short-term market temptations or panic during the investment process. The long-term risk preference value measures the user's attitude and tolerance towards risk over a longer time horizon.

[0061] The second feature parameter is a key indicator representing the user's long-term behavioral status, estimated by the second filter based on planned inflow data. Combining these second feature parameters sequentially into a second behavioral feature vector means arranging multiple second feature parameters, such as discipline intensity and long-term risk preference, into a corresponding vector according to a preset order or structure. This vector comprehensively and quantitatively describes the user's behavioral characteristics in long-term investment, providing a basis for subsequent asset allocation adjustments.

[0062] Set a longer evaluation window (e.g., the past 12 months). Statistics on all planned investment instructions set by the user (e.g., monthly investment of Y yuan on the Xth day). Calculation of the actual number of executions. With planned number of times The ratio as the execution rate Calculate the planned inflow amount for each actual execution. With the planned amount deviation Calculate the average of all deviations within the evaluation window L. The intensity of discipline ,in, All are preset weighting coefficients.

[0063] Get the past The month-end allocation ratio of user accounts in major asset classes (such as stocks, bonds, and cash) for each monitoring period (e.g., 24 months). The standard deviation of the time series of month-end allocation ratios for equity assets is obtained. Based on the historical returns of each asset class, the monthly return volatility of the user's historical asset portfolio is obtained. The long-term risk preference value .

[0064] In other embodiments, this application further proposes that the preset asset category refers to a pre-divided basic asset classification representing different market risk and return, and the allocation benchmark ratio refers to the asset allocation ratio preset for different preset asset categories; the risk level within the corresponding preset monitoring period is obtained according to the preset mapping rule and the long-term risk preference value in the second behavioral feature vector, including conservative, stable, and aggressive.

[0065] For different preset asset classes under different risk levels, corresponding basic allocation ratios are set, based on the discipline intensity in the second behavioral feature vector. Obtain the attenuation coefficient within the corresponding preset monitoring period. ; Obtain the updated configuration benchmark ratio for a single preset asset class within the corresponding preset monitoring period. ,in, A constant greater than 0 The default asset class configuration baseline ratio before the update This is the basic allocation ratio for the preset asset class under the risk level within the corresponding preset monitoring period.

[0066] The preset asset categories refer to pre-defined basic asset classifications representing different market risk and return profiles. These categories can be divided based on dimensions such as asset liquidity, risk level, and return potential, such as stocks, bonds, money market instruments, and real estate investment trusts (REITs). The preset mapping rule is used to convert long-term risk preference values ​​into corresponding risk levels. This rule uses a piecewise function or threshold judgment method. For example, when the long-term risk preference value is less than a first preference threshold, it is mapped to a conservative risk level; when it is greater than a second preference threshold, it is mapped to an aggressive risk level; when it is between the two preference thresholds (inclusive), it is mapped to a moderate risk level. The first preference threshold is less than the second preference threshold.

[0067] The risk level is a classification of investors' risk tolerance, including conservative, moderate, and aggressive. The base allocation ratio refers to the initial allocation ratio set for different preset asset classes under different risk levels. These ratios can be based on industry standards or internal research of financial institutions to pre-set a typical asset allocation portfolio for each risk level, for example: conservative: 20% stocks, 70% bonds, 10% cash; moderate: 40% stocks, 50% bonds, 10% cash; aggressive: 70% stocks, 20% bonds, 10% cash. The decay coefficient is a coefficient used to adjust the update speed of the allocation benchmark ratio, determined by the discipline intensity.

[0068] In other embodiments, this application further proposes a floating allocation ratio, which refers to a dynamic adjustment boundary that fluctuates around a preset asset class's allocation benchmark ratio, based on a behavioral resonance index within the same preset monitoring period. and sentiment index Adjust the preset basic fluctuation ratio within the corresponding preset monitoring period. ;

[0069] when At that time, the upper limit adjustment factor is obtained. ,when At that time, the upper limit adjustment factor will be adjusted. ,when At that time, the lower limit adjustment factor was obtained. ,when When, the lower limit adjustment factor is... ;in, Both are preset behavioral resonance thresholds. The first shrinkage coefficient is preset, and its value range is (0, 1].

[0070] According to the aforementioned sentiment index Obtain global adjustment factor ,in, The second contraction coefficient is preset, and the configured floating ratio includes an upper limit floating ratio. and lower limit fluctuation ratio Combined with the updated configuration benchmark ratio for a single preset asset category within the corresponding preset monitoring period Generate the asset allocation ratio for the preset asset class. ].

[0071] The asset allocation float ratio refers to the dynamic adjustment boundary that fluctuates around a pre-defined benchmark allocation ratio for asset classes. It is not a fixed value, but rather a range adjusted based on real-time changes in market conditions and individual behavior. Its purpose is to provide a degree of flexibility in asset allocation, allowing it to respond flexibly to short-term market sentiment and individual behavioral fluctuations while maintaining long-term strategy stability. For example, when market sentiment is high or individual behavior exhibits excessive optimism, the float ratio can be appropriately narrowed to avoid over-investment; conversely, when market sentiment is low or individual behavior exhibits excessive pessimism, the float ratio can be appropriately widened to capture potential investment opportunities.

[0072] The Behavioral Resonance Index characterizes the degree of alignment between an individual's short-term behavioral state and the macroeconomic market state, reflecting whether individual behavior resonates with market trends. The Sentiment Index, on the other hand, reflects an individual's emotional state within a specific monitoring period. These two indicators are used together to adjust a preset base fluctuation ratio. For example, a high Behavioral Resonance Index may indicate a high degree of consistency between individual behavior and market trends; in this case, the Sentiment Index can be used to further determine whether to tighten or loosen the fluctuation ratio. Conversely, a high Sentiment Index may indicate significant emotional fluctuations, requiring more cautious adjustments to the fluctuation ratio.

[0073] The upper and lower limit adjustment factors are dynamically calculated based on a comparison between the behavioral resonance index and a preset threshold. These two factors are used to non-linearly adjust the base floating ratio. When the behavioral resonance index exceeds the preset threshold range, these factors will contract or expand the floating ratio according to the degree of deviation. For example, when the behavioral resonance index is much higher than the preset threshold... When the upper limit adjustment factor decreases, the upper limit fluctuation ratio is narrowed to limit overly optimistic allocation; when the behavioral resonance index is much lower than... When this happens, the lower limit adjustment factor decreases, thereby narrowing the lower limit fluctuation ratio to avoid overly pessimistic allocation. This mechanism makes the adjustment of the fluctuation ratio more refined and intelligent, and can better adapt to extreme situations in the market and individual behavior.

[0074] The global adjustment factor is calculated based on the sentiment index and a preset second contraction coefficient. This factor is used to make an overall adjustment to the allocation fluctuation ratio, reflecting the impact of individual emotional state on asset allocation flexibility. For example, when the sentiment index is high, it indicates that individual emotions fluctuate greatly. In this case, the global adjustment factor will decrease, thereby narrowing the overall fluctuation ratio to reduce the risk caused by emotional fluctuations. When the sentiment index is low, it indicates that individual emotions are relatively stable. In this case, the global adjustment factor will be close to 1, keeping the allocation fluctuation ratio at a relatively loose level.

[0075] The upper and lower floating ratios are the final calculated dynamic fluctuation ranges used to determine the asset allocation ratio. They are obtained by multiplying the base floating ratio, the upper adjustment factor, the lower adjustment factor, and the global adjustment factor, respectively. The upper floating ratio represents the maximum upward fluctuation allowed from the benchmark ratio, and the lower floating ratio represents the maximum downward fluctuation allowed. The asset allocation ratio is the final generated investment range for a specific preset asset class. This range provides a dynamic investment interval that adapts to changes in the market and individual behavior, guiding the robo-advisory system in specific asset buying and selling operations.

[0076] In other embodiments, this application further proposes obtaining the behavioral resonance intensity within a preset monitoring period based on a behavioral resonance index, and updating the calculation parameters of the behavioral resonance index based on the statistical characteristics of the behavioral resonance intensity within multiple consecutive preset monitoring periods. The behavioral resonance intensity... If personal behavioral data carrying emotional outflow exists within a single preset monitoring period, it is determined that an emotional transaction occurred within that preset monitoring period; otherwise, it is determined that no emotional transaction occurred.

[0077] The mean of the behavioral resonance intensity within each of the preset monitoring periods in which emotional trading occurs is obtained, denoted as . The mean of the behavioral resonance intensity within each preset monitoring period in which no emotional trading occurred is obtained, and is denoted as . Obtain the difference value of the latest preset monitoring period among the multiple consecutive preset monitoring periods. ;

[0078] The preset adjustment parameters in the behavioral resonance index of the latest preset monitoring period are updated based on the difference value, and the preset adjustment parameters before the update are updated. Based on the above, increase and decrease by a preset step size respectively, and obtain the difference value after increasing the preset step size respectively. The difference value after reducing the preset step size ,exist In the process, the preset adjustment parameter that maximizes the difference value is selected as the updated adjustment parameter. .

[0079] Behavioral resonance intensity is a linear transformation of the behavioral resonance index. By subtracting 0.5 to center it, positive values ​​represent stronger behavioral resonance, and negative values ​​represent weaker behavioral resonance, thus providing a more intuitive measure of the degree and direction of behavioral resonance. Based on this, the system obtains the average behavioral resonance intensity for each preset monitoring period in which emotional trading occurs. This represents the average level of behavioral resonance intensity during the monitoring period in which emotional trading occurs. This helps to quantify the average degree of matching between individual behavior and market conditions when emotional trading occurs.

[0080] Simultaneously, the system obtains the average value of behavioral resonance intensity within each preset monitoring period in which no emotional trading occurs across multiple consecutive preset monitoring periods. This represents the average level of behavioral resonance intensity over a monitoring period in which no emotional transactions occurred, providing a benchmark for comparison with... A comparison is made to obtain the difference value of the latest preset monitoring period across multiple consecutive preset monitoring periods, in order to assess the impact of emotional trading on the intensity of behavioral resonance. This difference value is a key indicator for evaluating the effectiveness of the behavioral resonance index calculation parameters, and the goal is to maximize this difference so that the behavioral resonance index can better distinguish emotional trading.

[0081] The preset adjustment parameters in the Behavioral Resonance Index for the latest preset monitoring period are updated based on the difference values. This aims to more accurately reflect the match between individual behavior and market conditions, particularly in distinguishing emotional trading, by adjusting the preset adjustment parameters in the Behavioral Resonance Index calculation formula. The preset adjustment parameters are increased and decreased by one preset step size respectively, based on the previous preset adjustment parameters. And adjust the two new preset parameters respectively. , The behavioral resonance index is recalculated, and then the difference value after increasing and decreasing the preset step size is calculated. This tentative adjustment is to evaluate the impact of different parameter values ​​on the difference value, so as to find a parameter that maximizes the difference value.

[0082] exist In the process, the preset adjustment parameter that maximizes the difference is selected as the updated adjustment parameter. This is the decision-making step for parameter updates, ensuring that each update aims to maximize the distinction between emotional and non-emotional trading, thereby improving the effectiveness of the behavioral resonance index. Through comparison... Choose the largest of the three values ​​and set its corresponding preset adjustment parameter (i.e. , , This will be used as a new preset adjustment parameter for the next preset monitoring cycle.

[0083] In another embodiment, this application also provides an adaptive asset allocation method for smart investment advisors based on multi-source fusion Kalman filtering, such as... Figure 2 As shown, it includes the following steps:

[0084] Simultaneously acquire personal behavior data and macro market data within a preset monitoring period;

[0085] Generate corresponding behavior labels for the personal behavior data, input the personal behavior data carrying the behavior labels into a preset first filter, output a first feature parameter representing the short-term behavior state, and generate a first behavior state vector.

[0086] A market state vector representing the market state is generated based on the macro market data, and a behavioral resonance index representing the degree of matching between the first behavioral state vector and the market state vector is calculated based on the first behavioral state vector and the market state vector.

[0087] Personal behavioral data carrying behavioral tags is input into a preset second filter to output a second feature parameter representing the long-term behavioral state and generate a second behavioral state vector.

[0088] The configuration benchmark ratio for the preset asset category is updated according to the second behavior state vector, the configuration floating ratio around the configuration benchmark ratio is updated according to the behavior resonance index and the first behavior state vector, and the corresponding asset configuration ratio is generated.

[0089] The behavioral resonance intensity within a preset monitoring period is obtained based on the behavioral resonance index, and the calculation parameters of the behavioral resonance index are updated based on the statistical characteristics of the behavioral resonance intensity within multiple consecutive preset monitoring periods.

[0090] Through the aforementioned technical solution, this application achieves in-depth analysis and dynamic response to investor behavior, maintaining the stability of long-term allocation benchmarks while enhancing the flexibility of short-term adjustments. Specifically, the long-term behavioral state vector output by the preset second filter ensures robust updates to the allocation benchmark ratios, avoiding interference from short-term market fluctuations. The combination of the behavioral resonance index and the first behavioral state vector enables asymmetric adjustment of the allocation floating ratios, automatically applying reverse constraints when investor sentiment and market sentiment are highly resonant, effectively suppressing irrational trading behavior. This two-layer architecture not only overcomes the limitation of existing technologies in distinguishing between planned and emotional behavior but also significantly improves the personalization and market adaptability of asset allocation strategies through a closed-loop optimization mechanism, providing investors with more accurate and forward-looking intelligent investment advisory services.

[0091] In another embodiment, this application also provides a computer storage medium storing computer-executable instructions, which, when executed, implement the aforementioned intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering.

[0092] The above embodiments are only used to illustrate the technical methods of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of this application without departing from the spirit and scope of the technical methods of this application.

Claims

1. An adaptive asset allocation system for intelligent investment advisory services based on multi-source fusion Kalman filtering, characterized in that, Includes the following modules: The data acquisition module is used to simultaneously acquire personal behavior data and macro market data within a preset monitoring period; The first filtering module is used to generate corresponding behavior labels for the personal behavior data, input the personal behavior data carrying the behavior labels into a preset first filter, output a first feature parameter representing the short-term behavior state, and generate a first behavior state vector. The behavioral resonance module is used to generate a market state vector representing the market state based on the macro market data, and to calculate a behavioral resonance index representing the degree of matching between the first behavioral state vector and the market state vector. The second filtering module is used to input personal behavior data carrying behavior tags into a preset second filter to output a second feature parameter representing the long-term behavior state and generate a second behavior state vector. The configuration adjustment module is used to update the configuration benchmark ratio for a preset asset category according to the second behavior state vector, update the configuration floating ratio around the configuration benchmark ratio according to the behavior resonance index and the first behavior state vector, and generate the corresponding asset configuration ratio. The feedback optimization module is used to obtain the behavioral resonance intensity within a preset monitoring period based on the behavioral resonance index, and to update the calculation parameters of the behavioral resonance index based on the statistical characteristics of the behavioral resonance intensity within multiple consecutive preset monitoring periods. The process of calculating the behavioral resonance index includes: The elements in the first behavior state vector are weighted and summed to obtain the emotion index within the corresponding preset monitoring period. The emotion indices within multiple consecutive preset monitoring periods are then combined into an emotion index sequence. The market state vector is weighted and summed to obtain the market index for the corresponding preset monitoring period. The market indices for multiple consecutive preset monitoring periods are then combined into a market index sequence. Obtain the Pearson correlation coefficient between the sentiment index series and the market index series. Combined with preset adjustment parameters Obtaining the Behavioral Resonance Index And use it as the behavioral resonance index of the latest preset monitoring cycle in the multiple consecutive preset monitoring cycles, the value range of the preset adjustment parameter is [0.5, 1.5]; The process of updating the configuration baseline ratio includes: The preset asset class refers to the basic asset classification that represents different market risks and returns, and the allocation benchmark ratio refers to the asset allocation ratio preset for different preset asset classes. Based on the preset mapping rules and the long-term risk preference value in the second behavioral state vector, the risk level within the corresponding preset monitoring period is obtained, including conservative, stable, and aggressive. Corresponding basic allocation ratios are set for different preset asset classes under different risk levels, based on the discipline intensity in the second behavioral state vector. Obtain the attenuation coefficient within the corresponding preset monitoring period. ; Obtain the updated configuration benchmark ratio for a single preset asset class within the corresponding preset monitoring period. ,in, A constant greater than 0 The default asset class configuration baseline ratio before the update This is the basic allocation ratio for the preset asset class under the risk level within the corresponding preset monitoring period.

2. The intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering according to claim 1, characterized in that, The generation of the behavioral labels is achieved through a supervised learning classifier. The transaction time and transaction amount extracted based on personal behavioral data and the market volatility extracted based on macro market data are input into the supervised learning classifier to output the behavioral labels corresponding to the personal behavioral data, including planned inflows, event-driven inflows, and emotional outflows. The first filter is a Kalman filter with a first process noise covariance. Personal behavioral data carrying event-driven inflows and emotional outflows are input into the first filter to output a first feature parameter containing emotional sensitivity and impulsive tendency values. The first feature parameter is then combined sequentially into a first behavioral state vector.

3. The intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering according to claim 1, characterized in that, Based on the macro market data, multi-dimensional market characteristic indicators are extracted, including sentiment indicators representing market sentiment, volatility indicators representing market volatility, and liquidity indicators representing market liquidity. After normalization, each market characteristic indicator is combined in sequence to form a corresponding market state vector.

4. The intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering according to claim 2, characterized in that, The second filter is a Kalman filter with a second process noise covariance, which is smaller than the first process noise covariance. Personal behavior data carrying planned inflows are input into the second filter to output a second feature parameter containing discipline intensity and long-term risk preference value. The second feature parameter is then combined sequentially into a second behavior state vector.

5. The intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering according to claim 1, characterized in that, The configuration floating ratio refers to the dynamic adjustment boundary that fluctuates around the preset asset class's configuration benchmark ratio, based on the behavioral resonance index within the same preset monitoring period. and sentiment index Adjust the preset basic fluctuation ratio within the corresponding preset monitoring period. ; when At that time, the upper limit adjustment factor is obtained. ,when At that time, the upper limit adjustment factor will be adjusted. ,when At that time, the lower limit adjustment factor was obtained. ,when When, the lower limit adjustment factor is... ;in, Both are preset behavioral resonance thresholds. The first shrinkage coefficient is preset, and its value range is (0, 1]. According to the aforementioned sentiment index Obtain global adjustment factor ,in, The second contraction coefficient is preset, and the configured floating ratio includes an upper limit floating ratio. and lower limit fluctuation ratio Combined with the updated configuration benchmark ratio for a single preset asset category within the corresponding preset monitoring period Generate the asset allocation ratio for the preset asset class. ].

6. The intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering according to claim 1, characterized in that, The behavioral resonance intensity If personal behavioral data carrying emotional outflow exists within a single preset monitoring period, it is determined that an emotional transaction occurred within that preset monitoring period; otherwise, it is determined that no emotional transaction occurred. The mean of the behavioral resonance intensity within each of the preset monitoring periods in which emotional trading occurs is obtained, denoted as . The mean of the behavioral resonance intensity within each preset monitoring period in which no emotional trading occurred is obtained, and is denoted as . Obtain the difference value of the latest preset monitoring period among the multiple consecutive preset monitoring periods. ; The preset adjustment parameters in the behavioral resonance index of the latest preset monitoring period are updated based on the difference value, and the preset adjustment parameters before the update are updated. Based on the above, increase and decrease by a preset step size respectively, and obtain the difference value after increasing the preset step size respectively. The difference value after reducing the preset step size ,exist In the process, the preset adjustment parameter that maximizes the difference value is selected as the updated adjustment parameter. .

7. An adaptive asset allocation method for intelligent investment advisors based on multi-source fusion Kalman filtering, characterized in that, Includes the following steps: Simultaneously acquire personal behavior data and macro market data within a preset monitoring period; Generate corresponding behavior labels for the personal behavior data, input the personal behavior data carrying the behavior labels into a preset first filter, output a first feature parameter representing the short-term behavior state, and generate a first behavior state vector. A market state vector representing the market state is generated based on the macro market data, and a behavioral resonance index representing the degree of matching between the first behavioral state vector and the market state vector is calculated based on the first behavioral state vector and the market state vector. Personal behavioral data carrying behavioral tags is input into a preset second filter to output a second feature parameter representing the long-term behavioral state and generate a second behavioral state vector. The configuration benchmark ratio for the preset asset category is updated according to the second behavior state vector, the configuration floating ratio around the configuration benchmark ratio is updated according to the behavior resonance index and the first behavior state vector, and the corresponding asset configuration ratio is generated. The behavioral resonance intensity within a preset monitoring period is obtained based on the behavioral resonance index, and the calculation parameters of the behavioral resonance index are updated based on the statistical characteristics of the behavioral resonance intensity within multiple consecutive preset monitoring periods. The process of calculating the behavioral resonance index includes: The elements in the first behavior state vector are weighted and summed to obtain the emotion index within the corresponding preset monitoring period. The emotion indices within multiple consecutive preset monitoring periods are then combined into an emotion index sequence. The market state vector is weighted and summed to obtain the market index for the corresponding preset monitoring period. The market indices for multiple consecutive preset monitoring periods are then combined into a market index sequence. Obtain the Pearson correlation coefficient between the sentiment index series and the market index series. Combined with preset adjustment parameters Obtaining the Behavioral Resonance Index And use it as the behavioral resonance index of the latest preset monitoring cycle in the multiple consecutive preset monitoring cycles, the value range of the preset adjustment parameter is [0.5, 1.5]; The process of updating the configuration baseline ratio includes: The preset asset class refers to the basic asset classification that represents different market risks and returns, and the allocation benchmark ratio refers to the asset allocation ratio preset for different preset asset classes. Based on the preset mapping rules and the long-term risk preference value in the second behavioral state vector, the risk level within the corresponding preset monitoring period is obtained, including conservative, stable, and aggressive. Corresponding basic allocation ratios are set for different preset asset classes under different risk levels, based on the discipline intensity in the second behavioral state vector. Obtain the attenuation coefficient within the corresponding preset monitoring period. ; Obtain the updated configuration benchmark ratio for a single preset asset class within the corresponding preset monitoring period. ,in, A constant greater than 0 The default asset class configuration baseline ratio before the update This is the basic allocation ratio for the preset asset class under the risk level within the corresponding preset monitoring period.

8. A computer storage medium storing computer-executable instructions, characterized in that, When the computer-executable instructions are executed, they implement the intelligent investment advisory asset allocation adaptive system based on multi-source fusion Kalman filtering as described in any one of claims 1-6.

Images