Default risk assessment method and system for trading entities in commodity trading markets

Abstract

The present invention discloses a method and system for assessing the default risk of trading entities in a commodity trading market, which integrates the relationship between entities in a trading network and the network spillover effect into the assessment of the default risk of trading entities. First, the system constructs a trading network based on market trading data and trading entity data; then, the single trading entity characteristics and trading network characteristics are used to construct a single trading entity assessment model and a trading network spillover effect assessment model; finally, the used single trading entity characteristics and trading network characteristics are integrated and selected, and a hybrid model is constructed based on the selected characteristics as the final model, which is applied to the system. The trading entity default risk assessment system based on this final model integrates the internal factors of the entity and the external market factors, eliminates the limitation of the traditional assessment method based only on the characteristics of a single trading entity, improves the assessment effect, and can better adapt to the dynamics of the trading market.

Description

Technical Field

[0001] The present invention belongs to the technical field of bulk commodity e-commerce supervision, and specifically relates to a method and system for assessing default risks of trading entities in a bulk commodity trading market. Background Art

[0002] In recent years, with the development of the electronic commodity trading market, the types of commodities traded have gradually involved many industries and fields such as chemicals, grain, steel, plastics, and precious metals. Various risk events have occurred frequently during its rapid development.

[0003] Commodity buyers, traders, manufacturers, etc., in commodity transactions, are subject to the possibility of defaulting due to their unwillingness or inability to fulfill the contractual conditions for various reasons, causing losses to counterparties. This risk is called default risk. In the procurement process, traders must prepay the supplier for goods, and the supplier will collect the payment and ship the goods. Once the goods are not shipped as agreed in the contract, or the supplier is maliciously deceiving, the company's prepaid accounts face a certain risk of impairment loss; in the sales process, due to the early release of the right to goods, if the subsequent payment cannot be recovered or is difficult to recover, the risk of loss of accounts receivable of traders is relatively high. During the period when traders have the right to goods, the costs involved in logistics distribution, warehousing management, loading and unloading in the trade process are generally paid in advance by traders, and recovered from customers when settling with downstream customers. In addition, in the agency model, traders incur capital occupation fees when purchasing for customers, which are paid by customers at the time of settlement. If problems arise during the transaction process that lead to unsmooth transactions, traders may bear the relevant costs themselves and face certain cost loss risks. Default risk is highly correlated with the credit level of financial institutions and is more sensitive to market fluctuations. Risk control for default risk needs to start from the root cause of its occurrence and achieve an organic combination of external supervision and internal control.

[0004] Traditional default risk prediction models are mostly targeted at single trading entities, focusing on the internal characteristics of the company and the macro market situation. In recent years, the electronic trading market for commodities has achieved the integration of online and offline resources and the synergy of the supply chain through the "e-commerce + modern logistics + supply chain financing" operation model. Therefore, the exploration of the relationship between the various entities in the electronic trading supply chain of commodities and the research on supply chain finance are also of great significance to the control and early warning of systemic risks. The supply chain connects suppliers and buyers, forming a huge trading network in the market. In such a trading network, an important financial relationship that maintains the various trading entities is commercial credit, that is, the inter-enterprise credit relationship formed by the deferred payment or advance payment of enterprises in normal business activities and commodity transactions. In essence, commercial credit is equivalent to short-term financing from the buyer to the seller, which can replace bank credit financing to a certain extent. On the one hand, commercial credit is an important financing channel for manufacturers (raw material buyers); at the same time, commercial credit is also an important means for suppliers (raw material providers) to maintain production-market relationships and protect future income. On the other hand, shocks or imbalances experienced by one company may spread along the supply chain: intermediaries may respond to delayed payments by downstream buyers by delaying payments to upstream suppliers, thus creating a cascading effect in the trading network.

[0005] Therefore, it is necessary to propose a default risk assessment method and system for trading entities in commodity trading networks, integrating the network spillover effect generated by commercial credit into the assessment of default risk of trading entities, in order to improve the assessment and prediction capabilities of default behavior of trading entities. Summary of the invention

[0006] In the electronic commodity trading market, various trading entities form inter-enterprise credit relationships, namely commercial credit relationships, due to deferred payments or advance payments in normal business activities and commodity transactions. This relationship is an important financing channel for manufacturers (raw material purchasers), and it is also an important means for suppliers (raw material providers) to maintain production-market relationships and protect future income. When the assets of a trading entity encounter shocks or imbalances, such shocks or imbalances are likely to spread through commercial credit relationships, thereby forming a cascading effect in the trading network. Traditional default risk assessment methods lack the examination of commercial credit channel transmission, that is, the lack of comprehensive consideration of the spillover effects of the trading network, which makes the assessment limited and inaccurate.

[0007] To solve the above problems, the present invention discloses a method and system for assessing the default risk of trading entities in a commodity trading market. First, a commodity trading network is constructed based on market trading data and trading entity information data in the commodity trading market. Then, single trading entity characteristics and trading network characteristics are used respectively to construct a single trading entity assessment model and a trading network spillover effect assessment model. Finally, the used single trading entity characteristics and trading network characteristics are integrated and selected, and a hybrid model is constructed based on the selected characteristics as the final model, which is applied to the system to assess the default risk of trading entities. This method solves the assessment limitations and accuracy problems caused by the fact that the default risk assessment of trading entities in the commodity trading market only considers the characteristics of the entity itself.

[0008] The specific technical solution of the risk assessment method is as follows:

[0009] The first part is the construction of the trading network. In the electronic commodity trading market, each trading entity may have one or more trading counterparties. The trading entity is regarded as a node, and each pair of trading entities with a trading relationship is connected by a directed edge. The direction of the edge is selected as the direction of cash flow, and the opposite direction of the edge is the direction of goods circulation. For each edge, appropriate transaction-related data is selected as the weight of the edge. From the above, a basic trading network represented by a weighted directed graph G(V, E, w) can be obtained. Considering that the propagation of default risk in the trading network also involves the time dimension, the trading network graph itself will also change with the changes in the trading relationship between trading entities. Therefore, we add the time dimension to the graph G to obtain the dynamic graph G that is finally used to characterize the trading network. t (V,E t , w t ), where the node set is fixed, and the edge set and edge weights vary in the time dimension.

[0010] The second part is the default risk assessment of a single trading entity. In this part, the system only uses the characteristics of the trading entity itself to assess the default risk of a single trading entity. This part of the assessment uses the assessment value given by the assessment model used by the exchange or regulatory agency as the main feature, and at the same time selects the characteristics related to the historical default of the target trading entity as input features. The supervised learning method is used to obtain the model and the default risk assessment value of the single trading entity as the input for the third part of the transaction network spillover effect assessment.

[0011] The third part is the evaluation of the spillover effect of the trading network. In this part, the system only uses the features derived from the trading network and the default risk assessment value of the single trading subject in the second part to evaluate the default risk of a single trading subject. In other words, this part of the evaluation only considers the risk brought by the counterparty to the target trading subject, which can be transmitted from any node in the network along the network to the target trading subject. This part mainly examines two characteristics of the target trading subject: vulnerability and personalized PageRank. The vulnerability feature characterizes the risk situation of the target company itself, considering all adjacent nodes within one hop from the target trading subject; the personalized PageRank feature characterizes the distance between the target company and the default node in the trading network, which is obtained by applying the adjusted personalized PageRank algorithm to the default node of the trading network. Finally, linear regression is applied to the input features to obtain the model and the evaluation value of the spillover effect of the trading network suffered by the single trading subject.

[0012] The fourth part is the construction of the hybrid model. In this part, the data-driven method is used to integrate and select the characteristics of single trading entities and trading networks, and a hybrid model is constructed based on the selected characteristics as the final model, which is then applied to the system to effectively evaluate the default risk of trading entities in the commodity trading network.

[0013] The beneficial effects of the present invention are:

[0014] (1) Eliminating the limitations of traditional default risk assessment methods Compared with traditional default risk assessment methods, this method takes into account the characteristics of the transaction network, eliminating the limitations of traditional methods that only assess based on the characteristics of a single transaction entity.

[0015] (2) To improve the effectiveness of default risk assessment for trading entities, the spillover effect of the trading network is evaluated and verified. Finally, a hybrid model is formed by combining the characteristics of a single trading entity and the characteristics of the trading network and applied to the system to effectively improve the effectiveness of default risk assessment for trading entities.

[0016] (3) Adapting to the dynamics of the trading market Taking into account the dynamic changes of the trading market and adding the time dimension when constructing the trading network can better characterize the development and propagation of default risks in the trading network and adapt to the dynamics of the trading market. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the main principles of the present invention.

[0018] Figure 2 Schematic diagram of the components of the trading network built for the invented technology. DETAILED DESCRIPTION

[0019] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0020] As shown in the figure, the method and system for assessing default risk of trading entities in a commodity trading market according to the present invention include the following steps:

[0021] (1) Construction of trading network: In the initial stage, the system constructs a commodity trading network based on market trading data and trading subject information data. The trading subjects are regarded as network nodes, and each pair of trading subjects with a trading relationship is connected by a directed edge. The direction of the edge is selected as the direction of cash flow in the corresponding transaction, and the opposite direction of the edge is the direction of commodity circulation, such as Figure 2 As shown above, the connectivity relationship between all trading entities in the market is modeled, that is, the basic representation of the trading network is a weighted directed graph G(V, E, w), where V is the set of trading entity nodes, E is the set of directed edges, and a directed edge (u, v) indicates that trading entity u is a customer of trading entity v, so the default risk can be propagated along the direction of the edge, as shown in Figure 2 As shown in Figure 2. w(u, v) represents the weight of the directed edge (u, v). For each edge, the total transaction amount between the two transaction entities involved is selected as the edge weight to characterize the strength of the connection between them. Considering that the propagation of default risk in the transaction network also involves the time dimension, the transaction network itself will also change with the changes in the transaction relationship between the transaction entities. Therefore, we add the time dimension to the graph G and obtain the dynamic graph G used to characterize the transaction network. t (V,E t , w t ), the node set is fixed, but the edge set and edge weights will change in the time dimension. The subsequent evaluation of the transaction network spillover effect will be based on the transaction network dynamic graph obtained in this step.

[0022] (2) Default risk assessment of a single trading entity: In the second stage, the system uses only the characteristics of the trading entity itself to assess the default risk of a single trading entity. The assessment model used in this stage mainly uses three different trading entity characteristics as input features. The first input feature is the assessment model used internally by the exchange or regulatory agency, which is based on the trading entity's own financial information such as the trading entity's debt situation to derive the default probability P(d). This internal assessment model represents an official, econometric method for describing the default risk of the trading entity. For this default probability P(d), we convert it into a logarithmic probability score Score by applying the logit function, as shown in the following formula:

[0023] Score = log P(d) / (1-household(d))

[0024] The second input feature is one of the most predictive features used by the internal evaluation model: the number of overdraft days of the target trading entity in the past three months. The third input feature Hist is a Boolean indicator variable, which also examines the historical default of the target trading entity: if the target trading entity has defaulted at any time in the past, its Hist is 1. From a statistical point of view, trading entities that have defaulted are more likely to fall into financial difficulties again, thereby inducing secondary default behavior.

[0025] In summary, a reduced single-transaction subject default risk assessment model is constructed in this stage, and its expression is as follows:

[0026] Y=f(Score, Overdraft, Hist)

[0027] Among them, Y is a binary variable, indicating whether the target trading subject will default in the next three months. Random forest is used as the learning algorithm of the model, and the model is trained and verified by applying a sliding time window to the historical data set to obtain the final model of this stage and the default risk assessment value of the single trading subject, which will serve as the input for the subsequent implementation of the transaction network spillover effect assessment.

[0028] (3) Evaluation of the spillover effect of the trading network: In the third stage, the system only uses the features derived from the trading network and the default risk assessment value of the single trading subject in the second part to evaluate the default risk of a single trading subject. In other words, the evaluation at this stage only considers the risk that the counterparty brings to the target trading subject. This risk can be transmitted from any node in the network along the network to the target trading subject, and is partly attributed to the vulnerability of the trading subject in dealing with the risks transmitted from upstream and downstream. In summary, the evaluation model used in this stage mainly examines two characteristics of the target trading subject: vulnerability Frag and personalized PageRank value PPR.

[0029] ●Fragmentation: While providing commercial credit to customers, commercial credit providers expose themselves to risks from customers. Fragility is a description of this phenomenon. The fragility characteristics used in this section consider the scope of one hop, that is, for a given target trading subject, consider the risk exposure caused by the commercial credit between it and its adjacent trading subjects within one hop. Fragility can also be understood as a measure of the reliability of the commercial credit between it and its adjacent trading subjects within one hop.

[0030] Assume that the risk that the trading subject exposes itself to by granting commercial credit is proportional to its accounts receivable AR (Account Receivables), that is, the accounts payable but not yet paid by the purchasing unit due to the sale of goods and products to the purchasing unit in the normal course of business. i , its accounts receivable AR i The larger the net sales S is, the greater the risk exposure in the commercial credit channel will be. The impact of delayed payment from downstream customers will be greater, and it is more likely to delay payment to its upstream suppliers, thus extending asset shocks to the upstream. i To characterize the size of the transaction entity, then S i The higher the value, the lower the commercial credit risk exposure of transaction subject i. The larger the size of the transaction subject, the less likely it is to be subject to loan constraints, the greater the possibility of obtaining bank loans, and thus better able to absorb asset shocks. It is also less likely to respond to downstream customers' delayed payments by delaying payments to upstream suppliers.

[0031] Similar logic also appears in the opposite direction of the supply chain, that is, the transmission of default risk from suppliers to trading entities. i To characterize the size of the company, then P i The higher the value, the lower the commercial credit risk exposure of the transaction subject i. The larger the purchase scale of a company, the stronger its market power, which is reflected in the greater possibility of obtaining deferred payment permission or other support measures from suppliers when it encounters asset shocks. On the other hand, for the transaction subject, its accounts payable AP i The larger the Account Payables, the more sensitive it is to the financial status of upstream suppliers.

[0032] In summary, for transaction subject i, its vulnerability to the impact of upstream suppliers is defined as s (i) Vulnerability of downstream customers c (i) is the weighted average sum of the default probabilities P(d) of the following adjacent nodes:

[0033]

[0034]

[0035] in, and are the inbound and outbound adjacent node sets of transaction subject i in the transaction network, respectively, and w ij is the weight of the edge between nodes i and j, P(d) jis the default probability of company j calculated by the model in the previous stage. The logit transformation is applied to the weighted average sum of the default probabilities of adjacent nodes to convert them into log odds.

[0036] ●Personalized PageRank value PPR: Personalized PageRank features characterize the distance between the target trading subject and the default node in the current trading network. The trading subject that defaults can cause asset shocks in the trading network and spread along the trading chain. It is assumed that the default risk starts from the default node and randomly walks along the trading network.

[0037] Considering that the closer the trading entity is to the default node, the more likely it is to be affected, a restart probability is introduced into the random walk: the random walk continues to walk along the trading network with a probability of α, and stops walking and restarts from the origin with a probability of 1-α.

[0038] Considering that the same node may be affected by multiple default nodes at the same time, the random walk is set to start (or restart) from any node in a given set. The set is represented by a distribution Q based on all nodes in the trading network, and only takes non-zero values ​​on nodes that have defaulted or are in default. We call Q the restart vector.

[0039] In summary, the calculated stationary distribution of random walk is taken as the model input. This stationary distribution exists and is unique, and is relatively easy to calculate using the PPR algorithm. For each node, starting from the restart vector, solve the following recursive formula:

[0040] PPR α =αPPR α M+(1-α)Q

[0041] Where M is the adjacency matrix of the trading network, Q is the restart vector, α is the shock parameter, and α∈(0,1). Finally, the calculated PPR α The values ​​are normalized as follows to reduce their bias at high-indegree nodes:

[0042]

[0043] The final transaction network spillover effect evaluation model can be expressed as follows:

[0044] Y=f(PPR,Frag c , Frag s , Frag c ·Frag s )

[0045] Among them, PPR is personalized PageRank, Frag c and Frag sConsidering the vulnerability of downstream customers and upstream suppliers, Frag c ·Frag s It is the dot product of two vulnerability features. The linear regression model is selected as the learning algorithm of the model to obtain the final model and the evaluation value of the transaction network spillover effect in this stage, which will be used as a reference for constructing the hybrid model in the subsequent implementation.

[0046] (4) Hybrid model construction: The system integrates the input features used by the single trading entity default risk assessment model in the second part and the trading network spillover effect assessment model in the third part, applies a supervised learning model, and optimizes the model parameters through a cross-validation process to obtain the final model and apply it to the system. The system gives the probability of each trading entity in the trading network to default in the next three months.

[0047] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method and system for assessing default risk of trading entities in a commodity trading market, characterized by: The risk assessment method and system include four parts: transaction network construction, single transaction subject default risk assessment, transaction network spillover effect assessment and hybrid model construction; the hybrid model-based system finally obtained by the method assesses the default risk of transaction subjects by combining two types of characteristics: internal characteristics of transaction subjects and external market characteristics; The transaction network spillover effect assessment is that the system only uses the characteristics derived from the transaction network and the default risk assessment value of the single transaction subject to assess the default risk of a single transaction subject; The assessment only considers the risk that the counterparty brings to the target trading entity. This risk is transmitted from any node in the network to the target trading entity along the network, and is partly attributed to the vulnerability of the trading entity in dealing with the risks transmitted from upstream and downstream. The assessment model used mainly examines two characteristics of the target trading entity: vulnerability and personalized PageRank value; The vulnerability described is that in providing commercial credit to customers, the commercial credit provider exposes itself to risks from customers; Vulnerability is a characterization of this phenomenon; For a given target trading entity, consider the risk exposure caused by the commercial credit between it and its adjacent trading entities within one hop; For transaction subject i, its accounts receivable AR i The larger the net sales S is, the greater the risk exposure in the commercial credit channel will be. The impact of delayed payment by downstream customers will be greater, and it is more likely to delay payment to its upstream suppliers, thus extending the asset shock to the upstream. i To characterize the size of the transaction entity, S i The higher it is, the lower the commercial credit risk exposure of transaction subject i; the larger the size of the transaction subject, the less likely it is to be subject to loan constraints, and the greater the possibility of obtaining bank loans, so it can better absorb asset shocks, and it is less likely to respond to downstream customers' deferred payments by delaying payments to upstream suppliers; Similar logic also appears in the opposite direction of the supply chain, that is, the transmission of default risk from suppliers to trading entities; i To characterize the size of the company, then P i The higher the AP is, the lower the commercial credit risk exposure of the transaction subject i is; the larger the purchase scale of the company, the stronger the market dominance it has, which is reflected in the greater possibility of obtaining deferred payment permission or other support measures from suppliers when it encounters asset shocks; on the other hand, for the transaction subject, its accounts payable AP i The larger the value is, the more sensitive it is to the financial status of the upstream supplier. In summary, for transaction subject i, we define its vulnerability to the impact of upstream suppliers as Frag s (i) Vulnerability of downstream customers c (i) is the weighted average sum of the default probabilities P(d) of the following adjacent nodes: in, and are the inbound and outbound adjacent node sets of transaction subject i in the transaction network, respectively, and w ij is the weight of the edge between nodes i and j, P(d) j is the default probability of company j calculated by the single trading entity default risk assessment model; the weighted average of the default probabilities of adjacent nodes is transformed into a logarithmic probability by applying logit transformation.

2. The method and system for assessing default risk of trading entities in a commodity trading market according to claim 1, characterized in that: The transaction network construction is that the system constructs the commodity transaction network based on the market transaction data and the transaction subject information data in the initial stage; the transaction subjects are taken as network nodes, and each pair of transaction subjects with a transaction relationship is connected by a directed edge, and the direction of the edge is selected as the direction of the cash flow in the corresponding transaction, and the opposite direction of the edge is the direction of the flow of goods; from the above, the modeling of the connectivity relationship between all transaction subjects in the market is obtained, that is, the basic representation of the transaction network is a weighted directed graph G (V, E, w), where V is the transaction subject node set, E is the directed edge set, and the directed edge (u, v) indicates that the transaction subject u is a customer of the transaction subject v, so that the default risk propagates along the direction of the edge; w (u, v) represents the weight of the directed edge (u, v); for each edge, the total transaction amount between the two transaction subjects involved is selected as the edge weight to characterize the strength of the connection relationship between them; considering that the propagation of default risk in the transaction network also involves the time dimension, the transaction network itself will also change with the change of the transaction relationship between the transaction subjects; therefore, the time dimension is added to the graph G to obtain the dynamic graph G used to characterize the transaction network t (V,E t , w t ), the node set is fixed, but the edge set and edge weights may change in the time dimension; the spillover effect of the transaction network is evaluated based on the dynamic graph of the transaction network obtained in this step.

3. The method and system for assessing default risk of trading entities in a commodity trading market according to claim 1, characterized in that: The single transaction subject default risk assessment is that the system uses only the transaction subject's own characteristics to assess the default risk of a single transaction subject; the assessment model used mainly uses three different transaction subject characteristics as input features; The first input feature is the valuation model used internally by the exchange or regulatory agency, which is the default probability P(d) of the trading subject based on its own financial information such as the debt situation of the trading subject. The internal valuation model represents the official, econometric characterization method for the default risk of the trading subject. The default probability P(d) is converted into a logarithmic probability score Score by applying the logit function, expressed as Score = logP(d) / (1-P(d)). The second input feature is one of the most predictive features among the input features used by the internal valuation model: the number of overdraft days of the target trading subject in the past three months. The third input feature Hi st is a Boolean indicator variable, which also examines the historical default of the target trading entity: if the target trading entity has defaulted at any time in the past, its Hist is 1; in summary, a reduced single trading entity default risk assessment model is constructed, expressed as Y = f (Score, Overdraft, Hist); where Y is a binary variable, indicating whether the target trading entity will default in the next three months; random forest is used as the learning algorithm of the model, and the model is trained and verified by applying a sliding time window to the historical data set, and the final model and evaluation value of single trading entity default risk assessment are obtained as the input for the evaluation of the spillover effect of the trading network.

4. The method and system for assessing default risk of trading entities in a commodity trading market according to claim 1, characterized in that: The personalized PageRank value described: the characteristic characterizes the distance between the target trading subject and the default node existing in the current trading network; the trading subject that defaults can cause asset shocks in the trading network and spread along the trading chain; it is assumed that the default risk starts from the default node and randomly walks along the trading network; considering that the closer the trading subject is to the default node, the more easily it is affected, and a restart probability is introduced to the random walk: the random walk continues to walk along the trading network with a probability of α, and stops walking and restarts from the origin with a probability of 1-α; considering that the same node may be affected by multiple default nodes at the same time, it is assumed that the random walk can start or restart from any node in a given set; the set is represented by a distribution Q based on all nodes in the trading network, and only takes non-zero values ​​on nodes that have defaulted or are in default; Q is called the restart vector. In summary, the calculated stationary distribution of random walk is taken as the model input. This stationary distribution exists and is unique, and it is easy to calculate through the PPR algorithm. For each node, starting from the restart vector, solve the recursive formula PPR α =αPPR α The solution of M+(1-α)Q; Where M is the adjacency matrix of the trading network, Q is the restart vector, α is the damping parameter, and α∈(0,1); Finally, the calculated PPR α The values ​​are standardized as follows To reduce its deviation at high in-degree nodes.

5. The method and system for assessing default risk of trading entities in a commodity trading market according to claim 1, characterized in that: The evaluation of the transaction network spillover effect is done by using the formula Y = f(PPR, Frag c , Frag s , Frag c ·Frag s ) indicates that PPR is personalized PageRank, Frag c and Frag s Considering the vulnerability of downstream customers and upstream suppliers, Frag c ·Frag s is the dot product of two vulnerability features; The linear regression model is selected as the learning algorithm of the model to obtain the final model of this stage and the evaluation value of the spillover effect of the transaction network.

6. The method and system for assessing default risk of trading entities in a commodity trading market according to claim 1, characterized in that: The hybrid model construction is to systematically integrate the input features used by the single trading entity default risk assessment model and the trading network spillover effect assessment model, apply a supervised learning model, optimize the model parameters through a cross-validation process, and thus obtain the final model for application in the system; the system gives the probability of each trading entity in the trading network committing a default within the next three months.

Images