Computerized process for automatically generating data for purchase or sale orders of financial assets.

Abstract

The invention relates to a computerized method for automatically generating purchase or sale order data for financial assets, comprising the steps of: a) selecting, from a digital library of chart patterns, at least two chart patterns so as to pre-configure a combination of chart patterns, b) generating, by means of a computer processing unit using a neural network model, a vector representation of the pre-configured combination of the selected chart patterns, c) receiving in real time, on a computer terminal, a data stream from a stock market chart of a financial asset, d) detecting, on the received stock market chart, by means of a computer processing unit, a combination of chart patterns, e) generating, from a computer processing unit using a neural network model,a vector representation of the combination of chart patterns detected in step d), f) compare the vector representation generated in step e) with the vector representation of step b), g) in case of correlation between the vector representation of step e) and the vector representation of step b), automatically generate a buy or sell order for the financial asset.

Description

Title of the invention: Computerized method for the automatic generation of purchase or sale order data for financial assets. Technical field.

[0001] The invention relates to a computerized method for automatically generating order data and more particularly data to trigger a purchase or sale of a financial asset.

[0002] This process works with the main financial assets such as: foreign exchange market (FOREX), stock market (Stocks), international index market (CAC40, DAX, DOW JONES, SPX500, EUROSTOCK50, NASDAQ, ...), bond market, commodities market (GOLD, Oil, Gas, ...). State of the art.

[0003] In the field of technical analysis of financial markets, numerous systems and methods have been developed to identify chart patterns on stock charts in order to anticipate future price movements of financial assets. Traditionally, analysts use charting tools and specialized software to manually identify chart patterns such as head and shoulders, triangles, double tops and bottoms, trend channels, etc. These tools often rely on mathematical and statistical indicators to assist the analyst in detecting these patterns.

[0004] Prior systems primarily use pattern recognition algorithms based on predefined rules. For example, some software uses data smoothing and filtering techniques to accentuate the characteristics of chart patterns and then applies specific detection rules to identify the desired patterns. While these methods can be effective, they are limited by their dependence on strict rules and fixed parameters, which can reduce their adaptability to variations in stock market data.

[0005] The invention aims to remedy all or part of the aforementioned drawbacks. In particular, an objective of the invention is to provide a method capable of triggering the purchase or sale of a financial asset with increased precision and efficiency. Presentation of the invention.

[0006] The solution proposed by the invention is a computerized method for automatically generating purchase or sale order data for financial assets, comprising the steps of:

[0007] a) select, from a digital library of chart patterns, at least two chart patterns so as to pre-parameterize a combination of chart patterns,

[0008] b) generate, by means of a computer processing unit using a neural network model, a vector representation of the pre-parameterized combination of the selected chart patterns,

[0009] c) receive, on a computer terminal, a data stream of a stock market chart of a financial asset,

[0010] d) detect, on the received stock market chart, by means of a computer processing unit, a combination of chart patterns,

[0011] e) generate, from a computer processing unit using a neural network model, a vector representation of the combination of chart patterns detected in step d),

[0012] f) compare the vector representation generated in step e) with the vector representation from step b),

[0013] g) in case of correlation between the vector representation of step e) and the vector representation of step b), automatically generate a purchase or sale order data for the financial asset.

[0014] The invention is based on the vector representation, or "embedding," of chart patterns. The use of embedding in the context of technical analysis of financial markets offers several significant advantages. First, embedding makes it possible to transform complex chart patterns into vector representations, facilitating their processing by computer processing units. This transformation enables the precise and efficient comparison of detected chart patterns with pre-parameterized ones, using vector similarity measures such as cosine distance or Pearson correlation. Furthermore, embedding makes it possible to detect subtle variations in chart patterns that might escape detection methods based on the aforementioned prior art techniques.The method according to the invention is particularly robust to the dynamic conditions of financial markets, resulting in more precise recognition and increased responsiveness during the automatic generation of buy or sell orders.

[0015] Other advantageous features of the invention will become apparent from the description. Each of these features may be considered alone or in combination with the notable features defined above. Each of these features contributes, where appropriate, to the resolution of specific technical problems defined further in the description and in which the notable features defined above do not necessarily participate. The latter may, where appropriate, be the subject of one or more divisional patent applications. In particular:

[0016] According to one embodiment, the neural network model used in step b) and step e) is a convolutional neural network (CNN) model and / or a recurrent neural network (RNN) model.

[0017] According to one embodiment, a combination of 2 to 5 chart patterns is selected in step a).

[0018] According to one embodiment, a combination of 2 to 5 chart patterns is detected in step d).

[0019] According to one embodiment, during step d), each chart pattern detection is associated with a confidence score so as to filter the results to retain only the detections whose score reaches or exceeds a predefined confidence threshold.

[0020] According to one embodiment, before the implementation of step f), the vector representations generated in step e) and in step b) are pre-normalized by eliminating differences in magnitudes and / or scale effects.

[0021] According to one embodiment, during step f), a processing unit executes a Euclidean similarity or cosine similarity algorithm to compare the vector representation generated in step e) with the vector representation of step b).

[0022] According to one embodiment, in step g), the correlation is evaluated by the processing unit configured to define a similarity threshold, and to trigger the automatic generation of the purchase or sales order data if the similarity between the two vector representations reaches or exceeds said threshold. Description of the implementation methods.

[0023] The process of the invention generates manipulations of physical elements, in particular signals (electrical or magnetic) and digital data, capable of being stored, transferred, combined, compared, ..., and enabling the achievement of a desired result.

[0024] The invention implements one or more computer applications executed by computer equipment or servers. For clarity, it should be understood in the context of the invention that "equipment or server does something" means "the computer application executed by a processing unit of the equipment or server does something." Similarly, "the computer application does something" means "the computer application executed by the processing unit of the equipment or server does something."

[0025] For the sake of clarity, the present invention may refer to one or more "logical computer processes." These correspond to the actions or results obtained by executing instructions from various computer applications. Therefore, for the purposes of this invention, "a logical computer process is suitable for doing something" means "the Instructions from a computer application executed by a processing unit do something.

[0026] For the sake of clarity, the following clarifications are provided for certain terms used in the description and claims: - "Computer resource" can be understood in a non-limiting way as: component, hardware, software, file, connection to a computer network, amount of RAM memory, hard drive space, bandwidth, processor speed, number of CPUs, etc. - “Computer server” can be understood in a non-limiting way as: computer device (hardware or software) comprising computing resources to perform the functions of a server and which offers services, computer, plurality of computers, virtual server on the internet, virtual server on the Cloud, virtual server on a platform, virtual server on a local infrastructure, server networks, cluster, node, server farm, node farm, etc. - “Processing unit” can be understood in a non-limiting way as: processor, microprocessors, CPU (for Central Processing Unit). - “Computer hardware” means one or more spare parts of a computer equipment and can be understood in a non-limiting way as hardware. - “Computer application” can be understood as: software, computer program, etc. - "Computer network" can be understood, without limitation, as: computer bus, personal area network (PAN), local area network (LAN, WLAN, etc.), wide area network (WAN), internet, intranet, extranet. A computer network is a set of interconnected computer equipment used to exchange information and / or data, securely or not, according to a communication protocol (ISDN, Ethernet, ATM, IP, CLNP, TCP, HTTP, etc.). - "Database" can be understood, without limitation, as a structured and organized set of data stored on media accessible by computer equipment and capable of being queried, read, and updated. Data can be inserted, retrieved, modified, and / or destroyed. The management of and access to the database can be ensured by a set of computer applications that constitute a database management system (DBMS). - As used here, unless otherwise indicated, the use of the ordinal adjectives "first", "second", etc., to describe an object simply indicates that different occurrences of similar objects are being mentioned and does not imply that the objects so described must be in any given sequence, whether in time, space, ranking, or any other way.

[0027] The method of the invention comprises a step a) consisting of selecting, from a digital library of chart patterns, at least two chart patterns in order to pre-configure a combination of chart patterns. A combination of 2 to 5 chart patterns is preferably selected. This number makes it possible to cover a wide range of market scenarios while maintaining the relevance of the technical analyses. Furthermore, by limiting the combination to 2 to 5 patterns, noise in the data is reduced, which allows for clearer and more relevant analyses, with optimization of computing resources.

[0028] The digital library can be constructed by compiling chart patterns, advantageously with data formatting. In one embodiment, the compilation of chart patterns is carried out by identifying and cataloging various chart patterns commonly used in technical analysis, such as head and shoulders, triangles, double tops and bottoms, trend channels, etc. Each chart pattern is preferably annotated with relevant characteristics, such as its structure, typical time period, and associated market conditions. The chart pattern data is advantageously normalized in terms of size, time scale, and amplitude of price changes. In one embodiment, each chart pattern is then converted into a suitable digital representation, for example, in the form of pixel matrices.Using a human-machine interface, a user can combine selected chart patterns by defining, for example, an order and time relationships between them. For instance, a user might choose to analyze a head and shoulders pattern followed by an ascending triangle. Specific parameters for each chart pattern in the combination, such as price thresholds, durations, and support and resistance levels, can also be adjusted by the user.

[0029] The method includes a second step b) consisting of generating, by means of a computer processing unit using a neural network model, a vector representation of the pre-parameterized combination of chart patterns selected in step a).

[0030] The neural network model is preferably a convolutional neural network (CNN) and / or a recurrent neural network (RNN). A CNN can be used to extract visual features of each chart pattern. Then, an RNN can be used to capture the temporal dynamics of the chart patterns. The resulting vector representation depicts the features and dynamics of each chart pattern.

[0031] The model is trained with a set of historical chart patterns. The training data is pre-labeled with the types of chart patterns and their characteristics.

[0032] Pre-parameterized chart patterns are transformed into a format suitable for input to the neural network. This may involve resizing images and / or formatting time series. The transformed chart patterns are then passed through the neural network using forward propagation to obtain the model outputs. Vector representations of the chart patterns are extracted from the output layer or intermediate layers of the neural network. If the chart patterns are processed individually, their representation vectors are concatenated to form a single vector representation of the combination. Alternatively, pooling techniques are applied to merge the vectors into a single representation.Storage Format: The generated vector representation is advantageously stored in a structured format, such as a vector database, possibly with associated metadata (e.g., chart pattern identifiers, pre-configuration parameters). By following these detailed steps, step b) can be implemented in a structured and efficient manner to generate accurate and robust vector representations of pre-configured combinations of chart patterns, thus facilitating subsequent comparison and automated decision-making steps within the framework of technical analysis of financial markets.

[0033] The aforementioned steps a) and b) can be implemented using a computer installation comprising one or more computer servers and / or necessary computer resources.

[0034] The method includes a third step (c) of receiving, in real time on a computer terminal, a data stream of a stock chart of a financial asset. The data stream can be received from one or more real-time stock market data providers (e.g., Bloomberg®, Reuters®, Alpha Vantage®, IEX Cloud®). The data can be opening prices, closing prices, volumes, or other data, over defined time intervals (e.g., minute, hourly, daily). In-memory queues can be used to manage the incoming data streams. A user interface displays the stock chart in real time.

[0035] The method includes a fourth step (d) consisting of detecting, on the stock market chart received in step (c), using a computer processing unit, a combination of chart patterns. Preferably, a combination of 2 to 5 chart patterns is detected, this number corresponding to the number of chart patterns selected in the aforementioned step (a). This reduces the processing load by only seeking to detect the most promising specific combinations, thereby increasing the reliability of the results and improving the calculation speed.

[0036] The processing unit is, for example, integrated into a dedicated computer server or a user terminal (e.g., a computer). In one embodiment, the processing unit uses a neural network model, preferably a convolutional neural network (CNN), to detect chart patterns. Sliding windows can be used on time series to detect chart patterns in different time segments, for example, segments of 10, 20, or 50 periods.

[0037] According to one embodiment, each chart pattern detection is associated with a confidence score that filters the results to retain only those detections that reach or exceed a predefined confidence threshold. In particular, when the neural network model detects a chart pattern, it can generate a confidence score. For example, a score of 0.9 indicates a very high probability that the detected pattern is correct, while a score of 0.3 indicates a low probability. A confidence threshold is then established to filter the detections. For example, the processing unit can be programmed to consider only detections with a confidence score greater than 0.8. Only chart patterns with a confidence score above the defined threshold are retained for subsequent steps in order to improve the overall accuracy of the process.

[0038] The method includes a fifth step e) consisting of generating, from a computer processing unit using a neural network model, a vector representation of the combination of chart patterns detected in step d). The processing unit is advantageously the same as that used for step d).

[0039] The neural network model is preferably a convolutional neural network (CNN) and / or a recurrent neural network (RNN). The model is preferably the one used in step b), with the chart patterns detected in step d) being treated as the chart patterns of said step b). The CNN is particularly effective at detecting and recognizing complex patterns, which is especially effective for identifying chart patterns on stock charts. The RNN allows for a better understanding of stock price movements over extended periods. The combined use of CNNs and RNNs thus improves the accuracy of chart pattern detection by taking into account both spatial and temporal characteristics.

[0040] The method includes a sixth step f) consisting of comparing the vector representation generated in step e) with the vector representation from step b). According to a preferred embodiment, this comparison is performed using vector distance and similarity calculation techniques. The processing unit is advantageously the same as that used for steps d) and e).

[0041] The vector representations generated in steps e) and b) are advantageously normalized beforehand to make them comparable, eliminating differences in magnitudes and / or scale effects. Different normalization methods can be used, such as Euclidean normalization, which consists of dividing each vector representation by its Euclidean norm, or Min-Max normalization, which consists of transforming the values ​​of each vector representation so that they fall within a given range, generally between 0 and 1. This normalization ensures that the vector representations are comparable on a common scale and eliminates biases due to different magnitudes, thus facilitating accurate comparison of the vectors and improving the accuracy of the similarity calculations.

[0042] The processing unit can execute a similarity algorithm to compare the two vector representations. A Euclidean similarity algorithm can, for example, be parameterized to measure the direct distance between two vector representations in a multidimensional space: the smaller the distance, the greater the similarity. A cosine similarity algorithm can also be used to measure the angle between the two vector representations. A smaller angle (or a cosine closer to 1) indicates greater similarity. The cosine similarity between two vector representations VI and V2 is notably given by the formula: sim (VI, V2) = V1.V2 / (IIV1II.IIV2II).

[0043] Euclidean and cosine similarity algorithms are particularly robust, reliable and fast for vector comparison, ensuring accurate and rapid results.

[0044] The method includes a seventh step (g) in which, following the result of step (f), if there is a correlation between the vector representation of step (e) and the vector representation of step (b), then a buy or sell order for the financial asset is automatically generated. "Correlation" refers to the measure of similarity between the two vector representations: the one generated in step (e) from chart patterns detected in real time on the stock chart, and the one generated in step (b) from pre-configured chart patterns. A high correlation indicates that the two vectors are very similar, suggesting that the combination of chart patterns detected in real time closely matches the pre-configured one.

[0045] To evaluate this correlation, the aforementioned processing unit advantageously defines a similarity threshold to determine whether the vector representations are sufficiently similar. For example, for cosine similarity, a threshold of 0.9 can be chosen, indicating a high correlation. If the similarity between the two If vector representations reach or exceed the defined threshold, then the processing unit triggers the automatic generation of the purchase or sales order data.

[0046] Once the buy or sell order data has been generated, it can be pre-formatted in a standard format recognized by electronic trading systems. This order is then transmitted via an application programming interface (API) to a trading platform or online broker. The order is transmitted in real time to minimize the delay between the decision and execution, which is crucial in dynamic trading environments. The order is then placed in the order book of the relevant exchange or trading platform. Once validated, the order is executed, resulting in the purchase or sale of the financial asset according to market conditions at that precise moment.

[0047] The arrangement of the various elements and / or means and / or steps of the invention, in the embodiments described above, should not be understood as requiring such an arrangement in all implementations. Other variations may be envisaged.

[0048] Furthermore, one or more features described only in one embodiment can be combined with one or more other features described only in another embodiment. Similarly, one or more features described only in one embodiment can be generalized to other embodiments, even if this or these features are described only in combination with other features.

Claims

Demands

1. A computerized method for automatically generating purchase or sale order data for financial assets, comprising the steps of: a) selecting, from a digital library of chart patterns, at least two chart patterns so as to pre-parameterize a combination of chart patterns, b) generating, by means of a computer processing unit using a neural network model, a vector representation of the pre-parameterized combination of the selected chart patterns, c) receiving in real time, on a computer terminal, a data stream of a stock chart of a financial asset, d) detecting, on the received stock chart, by means of a computer processing unit, a combination of chart patterns, e) generating, from a computer processing unit using a neural network model, a vector representation of the combination of chart patterns detected in step d).f) compare the vector representation generated in step e) with the vector representation from step b), g) if there is a correlation between the vector representation from step e) and the vector representation from step b), automatically generate a purchase or sale order for the financial asset.

2. A method according to claim 1, wherein the neural network model used in step b) and step e) is a convolutional neural network (CNN) model and / or a recurrent neural network (RNN) model.

3. A method according to any one of the preceding claims, wherein a combination of 2 to 5 chart patterns is selected in step a).

4. A method according to any one of the preceding claims, wherein a combination of 2 to 5 chart patterns is detected in step d).

5. A method according to any one of the preceding claims, wherein during step d), each chart pattern detection is associated with a confidence score so as to filter the results to retain only those detections whose score reaches or exceeds a predefined confidence threshold.

6. A method according to any one of the preceding claims, wherein prior to the implementation of step f), the vector representations generated in step e) and step b) are pre-normalized by eliminating differences in magnitudes and / or scale effects.

7. A method according to any one of the preceding claims, wherein during step f), a processing unit executes a Euclidean similarity or cosine similarity algorithm to compare the vector representation generated in step e) with the vector representation of step b).

8. A method according to claim 7, wherein in step g), the correlation is evaluated by the processing unit configured to define a similarity threshold, and to trigger the automatic generation of the purchase or sales order data if the similarity between the two vector representations reaches or exceeds said threshold.