Data processing method and system based on multilayer network technology

Abstract

The invention provides a data processing method and system based on a multilayer network technology, and relates to the technical field of data processing. According to the data processing method, a multilayer network technology based on institution return rates and emotion indexes is constructed, interdependence between different tail institutions can be carefully captured from the perspective of the institutions, risk infection between the institutions can be analyzed from the perspective of external markets of the institutions, and risk formation and infection mechanisms can be revealed more clearly; the topological structure and clustering characteristics of the multi-layer network are mainly analyzed, the internal evolution characteristics of the network are discussed, and the method is of great significance to mechanism and risk supervision; meanwhile, the risk contribution problem of each mechanism is researched in combination with a multi-layer network technology, so that the system importance mechanism under a complex network view angle can be comprehensively identified; the multi-layer network technology is beneficial for improving the prediction capability of systematic risks; the risk is infected in each index, and interactive infection is also carried out, so that the prediction accuracy is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a data processing method and system based on multi-layer network technology. Background Art

[0002] At present, the following single-layer network technologies are mainly used to measure institutional relevance and risk evolution:

[0003] The first is a single-layer network built based on the correlation between corporate businesses and internal assets. This correlation arises from direct transaction contracts between institutions and indirect correlations between institutions holding common assets. It is conducive to exploring the specific transmission path of risks, but it is impossible to discover the overall correlation between institutions.

[0004] The second is a single-layer information spillover network built based on institutional market data. Market data has the advantages of high frequency, timeliness and good availability. The linkage of market data such as yields takes into account the overall relationship between institutions. It is a portable and efficient construction method.

[0005] The third is a single-layer network built based on the emotional transmission mechanism, that is, the phenomenon of risk contagion is partially realized by influencing market sentiment. The contagion of market sentiment will increase the intensity of risk spread on the one hand, and accelerate the speed of risk spread in the network on the other hand. The denser the network, the higher the degree of contagion.

[0006] However, most existing research techniques focus on using single-layer network technology to measure institutional relevance and risk evolution, but lack consideration of the complexity and multiplicity of network relevance, which is reflected in:

[0007] Existing research is mostly based on single-layer complex network technology, ignoring the complexity and multiplicity of inter-institutional connections. The research on risk contagion is relatively one-sided, and risk measurement based on single-layer network connectivity may be biased or even distorted. In fact, risks may spread and spread within a single-layer network, or between different layers of networks.

[0008] In the few studies based on multiple network technology, the establishment of institutional multi-layer networks is mostly based on return correlation. The establishment of institutional multi-layer networks is relatively simple, and the specific return meaning of multiple correlations between institutions is unclear, which makes it impossible to provide operational suggestions for preventing and resolving systemic risks. In recent years, market sentiment, as an important part of behavioral theory research, has attracted the attention of many scholars. In particular, studies have shown that major events can cause market sentiment fluctuations, which in turn affect corporate returns. However, the current research on market price linkage and sentiment linkage is still in a relatively fragmented state. Summary of the invention

[0009] In view of the shortcomings of the prior art, the present invention provides a data processing method and system based on multi-layer network technology, which solves the problem of inaccurate prediction caused by the lack of consideration of the complexity and multiplicity of network correlation in measuring institutional correlation and risk evolution process.

[0010] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0011] A data processing method and system based on multi-layer network technology, the data processing method comprises the following steps:

[0012] S1. Construct a single-layer network based on institutional yields and sentiment indexes. On this basis, construct a multi-layer network with institutions as nodes, the relationship between institutional yields and the relationship between market sentiment indexes as edges;

[0013] S2, correlation network feature extraction;

[0014] S3. Consider the predictive ability of single-layer network characteristics and multi-layer network characteristics for risk at the institutional return layer and sentiment index layer respectively, and compare the predictive ability of single-layer network characteristics with that of multi-layer network characteristics.

[0015] Preferably, the institutional rate of return network is specifically manifested as:

[0016] For m institutions, their rate of return is R t There exists a stationary QVAR process:

[0017]

[0018] In the formula, R t =(R 1t ,R 2t ,…,R mt ) T is the m-dimensional institutional yield vector at time t;

[0019] τ is the quantile, satisfying 0<τ<1;

[0020] is the return vector of the institution at the τth quantile;

[0021] β′ 0,t (τ) is a constant term;

[0022] β′ t (τ)=[β′ ij,t (τ)] is an m×m matrix;

[0023] Element β′ ij,t(τ) is the impact of institution j at time (t-1) on the τth quantile of the return distribution of institution τ at time t;

[0024] Add L to the loss function in the above formula 1 Regularizing the penalty term, we get:

[0025]

[0026] In the formula, is the value to be estimated;

[0027] ρ τ (u) = u(τ-I(u<0)) is an asymmetric absolute value loss function;

[0028] ||β′ t (τ)|| 1 For L 1 -norm;

[0029] λ is a regularization parameter, and a larger value indicates a greater penalty.

[0030] Preferably, the emotion index network is specifically manifested as:

[0031] The directional spillover of institution j to financial institution i is represented by the proportion of institution j’s contribution to the h-step generalized forecast error variance of institution i;

[0032] Construct the VAR(p) model of each institution's sentiment index. The specific expression is:

[0033]

[0034] In the formula, Index t =(Index 1t ,Index 2t ,…,Index mt ) T is the m-dimensional sentiment index vector at time t;

[0035] p is the lag order, determined using the AIC and BIC criteria;

[0036] ε t is the random error vector;

[0037] VAR can be transformed into a vector moving average process, namely:

[0038]

[0039] In the formula, A i is an m×m matrix, satisfying When i = 0, A i is the unit matrix; when i<0, Ai is a 0 matrix;

[0040] Using the generalized variance decomposition method, the specific expression is as follows:

[0041]

[0042] In the formula, θ ij (H) represents the h-step generalized forecast error variance of the sentiment index of institution j for the sentiment index of institution i;

[0043] σ ii is the i-th diagonal element of the covariance matrix;

[0044] e j To select a column vector, that is, the jth element in the vector is 1 and the other elements are 0;

[0045] To calculate the overflow index value, ij (H) Perform normalization, that is:

[0046]

[0047] Based on the normalization process, we can know that and

[0048] At this time, the directional spillover or correlation from institution j to institution i is defined as:

[0049]

[0050] Based on this, we can obtain the correlation matrix β″ t (H) represents the sentiment index network.

[0051] Preferably, the multi-layer association network is specifically manifested as:

[0052] Two-layer sparse coefficient matrix

[0053]

[0054] In the formula, α=1, 2 represent the institutional yield layer and the sentiment index layer respectively;

[0055] This study considers a multi-layer tail association network Ω = {G 1 ,G 2},

[0056] V is the vertex (organization) set.

[0057] Preferably, the associated network feature extraction includes: network node feature extraction, network clustering coefficient extraction and average path length extraction;

[0058] The network node feature extraction is specifically manifested as follows:

[0059] We select PageRank centrality as the network node feature and define it as:

[0060]

[0061] In the formula, q is the damping coefficient, which is expressed as the probability of a random node jumping to the next node, and is usually taken as 0.85;

[0062] k is the current iteration number;

[0063] Out j is the out-degree of institution node j;

[0064] The network clustering coefficient extraction is specifically manifested as follows:

[0065] The node clustering coefficient describes the possibility that the neighbor nodes of a node in the network are also neighbors of each other. It can measure the degree of clustering of nodes and is defined as:

[0066]

[0067] In the formula, E i is the number of edges actually existing between neighboring nodes of node i;

[0068] n i is the number of neighboring nodes that node i has;

[0069] is the maximum possible number of edges between neighboring nodes of node i;

[0070] On this basis, the network clustering coefficient is further defined to measure the clustering degree of the entire network; the network clustering coefficient refers to the average value of the clustering coefficients of all nodes in the network, which is defined as:

[0071]

[0072] The average path length extraction is specifically performed as follows:

[0073] The average path length of a network usually reflects the efficiency of information transmission in the network. The shorter the average path length, the higher the efficiency of information transmission in the network. It is defined as:

[0074]

[0075] Where, d ijIt is the sum of the edges on the shortest path from node i to node j;

[0076] is the maximum possible number of edges between all nodes in the network.

[0077] Preferably, the single-layer network feature prediction capability is specifically manifested as:

[0078] Analyze the predictive ability of single-layer network characteristics on risks:

[0079] SysRisk′ t+h =α 0 +α 1 Network t +α 2 Controls t +ε t

[0080] Where SysRisk′ t+h represents the systemic risk of a single-layer network with a leading period of h;

[0081] α 0 , α′ 0 , α 1 , α′ 1 and α 2 , α′ 2 is the coefficient to be estimated;

[0082] Explanatory variables Network t Contains the clustering coefficient and average path length in a single-layer network at time t;

[0083] Select money supply M2, fiscal balance FB and domestic credit DC as control variables t ;

[0084] ε t is a random error term.

[0085] Preferably, the multi-layer network feature prediction capability is specifically manifested as:

[0086] Analyze the predictive ability of multi-layer network characteristics on risks:

[0087] Analyze the predictive ability of multi-layer network characteristics on risks:

[0088] SysRisk t+h =α 3 +α 4 Net t +α 5 Controls t +ε t

[0089] Where, SysRisk″ t+h represents the multi-layer network systemic risk leading h periods;

[0090] α 3 , α′ 3 , α 4 , α′ 4 and α 5 , α′ 5 is the coefficient to be estimated;

[0091] Explanatory variable Net t is the clustering coefficient and average path length in the multi-layer association network containing time t;

[0092] Controls t It is money supply M2, fiscal balance FB and domestic credit DC;

[0093] ε t is a random error term.

[0094] Preferably, the network feature prediction capability comparison is specifically manifested as follows:

[0095] Using the Clark and West test, the null hypothesis is that the predictions obtained in the two prediction models perform the same, and rejection of the null hypothesis of equal prediction accuracy means that the extended model that includes network features outperforms the simpler baseline model;

[0096] To examine the incremental value of the extended models that include network features for predicting risk, we focus on evaluating the predictive performance that can be achieved when incorporating these metrics relative to an otherwise identical baseline predictive model that excludes network features; thus, the form of the predictive regression for the baseline model excludes the network features. t :

[0097] SysRisk t+h =α 6 +α 7 Controls t +ε t

[0098] Where SysRisk t+h The single-layer network systemic risk SysRisk′ includes the leading h period t+h and the multi-layer network systemic risk SysRisk″ with a leading period of h t+h ;

[0099] α 6 , α′ 6 and α 7 , α′ 7 is the coefficient to be estimated;

[0100] By observing the p-value or R 2 On the one hand, we observe the predictive ability of single-layer and multi-layer networks relative to the baseline model, and on the other hand, we observe the difference in predictive ability between single-layer and double-layer networks.

[0101] A data processing system based on multi-layer network technology, the processing system comprising: a network construction module, a network feature extraction module and a capability prediction and comparison module;

[0102] The network construction module is used to construct a single-layer network based on institutional yields and sentiment indexes, and on this basis, a multi-layer network is constructed with institutions as nodes, and the relationship between institutional yields and the relationship between market sentiment indexes as edges;

[0103] The network feature extraction module is used to extract associated network features;

[0104] The capability prediction and comparison module is used to consider the risk prediction capabilities of single-layer network features and multi-layer network features of the institutional return layer and the sentiment index layer respectively, and compare the prediction capabilities of single-layer network features with those of multi-layer network features.

[0105] The present invention provides a data processing method and system based on multi-layer network technology. Compared with the prior art, it has the following beneficial effects:

[0106] In the present invention, the data processing method constructs a multi-layer network technology based on institutional yield and sentiment index, which can not only capture the interdependence between different tail institutions in detail from the perspective of the institution itself, but also analyze the risk contagion between institutions from the perspective of the external market of the institution. The network connection is more real and objective, has a clear return meaning, is highly interpretable, and can more clearly reveal the formation and contagion mechanism of risks; and focuses on analyzing the topological structure and clustering characteristics of the multi-layer network, and exploring the inherent evolution characteristics of the network, which is of great significance to both institutions and risk supervision; at the same time, combining the multi-layer network technology to study the risk contribution of each institution, it is helpful to comprehensively identify systemically important institutions from the perspective of complex networks; compared with single-layer network technology, multi-layer network technology helps to improve the ability to predict systemic risks; risks will not only be transmitted within each indicator, but also be transmitted interactively. It is reasonable and effective to use multi-layer network technology to conduct early warning tests on systemic risks in a certain period in the future, which greatly improves the accuracy of prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0108] Figure 1 Flow chart of a data processing method in an embodiment of the present invention. DETAILED DESCRIPTION

[0109] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0110] The embodiments of the present application provide a data processing method and system based on multi-layer network technology to solve the problem of inaccurate prediction due to the lack of consideration of the complexity and multiplicity of network correlation in measuring institutional correlation and risk evolution process.

[0111] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0112] Example:

[0113] like Figure 1 As shown, the present invention provides a data processing method based on multi-layer network technology, and the data processing method includes the following steps:

[0114] S1. Construction of multi-layer network based on institutional yield and sentiment index

[0115] A single-layer network based on institutional yields and sentiment indexes is constructed. On this basis, a multi-layer network is constructed with institutions as nodes, the relationship between institutional yields, and the relationship between market sentiment indexes as edges.

[0116] Institutional Yield Network:

[0117] For m institutions, let their rate of return R t There exists a stationary QVAR process:

[0118]

[0119] In the formula, R t =(R 1t ,R2t ,…,R mt ) T is the m-dimensional institutional yield vector at time t;

[0120] τ is the quantile, satisfying 0<τ<1;

[0121] is the return vector of the institution at the τth quantile;

[0122] β′ 0,t (τ) is a constant term;

[0123] β′ t (τ)=[β′ ij,t (τ)] is an m×m matrix;

[0124] Element β′ ij,t (τ) is the impact of institution j at time (t-1) on the τth quantile of the return distribution of institution i at time t;

[0125] In order to accurately identify and select all other institutions that have a significant impact on the τth quantile of the return distribution of institution i, the LASSO model used in this paper compresses the coefficients with weak correlation to 0, thereby achieving variable selection; L is added to the loss function of the above formula 1 Regularizing the penalty term, we get:

[0126]

[0127] In the formula, is the value to be estimated;

[0128] ρ τ (u) = u(τ-I(u<0)) is an asymmetric absolute value loss function;

[0129] ||β′ t (τ)|| 1 For L 1 -norm;

[0130] λ is a regularization parameter, and a larger value indicates a greater penalty.

[0131] Sentiment Index Network:

[0132] The directional spillover of institution j to financial institution i is represented by the proportion of institution j’s contribution to the h-step generalized forecast error variance of institution i;

[0133] Construct the VAR(p) model of each institution's sentiment index. The specific expression is:

[0134]

[0135] In the formula, Indext =(Index 1t ,Index 2t ,…,Index mt ) T is the m-dimensional sentiment index vector at time t;

[0136] p is the lag order, determined using the AIC and BIC criteria;

[0137] ε t is the random error vector;

[0138] Because Index t The covariance is stable, so VAR can be transformed into a vector moving average process, that is:

[0139]

[0140] In the formula, A i is an m×m matrix, satisfying When i = 0, A i is the unit matrix; when i<0, A i is a 0 matrix.

[0141] Using the generalized variance decomposition method, the specific expression is as follows:

[0142]

[0143] In the formula, θ ij (H) represents the h-step generalized forecast error variance of the sentiment index of institution j for the sentiment index of institution i;

[0144] σ ii is the i-th diagonal element of the covariance matrix;

[0145] e j To select a column vector, that is, the jth element in the vector is 1 and the other elements are 0;

[0146] To calculate the overflow index value, ij (H) Perform normalization, that is:

[0147]

[0148] Based on the normalization process, we can know that and

[0149] At this time, the directional spillover or correlation from institution j to institution i is defined as:

[0150]

[0151] Based on this, we can obtain the correlation matrix β″t (H) represents the sentiment index network.

[0152] Multi-layer association network:

[0153] Based on the above model, we can get a double-layer sparse coefficient matrix

[0154]

[0155] In the formula, α=1, 2 represent the institutional yield layer and the sentiment index layer respectively;

[0156] This study considers a multi-layer tail association network Ω = {G 1 ,G 2},

[0157] V is the vertex (organization) set.

[0158] S2. Extraction of correlation network features

[0159] Network Node Characteristics

[0160] In order to quantify the importance of a single institution in the tail association network, PageRank centrality is selected as the network node feature and is defined as

[0161]

[0162] In the formula, q is the damping coefficient, which is expressed as the probability of a random node jumping to the next node, and is usually taken as 0.85;

[0163] k is the current iteration number;

[0164] Put j is the out-degree of institution node j.

[0165] Overall network characteristics

[0166] For the overall characteristics of the network, the network clustering coefficient and average path length are selected.

[0167] Network clustering coefficient

[0168] The node clustering coefficient describes the possibility that the neighbor nodes of a node in the network are also neighbors of each other. It can measure the degree of clustering of nodes and is defined as:

[0169]

[0170] In the formula, E i is the number of edges actually existing between neighboring nodes of node i;

[0171] ni is the number of neighboring nodes that node i has;

[0172] is the maximum possible number of edges between neighboring nodes of node i;

[0173] On this basis, the network clustering coefficient is further defined to measure the clustering degree of the entire network; the network clustering coefficient refers to the average value of the clustering coefficients of all nodes in the network, which is defined as:

[0174]

[0175] Average path length

[0176] The average path length of a network usually reflects the efficiency of information transmission in the network. The shorter the average path length (APL), the higher the efficiency of information transmission in the network. It is defined as:

[0177]

[0178] Where, d ij It is the sum of the edges on the shortest path from node i to node j;

[0179] is the maximum possible number of edges between all nodes in the network.

[0180] S3. Analysis of the predictive power of network features

[0181] The predictive ability of single-layer network characteristics and multi-layer network characteristics (both including two types of network characteristics, the average clustering coefficient and the average path length) for risk are considered respectively at the institutional return layer and the sentiment index layer, and the predictive ability of single-layer network characteristics is compared with that of multi-layer network characteristics to observe whether the multi-layer network has better predictive ability than the single-layer network characteristics.

[0182] Single-layer network feature prediction capability

[0183] Analyze the predictive ability of single-layer network characteristics on risks:

[0184] SysRisk′ t+h =α 0 +α 1 Network t +α 2 Controls t +ε t

[0185] Where SysRisk′ t+hrepresents the systemic risk of a single-layer network with a leading period of h;

[0186] α 0 , α′ 0 , α 1 , α′ 1 and α 2 , α′ 2 is the coefficient to be estimated;

[0187] Explanatory variables Network t Contains the clustering coefficient (Clu) and average path length (APL) in a single-layer network at time t;

[0188] Select money supply M2, fiscal balance FB and domestic credit DC as control variables t ;

[0189] ε t is a random error term.

[0190] Multi-layer network feature prediction capabilities

[0191] Analyzing the predictive power of multi-layer network characteristics on risk

[0192] Analyze the predictive ability of multi-layer network characteristics on risks:

[0193] SysRisk t+h =α 3 +α 4 Net t +α 5 Controls t +ε t

[0194] Where, SysRisk″ t+h represents the multi-layer network systemic risk leading h periods;

[0195] α 3 , α′ 3 , α 4 , α′ 4 and α 5 , α′ 5 is the coefficient to be estimated;

[0196] Explanatory variable Net t is the clustering coefficient and average path length in the multi-layer association network containing time t;

[0197] Controls t It is money supply M2, fiscal balance FB and domestic credit DC;

[0198] ε tis the random error term;

[0199] Comparison of network feature prediction capabilities

[0200] Using the Clark and West test, the null hypothesis is that the predictions obtained in the two prediction models perform equally well, and rejection of the null hypothesis of equal prediction accuracy means that the extended model that includes network features outperforms the simpler baseline model.

[0201] To examine the incremental value of the extended model that includes network features for predicting risk, we focus on evaluating the predictive performance that can be achieved when incorporating these metrics relative to an otherwise identical baseline predictive model that excludes network features. t :

[0202] SysBisk t+h =α 6 +α 7 Controls t +ε t

[0203] Where SysRisk t+h The single-layer network systemic risk SysRisk′ includes the leading h period t+h and the multi-layer network systemic risk SysRisk″ with a leading period of h t+h ;

[0204] α 6 , α′ 6 and α 7 , α′ 7 is the coefficient to be estimated;

[0205] By observing the p-value or R 2 On the one hand, we observe the predictive ability of single-layer and multi-layer networks relative to the baseline model, and on the other hand, we observe the difference in predictive ability between single-layer and double-layer networks.

[0206] The present invention provides a data processing system based on multi-layer network technology, the data processing system comprising: a network construction module, a network feature extraction module and a capability prediction and comparison module;

[0207] The network construction module is used to construct a single-layer network based on institutional yields and sentiment indexes, and on this basis, a multi-layer network is constructed with institutions as nodes, and the relationship between institutional yields and the relationship between market sentiment indexes as edges;

[0208] The network feature extraction module is used to extract associated network features;

[0209] The capability prediction and comparison module is used to consider the risk prediction capabilities of single-layer network features and multi-layer network features of the institutional return layer and the sentiment index layer respectively, and compare the prediction capabilities of single-layer network features with those of multi-layer network features.

[0210] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0211] 1. In the embodiment of the present invention, the method constructs a multi-layer network technology based on institutional yields and sentiment indexes, which can not only capture the interdependence between institutions at different tails from the perspective of the institutions themselves, but also analyze the risk contagion between institutions from the perspective of the external market of the institutions. The network connection is more real and objective, has a clear meaning of returns, is highly interpretable, and can more clearly reveal the formation and contagion mechanism of risks.

[0212] 2. In the embodiment of the present invention, the method focuses on analyzing the topological structure and clustering characteristics of the multi-layer network and exploring the intrinsic evolution characteristics of the network, which is of great significance to both institutional and risk supervision; at the same time, the risk contribution of each institution is studied in combination with multi-layer network technology, which helps to comprehensively identify systemically important institutions from the perspective of complex networks.

[0213] 3. In the embodiments of the present invention, multi-layer network technology helps to improve the ability to predict systemic risks compared to single-layer network technology; risks will not only spread within each indicator, but also spread interactively. Therefore, it is reasonable and effective to conduct early warning tests on systemic risks in a certain period of time in the future through multi-layer network technology.

[0214] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0215] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data processing method based on multi-layer network technology, characterized in that: The data processing method comprises the following steps: S1. Construct a single-layer network based on institutional yields and sentiment indexes. On this basis, construct a multi-layer network with institutions as nodes, the relationship between institutional yields and the relationship between market sentiment indexes as edges; S2, correlation network feature extraction; S3. Consider the predictive ability of single-layer network characteristics and multi-layer network characteristics for risk at the institutional return layer and sentiment index layer respectively, and compare the predictive ability of single-layer network characteristics with that of multi-layer network characteristics.

2. The data processing method based on multi-layer network technology according to claim 1, characterized in that: The institutional yield network is specifically manifested as follows: For m institutions, their rate of return is R t There exists a stationary QVAR process: Q Rt (t)=β′ 0,t (t)+b′ t (t)R t-1 In the formula, R t =(R 1t ,R 2t ,…,R mt ) T is the m-dimensional institutional yield vector at time t; τ is the quantile, satisfying 0<τ<1; is the return vector of the institution at the τth quantile; β′ 0,t (τ) is a constant term; β′ t (τ)=[β′ ij,t (τ)] is an m×m matrix; Element β′ ij,t (τ) is the impact of institution j at time (t-1) on the τth quantile of the return distribution of institution i at time t; Adding the L1 regularization penalty term to the loss function above yields: In the formula, is the value to be estimated; ρ τ (u) = u(τ-I(u<0)) is an asymmetric absolute value loss function; ||β′ t (τ)||1 is the L1-norm; λ is a regularization parameter, and a larger value indicates a greater penalty.

3. The data processing method based on multi-layer network technology as claimed in claim 2, characterized in that: The emotion index network is specifically manifested as follows: The directional spillover of institution j to financial institution i is represented by the proportion of institution j’s contribution to the h-step generalized forecast error variance of institution i; Construct the VAR(p) model of each institution's sentiment index. The specific expression is: In the formula, Index t =(Index 1t ,Index 2t ,…,Index mt ) T is the m-dimensional sentiment index vector at time t; p is the lag order, determined using the AIC and BIC criteria; ε t is the random error vector; VAR can be transformed into a vector moving average process, namely: In the formula, A i is an m×m matrix, satisfying When i = 0, A i is the identity matrix; When i<0, A i is a 0 matrix; Using the generalized variance decomposition method, the specific expression is as follows: In the formula, θ ij (H) represents the h-step generalized forecast error variance of the sentiment index of institution j for the sentiment index of institution i; σ ii is the i-th diagonal element of the covariance matrix; e j To select a column vector, that is, the jth element in the vector is 1 and the other elements are 0; To calculate the overflow index value, ij (H) Perform normalization, that is: Based on the normalization process, we can know that and At this point, the directional spillover or association from institution j to institution i is defined as: Based on this, we can obtain the correlation matrix β t ″″(H) represents the sentiment index network.

4. The data processing method based on multi-layer network technology according to claim 3, characterized in that: The multi-layer association network is specifically manifested as follows: Two-layer sparse coefficient matrix In the formula, α=1, 2 represent the institutional yield layer and the sentiment index layer respectively; This study considers a multi-layer tail association network Ω = {G 1 ,G 2 }, V is the vertex (organization) set.

5. The data processing method based on multi-layer network technology according to claim 1, characterized in that: The associated network feature extraction includes: network node feature extraction, network clustering coefficient extraction and average path length extraction; The network node feature extraction is specifically manifested as follows: We select PageRank centrality as the network node feature and define it as: In the formula, q is the damping coefficient, which is expressed as the probability of a random node jumping to the next node, and is usually taken as 0.85; k is the current iteration number; Out j is the out-degree of institution node j; The network clustering coefficient extraction is specifically manifested as follows: The node clustering coefficient describes the possibility that the neighbor nodes of a node in the network are also neighbors of each other. It can measure the degree of clustering of nodes and is defined as: In the formula, E i is the number of edges actually existing between neighboring nodes of node i; n i is the number of neighboring nodes that node i has; is the maximum possible number of edges between neighboring nodes of node i; On this basis, the network clustering coefficient is further defined to measure the clustering degree of the entire network; the network clustering coefficient refers to the average value of the clustering coefficients of all nodes in the network, which is defined as: The average path length extraction is specifically performed as follows: The average path length of a network usually reflects the efficiency of information transmission in the network. The shorter the average path length, the higher the efficiency of information transmission in the network. It is defined as: Where, d ij It is the sum of the edges on the shortest path from node i to node j; is the maximum possible number of edges between all nodes in the network.

6. The data processing method based on multi-layer network technology according to claim 1, characterized in that: The single-layer network feature prediction capability is specifically manifested as follows: Analyze the predictive ability of single-layer network characteristics on risks: SysRisk′ t+h =α0+α1Network t +α2Controls t +ε t Where SysRisk′ t+h represents the systemic risk of a single-layer network with a leading period of h; α0, α′0, α1, α′1, α2, α′2 are coefficients to be estimated; Explanatory variables Network t Contains the clustering coefficient and average path length in a single-layer network at time t; Select money supply M2, fiscal balance FB and domestic credit DC as control variables t ; ε t is a random error term.

7. The data processing method based on multi-layer network technology according to claim 6, characterized in that: The multi-layer network feature prediction capability is specifically manifested as follows: Analyze the predictive ability of multi-layer network characteristics on risks: SysRisk″ t+h =α3+α4Net t +α5Controls t +ε t Where, SysRisk″ t+h represents the multi-layer network systemic risk leading h periods; α3, α′3, α4, α′4, α5, α′5 are coefficients to be estimated; Explanatory variable Net t is the clustering coefficient and average path length in the multi-layer association network containing time t; Controls t It is money supply M2, fiscal balance FB and domestic credit DC; ε t is a random error term.

8. The data processing method based on multi-layer network technology according to claim 7, characterized in that: The network feature prediction capability comparison is specifically shown as follows: Using the Clark and West test, the null hypothesis is that the predictions obtained in the two prediction models perform the same, and rejection of the null hypothesis of equal prediction accuracy means that the extended model that includes network features outperforms the simpler baseline model; To examine the incremental value of the extended models that include network features for predicting risk, we focus on evaluating the predictive performance that can be achieved when incorporating these metrics relative to an otherwise identical baseline predictive model that excludes network features; thus, the form of the predictive regression for the baseline model excludes the network features. t : SysRisk t+h =α6+α7Controls t +ε t Where SysRisk t+h The single-layer network systemic risk SysRisk′ includes the leading h period t+h and the multi-layer network systemic risk SysRisk″ with a leading period of h t+h ; α6, α′6, α7, α′7 are coefficients to be estimated; By observing the p-value or R 2 On the one hand, we observe the predictive ability of single-layer and multi-layer networks relative to the baseline model, and on the other hand, we observe the difference in predictive ability between single-layer and double-layer networks.

9. A data processing system based on multi-layer network technology, characterized in that: The data processing system includes: a network construction module, a network feature extraction module and a capability prediction and comparison module; The network construction module is used to construct a single-layer network based on institutional yields and sentiment indexes, and on this basis, a multi-layer network is constructed with institutions as nodes, and the relationship between institutional yields and the relationship between market sentiment indexes as edges; The network feature extraction module is used to extract associated network features; The capability prediction and comparison module is used to consider the risk prediction capabilities of single-layer network features and multi-layer network features of the institutional return layer and the sentiment index layer respectively, and compare the prediction capabilities of single-layer network features with those of multi-layer network features.

Images