Portfolio optimization system based on quantification strategies

Abstract

The application discloses a portfolio optimization system based on a quantification strategy, and belongs to the technical field of blockchains and quantification investment. The system comprises a data acquisition module, a graph construction module, a flow quantification module and a holding optimization module. The data acquisition module acquires holding structure data of a portfolio and on-chain transaction records of corresponding digital assets. The graph construction module constructs an asset correlation graph according to the holding structure data, and decomposes the asset correlation graph into multiple time sequence subgraphs with time stamps in the on-chain transaction records as a trigger sequence. The flow quantification module calculates transaction flow direction weights between nodes for each time sequence subgraph, and generates a quantification edge set reflecting the direction of fund transmission between assets. The collection identification module filters out asset nodes with fund inflow exceeding a preset threshold according to dynamic fund transmission paths, and marks the asset nodes as fund collection points. The holding optimization module adjusts the holding proportion of each asset in the portfolio according to the distribution density of all fund collection points, and obtains an optimized holding configuration.

Description

Technical Field

[0001] This invention relates to the fields of blockchain technology and quantitative investment technology, specifically to a portfolio optimization system based on quantitative strategies. Background Technology

[0002] In digital asset portfolio management, optimizing portfolio structure hinges on accurately perceiving the true state of fund flows between assets. Existing solutions largely rely on historical asset price sequences, volatility, and correlation matrices of asset returns to construct optimization models, adjusting portfolio weights through classic methods such as mean-variance analysis or risk parity. These solutions depend on quantitative indicators at the transaction outcome level, failing to trace the complete path and dynamic process of fund transfers between assets. In a blockchain environment, on-chain transaction records contain fine-grained information such as the address, amount, and timestamp of each asset transfer, accurately reconstructing fund transmission behavior. However, existing portfolio optimization techniques do not deeply utilize the temporal transmission characteristics of on-chain transaction data, still treating each asset as an independent node with only statistical correlations, ignoring the changes in the direction and intensity of fund flows between assets. This makes it difficult for optimization results to respond promptly to market fund aggregation trends, especially during periods of rapid fund rotation, where portfolio adjustments lag behind actual changes in fund preferences. To address the aforementioned issues, it is necessary to solve how to extract dynamic paths reflecting the direction of fund transmission from on-chain transaction records, and how to identify assets where funds converge based on these dynamic paths, and transform their distribution characteristics into quantitative basis for portfolio structure adjustments. Solving these two problems will allow the portfolio optimization process to directly anchor to the endogenous laws of on-chain fund flows, rather than remaining at the level of indirect fitting of price appearances. Summary of the Invention

[0003] This invention provides a portfolio optimization system based on quantitative strategies. It aims to utilize on-chain transaction records of digital assets to construct a dynamic path that can depict the direction of fund transmission, and automatically adjust the portfolio holding ratio based on the distribution density of fund convergence points identified by the path, so that the optimized holding configuration can respond in real time to the real trend of fund flow between assets.

[0004] To achieve the above objectives, this invention provides the following technical solution: This invention provides a portfolio optimization system based on quantitative strategies, which integrates portfolio structure data with on-chain transaction records of digital assets to achieve quantitative analysis and automatic optimization of portfolio fund flows. The system includes a data acquisition module, a graph construction module, a flow vectorization module, a path splicing module, a convergence identification module, and a portfolio optimization module. The data acquisition module acquires portfolio structure data and corresponding on-chain transaction records of digital assets. The graph construction module constructs an asset association graph based on the portfolio structure data and decomposes the asset association graph into multiple time-series subgraphs using timestamps in the on-chain transaction records as trigger sequences. The flow vectorization module calculates the transaction flow weights between nodes in each time-series subgraph and generates quantitative edge sets reflecting the direction of fund transmission between assets based on these weights. The path splicing module splices the quantitative edge sets corresponding to all time-series subgraphs in chronological order to form a dynamic fund transmission path for the portfolio, thus fully presenting the fund transmission chain in the time dimension. The aggregation and identification module filters out asset nodes with inflows exceeding a preset threshold based on dynamic fund transmission paths and marks these asset nodes as fund aggregation points to accurately locate key assets in the investment portfolio where funds are concentrated. The portfolio optimization module adjusts the holding ratio of each asset in the investment portfolio based on the distribution density of all fund aggregation points to obtain an optimized portfolio allocation, tilting the portfolio towards fund aggregation and improving the matching degree between the allocation and market fund trends.

[0005] As a technical solution of this invention, the data acquisition module reads the asset code list and the nominal holding amount of each asset in the portfolio management account as portfolio structure data, and queries the blockchain data nodes to retrieve all on-chain transfer records of each digital asset in the asset code list within the most recent complete trading cycle as on-chain transaction records. In this way, it ensures that the acquired data contains both static portfolio information and dynamic on-chain transaction information.

[0006] In a preferred embodiment of the present invention, the graph construction module uses each digital asset in the holding structure data as a node and the historical synchronous transaction frequency between two digital assets as the edge weight to construct an asset association graph. Then, it determines multiple time slices according to the chronological order of the timestamps in the on-chain transaction records. Within each time slice, it extracts the nodes in the asset association graph where transactions occurred within that time slice and the edges connecting these nodes, forming the temporal subgraph corresponding to that time slice. In this way, the original composite association graph is decomposed into a series of time-ordered subgraphs, facilitating the capture of phased changes in fund flows.

[0007] As a further improvement of this invention, the flow vectorization module calculates the total transaction amount of each directed edge in each time series subgraph, and divides the total transaction amount by the sum of all outgoing transaction amounts of the starting node of the directed edge within the time series subgraph to obtain the transaction flow weight of the directed edge. Then, all directed edges with transaction flow weights greater than a preset flow threshold are extracted to form the quantized edge set of the time series subgraph. This processing eliminates insignificant transaction flows, retaining only edges with substantial transmission effects. The generated quantized edge set can effectively reflect the main flow trends of funds in each time slice.

[0008] Preferably, the path splicing module arranges all time-series subgraphs according to the chronological order of their corresponding time slices, obtaining a time-series sorting list. Based on this list, it matches the endpoint node of the directed edge set in the quantized edge set of each time-series subgraph with the starting node of the quantized edge set in the next time-series subgraph. The quantized edge sets in the two successfully matched time-series subgraphs are then connected sequentially to form a dynamic capital transmission path. Thus, the previously scattered capital flow information in various time slices is linked into a coherent path, fully depicting the transmission process of capital between assets.

[0009] In one technical solution of this invention, the aggregation and identification module accumulates the total inflow of funds from each directed edge pointing to the same asset node along the direction of the dynamic fund transmission path, as the cumulative fund inflow of that asset node. The cumulative fund inflow is then compared with a preset threshold. When the cumulative fund inflow exceeds the preset threshold, the asset node is marked as a fund aggregation point. To improve the rationality of the threshold setting, the preset threshold can be dynamically calculated based on the average and standard deviation of the inflow of funds from all asset nodes in the dynamic fund transmission path, making the identification of fund aggregation points more adaptable to market fluctuations and the fund distribution characteristics of different cycles. As a further limitation, only the top N nodes in terms of inflow volume can be selected as fund aggregation points, where N is a preset positive integer, thereby focusing on the most core fund aggregation assets.

[0010] In another preferred embodiment of the present invention, the portfolio optimization module calculates the graph distance between each capital aggregation point and its adjacent capital aggregation points in the asset association graph, obtaining a distance set. Based on the minimum and maximum distances in the distance set, it determines the normalized value of the distribution density. Then, based on the normalized value of each capital aggregation point, it increases the holding ratio of the asset corresponding to that capital aggregation point in the investment portfolio and correspondingly decreases the holding ratio of assets corresponding to non-capital aggregation points, resulting in an optimized portfolio allocation. This optimization method assigns higher weights to areas with higher capital concentration and lower weights to non-aggregation areas, enhancing the portfolio's ability to follow the mainstream capital allocation in the market.

[0011] To further optimize on-chain execution of the configuration, after obtaining the optimized position configuration, the system converts it into multiple sub-configuration schemes. Each sub-configuration scheme corresponds to the on-chain address of a digital asset. These sub-configuration schemes are sent sequentially to their respective on-chain addresses for signature confirmation according to their timestamps. After receiving all signature confirmations, a summary execution instruction containing all sub-configuration schemes is generated and submitted to the blockchain network. This process achieves fully closed-loop automated processing from position analysis to configuration optimization to on-chain execution.

[0012] The technical effects and advantages provided by the present invention in the above technical solution are as follows: To address the challenge of extracting dynamic fund transfer paths from on-chain transaction records, a graph construction module generates an asset association graph based on position structure data. Using transaction timestamps as trigger sequences, the graph is decomposed into multiple temporal subgraphs, allowing for refined segmentation of asset transaction relationships over time. Building upon this, a flow vectorization module calculates the transaction flow weights between nodes in each temporal subgraph. These weights are determined by the proportion of a unidirectional transaction amount to the total outflow amount from the starting node, filtering out weak fund transfer noise and retaining only transaction edges with significant transfer strength to form a quantified edge set. A path stitching module then matches and connects the quantified edge sets of each temporal subgraph according to chronological order, forming a complete dynamic fund transfer path. This process continuously reconstructs the temporal trajectory of directed fund flows between different assets from discrete on-chain transaction records. Each directed edge on the path carries weight information reflecting the transfer strength, intuitively revealing which assets funds gradually migrate from and to, and the changes in the intensity of the migration. This allows investment decisions to no longer rely on indirect correlations with asset prices, but rather directly observe the dynamic transfer structure of fund behavior. To address the issue of adjusting portfolios based on fund aggregation behavior, a aggregation identification module accumulates the total inflow of funds into the same asset node along the dynamic fund transmission path. A threshold dynamically calculated by the system based on the mean and standard deviation of inflows across all asset nodes is used as the criterion to select asset nodes with significantly higher inflows than the overall level as fund aggregation points. Subsequently, the portfolio optimization module calculates the spectral distances between fund aggregation points in the asset correlation graph, forming a distance set. The minimum and maximum distances are used to determine the normalized value of the distribution density, thereby increasing the holding ratio of assets corresponding to fund aggregation points in the portfolio while decreasing the holding ratio of assets outside of fund aggregation points. This adjustment method directly maps the degree of fund aggregation towards specific assets and the aggregation characteristics of aggregation points in the graph to changes in portfolio weights. This allows the portfolio optimization results to accurately follow the actual aggregation trend of funds, avoiding the lag problem where large-scale fund migration has occurred while the portfolio structure is still dominated by historical statistical relationships. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0014] Figure 1 This is a schematic diagram of a portfolio optimization system based on quantitative strategies; Figure 2 This is a flowchart of the data acquisition module acquiring position structure data and on-chain transaction records; Figure 3 These are histograms and cumulative distribution curves of digital asset transaction amounts; Figure 4 This is a schematic diagram of the results of filtering the directed edge transaction flow weights in the time-series subgraph; Figure 5 This is a distribution chart of the cumulative capital inflow at the capital aggregation point; Figure 6 It is a chart indicating the adjustment of the asset holding ratio of the investment portfolio and the points where funds converge. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] See Figure 1This invention provides a portfolio optimization system based on quantitative strategies. The system includes a data acquisition module, a graph construction module, a flow vectorization module, a path splicing module, a convergence identification module, and a position optimization module. The data acquisition module acquires the portfolio's position structure data and the corresponding on-chain transaction records of digital assets. The graph construction module constructs an asset association graph based on the position structure data and decomposes the asset association graph into multiple time-series subgraphs using timestamps in the on-chain transaction records as trigger sequences. The flow vectorization module calculates the transaction flow weights between nodes in each time-series subgraph and generates quantitative edge sets reflecting the direction of fund transmission between assets based on these weights. The path splicing module splices the quantitative edge sets corresponding to all time-series subgraphs in chronological order to form a dynamic fund transmission path for the portfolio. The convergence identification module filters out asset nodes with fund inflows exceeding a preset threshold based on the dynamic fund transmission path and marks these asset nodes as fund convergence points. The position optimization module adjusts the holding ratio of each asset in the portfolio based on the distribution density of all fund convergence points to obtain an optimized position configuration. Example

[0017] In specific implementation, please refer to Figure 2 The data acquisition module extracts a list of asset codes and the nominal holdings of each asset from the portfolio management account's database, using these as portfolio structure data. Each asset code in the asset code list uniquely identifies a digital asset; the asset code can be a trading pair symbol or contract address of the digital asset on an exchange or blockchain network. The nominal holdings represent the quantity of the corresponding digital assets currently held in the portfolio, measured in the smallest unit of the digital asset.

[0018] In practice, the data acquisition module sends query requests to the blockchain data nodes based on the asset code list. It retrieves all on-chain transfer records for each digital asset in the asset code list within the most recent complete transaction cycle, and uses these records as the on-chain transaction records. The blockchain data nodes, which are either full nodes or archive nodes in the blockchain network, provide a historical transaction data query interface. The query request includes the on-chain address parameter corresponding to the asset code and the time range parameter.

[0019] In some embodiments, the most recent complete transaction cycle is determined as follows: The current system time at which the data acquisition module performs the acquisition operation is obtained, and a duration span of a complete transaction cycle is set. The duration span of a complete transaction cycle can be the previous natural week or the previous natural month. If the previous natural week is used, the start and end timestamps of the natural week preceding the start time of the current system time are calculated, and the time period between the start and end timestamps is taken as the most recent complete transaction cycle. If the previous natural month is used, the start and end timestamps of the natural month preceding the start time of the current system time are calculated, and the time period between the start and end timestamps is taken as the most recent complete transaction cycle. A natural week begins at 00:00 on Monday, and a natural month begins at 00:00 on the first day of each month. Coordinated Universal Time (UTC) is used for the timestamps.

[0020] In practice, each on-chain transfer record includes a sending address field, a receiving address field, a transaction amount field, and a timestamp field. The sending address field records the on-chain address from which the digital asset is transferred, the receiving address field records the on-chain address from which the digital asset is transferred, the transaction amount field records the quantity of digital assets transferred, and the timestamp field records the block time at which the transaction was confirmed by the blockchain network. The data acquisition module stores each on-chain transfer record along with its corresponding asset code for use by subsequent modules.

[0021] See Figure 3 In the graph, the horizontal axis represents transaction amount in digital asset units, the left side of the vertical axis represents the number of transactions, and the right side of the vertical axis represents the cumulative distribution probability. The blue histogram reflects the distribution of the number of transactions within different transaction amount ranges, the orange curve is the cumulative distribution curve of transaction amount, and the black dashed line represents the average transaction amount.

[0022] As shown in the graph, transaction amounts are concentrated in a lower range, with the left bar being relatively tall. The highest number of transactions occurs in the range of 0 to 5000 digital asset units, peaking at nearly 500 transactions. As transaction amounts increase, the number of transactions decreases rapidly, with very few exceeding 5000 digital asset units, and the histogram quickly approaches zero. When transaction amounts exceed tens of thousands to hundreds of thousands of digital asset units, the number of transactions is almost zero, indicating that large transactions are extremely rare.

[0023] The cumulative distribution curve shows a steep upward trend, rapidly approaching 1 in the lower range of transaction amounts, indicating that the vast majority of transaction amounts are concentrated in the smaller range. After the cumulative distribution reaches about 95% or more, the curve tends to flatten out, indicating that ultra-high amount transactions have a limited contribution to the overall cumulative transaction amount.

[0024] The average transaction amount indicated by the black dashed line is in the lower range, reflecting that most transactions are small and frequent, with an average transaction amount of approximately 2,000 digital asset units. Example

[0025] In practice, the graph construction module reads the portfolio structure data, treats each digitized asset in the portfolio structure data as a node, and constructs a set of nodes for the asset association graph. The graph construction module assigns an edge to each pair of nodes, and the edge attributes include edge weight, which is used to represent the strength of the association between the two nodes.

[0026] The graph construction module calculates edge weights by counting the historical synchronization transaction frequency between two digital assets. The historical synchronization transaction frequency is defined as the number of times two digital assets are jointly transferred in the same transaction within a defined observation window. The observation window is a preset time interval with a fixed duration, which can be set to 30 calendar days. For a pair of nodes in the asset association graph, corresponding to the first and second digital assets respectively, the graph construction module traverses all on-chain transfer records within the observation window, packaging the digital assets transferred in each transaction into an asset set. If the first and second digital assets exist simultaneously in the same asset set, the historical synchronization transaction frequency count between the first and second digital assets is incremented by 1. After the traversal is complete, the final count is used as the edge weight between this pair of nodes. If the historical synchronization transaction frequency is 0, no edge is added between this pair of nodes in the asset association graph. The asset association graph is stored in the form of an adjacency list. Each node in the adjacency list is associated with a list storing all neighboring nodes connected to that node and their edge weights.

[0027] In implementation, the graph construction module obtains all timestamps from the on-chain transaction records, sorts the timestamps chronologically, and obtains a timestamp sequence. The graph construction module then divides the timestamp sequence into multiple time slices based on a preset time slice span. The time slice span is set as follows: The overall time span of the on-chain transaction records is counted, and the average number of transactions per day is calculated. If the average number of transactions per day is greater than a preset transaction frequency threshold, the time slice span is set to 1 hour; if the average number of transactions per day is not greater than the preset transaction frequency threshold, the time slice span is set to 1 calendar day. The preset transaction frequency threshold is set to 10,000 transactions per day. If a 1-hour time slice span is used, time slices are generated starting from the smallest timestamp in the timestamp sequence, with each hour as an interval. Each time slice corresponds to a start timestamp and an end timestamp. If a 1-calendar day time slice span is used, time slices are generated starting from the smallest date in the timestamp sequence, with each calendar day as an interval.

[0028] Within each time slice, the graph construction module scans on-chain transaction records, identifies nodes where transactions occurred within that time slice, and extracts nodes from these nodes that exist in the asset association graph, forming a valid node set. For each pair of nodes in the valid node set, if an edge exists between these two pairs of nodes in the asset association graph, then this edge and the two nodes connecting it are extracted to form a node and an edge in the time-series subgraph corresponding to that time slice. This process is repeated until all node pairs in the valid node set are extracted, forming the time-series subgraph corresponding to that time slice. The time-series subgraph is a directed graph, and the direction of the edges is determined based on the sending and receiving addresses in the on-chain transfer records. The direction of the directed edges points from the digital asset node corresponding to the sending address to the digital asset node corresponding to the receiving address.

[0029] In practice, the flow vectorization module calculates the total transaction amount for each directed edge in each time-series subgraph. The calculation is as follows: All on-chain transfer records within the corresponding time slice of the time-series subgraph are retrieved. For each on-chain transfer record, the digital asset node to which the sending address belongs is the starting node, and the digital asset node to which the receiving address belongs is the ending node. If a directed edge exists in the time-series subgraph pointing from the starting node to the ending node, the transaction amount of that on-chain transfer record is added to the total transaction amount of that directed edge. This process is repeated until the total transaction amount for all directed edges in the time-series subgraph is calculated.

[0030] The flow vectorization module calculates the transaction flow weight for each directed edge in the time-series subgraph. The transaction flow weight is calculated as follows: for a directed edge, the total transaction amount of this directed edge is divided by the sum of all outgoing transaction amounts of the starting node within the time-series subgraph. The sum of all outgoing transaction amounts of the starting node within the time-series subgraph is the sum of the transaction amounts of all on-chain transfer records in which the starting node is the sender. The formula for calculating the transaction flow weight is:

[0031] in, Indicates from node Pointing to node The weight of the directed edge transaction flow. Indicates from node Pointing to node The total transaction amount of the directed edges. Represents a node The set of all endpoint nodes of directed edges that start from a node and point to other nodes in this temporal subgraph. Represents a node The sum of all outgoing transactions within this time series subgraph. For nodes Points to any endpoint node.

[0032] The flow vectorization module compares the transaction flow weights with a preset flow threshold. The preset flow threshold is set to a fixed value of 0.1. This is based on the principle that fund flows with a transaction flow weight less than 0.1 are considered non-primary transmission directions, and the primary direction of fund flow is retained after filtering. If the transaction flow weight of a directed edge is greater than the preset flow threshold, this directed edge is extracted and added to the quantized edge set of the time series subgraph. If the transaction flow weight is not greater than the preset flow threshold, this directed edge is discarded. After all directed edges in the time series subgraph have been compared and extracted, the resulting quantized edge set is used as the quantized output of the time series subgraph. In another implementation, the preset flow threshold is determined based on the distribution of the transaction flow weights of all directed edges in the time series subgraph. Specifically, the upper quartile of all transaction flow weights is calculated, and the preset flow threshold is set to the upper quartile value to retain edges with a higher proportion in fund transmission.

[0033] See Figure 4 In the diagram, the horizontal axis represents the directed edge index, ranging from 0 to 500, and the vertical axis represents the transaction flow weight, ranging from 0 to 1. In the legend, red solid dots indicate "retained edges," gray crosses indicate "discarded edges," and a dashed line represents the preset flow threshold of 0.1. According to Embodiment 2, the flow vectorization module calculates the transaction flow weight of each directed edge and compares it with the preset flow threshold of 0.1, selecting directed edges with weights greater than 0.1 as retained edges, while the remaining edges are discarded.

[0034] The dashed line in the diagram crosses the transaction flow weight at 0.1, dividing the point set into upper and lower parts. The red solid points are clearly concentrated in the range above 0.1, with transaction flow weights mostly distributed between 0.1 and 0.8, showing a strong capital transmission weight, which meets the selection criteria for retaining edges. The gray crosses are clustered in the range of 0 to 0.1, with lower transaction flow weights, representing filtered non-primary transmission directions of capital flow.

[0035] Example 3: In practice, the path stitching module acquires all time-series subgraphs and the corresponding time slices for each subgraph. The module then arranges all time-series subgraphs according to the chronological order of their respective time slices. The order of the time slices is determined by their start timestamps, with subgraphs starting at smaller timestamps appearing first, followed by those starting at larger timestamps. If two time slices have the same start timestamp, their order is determined by their end timestamps, with the subgraph ending at a smaller timestamp appearing first. After arrangement, a time-series sorting list is obtained, containing the identifier and sorting number of each time-series subgraph.

[0036] In practice, the path stitching module extracts two adjacent time-series subgraphs in the order of their sequence numbers from the time-series sorting list, denoted as the preceding and following time-series subgraphs, respectively. Each preceding time-series subgraph corresponds to a quantized edge set, and each following time-series subgraph corresponds to a quantized edge set. The path stitching module extracts the endpoint node of each directed edge in the quantized edge set of the preceding time-series subgraph, forming the preceding endpoint node set. Similarly, the path stitching module extracts the starting node of each directed edge in the quantized edge set of the following time-series subgraph, forming the following starting node set.

[0037] The path concatenation module matches the node names of each endpoint node in the preceding endpoint node set with the node names of each starting node in the following starting node set. The specific process of node name matching is as follows: it determines whether the node name string of an endpoint node in the preceding endpoint node set is exactly the same as the node name string of a starting node in the following starting node set. The node name string is a unique identifier for the asset code of the digital asset within the entire system. If the two node name strings are exactly the same, the node name match is considered successful. If the two node name strings are not exactly the same, the node name match is considered unsuccessful.

[0038] In practical implementation, when a node name match is successful, the path splicing module connects the directed edges in the quantized edge set of the preceding time-series subgraph that end at the successfully matched endpoint node with the directed edges in the quantized edge set of the following time-series subgraph that start at the successfully matched starting node in sequence. The sequential connection method is as follows: the endpoint node of the directed edge in the quantized edge set of the preceding time-series subgraph overlaps with the starting node of the directed edge in the quantized edge set of the following time-series subgraph, merging them into a single node. This ensures that the directed edges in the quantized edge sets of the preceding and following time-series subgraphs are connected end-to-end at the same node, forming a continuous directed path segment.

[0039] In some embodiments, if two or more starting nodes in the subsequent starting node set successfully match the same ending node in the preceding ending node set, the path splicing module generates multiple branch paths at the matching node. Specifically, after the directed edge pointing to the matching node in the quantized edge set of the preceding time-series subgraph, each successfully matched directed edge whose starting node is in the quantized edge set of the subsequent time-series subgraph is connected to form multiple parallel directed path segments.

[0040] In some embodiments, if a terminal node in the preceding set of terminal nodes does not have a matching starting node in the following set of starting nodes, the path splicing module treats the directed edges in the quantized edge set of the preceding time series subgraph that terminate at that terminal node as truncated paths. The truncated paths are marked as terminated path segments in the dynamic fund transmission path, and the terminated path segments do not extend into subsequent time series subgraphs.

[0041] The path splicing module performs the aforementioned node name matching and edge connection operations on each pair of adjacent time series subgraphs in the time series sorting list, and finally integrates all the connected directed path segments to form a dynamic fund transmission path. The dynamic fund transmission path is a directed acyclic graph, which contains all directed edges in all time series subgraphs whose transaction flow weight is greater than a preset flow threshold and their connection relationships.

[0042] In practical implementation, the aggregation and identification module acquires the dynamic fund transfer path, which contains multiple asset nodes and directed edges connecting them. Each directed edge has a corresponding transaction amount attribute. Following the direction of the dynamic fund transfer path, the aggregation and identification module identifies the asset node pointed to by each directed edge and adds the transaction amount of the directed edge to the cumulative fund inflow of the pointed-to asset node. After traversing all directed edges in the dynamic fund transfer path, the aggregation and identification module obtains the cumulative fund inflow of each asset node. The formula for calculating the cumulative fund inflow is:

[0043] in, Represents asset nodes The cumulative amount of funds inflow, This represents all nodes pointing to assets in the dynamic fund transfer path. The set of directed edges, Represents a directed edge The transaction amount To point to asset nodes Any directed edge of .

[0044] In practice, the aggregation and identification module compares the cumulative capital inflow of each asset node with a preset threshold. The preset threshold is a pre-defined amount of capital, set by multiplying the average daily trading amount of all digital assets in the investment portfolio over the most recent complete trading period by an adjustment coefficient of 1.5. This coefficient is used to distinguish between normal capital flows and aggregation behavior, as inflows exceeding 1.5 times the average daily trading amount are considered abnormal aggregation. If the cumulative capital inflow of an asset node exceeds the preset threshold, the aggregation and identification module marks that asset node as a aggregation point. If the cumulative capital inflow of an asset node does not exceed the preset threshold, it is not marked. After comparing and marking all asset nodes, the aggregation and identification module outputs a list of aggregation points, containing the node names and cumulative capital inflows of all asset nodes marked as aggregation points.

[0045] See Figure 5In the graph, the horizontal axis represents the ranking of asset nodes, and the vertical axis uses a logarithmic scale to represent the cumulative capital inflow of the corresponding asset node, in digital asset units. The blue solid line curve shows the trend of the cumulative capital inflow of all asset nodes as the ranking changes. The curve shows a decreasing trend overall, indicating that the higher the ranking of the asset node, the greater the cumulative capital inflow. The black dashed line is a preset threshold line, which corresponds to a certain amount of capital inflow and is used to distinguish between normal capital flow and capital accumulation behavior.

[0046] In the diagram, red stars mark fund aggregation points, all located where the cumulative fund inflow curve exceeds a preset threshold. The red stars are distributed between approximately 1 and 50 on the horizontal axis, concentrated in the region with higher cumulative fund inflows. This reflects that the aggregation identification module, based on the comparison between cumulative fund inflows and the preset threshold, selects the top 50 asset nodes as fund aggregation points. This selection process corresponds to the steps in Example 3 where the aggregation identification module compares cumulative fund inflows with the preset threshold and marks fund aggregation points, as well as the operation of sorting the asset nodes in the threshold-exceeding node set in descending order and selecting the top N nodes.

[0047] The cumulative capital inflow within the top 50 range shows a significant sharp drop, indicating that capital accumulation is mainly concentrated in a few asset nodes, which aligns with the principle that capital accumulation points should be assets with significantly higher capital inflows than other nodes. Beyond this range, the cumulative capital inflow is significantly lower than the preset threshold, and there are no capital accumulation point markers, reflecting the capital inflow distribution characteristics of the dynamic capital transmission path under the combined action of the path splicing module and the aggregation identification module.

[0048] Example 4: In practice, the aggregation and identification module dynamically calculates a preset threshold before screening fund aggregation points. The module obtains the cumulative fund inflow for all asset nodes in the dynamic fund transmission path. The cumulative fund inflow is calculated by summing the total inflow of funds along each directed edge pointing to the same asset node along the direction of the dynamic fund transmission path. Finally, the module calculates the average cumulative fund inflow for all asset nodes. The cumulative fund inflows across all asset nodes are summed and then divided by the total number of asset nodes. The total number of asset nodes represents the number of nodes appearing in the dynamic fund transmission path. The aggregation and identification module calculates the standard deviation of the inflow funds. For each asset node, calculate the cumulative capital inflow and the average inflow amount. The difference is calculated by squared, summing the squared differences for all asset nodes, dividing by the total number of asset nodes, and then taking the arithmetic square root to obtain the standard deviation of inflow funds. The aggregation and identification module obtains the adjustment coefficient. Adjustment coefficient Set to a fixed value of 1.5. Adjustment coefficient. The value of 1.5 is set based on the assumption of a normal distribution. The corresponding cumulative probability is approximately 93.3%, which can filter out fund inflow nodes that are significantly above average, while avoiding the omission of important fund aggregation points due to excessively high threshold settings. The aggregation identification module will use the preset threshold. Calculated as The aggregation and identification module compares the cumulative fund inflow of each asset node with a preset threshold. A comparison is made when the cumulative capital inflow exceeds a preset threshold. At that time, the asset node is marked as a fund aggregation point.

[0049] In practice, the position optimization module obtains the asset association graph and the list of fund aggregation points output by the aggregation identification module. The asset association graph is an undirected graph with each digitized asset in the position structure data as a node and the historical synchronous transaction frequency between two digitized assets as the edge weight. The position optimization module traverses each fund aggregation point in the fund aggregation point list to determine the adjacent fund aggregation points for each fund aggregation point. The method for determining adjacent fund aggregation points is as follows: in the asset association graph, starting from the fund aggregation point, visit the neighboring nodes directly connected to the fund aggregation point along the edges. If a neighboring node belongs to the fund aggregation point list, then the neighboring node is marked as the adjacent fund aggregation point of the fund aggregation point. If the fund aggregation point does not have any directly connected neighboring nodes in the asset association graph that belong to the fund aggregation point list, then the position optimization module uses the fund aggregation point as the starting point and performs a breadth-first search in the asset association graph. The breadth-first search expands outward layer by layer according to the connection relationship between the nodes, and the search depth increases by 1 for each edge traversed. The first node found that belongs to the fund aggregation point list is taken as the adjacent fund aggregation point of the fund aggregation point.

[0050] The position optimization module calculates the graph distance between the fund aggregation point and each of its adjacent fund aggregation points. The graph distance is calculated using the number of edges traversed by the shortest path between nodes. When adjacent fund aggregation points are connected by a direct edge, the graph distance is 1. When adjacent fund aggregation points are obtained through breadth-first search, the graph distance equals the search depth, i.e., the number of edges traversed from the fund aggregation point to its adjacent fund aggregation points. For fund aggregation points with multiple adjacent fund aggregation points, the position optimization module calculates the arithmetic mean of the graph distances between the fund aggregation point and each of its adjacent fund aggregation points to obtain the mean distance of the fund aggregation point. ,in, The serial number is used to mark the collection point of funds. The integers are positive integers, numbered sequentially starting from 1. If a fund aggregation point has only one adjacent fund aggregation point, then the distance to the mean is... It is directly equal to the spectral distance between the fund aggregation point and the unique adjacent fund aggregation point.

[0051] The position optimization module collects the average distances from all fund convergence points, forming a set of average distances. The module then searches this set for the minimum average distance. and the mean of the maximum distance If the minimum distance mean Equal to the mean of the maximum distance If the mean distance between all fund aggregation points is the same, then the normalized distribution density of all fund aggregation points is set to 0.5. If the minimum mean distance... Not equal to the mean of the maximum distance For each fund collection point, the normalized distribution density is calculated using the following formula:

[0052] in, Indicates the first Normalized distribution density of each fund aggregation point; Indicates the first The average distance between each fund convergence point; This represents the minimum distance mean in the set of distance means; This represents the maximum distance mean in the set of distance means.

[0053] Normalized distribution density The range of values ​​is Normalized distribution density The smaller the value, the higher the value. The closer a fund aggregation point is to other fund aggregation points in terms of their spectral distance, the higher its distribution density; the normalized value of the distribution density... The larger the value, the higher the value. The greater the distance between a fund aggregation point and other fund aggregation points in the graph, the lower the distribution density.

[0054] The portfolio optimization module adjusts the holding ratio of each asset in the portfolio based on the normalized distribution density value. Before adjustment, each digital asset in the portfolio has an initial holding ratio, which is calculated from the nominal holding amount. The calculation method is as follows: multiply the nominal holding amount of each digital asset by the current market price of that digital asset to obtain the market value of the holdings; divide the market value of each digital asset by the total market value of the portfolio to obtain the initial holding ratio of each digital asset; the sum of all initial holding ratios is 1. The portfolio optimization module calculates the increase in holding ratio for each digital asset corresponding to each fund aggregation point in the fund aggregation point list. The calculation method for the increase in holding ratio is as follows: ,in, Indicates the first The increase in the proportion of digital assets held at each fund aggregation point Indicates the first The initial holding ratio of digital assets corresponding to each fund aggregation point To improve the strength coefficient. Set to a fixed value of 0.2 to increase the strength coefficient. The setting of 0.2 is to control the adjustment of the holding ratio of a single asset in a single optimization to no more than one-fifth of its initial holding ratio, thus avoiding drastic changes in the portfolio's holding structure. The adjusted holding ratio of the digital asset corresponding to the fund convergence point is... .

[0055] The portfolio optimization module sums up the increase in the holding ratio of digital assets corresponding to all fund aggregation points to obtain the total increase. , Total lift To maintain a constant total holding ratio of 1, deductions need to be made from the digital assets corresponding to non-fund aggregation points. Non-fund aggregation points refer to all digital assets in the portfolio other than those listed in the fund aggregation point list. The portfolio optimization module obtains the initial holding ratio of the digital assets corresponding to each non-fund aggregation point. ,in, This is a serial number marker for non-fund aggregation points. The position optimization module divides the initial holding ratio of digital assets corresponding to each non-fund aggregation point by the sum of the initial holding ratios of digital assets corresponding to all non-fund aggregation points to obtain the allocation ratio for each non-fund aggregation point. Then, it multiplies the allocation ratio by the total increase. The holding ratio reduction for each non-fund aggregation point is obtained. The adjusted holding ratio for non-fund aggregation points is: .

[0056] The portfolio optimization module merges the adjusted holding ratios of all digital assets corresponding to all fund aggregation points with the adjusted holding ratios of all digital assets corresponding to non-fund aggregation points to form an optimized portfolio allocation. The optimized portfolio allocation includes the asset code and adjusted holding ratio of each digital asset in the portfolio. Based on the adjusted holding ratio and the total size of the portfolio, the adjusted nominal holding of each digital asset is further calculated.

[0057] See Figure 6In the chart, the horizontal axis represents asset ranking, based on the initial holding ratio of each digital asset in the portfolio, arranged from largest to smallest. The higher the ranking, the higher the initial holding ratio of that asset. The vertical axis represents the holding ratio of the assets, using a logarithmic scale, covering a range of approximately 0.001 to 0.1. The legend shows that the dashed line represents the holding ratio before adjustment, the solid line represents the holding ratio after adjustment, and the red triangle marks the capital convergence point.

[0058] As can be seen from the graph, the initial holding ratio decreases with the asset ranking, and the curve shows a clear segmented step-like downward trend. The holding ratio is relatively high for the first few assets, and the holding ratio changes significantly within the asset ranking range of approximately 0 to 100. After the ranking exceeds 100, the holding ratio tends to level off, approaching the lower limit of 0.001.

[0059] The capital aggregation points marked by red triangles are concentrated in the asset ranking range with a high holding ratio, mainly distributed among the top 10 to 20 assets. For these capital aggregation points, the adjusted holding ratio is significantly higher than the unadjusted holding ratio, indicating that the portfolio optimization module increased the holding ratio of these assets based on the distribution density of the capital aggregation points. The specific increase is related to the normalized value of the distribution density of the capital aggregation points; the asset holding ratios corresponding to capital aggregation points closer to other capital aggregation points (high distribution density) increase more significantly.

[0060] In the mid-to-late ranking asset range (ranked above 100), the holding ratio curves before and after the adjustment basically overlapped, and no capital convergence point was marked, indicating that the holding optimization module did not significantly adjust the holding ratio of low-weight assets, thus maintaining the stability of the initial holding structure.

[0061] Example 5: In practice, after obtaining the optimized portfolio configuration, the portfolio optimization module sends it to the configuration execution module. The configuration execution module reads the optimized configuration, which includes the asset code and adjusted holding ratio of each digital asset in the portfolio. The configuration execution module then generates a sub-configuration scheme for each digital asset. The sub-configuration scheme is generated by obtaining the asset code of the digital asset and querying the corresponding on-chain address from the portfolio management account based on the asset code. The on-chain address is the storage address of the digital asset in the blockchain network, used for receiving and sending digital assets. The sub-configuration scheme includes an asset code field, an on-chain address field, an operation direction field, and a target holding amount field. The value of the "Operation Direction" field is based on a comparison between the adjusted nominal holdings and the current nominal holdings of the digital asset. If the adjusted nominal holdings are greater than the current nominal holdings, the operation direction is set to "increase," and the operation quantity is the difference between the adjusted and current nominal holdings. If the adjusted nominal holdings are less than the current nominal holdings, the operation direction is set to "decrease," and the operation quantity is the difference between the current and adjusted nominal holdings. If the adjusted nominal holdings are equal to the current nominal holdings, no sub-configuration scheme is generated. The "Target Holding" field should be filled with the adjusted nominal holdings. The configuration execution module sorts all generated sub-configuration schemes according to the timestamp of the corresponding digital asset's most recent appearance in the on-chain transaction record, with earlier timestamps appearing first.

[0062] In implementation, the configuration execution module sends each sub-configuration scheme to its corresponding on-chain address for signature confirmation in the ordered sequence. The sending method involves submitting the sub-configuration scheme as a message to be signed to the wallet client belonging to the on-chain address via the blockchain network's application programming interface (API). Upon receiving the message, the wallet client digitally signs the hash value of the sub-configuration scheme using the private key corresponding to the on-chain address. The digital signature algorithm used is the elliptic curve digital signature algorithm specified by the blockchain network. After signing, the wallet client returns a signature confirmation result, which includes the hash value of the sub-configuration scheme, the digital signature string, and the signature timestamp. After each sub-configuration scheme is sent, the configuration execution module enters a waiting state, continuously listening for the wallet client's response until it receives the signature confirmation result for that sub-configuration scheme or reaches the preset timeout period. The preset timeout period is set to 120 seconds, based on the fact that the average confirmation time for a single transaction on the blockchain network is approximately 60 seconds, and 120 seconds can cover the waiting time of two network latency peaks. If the preset timeout period is not reached and no signature confirmation result is received, the configuration execution module marks the sub-configuration scheme as a signature failure and suspends the sending of subsequent sub-configuration schemes.

[0063] In practice, after receiving the signature confirmation results of all sub-configuration schemes, the configuration execution module verifies the validity of each signature confirmation result. The verification method is as follows: using the elliptic curve digital signature verification algorithm of the blockchain network, the hash value of the sub-configuration scheme, the digital signature string, and the public key corresponding to the on-chain address are input to determine whether the signature is valid. If the signature confirmation results of all sub-configuration schemes pass verification, the configuration execution module generates a summary execution instruction. The summary execution instruction is a structured data object containing an instruction version number, a generation timestamp, a list of sub-configuration schemes, and a signature list. The instruction version number marks the data structure version of the summary execution instruction and is set as a string constant. The generation timestamp is set to the system time when the summary execution instruction was generated. The list of sub-configuration schemes contains all sub-configuration schemes that have passed signature verification. The signature list contains a digital signature string corresponding one-to-one with each sub-configuration scheme in the list. The configuration execution module serializes the summary execution instruction into a preset data format, which can be JSON. The configuration execution module sends the serialized summary execution instruction to the blockchain network through the blockchain network's transaction commit interface. In a blockchain network, aggregated execution instructions are encapsulated into a transaction. This transaction calls the batch adjustment function of the asset portfolio adjustment smart contract, passing the on-chain address, operation direction, and operation quantity of each sub-configuration scheme as function parameters. Nodes in the blockchain network execute this transaction, performing the corresponding asset transfer operation for each sub-configuration scheme to complete the adjustment of the holdings configuration.

[0064] In its implementation, the aggregation and identification module performs a sorting and filtering operation when marking fund aggregation points. The module acquires all asset nodes whose cumulative fund inflow exceeds a preset threshold and groups these nodes into a set of nodes exceeding the threshold. The module then reads the cumulative fund inflow of each asset node in the set and sorts them in descending order based on the magnitude of the cumulative fund inflow. The descending sorting process is as follows: the asset node with the largest cumulative fund inflow is ranked first, the second largest is ranked second, and so on, until all asset nodes in the set of nodes exceeding the threshold are assigned a sorting number. The module acquires a preset positive integer N, which is set to 5. This is based on the premise that the total number of digital assets in the investment portfolio is typically between 20 and 50. Selecting the top 5 asset nodes as fund aggregation points can cover the small portion of assets with the most concentrated fund aggregation in the portfolio, while avoiding the inclusion of too many secondary fund aggregation points that would lead to overly dispersed portfolio adjustments. The sorting numbers are compared with N, and asset nodes with sorting numbers less than or equal to N are selected as the final fund aggregation points. If the total number of asset nodes in the set of nodes exceeding the threshold is less than N, the aggregation and identification module will use all asset nodes in the set of nodes exceeding the threshold as the final fund aggregation point.

[0065] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A portfolio optimization system based on quantitative strategies, characterized in that, The system includes: The data acquisition module acquires the portfolio's holding structure data and the corresponding on-chain transaction records of digital assets; The graph construction module constructs an asset association graph based on the holding structure data, and decomposes the asset association graph into multiple time-series subgraphs using the timestamps in the on-chain transaction records as the trigger sequence. The flow vectorization module calculates the transaction flow weights between nodes for each time series subgraph, and generates a quantitative edge set reflecting the direction of fund transmission between assets based on the transaction flow weights. The path splicing module splices together the quantitative edge sets corresponding to all time series subgraphs in chronological order to form a dynamic capital transmission path for the investment portfolio. The aggregation and identification module filters out asset nodes whose capital inflow exceeds a preset threshold based on the dynamic capital transmission path, and marks the asset nodes as capital aggregation points. The portfolio optimization module adjusts the holding ratio of each asset in the portfolio based on the distribution density of all fund aggregation points to obtain an optimized portfolio configuration.

2. The portfolio optimization system based on quantitative strategies according to claim 1, characterized in that, The acquisition of portfolio holdings structure data and corresponding on-chain transaction records of digital assets includes: Read the asset code list and the nominal holdings of each asset from the portfolio management account as the portfolio structure data; The on-chain transaction records are obtained by querying the blockchain data nodes for all on-chain transfer records of each digital asset in the asset code list within the most recent complete transaction cycle.

3. The portfolio optimization system based on quantitative strategies according to claim 1, characterized in that, The step involves constructing an asset association graph based on the holding structure data, and using the timestamps in the on-chain transaction records as trigger sequences to decompose the asset association graph into multiple time-series sub-graphs, including: Using each digital asset in the portfolio structure data as a node and the historical synchronous transaction frequency between two digital assets as the edge weight, the asset association graph is constructed. Multiple time slices are determined according to the chronological order of the timestamps in the on-chain transaction records; Within each time slice, nodes in the asset association graph where transactions occurred within that time slice and the edges connecting those nodes are extracted to form the time-series subgraph corresponding to that time slice.

4. The portfolio optimization system based on quantitative strategies according to claim 3, characterized in that, For each time-series subgraph, the transaction flow weights between nodes are calculated, and based on these weights, a quantitative edge set reflecting the direction of fund transmission between assets is generated, including: For each time series subgraph, the total transaction amount of each directed edge is calculated, and the total transaction amount is divided by the sum of all outgoing transaction amounts of the starting node of the directed edge in the time series subgraph to obtain the transaction flow weight of the directed edge. All directed edges whose transaction flow weight is greater than a preset flow threshold are extracted to form the quantized edge set of the time series subgraph.

5. The portfolio optimization system based on quantitative strategies according to claim 4, characterized in that, The step of stitching together the quantized edge sets corresponding to all time-series subgraphs in chronological order to form the dynamic fund transmission path of the investment portfolio includes: Arrange all time series subgraphs in chronological order of their corresponding time slices to obtain a time series sorting list; According to the time-series sorting list, the endpoint node of the directed edge in the quantization edge set of each time-series subgraph is matched with the starting node in the quantization edge set of the next time-series subgraph by node name. The quantized edge sets in the two successfully matched time series subgraphs are connected sequentially to form the dynamic capital transmission path.

6. The portfolio optimization system based on quantitative strategies according to claim 5, characterized in that, The step of filtering out asset nodes with capital inflows exceeding a preset threshold based on the dynamic capital transmission path and marking these asset nodes as capital aggregation points includes: Along the direction of the dynamic fund transmission path, the total amount of funds flowing into each directed edge pointing to the same asset node is accumulated, which is taken as the cumulative fund inflow of that asset node. The cumulative capital inflow is compared with the preset threshold. When the cumulative capital inflow is greater than the preset threshold, the asset node is marked as the capital aggregation point.

7. The portfolio optimization system based on quantitative strategies according to claim 6, characterized in that, The preset threshold is dynamically calculated based on the average and standard deviation of the inflow of funds to all asset nodes in the dynamic fund transmission path.

8. The portfolio optimization system based on quantitative strategies according to claim 6, characterized in that, The step of adjusting the holding ratio of each asset in the investment portfolio based on the distribution density of all fund aggregation points to obtain the optimized portfolio allocation includes: In the asset association map, the map distance between each fund aggregation point and its adjacent fund aggregation points is calculated to obtain a distance set; Based on the minimum and maximum distances in the distance set, determine the normalized value of the distribution density; Based on the normalized value of each fund aggregation point, the holding ratio of the asset corresponding to the fund aggregation point in the portfolio is increased, and the holding ratio of the asset corresponding to the non-fund aggregation point is decreased accordingly to obtain the optimized portfolio allocation.

9. The portfolio optimization system based on quantitative strategies according to claim 1, characterized in that, After obtaining the optimized portfolio allocation, the following is also included: The optimized holding configuration is converted into multiple sub-configuration schemes, each sub-configuration scheme corresponding to the on-chain address of a digital asset; According to the order of the timestamps, the multiple sub-configuration schemes are sent sequentially to the corresponding on-chain addresses for signature confirmation; After receiving all signature confirmation results, a summary execution instruction containing all sub-configuration schemes is generated and submitted to the blockchain network.

10. The portfolio optimization system based on quantitative strategies according to claim 1, characterized in that, Asset nodes whose capital inflow exceeds a preset threshold are further sorted in descending order according to the amount of inflow, and only the top N nodes are selected as the capital aggregation points, where N is a preset positive integer.

Images