Intelligent self-trade-area tax analysis system fusing multi-dimensional economic indicators

Abstract

The invention relates to the technical field of data processing and economic analysis, and discloses a self-trade area tax intelligent analysis system fusing multi-dimensional economic indicators, which comprises a data fusion and factor extraction module, a causal pedigree identification module, a dynamic baseline modeling module, a compliance risk monitoring module, a pedigree adaptive reconstruction module and an anti-fact decision deduction module. The method comprises the following steps: firstly, extracting potential core factors in multi-dimensional economic indexes, constructing a causal conduction pedigree for reflecting a conduction relationship between the factors and tax indexes, and establishing a dynamic baseline model on the basis of the causal conduction pedigree; in the operation stage, the system evaluates the state of the model in real time by monitoring the structure sudden change risk and the systematic disturbance risk in parallel. According to the method, a closed-loop mechanism of risk monitoring and model self-adaption is established, so that the system can dynamically respond to external impact, the robustness and long-term effectiveness of analysis and prediction are guaranteed, and reliable support is provided for prospective decision making.

Description

Technical Field

[0001] This invention relates to the field of data processing and economic analysis technology, specifically to a free trade zone tax intelligent analysis system that integrates multi-dimensional economic indicators. Background Technology

[0002] When analyzing and forecasting tax revenues in economies such as free trade zones, econometric models or data mining models are typically used to construct quantitative relationships between tax indicators and other economic variables by analyzing historical data. The core of these models is based on the stable transmission structures between economic variables revealed by historical data.

[0003] However, the economic environment of free trade zones is highly open and dynamic, making them susceptible to domestic and international macroeconomic policies, industrial restructuring, or unforeseen events. When these external shocks alter the underlying logic of economic operation, the transmission relationships based on historical data and upon which existing models rely become inapplicable, leading to a significant decrease in model prediction accuracy. Existing analytical methods typically lack mechanisms for real-time assessment of the model's applicability, making it difficult to proactively identify whether the economic structure upon which they are based has changed.

[0004] Furthermore, even if the model structure remains effective, short-term external factors not incorporated into the model can still cause systemic disturbances to tax revenue. Existing technologies often struggle to effectively distinguish between such temporary disturbances and fundamental model structural failures. Therefore, when facing complex changes in the external environment, traditional models, due to their static nature, cannot self-correct or dynamically adapt, limiting their ability to provide timely and reliable risk warnings and decision support. Therefore, this invention provides a free trade zone tax revenue intelligent analysis system that integrates multi-dimensional economic indicators to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a free trade zone tax revenue intelligent analysis system that integrates multi-dimensional economic indicators, thus solving the problems mentioned in the background section.

[0006] The first aspect of this invention provides a free trade zone tax revenue intelligent analysis system that integrates multi-dimensional economic indicators, comprising: a processor; The memory stores computer-executable instructions, which the processor executes to achieve the following: The data fusion and factor extraction module is used to receive multi-source heterogeneous multi-frequency economic indicator data and extract potential core factors. The causal spectrum identification module is used to identify the causal relationship between the potential core factors and tax indicators in order to construct a causal transmission spectrum. The dynamic baseline modeling module is used to construct and update a dynamic baseline model for predicting tax revenue indicators based on the causal transmission spectrum. The compliance risk monitoring module is used to receive real-time data and output risk signals based on the dynamic baseline model by monitoring structural mutation risk and systemic disturbance risk in parallel. The causal transmission spectrum adaptive reconstruction module is used to perform local causal rediscovery on the causal transmission spectrum in response to the risk signal to generate a reconstructed causal transmission spectrum, and to trigger the dynamic baseline modeling module to update the dynamic baseline model. The counterfactual decision-making module is used to respond to user-submitted intervention commands, perform simulations based on the current dynamic baseline model, and output a decision-making report.

[0007] A second aspect of this invention provides a method for intelligent analysis of tax revenue in free trade zones that integrates multi-dimensional economic indicators. This method includes the following steps: Perform data fusion and factor extraction to receive multi-source heterogeneous multi-frequency economic indicator data and extract potential core factors; Perform causal spectrum identification to identify the causal relationship between the potential core factors and tax indicators, thereby constructing a causal transmission spectrum; Perform dynamic baseline modeling to construct a dynamic baseline model based on the causal transmission spectrum; Perform compliance risk monitoring to receive real-time data, and monitor structural mutation risk and systemic disturbance risk in parallel based on the dynamic baseline model, thereby outputting risk signals; Perform adaptive reconstruction of the causal transmission spectrum in response to the risk signal, perform local causal rediscovery on the causal transmission spectrum, and update the dynamic baseline model; Perform counterfactual decision simulations in response to user-submitted intervention commands, perform simulations based on the current dynamic baseline model, and output a decision simulation report.

[0008] The compliance risk monitoring module or the steps for performing compliance risk monitoring include: Recursive estimation is performed using the real-time data and dynamic baseline model to obtain time-varying parameter sequences and predicted residual sequences; Based on the stability analysis of the time-varying parameter sequence, the risk of structural mutation is detected; Based on the anomaly analysis of the predicted residual sequence, the risk of systemic disturbance is detected.

[0009] The detection of the structural mutation risk includes: The cumulative sum test is performed on the recursively estimated time-varying parameter sequence, and the presence of structural breakpoints is determined by whether the statistic exceeds a critical value.

[0010] The detection of the systemic disturbance risk includes: Calculate the variance of the predicted residual sequence within the sliding time window and compare the variance with a preset risk threshold determined based on historical data.

[0011] The phylogenetic adaptive reconstruction module or the step of performing phylogenetic adaptive reconstruction includes the following steps: Triggering and location are performed based on the risk signals to identify risk transmission paths or failure nodes within the causal transmission spectrum; Local causal rediscovery is performed on a subset of variables involved in the risk transmission path or failure node to generate a reconstructed local causal structure; The reconstructed local causal structure is integrated into the cause-effect transmission spectrum to form the reconstructed causal transmission spectrum, and the dynamic baseline model is updated using the reconstructed causal transmission spectrum.

[0012] The data fusion and factor extraction include: applying a mixed data sampling method to unify multi-frequency economic indicator data to a common frequency, and applying a dynamic factor model to the data after unifying the frequency to extract the potential core factors.

[0013] The dynamic baseline modeling includes: constructing the dynamic baseline model using a time-varying parameter vector autoregressive model, wherein the causal transmission spectrum is used to determine the variable composition of the model and the prior constraints of the time-varying parameters.

[0014] The counterfactual decision-making deduction includes: Based on the intervention instructions (e.g., adjusting specific regulatory policies or economic variables), modify one or more input variables or parameters of the dynamic baseline model; Perform multi-step forward prediction based on the modified model to generate simulated trajectories; The intervention effect is quantitatively evaluated by comparing the simulated trajectory with the baseline trajectory without intervention, and the decision projection report is generated.

[0015] The multi-frequency economic indicator data includes at least two of the following: daily logistics throughput data, monthly tax revenue data by tax type, and quarterly regional GDP data.

[0016] This invention provides a smart tax analysis system for free trade zones that integrates multi-dimensional economic indicators. It offers the following advantages: 1. This invention constructs a closed loop of risk response and model self-correction by setting up a compliance risk monitoring module and a spectrum adaptive reconstruction module. When the system detects structural mutation risk or systemic disturbance risk, it can not only output risk signals, but also trigger the local causal rediscovery and reconstruction of the causal transmission spectrum, and update the dynamic baseline model accordingly. This mechanism enables the system to actively adapt to changes in the external economic environment, solves the problem of traditional static models failing when the structure changes, and thus ensures the long-term effectiveness and robustness of analysis and prediction.

[0017] 2. The risk monitoring method proposed in this invention can monitor structural mutation risks based on model time-varying parameters and systemic disturbance risks based on model prediction residuals in parallel. The dual monitoring mechanism enables the system to distinguish whether the failure is due to the failure of the internal transmission relationship of the model or the systemic shock caused by unmodeled external factors. Compared with the single threshold alarm method, it provides a preliminary diagnosis of the source of risk, provides a more targeted basis for subsequent adaptive reconstruction of the spectrum, and improves the accuracy of the system's self-repair.

[0018] 3. This invention incorporates a counterfactual decision-making simulation module, which performs simulations based on a dynamic baseline model that is monitored and updated in real time. Since the baseline model can continuously reflect the current economic structure through an adaptive mechanism, the decision-making simulation based on it can more accurately simulate the transmission effects of different intervention commands, providing reliable technical support for management departments to conduct forward-looking policy effect evaluation and quantitative decision-making. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the system architecture of the present invention; Figure 2 This is a flowchart of the method of the present invention; Figure 3 This is a schematic diagram of data fusion and dynamic factor extraction according to the present invention; Figure 4 This is a flowchart illustrating the online monitoring and adaptive phase of the present invention.

[0020] Among them, 110 is the data fusion and factor extraction module; 120 is the causal genealogy identification module; 130 is the dynamic baseline modeling module; 140 is the compliance risk monitoring module; 150 is the genealogy adaptive reconstruction module; and 160 is the counterfactual decision inference module. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see the appendix Figure 1 -Appendix Figure 4 This invention provides a free trade zone tax revenue intelligent analysis system that integrates multi-dimensional economic indicators, including: Data fusion and factor extraction module 110; Causal genealogy identification module 120; Dynamic baseline modeling module 130; Compliance risk monitoring module 140; Spectral adaptive reconstruction module 150; Counterfactual decision-making simulation module 160.

[0023] The data fusion and factor extraction module 110 is used to preprocess and align the original multidimensional heterogeneous data related to the free trade zone, and extract core dynamic factors. The causal genealogy identification module 120, whose input is connected to the data fusion and factor extraction module 110, is used to identify the initial causal transmission genealogy based on the extracted factors and target tax variables using non-Gaussianity. The dynamic baseline modeling module 130, whose input is connected to the causal lineage identification module 120, is used to estimate the parameters of the structure vector autoregression (SVAR) model based on the initial lineage and establish the initial baseline model. The compliance risk monitoring module 140, whose input is connected to the dynamic baseline modeling module 130, is used to monitor the parameter stability and residual fluctuations of the baseline model when the system is running online, in order to identify structural mutation risks and systemic disturbance risks, and output risk signals. The adaptive causal reconstruction module 150 has its input connected to the compliance risk monitoring module 140 and its output connected to the dynamic baseline modeling module 130. This adaptive causal reconstruction module 150 is triggered upon receiving a risk signal to perform causal rediscovery on the local causal lines that have experienced anomalies, and to generate updated causal lines for the dynamic baseline modeling module 130 to update the baseline model. The counterfactual decision-making simulation module 160, whose input is connected to the dynamic baseline modeling module 130 (or the genealogy adaptive reconstruction module 150), is used to call the current baseline model and execute the policy intervention simulation input by the user.

[0024] All of the above modules are deployed in the analysis engine layer. The system also includes a data layer for storing raw data, extracted factors, and multi-version baseline models generated by the genealogy adaptive reconstruction module 150. The system also includes an application layer for providing users with compliance risk warning interfaces and administrative decision support interfaces, and for receiving user intervention instructions.

[0025] See attached document Figure 2 , Figure 2 This is a flowchart of an adaptive tax analysis and regulation method according to an embodiment of the present invention.

[0026] The method flow includes an initialization phase, an online monitoring and adaptation phase, and a decision support phase.

[0027] The initialization phase (offline modeling phase) includes step S100, which is used to establish an initial baseline model; In step S110, the data fusion and factor extraction module 110 processes historical data and extracts core dynamic factors. ; Step S120: Causal spectrum identification module 120 identifies the initial causal transmission spectrum. ; Step S130: The dynamic baseline modeling module 130 establishes the initial baseline model. ; The online monitoring and adaptation phase includes step S200, which is used to monitor risks in real time and enable the model to iterate itself. The system proceeds to step S210 to receive new real-time data. ; Step S220, the compliance risk monitoring module 140, based on the current baseline model (In the first loop) for To determine if there is a risk, testing will be conducted. Step S230: Perform risk assessment. If no risk is identified ( The process returns to S210 to continue monitoring data for the next cycle. If a risk is identified... The process proceeds to step S240; In step S240, the spectral adaptive reconstruction module 150 is triggered to locate the risk signal. The baseline model is partially reconstructed and updated to generate a new generation of baseline models. Then, the process returns to S210, using Continue monitoring as a new benchmark; The decision support phase (on-demand triggering phase) includes step S300; In step S310, the system receives a policy intervention instruction input by the user; In step S320, the counterfactual decision inference module 160 calls the current baseline model maintained in step S200. ; Step S330, Counterfactual decision-making module 160 performs graph modification and Operator intervention simulates the net effect of policy transmission; Step S340: The system outputs a decision deduction report.

[0028] This invention also provides a method for intelligent analysis of tax revenue in free trade zones that integrates multi-dimensional economic indicators, including the following steps: S100, Execute the system initialization phase. This phase establishes an initial baseline model based on historical data, including the following sub-steps: S110 executes the data fusion and factor extraction module to process multidimensional heterogeneous historical data in order to extract core dynamic factors; Data standardization; the system receives data related to the free trade zone. The system uses multi-source heterogeneous time series data, which includes indicators at different frequencies, such as daily logistics throughput data, daily cross-border capital flow data, monthly tax revenue data by tax type, monthly FD (foreign direct investment) data, and quarterly regional GDP data. The system preprocesses these observed indicators, including time alignment, missing value imputation, and standardization. For missing value imputation and standardization, those skilled in the art can use standard time series imputation methods (such as spline interpolation) and standardization methods (such as Z-Score) to process them. These are well-known techniques in the field and will not be elaborated here. Dynamic factor extraction, for processing To address the high dimensionality and high collinearity of the observation indicators, this embodiment employs a Dynamic Factor Model (DFM) for dimensionality reduction. This model assumes... dimensional observation index vector It is by indivual( Unobservable common latent factors and Heterogeneity perturbation of dimensions Driven by, its state-space form can be represented as: ; in, yes A dimensional vector of observation indicators; yes The common potential factor vector of the dimension (e.g., representing core economic conditions such as "trade activity" and "financial climate"). yes dimensional factor loading matrix, It is a heterogeneous perturbation vector; Multi-frequency data fusion in factor estimation In order to integrate information from high-frequency data (such as daily logistics) into the estimation of low-frequency factors (such as monthly economic conditions), this embodiment adopts the idea of ​​mixed data sampling (MIDAS) regression. Specifically, in estimating monthly factors At that time, it is allowed to dynamically change High-frequency observations ( For high-frequency time indexing, a weighted polynomial (such as the Almon polynomial or the Beta polynomial) is used. Weighted calculation ensures that high-frequency information is not lost; S120, execute the causal spectrum identification module, and identify the initial causal transmission spectrum based on the core dynamic factors and target tax variables; Construct a state vector, which is the result of extracting the state vector in step S110. Core Factors and Core target variable of dimension (For example, value-added tax revenue, corporate income tax revenue, etc.) are merged to form a dimension( The combined state vector : ; Construct a structured vector autoregressive (SVAR) model for identification. The instantaneous and lagged causal relationships among the variables are constructed as follows: SVAR model of order: ; in, yes A dimensional instantaneous causal structure matrix, whose off-diagonal elements Defined variables right The instantaneous causal effects; ( )yes The lag coefficient matrix of dimension 1 defines the lag. The impact of variables in the current period on variables in the current period; yes A structural impact vector of dimension, assuming that its components Independent of each other, the core technical objective of this step is identification. matrix; Identification based on non-Gaussianity, traditional SVAR identification Relying on subjective economic theory constraints, this embodiment utilizes real economic shocks ( Due to its characteristic of not following a Gaussian distribution, the linear non-Gaussian acyclic model (LiNGAM) or independent component analysis (ICA) algorithm is used for identification. First, a simplified VAR model is estimated: ; Obtain its simplified residual In the SVAR model: and ; because It is a non-Gaussian independent component. That is The linear mixture, in this embodiment for Simplified residuals of dimension The ICA algorithm is applied to find a dismixing matrix. , so that: In ; To maximize the independence of each component, according to theory, this unmixing matrix... That is Consistent estimation of a matrix (after appropriate permutations and scaling) Therefore, the instantaneous causal structure matrix It was uniquely identified, The non-zero element pattern of the matrix defines the initial causal transmission spectrum. ; S130, execute the dynamic baseline modeling module to construct the initial baseline model based on the initial causal transmission spectrum; Model parameter estimation, the instantaneous causal structure matrix identified in step S120 Under constraints, SVAR model This transforms the system into a recognizable system of simultaneous equations. In this embodiment, the equation-by-equation least squares (OLS) or generalized least squares (GLS) method is used to transform the lag parameter matrix. By performing a consistent estimate, we obtain And estimate the (diagonal) covariance matrix of the structural impact. ; Establish a dynamic baseline and estimate the complete set of parameters. Together with the graph structure They are collectively defined as the system's "initial baseline model". ,based on The system further calculates its standard dynamic characteristics as a benchmark for subsequent online monitoring. These dynamic characteristics include: Impulse Response Function (IRF): Used to quantize when... When a structural shock (such as a "trade activity" shock) causes a one-unit change in the future... Expect All variables (especially taxes) The dynamic transmission path, time delay, and intensity of ( ); Prediction Error Variance Decomposition (FEVD): Used for quantification Each variable in In the step prediction error variance, the structural impact is composed of various different results. The contribution ratio, the calculation of IRF and FEVD mentioned above involves converting the SVAR model into its vector moving average (VMA) form, and the specific calculation method is a well-known technique in the field, which will not be elaborated here.

[0029] S200, the online monitoring and adaptation phase, is used for real-time risk monitoring and model iteration, and includes the following sub-steps: S210, Receive new real-time data. Step S210: The system receives new real-time data. ; S220, Execute the compliance risk monitoring module, perform detection based on the current baseline model (the initial baseline model established in S100 during initial execution) to generate risk judgment signals, including the following sub-steps: S221, Perform recursive estimation; the system is based on the current baseline model. (During initial execution, Established in S130 ), for the new data received The process involves recursive estimation methods, such as recursive least squares (RLS) or a Kalman filter, to adjust the hysteresis parameter matrix in the SVAR model. By updating the parameters periodically, a real-time estimate sequence can be obtained. Meanwhile, the system computation model Real-time one-step prediction of residuals Its definition is: ; in Based on Information and models of the time and prior right One-step prediction value; S222, detecting structural mutation risk ( This step is used to monitor the stability of the fundamental transmission laws in the economy, and its monitoring object is the real-time parameter estimation sequence generated in S221. In one embodiment, the system for Pre-defined key coefficients in the sequence (e.g., those reflecting the correlation between "logistics" and "profit") Apply structural mutation testing; This test can be performed using the CUSUM test or the CUSUM-of-Squares test. Taking the CUSUM test as an example, when its test statistic... (Calculated based on recursive residuals or parameter changes) Breaking through the preset confidence boundary When a structural mutation occurs, the system determines that a structural mutation risk signal has been generated: ; in This indicates that no structural mutation has occurred; This indicates that a risk of structural mutation has been detected. S223, Detecting systemic disturbance risks ( This step is used to monitor for any models not included in the current baseline model. The novel, persistent influencing factors are captured, and their monitoring targets are the residual sequences generated in S221. ; The system for residual sequences (or its specific components) White noise test is applied to determine whether there is serial correlation. This test can be performed using the Ljung-Box Q test, which is calculated... hysteresis Statistic: ; in This is the sample window length used for calculation. It is a residual of The autocorrelation coefficient of the first-order sample, when Greater than at the significance level Below, the degrees of freedom are ( The critical value of the chi-square distribution (where SVAR lag order is the lag order) At that time, the system determines that the residuals have significant serial correlation and generates a systematic disturbance risk signal: ; in This indicates that the residual is white noise; This indicates that a systemic disturbance risk has been detected; S230: Based on the risk assessment signal, a judgment is made. If no risk is found, the process returns to step S210 to continue monitoring. The system performs a risk assessment and generates a total risk signal based on the outputs of S222 and S223. ; like This indicates that no risk was detected, and the process returns to step S210 to continue monitoring for the next cycle. like This indicates that a risk has been detected, and the process proceeds to step S240; S240, If there is a risk, execute the phylogenetic adaptive reconstruction module to perform local reconstruction of the baseline model to generate an updated baseline model, and use the updated baseline model as the current baseline model. The process returns to step S210. In step S240, the spectral adaptive reconstruction module 150 is executed. When triggered, it includes the following sub-steps: S241, Execution triggering and localization, the genealogy adaptive reconstruction module 150 first sets the current baseline model The system archives its parameters in a versioned manner, and then, based on risk signals... To identify the source of the risk; like The system locates the parameter in S222 that has mutated (e.g., The variable corresponding to ) and ; like The system located the residual in S223. Variables with significant sequence correlation The system locates these variables and their positions in the current causal spectrum. The first-order (or second-order) adjacency variables in the graph are collectively defined as a "local subgraph". and its corresponding variable subset ; S242, Perform Local Causal Rediscovery; the system does not apply to the SVAR model. All The system performs a global reconstruction of the dimensional variables, but only targets those located in S241. A subset of variables is retrieved, along with the latest time window containing the mutation event. Within this subset of data, the system re-executes the causal structure discovery algorithm (e.g., LiNGAM or ICA algorithm) described in step S120 to identify whether the local economic structure has evolved into a new causal relationship, thereby obtaining a new local instantaneous causal structure. ; S243, Perform model iteration; the system will then incorporate the newly discovered local causal structures from S242. Merge back to the original In this process, a new generation of (updated) causal transmission systems is generated. Subsequently, the system re-executes step S130 (dynamic baseline modeling), that is, in Under structural constraints, re-estimate all parameters of the SVAR model ( and ), to obtain a new generation of baseline models ,Should Set as the new "current baseline model" to replace This information is then used in the next round of steps S210 and S220 of the S200 phase, after which the process returns to S210.

[0030] S300, the decision support phase, responds to user-triggered actions and includes the following sub-steps: S310 responds to policy intervention commands input by the user; In step S310, the counterfactual decision inference module 160 responds to a policy intervention instruction input by the user, which is an administrative decision to be simulated, such as a new tax incentive policy or regulatory measure. The counterfactual decision inference module 160 translates this instruction into one or more intervention operations on specific variables in the SVAR model, denoted as... ,in yes dimensional state vector The corresponding strategy variable (or its proxy variable) in the middle, It is the new value set by the policy instruction; S320, Execute the counterfactual decision inference module, and call the current baseline model maintained by step S200; In step S320, the counterfactual decision inference module 160 calls the "current baseline model" maintained in step S200 (online monitoring and adaptation phase). ,Should It is the latest model updated through step S240 (phylogenetic adaptive reconstruction). Or the initial model before refactoring is triggered. ,this This reflects the latest economic causal structure currently identified by the system. and the corresponding parameter set ; S330, based on policy intervention directives, performs map modifications and intervention simulations on the current baseline model to calculate the net effect of policy transmission; In step S330, the system performs the intervention operation based on step S310. For the current baseline model The modification is implemented to calculate the net effect of policy transmission. This step specifically includes: S331, execute map modification, system based on Operation, modify 5VAR model The structure, the system is located at the first Variables The corresponding number One equation; S332, the system will... The equations (i.e., those used in the original model to describe) The endogenous dynamic equation is replaced with an exogenous given equation: ; This operation is in the causal spectrum. The meaning above is to cut off all pointers. Endogenous causal connections (i.e., in the original model) and All effects in the matrix The coefficients no longer take effect in the simulation, making This becomes an exogenous fixed value during the simulation, thus yielding a new, intervened SVAR model. ; S333, Implementation Intervention Scenario Simulation, the system in In terms of model, based on Initial conditions at time Calculate the future by forward recursion Expect( The expected path of the state vector (for the number of prediction periods) ; S340, Output a decision projection report containing the net effect. In step S340, the system performs net effect assessment and report generation, which specifically includes: S341, Calculate the baseline scenario path, with the system operating on the current baseline model without intervention. Above, based on the same as in S333 Initial conditions at time Recursive calculation of the future Expected path in the period ; S342, Calculate the net effect, for Any target variable in (For example, revenue from a certain type of tax) The system calculates its future Net effect of the period It is calculated as the difference between the intervention path and the baseline path; ; S343, Counterfactual decision deduction module 160 generates and outputs a decision deduction report, which in one embodiment includes: various target variables (especially tax variables). The net effect of Time series diagrams and intervention pathways With baseline path The comparison diagram, and based on the current causal spectrum Analysis of policy transmission channels.

Claims

1. A free trade zone tax revenue intelligent analysis system integrating multi-dimensional economic indicators, characterized in that: include: processor; A memory, wherein computer-executable instructions are stored thereon, and the processor executes the instructions to perform: The data fusion and factor extraction module is used to receive multi-source heterogeneous multi-frequency economic indicator data and extract potential core factors. The causal spectrum identification module is used to identify the causal relationship between the potential core factors and tax indicators in order to construct a causal transmission spectrum. The dynamic baseline modeling module is used to construct and update a dynamic baseline model for predicting tax revenue indicators based on the causal transmission spectrum. The compliance risk monitoring module is used to receive real-time data and output risk signals based on the dynamic baseline model by monitoring structural mutation risk and systemic disturbance risk in parallel. The causal transmission spectrum adaptive reconstruction module is used to perform local causal rediscovery on the causal transmission spectrum in response to the risk signal to generate a reconstructed causal transmission spectrum, and to trigger the dynamic baseline modeling module to update the dynamic baseline model. The counterfactual decision-making module is used to respond to user-submitted intervention commands, perform simulations based on the current dynamic baseline model, and output a decision-making report.

2. The free trade zone tax revenue intelligent analysis system integrating multi-dimensional economic indicators as described in claim 1, characterized in that, The compliance risk monitoring module uses real-time data and a dynamic baseline model to perform recursive estimation to obtain time-varying parameter sequences and predicted residual sequences; it detects structural mutation risks based on stability analysis of the time-varying parameter sequences; and it detects systemic disturbance risks based on anomaly analysis of the predicted residual sequences.

3. The free trade zone tax revenue intelligent analysis system integrating multi-dimensional economic indicators as described in claim 1, characterized in that, The adaptive reconstruction module performs triggering and localization based on the risk signal to identify risk transmission paths within the causal transmission spectrum; performs local causal rediscovery on the risk transmission paths to generate reconstructed local causal structures; and integrates the reconstructed local causal structures into the cause-effect transmission spectrum to form the reconstructed causal transmission spectrum.

4. The free trade zone tax revenue intelligent analysis system integrating multi-dimensional economic indicators as described in claim 1, characterized in that, The data fusion and factor extraction module is used to apply dynamic factor models and mixed data sampling methods to extract the potential core factors; The dynamic baseline modeling module is used to construct the dynamic baseline model using a time-varying parameter vector autoregressive model.

5. A method for intelligent analysis of tax revenue in free trade zones that integrates multi-dimensional economic indicators, applied to any of the free trade zone tax revenue intelligent analysis systems that integrate multi-dimensional economic indicators as described in claims 1-4, characterized in that, Includes the following steps: Perform data fusion and factor extraction to receive multi-source heterogeneous multi-frequency economic indicator data and extract potential core factors; Perform causal spectrum identification to identify the causal relationship between the potential core factors and tax indicators, thereby constructing a causal transmission spectrum; Perform dynamic baseline modeling to construct a dynamic baseline model based on the causal transmission spectrum; Perform compliance risk monitoring to receive real-time data, and monitor structural mutation risk and systemic disturbance risk in parallel based on the dynamic baseline model, thereby outputting risk signals; Perform adaptive reconstruction of the causal transmission spectrum in response to the risk signal, perform local causal rediscovery on the causal transmission spectrum, and update the dynamic baseline model; Perform counterfactual decision simulations in response to user-submitted intervention commands, perform simulations based on the current dynamic baseline model, and output a decision simulation report.

6. The intelligent analysis method for tax revenue in free trade zones integrating multi-dimensional economic indicators as described in claim 5, is characterized in that, The steps for performing compliance risk monitoring include: Recursive estimation is performed using the real-time data and dynamic baseline model to obtain time-varying parameter sequences and predicted residual sequences; Based on the stability analysis of the time-varying parameter sequence, the risk of structural mutation is detected; Based on the anomaly analysis of the predicted residual sequence, the risk of systemic disturbance is detected.

7. The intelligent analysis method for tax revenue in free trade zones integrating multi-dimensional economic indicators as described in claim 6, characterized in that, The steps for detecting the risk of structural mutation include: A cumulative sum test is performed on the recursively estimated time-varying parameter sequence to determine whether there are structural breakpoints.

8. The intelligent analysis method for tax revenue in free trade zones that integrates multi-dimensional economic indicators according to claim 6, characterized in that, The steps for detecting the systemic disturbance risk include: Calculate the variance of the predicted residual sequence within the sliding time window and compare the variance with a preset risk threshold.

9. The intelligent analysis method for tax revenue in free trade zones integrating multi-dimensional economic indicators as described in claim 5, characterized in that, The steps of performing adaptive reconstruction of the genealogy include: Triggering and location are performed based on the risk signals to identify risk transmission paths within the causal transmission spectrum; Perform local causal rediscovery on the aforementioned risk transmission path; The reconstructed local causal structure is integrated into the cause-effect transmission spectrum to update the causal transmission spectrum and the dynamic baseline model.

10. The intelligent analysis method for tax revenue in free trade zones that integrates multi-dimensional economic indicators according to claim 5, characterized in that, The steps for performing the counterfactual decision deduction include: According to the intervention instructions, modify one or more input variables of the dynamic baseline model; Perform multi-step forward prediction to generate a simulated trajectory; The decision-making report is generated by comparing the simulated trajectory with the baseline trajectory.

Images