An adaptive detection method and system for insider trading in commodities

Abstract

The present invention discloses an adaptive detection method and system for insider trading of commodities. The system extracts historical trading behavior data of trading users, uses multiple trading indicators to characterize, and identifies abnormal subsequences in trading indicators through a time series anomaly detection method to determine abnormal time intervals. When performing time series anomaly detection, the time series is divided into a set to be detected and a detected set. During the iteration process, the time interval most likely to cause insider trading is selected each time, and then the time series is adaptively divided into time intervals that need to be detected and detected. The similarity of trading behavior in each time interval is measured through a self-supervised learning method, and the degree of abnormality is given to the detected time interval. Based on this adaptive detection method and system for insider trading behavior, it helps regulatory authorities to better supervise the market and maintain the stability and fairness 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 an adaptive detection method and system for bulk commodity insider trading behavior. Background Art

[0002] With the development of information technology, the electronic trading market for commodities is developing rapidly. It provides integrated services such as electronic trading, financing, settlement, delivery and price forecasting for producers, sellers and commodity trading through electronic trading systems, which promotes the commodity circulation in the commodity trading market. In the electronic trading process of commodities, there is a possibility of insider trading risks. For example, some financial institutions or regulatory staff can know insider information that ordinary external investors cannot know because of their status or position. Through insider information, they can conduct corresponding transactions for themselves or others to seek benefits. Since there are many trading behaviors, insider trading behaviors are hidden in them. Therefore, it is very important to help regulators dig out the time period when insider trading behaviors may occur through system algorithms.

[0003] When insider trading occurs, insider information only affects the price of certain commodities. In order to seek profits, insider traders will buy and sell certain specific commodities based on the insider information to seek additional benefits for themselves. This kind of buying and selling behavior is different from normal buying and selling behavior. Usually, a large amount of money is invested in a certain commodity in a short period of time, thereby obtaining higher returns than usual. Therefore, the system analyzes whether there is an abnormal trading behavior pattern based on the trading data of the trading user to determine whether it is insider trading. Therefore, the trading behavior is converted into multiple trading indicators, and then the abnormal subsequences in the indicators are identified to determine the abnormal time interval.

[0004] Time series anomaly detection refers to finding some continuous subsequences from the time series that are different from most other subsequences. Corresponding to the insider trading behavior detection method and system, the abnormal subsequences represent the intervals where insider trading occurs, and most subsequences represent the intervals where normal trading occurs. The self-supervised learning method is used to measure the similarity between different subsequences to determine whether some subsequences are different from other subsequences, and identify possible abnormal sequences. Since the time interval where insider trading occurs only accounts for a part of the time interval where all transactions occur, the system divides the regulated time interval into several sub-intervals. If the time interval is divided into smaller ones, it will take a lot of time to detect due to the large number of transactions and large data scale. If it is divided into larger time intervals, although the possible time interval of insider trading can be detected quickly, due to the large time interval given, the regulators need to spend a lot of time to check the trading behavior in the time interval, thus reducing the regulatory effect. Summary of the invention

[0005] To solve the above problems, the present invention discloses an adaptive detection method and system for insider trading in commodities, which can adjust the detection time interval according to the length of the trading time interval, thereby providing the results of insider trading detection through the calculation of trading indicators and the discovery of abnormal subsequences, helping regulators to better supervise insider trading behavior.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] An adaptive detection method and system for insider trading of commodities, the specific technical solution is as follows:

[0008] After the system is initialized, for a given regulatory scope, the transaction behavior data of the traders who need to be supervised in the time interval that needs to be supervised is taken out from the database and added to the set to be detected S. Then, from the set to be detected, the shortest time interval with the maximum degree of abnormality is selected. This ensures that each detection selects the time interval where insider trading is likely to occur as much as possible, and selecting the shortest possible time interval can ensure that the time interval where insider trading is likely to occur is given more quickly. After the time interval for detection is selected, the time interval will be divided into N time intervals. Since the length of the time interval after segmentation needs to be an integer, the length of the interval after segmentation is taken as Each time interval contains several transactions. For example, in the example, the trading user is in the time interval of commodity price fluctuation [t a , t b ), at t1 and t2, two buying actions of quantities v1 and v2 were performed respectively, and at t3, a selling action of quantity v3 was performed.

[0009] The system can obtain a variety of trading indicators by analyzing the trading behavior data within the time interval. The system uses four indicators, namely, the rate of increase of capital investment, the concentration of investment funds, the frequency of transactions and the rate of return on investment, to represent the trading characteristics of the trading behavior within the time interval. They correspond to the behavior that when insider trading occurs, insiders may invest a large amount of funds, concentrate a large amount of funds on a certain commodity, and frequently trade to obtain a large amount of income. Since the insider trading process may span the divided time interval, the sequence anomaly detection method is used for anomaly detection in the insider trading scenario. Through the self-supervised learning method, the k-distance between the subsequence and other similar subsequences is calculated to detect the outliers of the subsequence.

[0010] Since the insider trading behavior pattern is different from the normal trading behavior pattern in the trading behavior pattern of the trading users, and the more different the sequence composed of the trading indicators is from the sequence composed of the normal indicators, the greater the abnormal value of the subsequence. After obtaining the abnormal value of each subsequence. It will determine whether each subsequence meets the interval length requirement. If it does, the subsequence will be added to the detected set, otherwise the subsequence will be added to the set to be detected and the degree of abnormality will be marked. By continuously iteratively selecting the shortest time series with the maximum degree of abnormality in the set to be detected, until the running time limit is exceeded during the iteration or the set to be detected is empty, the detection method is terminated. The system returns the detected set to display the interval where insider trading is most likely to occur to the regulatory authorities.

[0011] The beneficial effects of the present invention are:

[0012] (1) Improving the effectiveness of insider trading behavior detection Based on the commodity trading model, the system considers the differences between insider trading behavior and normal trading behavior in the trading behavior data used for trading, and characterizes trading behavior using multiple trading indicators to help the subsequent use of abnormal time series detection methods and solve the problem of previously considering only a single indicator for modeling.

[0013] (2) Reduce the time required for detection methods. The system adaptively adjusts the time intervals during insider trading detection and detects the time intervals most likely to contain insider trading first, thus resolving the problem of directly using the target time interval to divide the length, which results in excessively long detection times in large-scale trading scenarios.

[0014] (3) Maintaining the fairness and stability of the commodity trading market In the commodity trading market, insider trading allows some people to obtain illegal benefits, which is unfair to ordinary investors who cannot obtain insider information. It also damages the confidence of ordinary stock investors in further trading in the trading market and affects the normal development of the trading market. The adaptive detection method and system of insider trading can help regulatory authorities better combat insider trading. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is the main principle diagram of the present invention.

[0016] Figure 2 It is a schematic diagram of a trading user's trading behavior on a price-fluctuating commodity within a time range. DETAILED DESCRIPTION

[0017] 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.

[0018] As shown in the figure, the adaptive detection method and system for commodity insider trading described in the present invention includes the following steps:

[0019] (1) In the initial stage, the system extracts historical transaction data of commodity trading users within the regulatory scope. a , t b ), the transaction behavior data of N trading accounts that have traded the jth commodity are extracted using the set A = {a1, a2, ..., a N} to represent these trading accounts. Assume that trading account a i ∈A has traded in M ​​types of goods, so the trading behavior on each type of goods is represented by a set Assuming that K transactions are conducted on the jth commodity, the kth transaction can be expressed as in, Indicates trading account a i The time of the kth transaction on the jth commodity, Indicates trading account a i The price of the commodity at the kth transaction on the jth commodity, Indicates trading account a i The number of items in the kth transaction on the jth item, Indicates trading account a i The transaction type for the kth transaction on the jth commodity. Represents a sell transaction represents a buying transaction. Before making K transactions, the user may hold a certain number of commodities in advance, so α The total number of items held at a given moment is expressed as The price of the product at this time is The total value of the goods initially held is expressed as Similarly, after K transactions, it is still possible to hold a certain amount of the product, so t b The total number of items held at a given moment is expressed as The price of the product at this time is The total value of the goods held at the end is expressed as Add the transaction behavior of each trading account to the set to be detected S, which is defined as S = {<[t l , t r ), B i ,δ>|t a ≤t l ≤t r ≤t b , 1≤i≤N}, where [t l , t r ) represents the time interval of the transaction behavior, B i represents the trading behavior of the ith trading account in this interval, δ represents the abnormal value of the trading behavior, and the initial abnormal value is 0.

[0020] (2) The system sorts the transaction behaviors in the set to be detected. The larger the outlier value δ is, the higher the sorting priority is. For transaction behaviors with the same outlier value, the interval length t r -t l The smaller the value, the higher the sorting priority. Select the transaction with the highest priority to divide the time interval, and divide [t l , t r ) is divided into N segments, then the length of the time interval is Therefore, the time interval sequence obtained by division is expressed as S = {[t l , t l +T],[t l +T,t l +2T],...,[t l +(N-1)T,t r ]In the time interval [t x , t x+T ) The transaction behavior on the jth commodity can be expressed as

[0021] (3) The system converts the trading behavior within the divided time into trading indicators. In order to distinguish the trading characteristics of insider trading from normal trading, four trading indicators are used: capital investment increase rate, investment capital concentration, trading frequency and investment return rate.

[0022] Capital investment increase rate: It represents the ratio of the value of additional funds invested over a period of time to the value of the original capital investment.

[0023]

[0024] The numerator represents the transaction account a i On the jth commodity [t x , t x+T ) The total value of all purchased commodities during the time interval. The denominator represents the trading account a i On the jth commodity at t x The total value held at a given moment. If the trading account suddenly adds funds to invest in this commodity, it may be insider trading caused by insider information.

[0025] Transaction frequency: It represents the ratio of the total number of transactions for the product over a period of time.

[0026]

[0027] The numerator represents the transaction account a i On the jth commodity [t x , t x+T ) The denominator represents the total length of the time interval. If the trading account rarely trades on this commodity but suddenly trades a large number of times in a certain period of time, it may be insider trading caused by insider information.

[0028] Investment fund concentration: It represents the trading account a i The ratio of the value of the jth commodity to the funds invested in all commodities.

[0029]

[0030] If a trading account has always adopted a diversified investment strategy in the past, but changes to a concentrated investment strategy in a certain period of time, it means that insider trading may be caused by insider information.

[0031] Return on investment: It represents the ratio between the net income over a period of time and the investment cost at a certain point in time. A high return on investment means that the investment can obtain a higher return. x , t x+T ) time interval is expressed as The sum of the previous items is expressed as the trading account a i On the jth commodity [t x , t x+T ) The total income from all selling transactions during the time interval t x+r The total value of the commodity held at all times. x , t x+T ) The investment cost of the time interval is expressed as The sum of the previous items is expressed as the trading account a i On the jth commodity [t x , t x+T ) The total cost of all buy transactions within the time interval, the latter term represents the total cost of all buy transactions within the time interval t x The total value of the item is always kept.

[0032] Through the above two items, [t x , t x+T ) time interval is expressed as

[0033]

[0034] This indicator represents the trading account a i In [t x , t x+T ) is the ratio of the net income obtained through the j-th commodity trading behavior to the total investment cost during the time interval. If the investment return rate of the trading account during this time interval has a sudden increase compared with the previous one, it means that there may be insider trading caused by insider information.

[0035] After calculating the index, [t x , t x+T ) within the transaction, use To represent the trading characteristics of this time interval. Finally, a trading indicator sequence with a length of N is obtained.

[0036] (4) Detect abnormal subsequences in the transaction indicator sequence calculated in step (3). By moving the window with a length of n, Nn subsequences can be obtained. The set of all subsequences can be expressed as:

[0037]

[0038] in Expressed as t p The subsequence of eigenvalues ​​starting with length n can be expressed as:

[0039]

[0040] The degree of abnormality of a sequence is determined by whether the subsequences are similar. Therefore, the Eros distance is used to measure the similarity between two subsequences. Assume that p and q are two subsequences, and the subsequences contain 4 trading indicators. Therefore, they can be represented as an n×4 matrix, that is, each row represents the time series of an indicator. Perform singular value decomposition (SVD) on the covariance matrix to obtain the right eigenvector matrix V p =(a1, a2, ..., a n ), V q =(b1, b2, ..., b n ), then the distance between p and q is defined as

[0041]

[0042] Where w is the weight vector, which is obtained by normalizing the average vector of the eigenvalue vector. When using the distance definition, for subsequences with overlapping windows, the similarity between them is usually not compared to avoid the deviation caused by self-similarity. Different subsequences have different offsets and amplitudes, and they must be standardized before comparing the similarity. After defining the distance formula, the k-nearest neighbor distance k-distance(p) of the subsequence p can be defined: 1. There are at least k subsequences, o′∈SS\{p}, such that dist(p, o′)≤dist(p, o). 2. There are at least k-1 subsequences, o′∈SS\{p}, "such that" dist(p, o′)<dist(p, o).

[0043] The set of k nearest neighbor subsequences of subsequence p can be expressed as N k-dist(p) ={q∈SS\{p}|dist(p,q)≤k-distance(p)}. The k-nearest neighbor reachable distance between subsequences p and o is expressed as reach-dist k (p, o) = max{k-distacnce(o), dist(p, o)}. The local reachability density of a subsequence p is defined as

[0044]

[0045] By comparing the local reachability density of subsequence p with the reported reachability density of surrounding neighbor subsequences, the local anomaly factor of subsequence p can be defined.

[0046]

[0047] The larger the local anomaly factor of subsequence p, the smaller the local reachability density of subsequence p, and the smaller the local anomaly factor of subsequence N k-dist(p)The local reachable density of the set is large. This means that the subsequences of the neighbors around p are similar to each other, but there is a lack of subsequences similar to p. Corresponding to the trading behavior of trading users, normal trading behavior is similar, while insider trading behavior is different from normal trading behavior.

[0048] (5) Each subsequence of the outliers obtained in step (4) is judged. If the time interval length of the subsequence meets the interval length requirements of the regulatory department, the subsequence and its outliers are added to the detected set. Otherwise, it means that the subsequence needs further testing, so repeat steps (2) to (4). When the set to be tested is empty or the detection method exceeds the specified running time, execute step (6).

[0049] (6) Return the labeled subsequences in the detected set. Each subsequence represents the degree of abnormality of the trading behavior within the time interval. The larger the value, the more likely it is that insider trading has occurred.

[0050] It should be noted that the above content only illustrates the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.

Claims

1. An adaptive detection method for commodity insider trading, characterized by: This detection method is based on the trading behavior data of trading users. According to the characteristics of insider trading behavior, the trading indicators are calculated, and the abnormal trading behavior of the indicators is detected by adaptively adjusting the time interval size. The insider trading behavior indicator is the i On the jth commodity [t x , t x+T ) Definition of trading behavior in the time interval Trading indicators for insider trading behavior: Capital investment increase rate: It represents the ratio of the value of additional funds invested in a period of time to the value of the original capital investment; The numerator represents the transaction account a i On the jth commodity [t x , t x+T ) The total value of all purchased commodities during the time interval; the denominator represents the trading account a i On the jth commodity at t x The total value held at any given moment; if the trading account suddenly adds funds to invest in this commodity, it may indicate insider trading caused by insider information; trading frequency: it represents the ratio of the total number of transactions in the commodity over a period of time; The numerator represents the transaction account a i On the jth commodity [t x , t x+T ) The total number of all transactions in the time interval; the denominator represents the total length of the time interval; if the trading account rarely trades on the commodity but suddenly conducts a large number of transactions in a certain period of time, it may indicate insider trading caused by insider information; Investment fund concentration: It represents the trading account a i The ratio of the value of the jth commodity to the funds invested in all commodities; If the trading account has always adopted a diversified investment strategy in the past, but has changed to a concentrated investment strategy in a certain period of time, it means that there may be insider trading caused by insider information; Return on investment: It represents the ratio between the net income over a period of time and the investment cost at a certain point in time; a high return on investment means that the investment can obtain higher returns; [t x , t x+T ) time interval investment income is expressed as The sum of the previous items is expressed as the trading account a i On the jth commodity [t x , t x+T ) The total income from all selling transactions during the time interval t x+T The total value of the commodity held at all times; x , t x+T ) The investment cost of the time interval is expressed as The sum of the previous items is expressed as the trading account a i On the jth commodity [t x , t x+T ) The total cost of all buy transactions within the time interval, the latter term represents the total cost of all buy transactions within the time interval t x Always hold the total value of the commodity; through the above two items, [t x , t x+T ) time interval is expressed as This indicator represents the trading account a i In [t x , t x+T ) is the ratio of the net income obtained through the j-th commodity trading behavior to the total investment cost within the time interval; if the investment return rate of the trading account within this time interval has a sudden increase compared with the previous one, it means that there may be insider trading caused by insider information; [t x , t x+T ) within the transaction, use To represent the trading characteristics of this time period.

2. The method for adaptively detecting insider trading in commodities according to claim 1, characterized in that: The transaction behavior data of the transaction user refers to the system's data on the transaction behavior of the transaction user in the time interval [t a , t b ), the transaction data of N trading accounts that have traded the jth commodity are extracted, using the set A = {a1, a2, ..., a N } to represent these trading accounts; assuming that trading account a i ∈A has traded in M ​​types of goods, so the trading behavior on each type of goods is represented by a set Assuming that K transactions are conducted on the jth commodity, the kth transaction is expressed as in, Indicates trading account a i The time of the kth transaction on the jth commodity, Indicates trading account a i The price of the commodity at the kth transaction on the jth commodity, Indicates trading account a i The number of items in the kth transaction on the jth item, Indicates trading account a i The transaction type for the kth transaction on the jth commodity; Represents a sell transaction represents the transaction behavior of buying; before making K transactions, the user may hold a certain number of commodities in advance, so t a The total number of items held at a given moment is expressed as The price of the product at this time is The total value of the goods initially held is expressed as Similarly, after K transactions, it is still possible to hold a certain amount of the product, so t b The total number of items held at a given moment is expressed as The price of the product at this time is The total value of the goods held at the end is expressed as Add the transaction behavior of each trading account to the set to be detected S, which is defined as S = {<[t l , t r ), B i ,δ>|t a ≤t l ≤t r ≤t b , 1≤i≤N}, where [t l , t r ) represents the time interval of the transaction behavior, B i represents the trading behavior of the ith trading account in this interval, and δ represents the outlier of the trading behavior.

3. The adaptive detection method for insider trading of commodities according to claim 1 is characterized by: The system divides all time intervals that need to be detected into two sets: the set to be detected and the set that has been detected. In each iteration, the transaction behaviors in the set to be detected are sorted. The larger the outlier value δ is, the higher the sorting priority is. For transaction behaviors with the same outlier value, the interval length t r -t l The smaller it is, the higher the sorting priority; select the transaction with the highest priority to divide the time interval, and divide [t l , t r ) is divided into N segments, then the length of the time interval is T = t r -t l / N; therefore, the time interval sequence obtained by division is expressed as S = {[t l , t l +T],[t l +T,t l +2T],...,[t l +(N-1)T,t r ]In the time interval [t x , t x+T ) The transaction behavior on the jth commodity is expressed as Anomaly detection is performed on each time interval sequence obtained by division.

4. The adaptive detection method for insider trading of commodities according to claim 3 is characterized by: Using a window of length n to move the time series of trading indicators, we can get Nn subsequences. The set of all subsequences is expressed as in Expressed as t p The subsequence of eigenvalues ​​starting with length n is represented by Because the degree of abnormality of the time series is determined by whether the subsequences are similar, the Eros distance is used to measure the similarity between two subsequences. Assume that p and q are two subsequences, and the subsequences contain 4 trading indicators, so they are represented as an n×4 matrix, that is, each row represents the time series of an indicator; perform singular value decomposition (SVD) on the covariance matrix to obtain the right eigenvector matrix V p =(a1, a2, ..., a n ), V q =(b1, b2, ..., b n ), then the distance between p and q is defined as Where w is a weight vector, which is obtained by normalizing the mean vector of the eigenvalue vector. When using the distance definition, for subsequences with overlapping windows, the similarities between them are usually not compared to avoid the deviation caused by the b similarity. Different subsequences have different offsets and amplitudes, which need to be standardized before the similarity comparison.

5. The method for adaptively detecting insider trading in commodities according to claim 4, characterized in that: k-distance(p) of a subsequence p:

1. There are at least k subsequences, o′∈SS\{p}, such that dist(p, o′)≤dist(p, o); 2. There are at least k-1 subsequences, o′∈SS\{p}, "such that" dist(p, o′)<dist(p, o); The set of k-nearest neighbor subsequences of a subsequence p is denoted by N k-dist(p) = {q∈SS\{p}|dist(p,q)≤k-distance(p)}; the k-nearest neighbor reachable distance between subsequences p and o is expressed as reach-dist k (p, o) = max{k-distacnce(o), dist(p, o)}; the local reachability density of the subsequence p is defined as By comparing the local reachability density of subsequence p with the reported reachability density of surrounding neighbor subsequences, the local anomaly factor of subsequence p is defined; The larger the local anomaly factor of subsequence p, the smaller the local reachability density of subsequence p, and the smaller the local anomaly factor of subsequence N k-dist(p) The local reachable density of the set is large; this means that the subsequences of the neighbors around p are similar to each other, but lack subsequences similar to p; corresponding to the trading behavior of trading users, normal trading behavior is similar, while insider trading behavior is different from normal trading behavior.

6. The adaptive detection method for insider trading of commodities according to claim 1, characterized in that: The system divides the time series with the largest outlier in the detection set and performs anomaly detection; it judges each subsequence that obtains the outlier. If the time interval length of the subsequence has met the interval length requirements of the regulatory department, the subsequence and its outlier are added to the detected set; otherwise, it means that the subsequence needs further detection and is added to the detection set; repeat until the detection set is empty or the detection method exceeds the specified running time; finally, the marked subsequences in the detection set are returned. Each subsequence represents the degree of abnormality of the trading behavior in the time interval. The larger the value, the more likely it is that insider trading will occur.

Images