Method and system for predicting asset portfolio

Abstract

An asset portfolio prediction method according to one disclosure of the present invention is a method for predicting an asset portfolio performed by a computing system, the method comprising the steps of: determining a target prediction task which is data specifying a target task; collecting raw data for performing the determined target prediction task; arranging the collected raw data in time series according to a specific time axis; training a predetermined time series prediction model on the basis of the arranged raw data and the determined target prediction task; predicting portfolio asset weight data for each date for a predetermined prediction unit period on the basis of the trained time series prediction model; generating single portfolio asset weight data by processing the predicted portfolio asset weight data for each date at every predetermined rebalancing time point; and providing the generated single portfolio asset weight data.

Description

Asset portfolio forecasting method and system

[0001] The present disclosure relates to a method and system for predicting asset portfolios. More specifically, it relates to a method and system for predicting asset portfolio weights based on adjustable targets and multi-faceted data.

[0002] In the financial investment environment, it is crucial to simultaneously ensure predictive accuracy, stability, and purposefulness of portfolio asset allocation.

[0003] However, existing traditional portfolio construction methods (e.g., Mean-Variance Optimization (MVO) method) rely on past data to calculate the covariance matrix and derive the weights of each asset through the inverse matrix operation, which has the problem that numerical instability of the covariance matrix may occur depending on the data collection period or market volatility.

[0004] Furthermore, statistical regression models based on past returns or machine learning-based models that use only limited features have limitations in that they cannot immediately adapt to rapid changes in the market environment or unexpected macroeconomic variables, and as a result, prediction errors can accumulate, hindering portfolio management performance.

[0005] Moreover, existing portfolio optimization models typically only predict future returns and calculate asset weights based on these. Therefore, in actual investment management, they cannot directly provide a time series representation of how asset weights should change at each rebalancing cycle, which may limit the establishment of practical rebalancing strategies.

[0006] Failure to comprehensively consider the expected fluctuations in asset weights can make consistent and stable asset allocation difficult, making it difficult to effectively respond to unexpected market risks.

[0007] Moreover, existing prediction models do not provide sufficient supporting information (e.g., which variables are causally related to the results) that investors or operators can use to verify the reliability of the prediction results, making it difficult to ensure reliability when utilizing the prediction results in actual investment strategies.

[0008] In addition, flexible investment strategy responses to specific events or variable changes are required by performing user-customized scenarios (e.g., what-if simulations). However, existing systems lack the ability to immediately re-forecast and provide supporting data in response to such assumption changes, which can hinder investor decision-making efficiency.

[0009] Furthermore, in the past, the feedback loop structure that continuously monitors the performance of the predicted asset weight according to the rebalancing cycle and updates and renews the model based on the performance was also inadequate, which limited adaptability to market changes and stability of long-term operating performance.

[0010] Therefore, in one embodiment according to the present disclosure, an improved technical paradigm is introduced to overcome the limitations of existing technologies, such as dynamically predicting asset portfolio weights based on adjustable targets and multi-faceted data, providing a basis for such predictions, evaluating operating performance accordingly, and performing model feedback updates.

[0011] One embodiment of the present disclosure is designed to solve the problems of the prior art as described above, and the purpose is to provide a method and system for predicting asset portfolio weights using a deep learning model based on a Temporal Fusion Transformer (TFT) by utilizing adjustable targets and multi-faceted data.

[0012] However, the technical problems that the present disclosure and the embodiments of the present disclosure seek to solve are not limited to the technical problems described above, and other technical problems may exist.

[0013] A method for predicting an asset portfolio according to one disclosure of the present invention is a method for predicting an asset portfolio performed by a computing system, comprising: determining a target prediction task, which is data specifying a target task; collecting raw data for performing the determined target prediction task; arranging the collected raw data in time series according to a specific time axis; training a predetermined time series prediction model based on the aligned raw data and the determined target prediction task; predicting daily portfolio asset weight data for a predetermined prediction unit period based on the trained time series prediction model; processing the predicted daily portfolio asset weight data at each predetermined rebalancing point to generate single portfolio asset weight data; and providing the generated single portfolio asset weight data.

[0014] In another aspect, the step of determining the target prediction task includes a step of obtaining task setting information, which is information including at least two or more data from among a predetermined domain, a current portfolio asset ratio, a rebalancing period, an asset pool, and investment propensity data.

[0015] In another aspect, the investment propensity data includes data specifying at least one of a profit maximizing propensity that allocates weight to assets with positive past cumulative returns, a risk minimizing propensity that allocates weight to assets with low past volatility, and a Sharpe ratio optimizing propensity that allocates weight in order of high Sharpe ratio for each asset.

[0016] In another aspect, the step of collecting the raw data includes a step of collecting at least one of predetermined price time series data, macroeconomic indicator data, market indicator data, and target impact variable data.

[0017] In another aspect, the target influence variable data is a variable that is correlated with at least one of the domain and asset data corresponding to the target prediction task above a predetermined standard.

[0018] In another aspect, the step of aligning the raw data in a time-series manner along a specific time axis includes the step of aligning the task setting information corresponding to the target prediction task along the time axis.

[0019] In another aspect, the step of training the time series prediction model includes the steps of inputting time series data from a predetermined point tp to a predetermined point t in a learning data set including the time-series-sorted raw data and the task setting information into the time series prediction model, the step of predicting the daily portfolio asset weight data from a predetermined point t+1 to a predetermined point t+h through the time series prediction model input with the time series data, the step of performing normalization post-processing so that the sum of the asset weights included in the portfolio asset weight data becomes '1' for each of the predicted daily portfolio asset weight data, the step of comparing the normalized daily portfolio asset weight data with the actual daily portfolio asset weight data from the predetermined point t+1 to a predetermined point t+h, the step of calculating a loss according to the result of the comparison, and the step of updating the parameters of the time series prediction model based on the calculated loss.

[0020] In another aspect, the step of predicting daily portfolio asset weight data for the above prediction unit period includes a step of performing normalization post-processing so that the sum of the weights of each asset included in the portfolio asset weight data becomes '1' for each of the predicted daily portfolio asset weight data.

[0021] In another aspect, the step of generating the single portfolio asset weight data includes a step of applying at least one rule among a rule for applying a predetermined operation to the daily portfolio asset weight data and a rule for determining whether the daily portfolio asset weight data satisfies a predefined criterion, to the daily portfolio asset weight data, and a step of generating the single portfolio asset weight data based on a result according to the applied rule.

[0022] In another aspect, the step of generating the single portfolio asset weight data further includes the step of generating the single portfolio asset weight data by selectively applying at least one of a predetermined clipping and smoothing operation if there are predetermined transaction costs and constraints on the weights of assets.

[0023] In another aspect, the method for predicting an asset portfolio according to one disclosure of the present invention further includes a step of providing basis data, which is data explaining how the single portfolio asset weight data was determined based on what causal relationship and data flow.

[0024] In another aspect, the asset portfolio prediction method according to one disclosure of the present invention further includes a step of evaluating the performance of the time series prediction model based on actual investment performance according to the single portfolio asset weight data, and a step of updating the parameters of the time series prediction model according to the evaluation result.

[0025] In another aspect, a method for predicting an asset portfolio according to one disclosure of the present invention further includes a step of obtaining a target label data set, which is data that generates daily portfolio asset weight data corresponding to a prediction period of a time series prediction model based on a predetermined existing prediction model, and a step of training the time series prediction model based on the obtained target label data set.

[0026] In another aspect, the step of training the time series prediction model based on the correct answer label data set includes the step of comparing each of the daily portfolio asset weight data included in the correct answer label data set with each of the daily portfolio asset weight data predicted by the time series prediction model on a 1:1 basis, the step of calculating a loss according to the result of the comparison, and the step of updating the parameters of the time series prediction model based on the calculated loss.

[0027] In another aspect, the step of obtaining the correct answer label data set includes a step of obtaining the correct answer label data set for each purpose based on an existing prediction model optimized for prediction for each purpose, and the step of training the time series prediction model based on the correct answer label data set includes a step of determining a correct answer label data set corresponding to a task purpose based on at least one of a predetermined investment tendency, a market situation, and a domain from among the correct answer label data sets obtained for each purpose, and a step of training the time series prediction model based on the determined correct answer label data set.

[0028] In another aspect, the step of providing the single portfolio asset weight data includes the step of providing personalized portfolio rebalancing recommendation information based on the single portfolio asset weight data through a user interface of a robo-advisor service.

[0029] In another aspect, a method for predicting an asset portfolio according to one disclosure of the present invention further includes a step of obtaining a What-if scenario assuming a change in a specific macroeconomic indicator or market indicator through user input, and a step of generating changed single portfolio asset weight data and changed basis data through the time series prediction model according to the obtained scenario and providing the data to a fund management support system.

[0030] In another aspect, the target prediction task includes a predetermined investment theme or factor, and the single portfolio asset weight data is used as input to determine the constituent portfolio of a new exchange-traded fund (ETF) product.

[0031] In another aspect, the step of training the time series prediction model includes a step of training the time series prediction model, which is a deep learning model based on a Temporal Fusion Transformer (TFT), based on the sorted raw data and the determined target prediction task.

[0032] Meanwhile, an asset portfolio prediction system according to one disclosure of the present invention comprises: at least one memory; and at least one processor for predicting an asset portfolio by reading at least one application stored in the memory; wherein instructions of the processor include instructions for performing the steps of: determining a target prediction task, which is data specifying a target task; collecting raw data for performing the determined target prediction task; arranging the collected raw data in time series according to a specific time axis; learning a predetermined time series prediction model based on the aligned raw data and the determined target prediction task; predicting daily portfolio asset weight data for a predetermined prediction unit period based on the learned time series prediction model; generating single portfolio asset weight data by processing the predicted daily portfolio asset weight data at each predetermined rebalancing point in time; and providing the generated single portfolio asset weight data.

[0033] In another aspect, an asset portfolio prediction system according to one disclosure of the present invention comprises: a plurality of neurons configured as an array including at least one register, at least one programmable logic, and at least one input interface; a plurality of synaptic circuits storing synaptic weights that adjust connection strengths between the plurality of neurons; and at least one routing network that controls data flow between the plurality of neurons; wherein each of the plurality of neurons is connected to at least one other neuron through the routing network to set a transmission path of the weights, and further comprises a field programmable gate array (FPGA) implementation for a predetermined artificial neural network.

[0034] In another aspect, an asset portfolio prediction system according to one disclosure of the present invention comprises: a plurality of neurons organized in an array including at least one register, at least one microprocessor, and at least one input; and a plurality of synaptic circuits storing synaptic weights that adjust connection strengths between the plurality of neurons; wherein each of the plurality of neurons further comprises an application specific integrated circuit (ASIC) for a predetermined artificial neural network connected to at least one other neuron through one of the plurality of synaptic circuits.

[0035] The asset portfolio prediction method and system according to one embodiment of the present disclosure predicts asset portfolio weights using a deep learning model based on a Temporal Fusion Transformer (TFT) by utilizing adjustable targets and multi-faceted data, thereby having the effect of flexibly responding to changes in market conditions and establishing and providing an optimal rebalancing strategy that suits the investment purpose.

[0036] However, the effects that can be obtained from the present disclosure are not limited to the effects mentioned above, and other effects that are not mentioned can be clearly understood from the description below.

[0037] FIG. 1 illustrates an example block diagram of a computing system implementing an asset portfolio prediction service according to one embodiment of the present disclosure.

[0038] FIG. 2 illustrates an example block diagram of a computing device implementing an asset portfolio prediction service according to one embodiment of the present disclosure.

[0039] FIG. 3 illustrates an example block diagram of another aspect of a computing device implementing an asset portfolio prediction service according to one embodiment of the present disclosure.

[0040] FIG. 4 illustrates an example of a flowchart for explaining an asset portfolio prediction method according to one embodiment of the present disclosure.

[0041] FIG. 5 illustrates an example of a conceptual diagram of a portfolio prediction method based on a time series prediction model according to one embodiment of the present disclosure and an existing portfolio composition method.

[0042] FIG. 6 illustrates an example of the results of predicting various sector-specific portfolio weights of a time series prediction update model according to one embodiment of the present disclosure.

[0043] Figure 7 illustrates an example of a graph comparing the performance of various types of portfolio management in terms of cumulative returns, showing that a method of predicting portfolio weights reflecting a predetermined external indicator (variable) according to one embodiment of the present disclosure shows the best performance.

[0044] The present disclosure is capable of various modifications and embodiments. Therefore, specific embodiments are illustrated in the drawings and described in detail in the detailed description. The effects and features of the present disclosure, and methods for achieving them, will become clear with reference to the embodiments described in detail below together with the drawings. However, the present disclosure is not limited to the embodiments disclosed below and may be implemented in various forms. In the following embodiments, terms such as "first" and "second" are not used in a limiting sense but are used for the purpose of distinguishing one component from another. Furthermore, the singular expression includes the plural expression unless the context clearly indicates otherwise. Furthermore, terms such as "include" and "have" indicate the presence of a feature or component described in the specification, and do not preemptively exclude the possibility that one or more other features or components may be added. Furthermore, the sizes of components in the drawings may be exaggerated or reduced for convenience of explanation. For example, the size and thickness of each component shown in the drawings are arbitrarily shown for convenience of explanation, and thus the present disclosure is not necessarily limited to what is shown.

[0045] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the attached drawings. When describing with reference to the drawings, identical or corresponding components are given the same reference numerals and redundant descriptions thereof will be omitted.

[0046]

[0047] [Example system providing asset portfolio prediction services]

[0048] Hereinafter, an exemplary system for implementing a service for predicting asset portfolio weights based on adjustable targets and multi-faceted data (hereinafter, “asset portfolio prediction service”) is described in detail with reference to the attached drawings.

[0049] FIG. 1 illustrates an example block diagram of a computing system implementing an asset portfolio prediction service according to one embodiment of the present disclosure.

[0050] Referring to FIG. 1, a computing system (1000) implementing the asset portfolio prediction service of the present disclosure includes a user computing device (110), a server computing system (130), and a training computing system (150), and each device and system are communicatively connected via a network (170).

[0051] The asset portfolio prediction method and system according to one embodiment of the present disclosure may 1) be implemented and provided locally by a user computing device (110), 2) be implemented and provided in the form of a web service by a server computing system (130) communicating with the user computing device (110), or 3) be implemented and provided by the user computing device (110) and the server computing system (130) in connection with each other.

[0052] At this time, in the embodiment, the user computing device (110) and / or the server computing system (130) can train the machine learning model (120 and / or 140) through interaction with a training computing system (150) communicatively connected via a network (170). The training computing system (150) may be separate from the server computing system (130) or may be part of the server computing system (130).

[0053] And at this time, the artificial intelligence model can be 1) trained directly locally by the user computing device (110), 2) trained by the server computing system (130) and the user computing device (110) interacting with each other through a network (170), and 3) trained by a separate training computing system (150) using various training techniques and learning techniques. And the artificial intelligence model trained by the training computing system (150) can be implemented in a manner of being provided / updated by transmitting the model to the user computing device (110) and / or the server computing system (130) through a network (170).

[0054] In some embodiments, the training computing system (150) may be part of a server computing system (130) or part of a user computing device (110).

[0055] -User Computing Device (110)

[0056] The user computing device (110) may include any type of computing device, such as a smart phone, a mobile phone, a digital broadcasting device, a personal digital assistant (PDA), a portable multimedia player (PMP), a desktop, a wearable device, an embedded computing device, and / or a tablet PC.

[0057] Additionally, in the embodiment, the user computing device (110) may further include a predetermined server computing device that provides an asset portfolio prediction service environment.

[0058] Such a user computing device (110) includes at least one processor (111) and memory (112).

[0059] Here, the processor (111) of the user computing device (110) may be composed of at least one or a plurality of processors electrically connected among a central processing unit (CPU), a graphics processing unit (GPU), application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors and / or other electrical units for performing functions.

[0060] In particular, depending on the embodiment, such a processor (111) may be configured based on a field programmable gate array (FPGA) implementation, which is a hardware technology for implementing a predetermined digital circuit, and / or an application specific integrated circuit (ASIC).

[0061] Here, a field programmable gate array (FPGA) can mean a flexible digital circuit that can be programmed according to user needs.

[0062] In an embodiment, the field programmable gate array implementation may include a register that temporarily stores data and controls the flow and timing of signals to maintain intermediate results of operations or state information to support synchronized operation of the FPGA; programmable logic that programs operations within the FPGA to perform specific functions or operations as logic circuits that can be configured according to user needs; and an input interface that receives signals from an external device or sensor as a path for receiving data from outside the FPGA and transmits them to the internal circuit.

[0063] By combining the above components, field programmable gate array implementations can provide flexible and diverse forms of digital circuits.

[0064] Meanwhile, an application-specific integrated circuit (ASIC) can refer to a custom integrated circuit that is specifically designed to perform a specific purpose or function.

[0065] As an example, the application-specific integrated circuit may include a register, which is a small memory device that temporarily stores and manages data and stores intermediate results of calculations or status information to support rapid processing of the operation of the ASIC; a microprocessor, which is a central processing unit that performs control and operations within the ASIC and, if necessary, performs various operations or generates control signals to coordinate the operation of the entire system; and an input block, which is an interface that receives data from the outside and receives data to be processed by the ASIC and transmits it to the inside, and receives various input data through a connection with a sensor or an external device.

[0066] By combining the above components, application-specific integrated circuits can perform specific tasks in an optimized manner.

[0067] Returning again, the memory (112) of the user computing device (110) may include one or more non-transitory / transitory computer-readable storage media such as RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, and combinations thereof, and may include web storage of a server that performs a memory storage function on the Internet. This memory (112) may store data (113) and commands (114) necessary for the at least one processor (111) to perform functional operations such as training an artificial intelligence model or executing an asset portfolio prediction service using an artificial intelligence model.

[0068] In one embodiment, the user computing device (110) can perform various deep learning for an asset portfolio prediction service in conjunction with a deep-learning neural network.

[0069] Here, the deep learning neural network according to the embodiment may include a convolutional neural network (CNN), an R-CNN (Regions with CNN features), a Fast R-CNN, a Faster R-CNN, a Mask R-CNN, etc., and may include any deep learning neural network that includes an algorithm capable of performing the embodiment described below, and the embodiment of the present disclosure does not limit or restrict such deep learning neural network itself.

[0070] At this time, depending on the embodiment, the deep learning neural network may be installed directly in the server computing system (130) or may operate as a device separate from the server computing system (130) to perform deep learning for the asset portfolio prediction service.

[0071] Additionally, in one embodiment, the user computing device (110) may store at least one machine learning model (120).

[0072] For example, the user computing device (110) may be various machine learning models such as multiple neural networks (e.g., deep neural networks) that perform an asset portfolio prediction method based on structured / quantitative data, or other types of machine learning models including nonlinear models and / or linear models, or may be configured as a combination thereof.

[0073] For example, the machine learning model may store linear regression, decision tree, random forest, gradient boosting, pre-trained language model, or / and deep learning model. The neural network may include at least one of feed-forward neural networks, recurrent neural networks (e.g., long short-term memory recurrent neural networks), convolutional neural networks, or / and other types of neural networks.

[0074] Additionally, according to an embodiment, the user computing device (110) may store a model to be used in each process and a prompt template that serves as the basis for input to the model in order to perform at least part of the process performed for the asset portfolio prediction method through a large-scale language model (LLM).

[0075] In one embodiment, a user computing device (110) may receive at least one machine learning model (120) from a server computing system (130) via a network (170), store the model in a memory (112), and then execute the stored machine learning model (120) by a processor (111) to perform an asset portfolio prediction service, etc.

[0076] In another embodiment, a user computing device (110) may perform operations through a machine learning model (140) including at least one machine learning model (140) in conjunction with a server computing system (130) and provide an asset portfolio prediction service to a user by communicating data related thereto to an external party.

[0077] For example, a user computing device (110) can perform an asset portfolio prediction service by having a server computing system (130) provide output for a user's input using a machine learning model (140) via the web.

[0078] Additionally, the artificial intelligence model may be implemented in such a way that at least some of the machine learning models (120 and / or 140) are executed on a user computing device (110) and the rest are executed on a server computing system (130).

[0079] Additionally, the user computing device (110) may include at least one input component (121) that detects user input.

[0080] For example, the user input component (121) may include a touch sensor (e.g., a touch screen and / or touch pad, etc.) that detects the touch of a user's input medium (e.g., a finger or a stylus), an image sensor that detects the user's motion input, a microphone, button, mouse, and / or keyboard that detects the user's voice input, etc.

[0081] Here, the image sensor may include an image processing module. Specifically, the image sensor may process still images or moving images obtained by an image sensor device (e.g., CMOS or CCD).

[0082] Additionally, the image sensor can process still images or moving images acquired through the image sensor device using an image recognition process (e.g., OCR, etc.) and / or an image processing module to extract necessary information and transmit the extracted information to the processor.

[0083] Additionally, the input component (121) may receive input for an external controller (e.g., mouse, keyboard, etc.) based on the interface module, and in this case may include an external output device (e.g., speaker).

[0084] At this time, the interface module may be configured to include at least one of a wired / wireless headset port, an external charger port, a wired / wireless data port, a memory card port, a port for connecting a device equipped with an identification module, an audio I / O (Input / Output) port, a video I / O (Input / Output) port, an earphone port, a power amplifier, an RF circuit, a transceiver, and other communication circuits.

[0085] Additionally, the external output device may include a display system that outputs various information related to the asset portfolio prediction service as graphic images.

[0086] Such a display system can be implemented by including at least one of a liquid crystal display (LCD), a thin film transistor-liquid crystal display (TFT LCD), an organic light-emitting diode (OLED), a flexible display, a 3D display, and an e-ink display.

[0087] Meanwhile, the user computing device (110) including the above-described components may further perform at least some of the functional operations performed by the server computing system (130) described below.

[0088] -Server Computing System (130)

[0089] The server computing system (130) can perform a series of processes to provide an asset portfolio prediction service.

[0090] In detail, in an embodiment, the server computing system (130) can provide the asset portfolio prediction service by exchanging data necessary to enable the asset portfolio prediction service process to be run on an external device, such as a user computing device (110), with the external device.

[0091] In more detail, in an embodiment, a server computing system (130) may provide an environment in which an application can operate on a user computing device (110).

[0092] To this end, the server computing system (130) may include application programs, data and / or commands for the application to operate, and may transmit and receive various data based thereon with the external device.

[0093] Additionally, the server computing system (130) includes at least one processor (131) and memory (132).

[0094] Here, the processor (131) of the server computing system (130) may be composed of at least one or a plurality of processors electrically connected among a central processing unit (CPU), a graphics processing unit (GPU), application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors and / or other electrical units for performing functions.

[0095] In particular, depending on the embodiment, such a processor (131) may be configured based on a Field Programmable Gate Array (FPGA) implementation, which is a hardware technology for implementing a predetermined digital circuit, and / or an Application Specific Integrated Circuit (ASIC). A detailed description thereof is omitted in accordance with the description of the FPGA and ASIC described above.

[0096] And the memory (132) may include one or more non-transitory / transitory computer-readable storage media such as RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. This memory (132) may store data (133) and instructions (134) necessary for the processor (131) to perform functional operations such as training an artificial intelligence model or executing an asset portfolio prediction service using an artificial intelligence model.

[0097] In one embodiment, the server computing system (130) may be implemented to include at least one computing device. For example, the server computing system (130) may be implemented to operate multiple computing devices according to a sequential computing architecture, a parallel computing architecture, or a combination thereof. Additionally, the server computing system (130) may include multiple computing devices connected via a network (170).

[0098] Additionally, the server computing system (130) may store at least one machine learning model (140). For example, the server computing system (130) may include a neural network and / or other multi-layer nonlinear model as the machine learning model (140). Exemplary neural networks may include feedforward neural networks, deep neural networks, recurrent neural networks, and convolutional neural networks.

[0099] In an embodiment, the server computing system (130) may further include a data store computing system (hereinafter, “data store”), which is a storage for continuously storing and managing raw data that is the basis of the asset portfolio prediction service.

[0100] These data stores can include various forms of data storage, ranging from file systems to cloud storage. For example, a data store can include at least one of the following: a relational database that uses a structured query language (SQL) to define and manipulate data; a NoSQL database designed for flexibility and scalability to handle unstructured and semi-structured data; a data warehouse, which is a system used for reporting and data analysis, centralizes large amounts of data from multiple sources and is optimized for querying and analysis; a data warehouse that stores large amounts of raw data in its native formats of structured, semi-structured, and unstructured data; and a database on a local storage device or Network Attached Storage (NAS) that stores data in files in a format typically accessible by a computer operating system.

[0101] -Training Computing System (150)

[0102] The training computing system (150) includes at least one processor (151) and memory (152).

[0103] Here, the processor (151) of the training computing system (150) may be composed of at least one or a plurality of processors electrically connected among a central processing unit (CPU), a graphics processing unit (GPU), application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors and / or other electrical units for performing functions.

[0104] In particular, depending on the embodiment, such a processor (151) may be configured based on a Field Programmable Gate Array (FPGA) implementation, which is a hardware technology for implementing a predetermined digital circuit, and / or an Application Specific Integrated Circuit (ASIC). A detailed description thereof is omitted in accordance with the description of the FPGA and ASIC described above.

[0105] And the memory (152) may include one or more non-transitory / transitory computer-readable storage media such as RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. This memory (152) may store data (153) and instructions (154) necessary for the processor (151) to perform learning of an artificial intelligence model, etc.

[0106] For example, the training computing system (150) may include a model trainer (160) that trains a machine learning model (120 and / or 140) stored in the user computing device (110) and / or server computing system (130) using various training or learning techniques, such as backpropagation of errors (according to the framework illustrated in FIG. 3).

[0107] For example, such a model trainer (160) may perform updates for one or more parameters of a machine learning model (120 and / or 140) for an asset portfolio prediction service in a backpropagation manner based on a defined loss function.

[0108] In some implementations, performing backward propagation of errors may include performing truncated backpropagation through time. The model trainer (160) may perform a number of generalization techniques (e.g., weight reduction, dropout, and / or knowledge distillation) to improve the generalization ability of the trained machine learning model (120 and / or 140).

[0109] Additionally, the model trainer (160) may train the machine learning model (120 and / or 140) based on a series of training data (161). Here, the training data (161) may include data in different formats, such as, for example, images, audio samples, and / or text. Examples of image types that may be used may include video frames, LiDAR point clouds, X-ray images, computed tomography scans, hyperspectral images, and / or various other forms of images.

[0110] Such training data (161) may be provided by a user computing device (110) and / or a server computing system (130). When the training computing device trains a machine learning model (120 and / or 140) on specific data of the user computing device (110), the machine learning model (120 and / or 140) may be characterized as a personalized model.

[0111] And the model trainer (160) includes computer logic utilized to provide the desired function.

[0112] Additionally, the model trainer (160) may be implemented as hardware, firmware, and / or software that controls a general-purpose processor. In one implementation, the model trainer (160) includes a program file stored in a storage device, which may be loaded into a memory (152) and executed by one or more processors (151). In another implementation, the model trainer (160) includes one or more sets of computer-executable data (153) and instructions (154) stored in a tangible computer-readable storage medium, such as a RAM hard disk or an optical or magnetic medium.

[0113] The network (170) includes, but is not limited to, a 3rd Generation Partnership Project (3GPP) network, a Long Term Evolution (LTE) network, a World Interoperability for Microwave Access (WIMAX) network, the Internet, a Local Area Network (LAN), a Wireless Local Area Network (Wireless LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), a Bluetooth network, a satellite broadcasting network, an analog broadcasting network, and / or a Digital Multimedia Broadcasting (DMB) network.

[0114] In general, communication over a network (170) may be performed using any type of wired and / or wireless connection, using various communication protocols (e.g., TCP / IP, HTTP, SMTP, and / or FTP), encodings or formats (e.g., HTML and / or XML), and / or protection schemes (e.g., VPN, Secure HTTP, and / or SSL).

[0115] FIG. 2 illustrates an example block diagram of a computing device implementing an asset portfolio prediction service according to one embodiment of the present disclosure.

[0116] As shown in FIG. 2, the computing device (100) included in the user computing device (110), the server computing system (130), and the training computing system (150) includes a plurality of applications (e.g., Application 1 to Application N). Each application may include a machine learning library and one or more machine learning models. For example, the applications may include an image processing (e.g., Detection, Classification, and / or Segmentation, etc.) application, a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, and / or a chat-bot application.

[0117] In an embodiment, the computing device (100) may include a model trainer (160) for training an artificial intelligence model, and may store and operate the trained artificial intelligence model to provide output data according to predetermined input data.

[0118] Each application of the computing device (100) may communicate with a number of other components of the computing device (100), such as, for example, one or more sensors, a context manager, a device state component, and / or additional components. In one embodiment, each application may communicate with each device component using an API (e.g., a public API). In one embodiment, the API used by each application may be specific to that application.

[0119] FIG. 3 illustrates an example block diagram of another aspect of a computing device implementing an asset portfolio prediction service according to one embodiment of the present disclosure.

[0120] Referring to FIG. 3, a computing device (200) includes a plurality of applications (e.g., Application 1 to Application N). Each application may communicate with a central intelligence layer. For example, the applications may include an image processing application, a text messaging application, an email application, a dictation application, a virtual keyboard application, and / or a browser application. In one embodiment, each application may communicate with the central intelligence layer (and models stored therein) using an API (e.g., a common API across all applications).

[0121] The central intelligence layer may include multiple machine learning models. For example, as illustrated in FIG. 3, at least some of the machine learning models may be provided to each application and managed by the central intelligence layer. In other implementations, two or more applications may share a single machine learning model. For example, in some implementations, the central intelligence layer may provide a single model to all applications. In some implementations, the central intelligence layer may be included within the operating system of the computing device (200) or implemented differently.

[0122] The central intelligence layer may communicate with a central device data layer. The central device data layer may be a centralized data repository for the computing device (200). As illustrated in FIG. 3, the central device data layer may communicate with a number of other components of the computing device (200), such as, for example, one or more sensors, a context manager, a device state component, and / or additional components. In some implementations, the central device data layer may communicate with each device component using an API (e.g., a private API).

[0123] The techniques described herein may refer to servers, databases, software applications, and other computer-based systems, as well as actions taken and information transmitted to or from such systems. It will be appreciated that the inherent flexibility of computer-based systems allows for a wide range of possible configurations, combinations, and division of labor and functionality between and among components. For example, the processes described herein may be implemented using a single device or component, or multiple devices or components operating in combination. Databases and applications may be implemented on a single system or in a distributed system across multiple systems. Distributed components may operate sequentially or in parallel.

[0124]

[0125] [How to Predict Asset Portfolios]

[0126] Hereinafter, a method for predicting asset portfolio weights based on adjustable targets and multi-faceted data (i.e., a method for implementing an asset portfolio prediction service) by a computing system (1000) according to one embodiment of the present disclosure will be described in detail with reference to the attached drawings.

[0127] FIG. 4 illustrates an example of a flowchart for explaining an asset portfolio prediction method according to one embodiment of the present disclosure.

[0128] Referring to FIG. 4, an asset portfolio prediction method according to an embodiment of the present disclosure may include a step of determining a target prediction task (S101), a step of collecting and preprocessing data for performing the determined target prediction task (S103), a step of performing first model learning based on the collected and preprocessed data (S105), a step of generating a target label data set based on the collected and preprocessed data (S107), a step of performing second model learning for training a time series prediction model based on the generated target label data set (S109), a step of acquiring portfolio prediction information based on the learned time series prediction model (S111), a step of generating and providing portfolio rebalancing information based on the acquired portfolio prediction information (S113), a step of providing basis data for the provided portfolio rebalancing information (S115), and a step of analyzing the performance of the provided portfolio rebalancing information to perform performance evaluation and feedback of the time series prediction model (S117).

[0129] Specifically, a computing system (1000) according to one embodiment of the present disclosure can determine a target prediction task. (S101)

[0130] Here, the target prediction task according to the embodiment may mean a task of predicting the proportion (weight) of portfolio assets for a given first domain, as data specifying a target task to be performed.

[0131] In detail, in the embodiment, the computing system (1000) can obtain various data (hereinafter, task setting information) required to determine a target prediction task according to a predetermined method.

[0132] At this time, the task setting information may include a predetermined first domain, current portfolio asset ratio or investment amount, entire outlook period, forecast unit period (rebalancing period), asset pool, and / or investment tendency data.

[0133] Here, the investment propensity according to the embodiment may include a profit maximization propensity that allocates weight to assets with positive past cumulative returns based on the expected return of the portfolio according to [Mathematical Formula 1] below, a risk minimization propensity that allocates weight to assets with low past volatility based on the portfolio volatility according to [Mathematical Formula 2] below, and a Sharpe ratio optimization propensity that allocates weight in order of high Sharpe ratio for each asset based on the Sharpe ratio according to [Mathematical Formula 3] below.

[0134] [Mathematical Formula 1]

[0135]

[0136]

[0137] Here, [Equation 1] refers to the average portfolio return over a specific period, means the length of the trading period, refers to the portfolio return adjusted for transaction costs at a given point in time t, refers to the total number of assets included in the portfolio, refers to the weight of the i-th asset at time t-1 (day), means the realized arithmetic rate of return of the i-th asset from time t-1 to time t, may refer to a constant transaction cost rate (fee) (e.g., 0.002%).

[0138] [Equation 2]

[0139]

[0140] Here, [Equation 2] means the expected portfolio return (see Equation 1), can mean the expected value of the square of the portfolio return.

[0141] [Equation 3]

[0142]

[0143] Here, [Equation 3] refers to the Sharpe ratio, which can mean the value obtained by dividing the expected portfolio return according to [Mathematical Formula 1] by the portfolio volatility according to [Mathematical Formula 2].

[0144] In more detail, in an embodiment, the computing system (1000) may obtain user input specifying the task setting information described above by utilizing a user interface including a predetermined input template and / or an interaction model based on a Large Language Model (LLM).

[0145] At this time, according to an embodiment, the computing system (1000) can analyze a conversation context (Cotext) based on user input based on an interaction model, and determine task setting information, elements necessary for determining a target prediction task, and / or a purpose based on the analyzed context.

[0146] Additionally, according to an embodiment, the computing system (1000) may generate additional queries based on the analyzed context to obtain additionally required data (e.g., data specifying elements necessary for determining a target prediction task).

[0147] Additionally, in an embodiment, the computing system (1000) can determine the target prediction task described above based on the acquired user input.

[0148] For example, the computing system (1000) can determine a target prediction task that predicts a portfolio proportion (weight) by reflecting an investment tendency according to the task setting information during a prediction unit period (rebalancing period) according to the task setting information for a first domain obtained based on a predetermined user input.

[0149] In this way, in the embodiment, the computing system (1000) can determine a target prediction task based on user input obtained through various methods (for example, a method based on a user interface including a predetermined input template and / or an interaction model based on a large language model (LLM)), thereby configuring a target prediction task that accurately reflects the user's intention, thereby improving the efficiency and prediction accuracy of subsequent data collection and model learning stages.

[0150] Additionally, a computing system (1000) according to one embodiment of the present disclosure can perform data collection and preprocessing to perform a determined target prediction task. (S103)

[0151] That is, in the embodiment, the computing system (1000) can collect and preprocess various raw data required to perform a target prediction task.

[0152] Hereinafter, for the purpose of effective explanation, the entire outlook period will be referred to as T, a given point in time within the entire outlook period will be referred to as t, a past point in time p periods from the given point in time t will be referred to as (tp), and a future point in time h periods from the given point in time t will be referred to as (t+h).

[0153] At this time, the p period and / or the h period may be set in a predetermined manner (e.g., based on user input and / or a specific optimization algorithm, etc.) or may be set according to a prediction unit period (rebalancing period) according to a target prediction task.

[0154] Additionally, for effective explanation, each period will be explained based on date below.

[0155] FIG. 5 illustrates an example of a conceptual diagram of a portfolio prediction method based on a time series prediction model according to one embodiment of the present disclosure and an existing portfolio composition method.

[0156] In detail, referring to FIG. 5, in an embodiment, the computing system (1000) can collect raw data from time point (tp) to time point (t-1).

[0157] In more detail, the computing system (1000) can collect price time series data including high, low, close, and volume for each asset included in a preset asset pool.

[0158] Additionally, the computing system (1000) can collect macroeconomic indicator data including major macroeconomic indicators such as WTI crude oil prices, U.S. Treasury yields, interest rates, inflation rates, and GDP growth rates.

[0159] Additionally, the computing system (1000) can collect market indicator data including volatility index (VIX), overall market index, etc.

[0160] At this time, according to an embodiment, the computing system (1000) can detect a target influence variable, which is a variable related to the target prediction task.

[0161] For example, the computing system (1000) can detect a target influence variable that has a correlation with a first domain and / or asset, etc. exceeding a predetermined standard according to a target prediction task.

[0162] And the computing system (1000) can collect raw data (hereinafter, target influence variable data) corresponding to the detected target influence variable.

[0163] To this end, the computing system (1000) can collect target analysis reports by utilizing a large language model (LLM) or keyword search engine based on keywords representing a first domain according to a target prediction task.

[0164] In addition, the computing system (1000) can define the target influence variables described above at a semantic level from the collected target analysis data, and extract the precedence relationship and weight between the defined variables.

[0165] And the computing system (1000) can obtain target influence variable data from target analysis data by reflecting the extracted precedence relationship and weight.

[0166] Returning again, in another embodiment, the computing system (1000) can align the raw data collected as above (i.e., price time series data, macroeconomic indicator data, market indicator data and / or target impact variable data) along the same time axis.

[0167] At this time, according to the embodiment, the computing system (1000) can align raw data collected at various points in time along a specific time axis by applying a sliding window method.

[0168] Here, the sliding window method according to the embodiment may mean a process of extracting and sorting data included in each window while moving a time window of a fixed length at regular intervals.

[0169] Thus, in the embodiment, the computing system (1000) can generate raw feature data, which is data that aligns various raw data from time point (tp) to time point (t-1) along a consistent time series axis.

[0170] At this time, according to an embodiment, the computing system (1000) may also generate low feature data in which the above-described task setting information is aligned along the time series axis.

[0171] In this way, the computing system (1000) in the embodiment can effectively support the resolution of numerical instability and past prediction error accumulation problems of covariance matrix inversion that occur in the traditional MVO (Mean-Variance Optimization) method by securing raw feature data that lists internal and external data from various angles according to a consistent time series, thereby providing objective and stable learning data that improves the efficiency and prediction accuracy of subsequent model learning.

[0172] In addition, referring to FIG. 5, a computing system (1000) according to one embodiment of the present disclosure can perform first model learning based on collected and preprocessed data. (S105)

[0173] In detail, in an embodiment, the computing system (1000) can perform first model learning to learn a predetermined time series prediction model using collected and preprocessed raw feature data and task setting information corresponding to the target prediction task described above (hereinafter, first model learning data set).

[0174] Here, the time series prediction model according to the embodiment may include a deep learning model (hereinafter, TFT model) based on a temporal fusion transformer (TFT).

[0175] At this time, unlike existing prediction models (e.g., traditional mean-variance optimization-based models that are not transformer-based, pre-trained statistical regression models, or machine learning models that can only process limited input features), the TFT model can simultaneously accept and process various types of categorical and continuous inputs, such as a given portfolio asset ratio, daily returns by asset, macroeconomic indicators, market indicators, and / or target impact variables, and can effectively encode past observations with variable lengths while automatically selecting important time points and features within the time series through a self-attention mechanism (feature selection).

[0176] Based on this, the time series prediction model according to the embodiment may be a deep learning model based on a time series fusion transformer that uses various types of data as input features as described above and outputs a time series of portfolio asset proportions at a predetermined future point in time according to the input features.

[0177] On the other hand, the existing prediction model according to the embodiment may be a deep learning model that is not transformer-based but is a pre-trained machine learning model and / or statistical regression model, and has the limited characteristic of processing only a small number of input features determined in advance and operating only under fixed parameter settings.

[0178] That is, the time series prediction model in the embodiment of the present disclosure may be a model capable of processing much richer and more diverse input features than existing prediction models, and thus may be a model capable of providing more precise and reliable portfolio asset ratio prediction results.

[0179] Returning again, in more detail, in the embodiment, the computing system (1000) can input time series interval data from time point (tp) to time point t among the low feature data included in the first model learning data set into the time series prediction model.

[0180] As an example, the computing system (1000) can select multiple learning samples separated by a predetermined time interval (e.g., a rebalancing period) from the entire raw feature data through a windowing method, and sequentially input each selected sample into a time series prediction model.

[0181] At this time, each of the input samples described above may include a predetermined portfolio asset ratio, daily returns by asset, macroeconomic indicators, market indicators, target influence variables, and / or task setting information, and may be aligned along a consistent time series axis and provided as an encoder of a time series prediction model.

[0182] In addition, in the embodiment, the computing system (1000) can predict portfolio asset weight time series data for the time series period from time (t+1) to time (t+h) through a time series prediction model that receives time series period data (hereinafter, past input features) from time (tp) to time t.

[0183] At this time, the time series prediction model in the embodiment can derive a highly reliable prediction value for each prediction unit period (rebalancing period) by considering a predetermined future point in time step by step through a deep learning algorithm including a self-attention mechanism (e.g., multi-head self-attention, etc.) and a gated skip connection, etc.

[0184] Here, the prediction results of the time series prediction model can be output in the form of a vector representing the proportion of each portfolio asset, and can be arranged by date along the time series axis.

[0185] Additionally, in the embodiment, the computing system (1000) can perform 3) normalization post-processing on the predicted portfolio asset weight time series data.

[0186] In detail, in an embodiment, the computing system (1000) can perform a normalization operation so that the sum of the weights of each asset of each portfolio output by the time series prediction model becomes '1'.

[0187] For example, the computing system (1000) predicts the portfolio asset weight predicted by the time series forecasting model for a specific date k. ) can be normalized according to the following [Mathematical Formula 4].

[0188] [Equation 4]

[0189]

[0190] Through this, in the embodiment, the computing system (1000) can provide result data that satisfies the fully invested condition in which the entire portfolio asset group can be invested completely and reasonably.

[0191] Additionally, in the embodiment, the computing system (1000) can perform 4) comparison between predicted and actual data.

[0192] That is, the computing system (1000) can compare the output values ​​from the time point (t+1) to (t+h), which is the prediction interval, with the actual portfolio asset weight time series from the time point (t+1) to (t+h) corresponding to the prediction interval in the first model learning data set on a 1:1 basis.

[0193] Additionally, in the embodiment, the computing system (1000) can perform 5) loss calculation and performance evaluation according to the comparison result.

[0194] As an example, the computing system (1000) can calculate the difference between the predicted value of a time series prediction model and the actual value using a defined loss function such as the mean squared error (MSE).

[0195] And the computing system (1000) can evaluate the current performance of the time series prediction model based on the calculated difference value.

[0196] At this time, the computing system (1000) can monitor model performance during the entire epoch by aggregating the loss values ​​calculated for each learning sample.

[0197] Additionally, in the embodiment, the computing system (1000) can perform 6) model update according to the calculated loss value.

[0198] In detail, in an embodiment, the computing system (1000) may update the parameters of the time series prediction model by utilizing a predetermined optimization technique (e.g., Adam optimizer, etc.) to minimize the loss calculated as above.

[0199] According to an embodiment, the computing system (1000) repeats the above process in epoch units, and when a predetermined condition (e.g., an early stopping condition, etc.) is satisfied, the learning can be terminated at the corresponding optimal point and the optimal model parameters can be determined.

[0200] Thus, in the embodiment, the computing system (1000) can perform first model learning to learn a time series prediction model using a first model learning data set including raw feature data and task setting information.

[0201] In this way, in the embodiment, the computing system (1000) can perform first model learning by predicting a portfolio asset proportion (weight) time series for a period (t+1) to (t+h) based on data for a predetermined period (tp) to t through a time series prediction model, and updating the corresponding time series prediction model to minimize the difference between the predicted value for the period (t+1) to (t+h) and the actual value (i.e., the portfolio asset proportion (weight) time series corresponding to the period (t+1) to (t+h) within the first learning data set), thereby training the output of the time series prediction model to follow the correct answer label according to the first model learning data set.

[0202] Through this, the computing system (1000) can eliminate numerical instability caused by the covariance inverse matrix operation that occurs during the prediction process of an existing prediction model (e.g., a model based on a traditional mean-variance optimization method, etc.), minimize the accumulation of prediction errors, and implement a portfolio weight (weight) prediction model that can immediately respond to market changes by flexibly integrating various external indicators.

[0203] In addition, in the embodiment, the computing system (1000) can implement a time series prediction model that achieves a desired goal (in the embodiment, profit maximization, risk minimization, and / or Sharpe ratio optimization, etc.) while maintaining stability and scalability not only for small-scale assets by sector but also for large-scale asset groups such as the S&P 500 through the learning structure described above.

[0204] Therefore, the computing system (1000) can overcome the limitations of existing optimization methods and provide a more reliable dynamic asset allocation solution in an actual investment environment.

[0205] In addition, a computing system (1000) according to one embodiment of the present disclosure can generate a target label data set based on collected and preprocessed data. (S107)

[0206] Here, the correct answer label data set according to the embodiment may mean data predicted using an existing prediction model for the portfolio asset weight (weight) for each day for a predetermined prediction period (in the embodiment, the period (t+1) to (t+h)).

[0207] That is, in the embodiment, the correct label data set may be data generated using an existing prediction model, which contains the proportion (weight) of each portfolio asset corresponding to each date included in the prediction period of the time series prediction model.

[0208] In an embodiment, the correct answer label data set as above can be used as target (correct answer) data in a second model learning based on a time series prediction model.

[0209] In detail, in an embodiment, the computing system (1000) can 1) calculate a summary index for each asset using raw data collected for a predetermined past period (in an embodiment, a period (tp) to t).

[0210] Here, the raw data according to the embodiment may refer to basic data collected to calculate the performance and risk level of portfolio assets, such as price time series data by asset (e.g., high price, low price, closing price, etc.), trading volume data, and / or volatility index (VIX).

[0211] In an embodiment, the computing system (1000) may obtain the above-described source data in various ways, such as extracting the source data from the above-described raw data (or raw feature data) or collecting the source data through linkage with a predetermined external server, etc.

[0212] In addition, the summary indicator according to the embodiment may mean a value including, for example, the average return of an asset, the volatility (variance) of the return, the correlation (covariance) between assets, and / or the Sharpe ratio indicating the profitability relative to risk, as a representative statistical indicator calculated based on the source data as described above.

[0213] As an example, the computing system (1000) can use a preset calculation algorithm for each summary indicator to calculate each summary indicator according to the collected source data from time point (tp) to time point t.

[0214] For example, the computing system (1000) calculates the daily return for each asset based on the price time series data for each asset, and then calculates the cumulative return, average return, standard deviation (volatility) and / or correlation coefficient between assets for a certain period of time based on this, and can derive the Sharpe ratio indicating the profitability relative to risk for each asset by utilizing the calculated data values.

[0215] In addition, in the embodiment, the computing system (1000) can 2) generate a purpose-specific correct answer label data set according to the calculated summary index.

[0216] Here, the purpose-specific correct answer label data set according to the embodiment may mean a correct answer label data set created by matching each purpose using an existing prediction model optimized for a specific objective purpose (for example, investment tendency and / or market conditions, etc.).

[0217] That is, in the embodiment, the computing system (1000) can use an existing prediction model with high prediction accuracy for each specific investment tendency (e.g., profit maximization, risk minimization, Sharpe ratio optimization, etc.) and / or market condition (e.g., bull market, bear market, adjustment market, etc.) to generate a set of correct answer labels optimized for each purpose (i.e., a purpose-specific correct answer label data set).

[0218] In detail, in an embodiment, the computing system (1000) can detect a summary indicator corresponding to a preset input feature for each existing prediction model, and input the detected summary indicator as an input feature to the corresponding existing prediction model.

[0219] For example, the computing system (1000) may input an input feature (e.g., a predetermined first summary indicator, etc.) corresponding to an existing prediction model, in which the prediction accuracy according to the first investment tendency (e.g., profit maximization) is optimized to a predetermined standard or higher, to generate a first correct answer label data set customized for the first investment tendency.

[0220] As another example, the computing system (1000) may input an input feature (e.g., a predetermined second summary index, etc.) corresponding to an existing prediction model, in which the prediction accuracy according to the second investment tendency (e.g., risk minimization) is optimized to a predetermined standard or higher, to generate a second correct answer label data set customized for the second investment tendency.

[0221] As another example, the computing system (1000) can input an input feature (e.g., a predetermined third summary index, etc.) corresponding to an existing prediction model whose prediction accuracy according to a third investment propensity (e.g., Sharpe ratio optimization) is optimized to a predetermined standard or higher, thereby generating a third correct answer label data set customized for the third investment propensity.

[0222] As another example, the computing system (1000) can input an input feature (e.g., a predetermined fourth summary indicator, etc.) corresponding to an existing prediction model whose prediction accuracy according to a first market situation (e.g., a bull market) is optimized to a predetermined standard or higher, thereby generating a fourth correct answer label data set customized for the first market situation.

[0223] As another example, the computing system (1000) can input an input feature (e.g., a predetermined fifth summary indicator, etc.) corresponding to an existing prediction model whose prediction accuracy according to a second market situation (e.g., a bear market) is optimized to a predetermined standard or higher, thereby generating a fifth correct answer label data set customized for the second market situation.

[0224] As another example, the computing system (1000) can input an input feature (e.g., a predetermined sixth summary index, etc.) corresponding to an existing prediction model whose prediction accuracy according to a third market situation (e.g., a market adjustment) is optimized to a predetermined standard or higher, thereby generating a sixth correct answer label data set customized for the third market situation.

[0225] According to an embodiment, the computing system (1000) may select one investment tendency among various investment tendencies and one market situation among various market situations, detect an existing prediction model optimized for both the selected investment tendency and the market situation, and input a predetermined input feature into the detected existing prediction model to generate a correct answer label data set optimized for the investment tendency and the market situation.

[0226] In addition, according to an embodiment, the computing system (1000) may determine an existing prediction model that will generate a correct answer label data set based on investment propensity data and / or market indicator data according to low feature data according to task setting information, and extract input features (in the embodiment, predetermined summary indicators) that match the determined existing prediction model and input them into the corresponding existing prediction model to generate a correct answer label data set suitable for user needs.

[0227] At this time, in the embodiment, the computing system (1000) can consistently connect the correct label data set output by the existing prediction model and the asset information by mapping each index of the summary indicator vector to a specific asset one-to-one according to the asset pool information defined in advance.

[0228] In this way, in the embodiment, the computing system (1000) can generate a correct answer label data set that matches various investment tendencies and / or market conditions using given data, and selectively utilize this when learning a time series prediction model later, thereby more easily and flexibly supporting learning a time series prediction model to implement a prediction optimized for the user's investment goal and / or market environment.

[0229] Additionally, the computing system (1000) according to one embodiment of the present disclosure can perform second model learning to train a time series prediction model based on the generated correct answer label data set. (S109)

[0230] In detail, in an embodiment, the computing system (1000) can perform second model learning to optimize a time series prediction model for a specific purpose by using the purpose-specific correct answer label data set generated as described above (i.e., an optimal correct answer label data set generated for each purpose using an existing prediction model with high prediction accuracy for a specific investment tendency (e.g., profit maximization, risk minimization, Sharpe ratio optimization, etc.) and / or market condition (e.g., bull market, bear market, adjustment market, etc.)).

[0231] In detail, in an embodiment, the computing system (1000) can 1) determine a correct label data set corresponding to the user's task purpose based on task setting information and / or raw feature data.

[0232] Here, the user's task objective according to the embodiment can be defined as a task of predicting the portfolio asset ratio (weight) up to a certain point in time based on the investment tendency, market conditions, and / or domain that matches the user's needs.

[0233] In more detail, in an embodiment, the computing system (1000) can detect a correct label data set (hereinafter, a target label data set) corresponding to a task purpose desired by a user from a correct label data set generated for a specific purpose (i.e., a purpose-specific correct label data set) based on the above-described task setting information and / or raw feature data.

[0234] Specifically, in the embodiment, the computing system (1000) can determine the user's task purpose by analyzing the investment propensity data and / or market indicator data included in the task setting information and / or low feature data.

[0235] For example, the computing system (1000) can determine the task purpose of a user who wants to predict the proportion (weight) of portfolio assets for a given first domain up to a specific point in time (for example, a rebalancing point in time) under a condition where the user's investment tendency is 'profit maximization' and the market situation is 'rising'.

[0236] Additionally, in the embodiment, the computing system (1000) can detect a correct label data set that most closely matches the determined user's task purpose and a predetermined standard.

[0237] For example, the computing system (1000) can detect a correct label data set generated through an existing prediction model optimized for 'profit maximization-rising market' as a correct label data set that best matches the task purpose of the corresponding user.

[0238] And in the embodiment, the computing system (1000) can determine the correct label data set detected as above as the target label data set.

[0239] At this time, according to the embodiment, the computing system (1000) may, through an interactive user interface and / or prompt-based question-and-answer, confirm whether the determined target label data set matches the user's actual task purpose and may also support modifying or re-determining the target label data set as necessary.

[0240] Additionally, in the embodiment, the computing system (1000) can perform 2) a comparison between the determined correct label data set (i.e., the target label data set) and the predicted data of the time series prediction model.

[0241] In detail, in the embodiment, the computing system (1000) can compare the daily portfolio asset weight data from time (t+1) to time (t+h) included in the target label data set determined as above (i.e., the correct label data set optimized for learning the user's task purpose) with the daily portfolio asset weight data from time (t+1) to time (t+h) predicted by the time series prediction model (i.e., the predicted data of the time series prediction model) by corresponding them 1:1 for each matching date.

[0242] At this time, the prediction data of the above time series prediction model may be data to which normalization post-processing (see the description of step S105) has been applied to the prediction value of the time series prediction model.

[0243] Additionally, in the embodiment, the computing system (1000) can perform 3) loss calculation and performance evaluation according to the comparison result.

[0244] As an example, the computing system (1000) can quantitatively calculate the difference between the predicted data of a time series prediction model and a target label data set using a defined loss function such as the mean squared error (MSE).

[0245] And the computing system (1000) can evaluate the current performance of the time series prediction model based on the calculated difference value.

[0246] At this time, the computing system (1000) can monitor model performance during the entire epoch by aggregating the loss values ​​calculated for each learning sample.

[0247] Additionally, in the embodiment, the computing system (1000) can perform 4) model update according to the calculated loss value.

[0248] In detail, in an embodiment, the computing system (1000) can perform second model learning to train the predicted value of the time series prediction model to follow the target label data set based on the difference value calculated as above.

[0249] Specifically, in an embodiment, the computing system (1000) may update internal parameters of a time series prediction model in a direction that minimizes the calculated loss value using a predetermined optimization technique (for example, Adam optimizer, RMSprop, etc.).

[0250] At this time, according to an embodiment, the computing system (1000) may repeatedly perform learning in epoch units until the corresponding loss value converges below a certain standard or a pre-set early stopping condition is satisfied.

[0251] Thus, in the embodiment, the computing system (1000) can perform second model learning to update the parameters of a time series prediction model based on a target label data set that matches a specific purpose, such as a user's investment goal and / or market conditions, so as to train the model to exhibit prediction performance optimized for the specific purpose.

[0252] In this way, in the embodiment, the computing system (1000) can additionally train a time series prediction model by utilizing a target label data set that matches the user's task purpose among the correct label data sets generated in advance for various purposes.

[0253] Through this, the computing system (1000) can provide a time series prediction model that performs a customized portfolio weight prediction that more immediately adapts to the user's investment strategy and / or market conditions and further improves the prediction reliability.

[0254] Additionally, a computing system (1000) according to one embodiment of the present disclosure can obtain portfolio prediction information based on a learned time series prediction model. (S111)

[0255] FIG. 6 illustrates an example of the results of predicting various sector-specific portfolio weights of a time series prediction update model according to one embodiment of the present disclosure.

[0256] Referring to FIG. 6, the portfolio prediction information according to the embodiment herein may mean the weight (weight) information for each portfolio asset for the first domain estimated (predicted) by inputting predetermined input data (e.g., various types of categorical and continuous input features such as predetermined portfolio asset ratios, daily returns by asset, macroeconomic indicators, market indicators, and / or target influence variables) into the time series prediction model (hereinafter, time series prediction update model) learned as described above, and performing the target prediction task described above.

[0257] At this time, the portfolio prediction information according to the embodiment may be information including time-series data on the portfolio asset proportion (weight) for each day for a predetermined prediction period (in the embodiment, the period (t+1) to (t+h)).

[0258] Additionally, the portfolio prediction information according to the embodiment may include data on the portfolio asset weight (weight) for each day with normalization post-processing (see the description of step S105) applied.

[0259] In detail, in an embodiment, the computing system (1000) can 1) input predetermined input data into a time series prediction update model.

[0260] In more detail, in an embodiment, the computing system (1000) can obtain predetermined input data based on user input and / or a preset algorithm.

[0261] As an example, the computing system (1000) can obtain various raw data corresponding to a predetermined past period (in the example, the period (tp) to t) as the above-described input data.

[0262] And in the embodiment, the computing system (1000) can input the acquired input data into a time series prediction update model.

[0263] In addition, in the embodiment, the computing system (1000) can 2) predict the portfolio asset weight time series data for the time series section up to a predetermined future point in time through a time series prediction update model that receives input data.

[0264] In detail, in an embodiment, the computing system (1000) can time-series predict portfolio asset weight data for each date included in a period up to a predetermined future point in time (in an embodiment, a point in time (t+h) that is a point in the future by a prediction unit period (rebalancing period)) through the time-series prediction update model described above.

[0265] That is, the computing system (1000) can obtain time series data of the proportion (weight) of each portfolio asset arranged by date along the time series axis for the period (t+1) to (t+h) through a time series prediction update model.

[0266] At this time, in the embodiment, the computing system (1000) can perform a prediction process based on a time series prediction update model by reflecting model parameters optimized according to first model learning and / or second model learning.

[0267] In this way, referring further to FIG. 5, unlike existing methodologies that estimate a return-risk structure based on predetermined input features (e.g., past returns, etc.) and calculate a portfolio asset ratio based on the estimated return-risk structure information, the computing system (1000) in the embodiment can directly predict and provide a portfolio asset ratio based on various types of input data using a time series prediction update model according to an embodiment of the present disclosure.

[0268] That is, the computing system (1000) outputs a portfolio asset ratio (weight) that can be rebalanced directly without an intermediate step as in the past through a time series prediction update model, thereby reducing numerical instability occurring in conventional covariance matrix estimation and / or inverse matrix calculation, and also minimizing the error accumulation problem.

[0269] Additionally, in the embodiment, the computing system (1000) can perform 3) normalization post-processing on the predicted portfolio asset weight time series data.

[0270] In detail, in an embodiment, the computing system (1000) can perform a normalization operation so that the sum of the weights of each asset in each portfolio predicted by the time series prediction model becomes '1'.

[0271] In the embodiment, a description of a specific method by which the computing system (1000) performs normalization post-processing is omitted by applying the description in step S105 described above.

[0272] Thus, in the embodiment, the computing system (1000) performs a target prediction task according to the user's task purpose (in the embodiment, investment tendency, market situation, and / or domain, etc.) based on predetermined input data, and obtains portfolio prediction information, which is the result data of predicting the daily portfolio asset ratio (weight) up to a specific future point in time for a predetermined first domain.

[0273] At this time, in the embodiment, the computing system (1000) can repeatedly perform the above-described step S111 in cycles of a prediction unit period (rebalancing period) according to preset task setting information.

[0274] In this way, in the embodiment, the computing system (1000) can continuously produce and provide portfolio prediction information suitable for the user's investment goals and market changes by reflecting various input data through a learned time series prediction model (i.e., a time series update model).

[0275] Accordingly, the computing system (1000) can support dynamic asset allocation more stably while flexibly responding to the user's investment goals and market changes, while eliminating the numerical instability of the optimization process through the existing prediction model.

[0276] Additionally, a computing system (1000) according to one embodiment of the present disclosure can generate and provide portfolio rebalancing information based on acquired portfolio prediction information. (S113)

[0277] Here, the portfolio rebalancing information according to the embodiment may mean information that integrates at least one daily portfolio asset weight (weight) data included in the portfolio forecast information into one according to a predetermined method.

[0278] That is, the portfolio rebalancing information according to the embodiment may be data that is reconstructed into a single portfolio asset proportion (weight) data representing the entire predetermined prediction unit period (rebalancing period) by processing at least one daily portfolio asset proportion (weight) data in the portfolio prediction information in a predetermined manner.

[0279] Accordingly, portfolio rebalancing information can provide optimal asset-specific weight data predicted through a time series forecast update model by considering a specific purpose (e.g., profit maximization, risk minimization, and / or Sharpe ratio optimization) at a specific point in time (in an embodiment, point (t+h) after the rebalancing period has elapsed).

[0280] At this time, portfolio rebalancing information according to the embodiment may be generated and provided for each preset prediction unit period (rebalancing period).

[0281] In detail, in an embodiment, the computing system (1000) may generate portfolio rebalancing information by applying a preset rule to at least one daily portfolio asset weight data included in the portfolio forecast information in conjunction with its own processor and / or a separate processing module.

[0282] As an example, the computing system (1000) can generate portfolio rebalancing information by applying a predetermined operation to the weight values ​​of each asset in the daily portfolio asset weight (weight) data predicted for a predetermined prediction period (rebalancing period).

[0283] Specifically, in an embodiment, the computing system (1000) can perform a predetermined operation based on the weight value for each asset in the predicted daily portfolio asset weight (weight) data for a predetermined prediction period (rebalancing period).

[0284] For example, the computing system (1000) can perform an average operation based on the weight value of each asset in the daily portfolio asset weight (weight) data.

[0285] Additionally, in the embodiment, the computing system (1000) can determine the integrated weight (weight) for each asset based on the result of the performed operation.

[0286] For example, the computing system (1000) can calculate an average of the weight values ​​for each asset according to the average operation performed above, and can determine the calculated average as the final integrated weight (weight) for each asset.

[0287] Additionally, in the embodiment, the computing system (1000) can perform a normalization operation so that the sum of the determined integrated weights (weights) for each asset becomes '1'.

[0288] For example, the computing system (1000) can perform normalization post-processing so that the sum of the integrated weight value of the first asset, the integrated weight value of the second asset, and the integrated weight value of the third asset becomes '1'.

[0289] At this time, a description of a specific method by which the computing system (1000) performs normalization post-processing in the embodiment is omitted by applying the description of step S105 described above.

[0290] Thus, in the embodiment, the computing system (1000) can generate portfolio rebalancing information by applying a predetermined operation to the weight values ​​of each asset in the predicted daily portfolio asset weight (weight) data for a predetermined prediction period (rebalancing period).

[0291] In another embodiment, the computing system (1000) may generate portfolio rebalancing information based on whether 2) a predetermined criterion defined in advance is met.

[0292] In detail, in an embodiment, the computing system (1000) can select data that satisfies a preset condition (e.g., portfolio asset weight data corresponding to a date with the highest or lowest weight of a specific asset, etc.) from among at least one daily portfolio asset weight data in the portfolio prediction information.

[0293] And in the embodiment, the computing system (1000) can set the selected portfolio asset weight data as portfolio rebalancing information.

[0294] That is, in the embodiment, the computing system (1000) can detect data that satisfies a predetermined condition among the daily portfolio asset weight (weight) data predicted for a predetermined prediction period (rebalancing period), and utilize the detected portfolio asset weight (weight) data as single portfolio asset weight (weight) data representing the rebalancing period (i.e., portfolio rebalancing information).

[0295] Returning to the above, the computing system (1000) that generated the portfolio rebalancing information in the embodiment can provide the generated portfolio rebalancing information in a predetermined manner (e.g., display output and / or linkage with an external application service, etc.).

[0296] As an embodiment, the computing system (1000) can integrate the above-described portfolio rebalancing information into the user interface of a robo-advisor service to recommend and provide personalized portfolio rebalancing information based on predetermined portfolio asset weight data.

[0297] In this case, the computing system (1000) can generate a differentiated recommendation message including a rebalancing recommendation time and a trading ratio for each asset by referencing profile data on the user's risk preference, investment period, transaction cost, etc., and deliver the message to the user in the form of a dashboard or mobile notification.

[0298] This allows users to intuitively understand and immediately implement a rebalancing strategy that is optimal for them.

[0299] In another embodiment, the computing system (1000) may, in conjunction with a predetermined fund management support system, obtain a What-if scenario that assumes a change in a specific macroeconomic indicator (e.g., GDP growth rate and / or interest rate fluctuation) or market indicator (e.g., S&P 500 fluctuation rate) through user input.

[0300] In addition, the computing system (1000) re-executes the time series prediction update model based on the acquired scenario parameters to newly generate changed portfolio asset weight data and corresponding basis data, and transmits this to the fund operation support system in real time, thereby enabling the operation manager to immediately verify the rebalancing effect in a virtual environment and reflect it in decision-making.

[0301] In another embodiment, the computing system (1000) may be configured to include a specific investment theme (e.g., green energy and / or healthcare, etc.) and / or a factor (e.g., low volatility and / or high dividend, etc.) when setting up a target prediction task.

[0302] At this time, the computing system (1000) can use predetermined portfolio asset weight data as input for determining the constituent portfolio of a new listed index fund (ETF) product.

[0303] Specifically, the computing system (1000) calculates the inclusion ratio of each ETF constituent asset based on the predicted weight information and transmits it to the ETF design module to be used as a composition guideline for a new index product.

[0304] Through this configuration process, the computing system (1000) can automatically design and propose new ETF products optimized for market trends and user investment themes.

[0305] Returning again, in this embodiment, the computing system (1000) processes the daily portfolio weight data predicted through the time series prediction update model into single portfolio asset weight data at each predetermined rebalancing cycle and provides it, thereby suggesting the optimal asset weight (weight) at the point in time when rebalancing is to be performed by considering the change and flow of the predicted portfolio asset weight for the period from the present point in time to the specific point in time when the portfolio asset weight is to be predicted (i.e., the rebalancing point in time).

[0306] That is, in the embodiment, the computing system (1000) can provide a customized portfolio asset ratio that effectively responds to situations such as unexpected market risks by supporting consistent and stable asset allocation while comprehensively considering the fluctuation flow of the predicted portfolio asset ratio.

[0307] Thus, in the embodiment, the computing system (1000) can further improve the usability and utility of the portfolio asset ratio (weight) estimated through the time series prediction update model in the actual asset management stage.

[0308] At this time, depending on the embodiment, the computing system (1000) may optionally apply additional clipping and / or smoothing operations to generate portfolio rebalancing information when there are constraints on a predetermined transaction cost and / or minimum and maximum weight of assets.

[0309] As an example, the computing system (1000) may perform a clipping operation to adjust the weight value of each asset within a preset allowable range to prevent the generated portfolio rebalancing information from being biased below a predefined minimum weight or above a predefined maximum weight for each asset.

[0310] Specifically, in the embodiment, the computing system (1000) may apply post-processing to truncate the value to the minimum or maximum value, respectively, if the calculated integrated weight per asset is smaller than the minimum value or larger than the maximum value, and normalize the value again so that the sum of the adjusted weights becomes '1'.

[0311] Through this, in the embodiment, the computing system (1000) can prevent excessive concentration or insignificant allocation of weight to a specific asset, and effectively reduce risk concentration and liquidity problems during actual portfolio management.

[0312] Additionally, in the embodiment, the computing system (1000) may optionally apply a smoothing operation to mitigate rapid fluctuations in the weight values ​​by asset or excessively sensitive reflection of rebalancing information according to specific date data.

[0313] For example, in the embodiment, the computing system (1000) can gently adjust the sudden change in weight by utilizing a moving average or exponential smoothing technique based on the weight data by asset during the most recent rebalancing period.

[0314] Accordingly, in the embodiment, the computing system (1000) can support minimizing transaction costs (turnover) and executing a stable portfolio rebalancing strategy.

[0315] Effect (In this way, in the embodiment, the computing system (1000) can provide safe and efficient portfolio rebalancing information that is more suitable for the actual investment operation environment by providing portfolio rebalancing information through additional data processing that further considers certain transaction costs and / or minimum and maximum weight constraints.

[0316] Additionally, a computing system (1000) according to one embodiment of the present disclosure can provide basis data for the provided portfolio rebalancing information. (S115)

[0317] Here, the basis data according to the embodiment may mean data explaining how the final portfolio asset weights (weights) calculated according to a cycle according to a predetermined prediction period (i.e., rebalancing period) were determined based on causal relationships and data flows.

[0318] For example, such supporting data may include causal associations with key input variables, market influencers, and / or target influence variables.

[0319] Specifically, in an embodiment, the computing system (1000) can generate a causal relationship graph at the feature level based on semantic causal relationship information and a structured data set for key features (e.g., price time series data, macro indicators, market indicators, and / or target influence variables) that have influenced portfolio rebalancing information.

[0320] In addition, in the embodiment, the computing system (1000) can interpret, through a predetermined causal relationship discovery model (Data-driven Causal Discovery), which variables have influenced and to what extent the portfolio rebalancing information produced at the present time is based on data collected at a past point in time and the causal flow between the data.

[0321] For example, the computing system (1000) can derive the influence of external macro indicators such as price fluctuations by major assets, interest rates, oil prices, and / or volatility index (VIX) on determining the weight of a specific asset group during a rebalancing period, along with weights, and can utilize this as basis data.

[0322] Additionally, in the embodiment, the computing system (1000) can provide the contribution and relative influence of each variable in numerical form along with the derived causal relationship graph.

[0323] Therefore, the computing system (1000) can support intuitive confirmation of the reason for producing portfolio rebalancing information and the flow of decision-making, rather than simply providing numerical suggestions.

[0324] Additionally, according to an embodiment, the computing system (1000) may provide a user input interface that can adjust the values ​​of key variables (e.g., interest rate hike scenarios and / or increased volatility in specific sectors, etc.) to perform what-if simulations based on provided basis data.

[0325] At this time, in the embodiment, the computing system (1000) updates a structured dataset based on a hypothetical scenario input by a user, and then re-executes the same causal relationship interpretation process as described above, thereby producing and providing portfolio rebalancing information and new basis data for the same in a changed environment.

[0326] In this way, in the embodiment, the computing system (1000) can increase the reliability of the provided portfolio rebalancing information and enhance user understanding and persuasiveness by providing not only simple predicted values ​​but also data-based causal interpretation and supporting information.

[0327] Additionally, in the embodiment, the computing system (1000) may support more sophisticated decision-making by providing additional functionality to perform predictive environment change simulations as needed.

[0328] Additionally, a computing system (1000) according to one embodiment of the present disclosure can perform performance evaluation and feedback of a time series prediction model by analyzing the performance of the provided portfolio rebalancing information. (S117)

[0329] In detail, in the embodiment, the computing system (1000) can monitor actual investment performance according to the provided portfolio rebalancing information and evaluate the performance according to various indicators.

[0330] As an example, the computing system (1000) can analyze and evaluate the performance of actual investments based on portfolio rebalancing information using indicators such as cumulative return, Sharpe ratio, max drawdown, and / or risk indicators (e.g., standard deviation).

[0331] At this time, in the embodiment, the computing system (1000) can calculate a performance (evaluation) index based on portfolio rebalancing information at each predetermined rebalancing point (cycle), and perform performance analysis and evaluation according to the calculated index.

[0332] According to an embodiment, the computing system (1000) may flexibly adjust the performance monitoring cycle according to a predetermined investment purpose, strategy, and / or user needs.

[0333] In addition, according to an embodiment, the computing system (1000) can provide the performance (evaluation) indicators calculated as above by visualizing them in a predetermined manner (e.g., in the form of a dashboard, etc.).

[0334] Additionally, in the embodiment, the computing system (1000) may perform a feedback process for adjusting parameters and / or input feature configuration of the time series prediction update model based on the performance (evaluation) indicators produced as described above.

[0335] In an embodiment, the computing system (1000) can compare and analyze the difference between the daily portfolio asset weight predicted through the time series prediction update model and the actual optimized weight (e.g., the portfolio asset weight obtained through the existing prediction model, etc.) for a predetermined task period after the first model learning and / or the second model learning are completed.

[0336] And the computing system (1000) can evaluate the performance of the time series prediction update model using analyzed data and predetermined performance (evaluation) indicators (e.g., cumulative return, Sharpe ratio, maximum drawdown and / or risk indicator, etc.).

[0337] At this time, in the embodiment, the computing system (1000) can support, when the performance evaluation result satisfies a predetermined standard (e.g., a specific performance (evaluation) indicator value is higher than a preset value, etc.), deploy the time series prediction update model to the production environment and reflect the portfolio asset weight (i.e., portfolio rebalancing information) calculated for each rebalancing cycle to the actual investment operation environment.

[0338] On the other hand, in the embodiment, if the performance evaluation result does not meet a predetermined standard (e.g., a specific performance (evaluation) indicator value is higher than a preset value, etc.), the computing system (1000) can automatically modify the parameters of the time series prediction update model or perform additional learning according to user settings and / or preset data processing algorithms.

[0339] In this way, in the embodiment, the computing system (1000) continuously monitors and evaluates the performance of the time series prediction update model, and performs a feedback process based on the evaluation results, thereby continuously improving the prediction stability and suitability for purpose of the time series prediction update model (i.e., a model that outputs time series data of the proportion (weight) of each portfolio asset considering various external indicators (variables), thereby continuously improving the effectiveness and profitability of the rebalancing strategy based thereon (see FIG. 7).

[0340] As described above, the asset portfolio prediction method and system according to one embodiment of the present disclosure can predict asset portfolio weights based on adjustable targets and multi-faceted data, thereby flexibly responding to changes in market conditions and establishing and providing an optimal rebalancing strategy that suits the investment purpose.

[0341] Meanwhile, the embodiments according to the present disclosure described above may be implemented in the form of program commands that can be executed through various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program commands, data files, data structures, etc., either singly or in combination. The program commands recorded on the computer-readable recording medium may be specially designed and configured for the present disclosure or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specifically configured to store and execute program commands, such as ROMs, RAMs, and flash memories. Examples of program commands include not only machine language codes generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc. Hardware devices may be changed into one or more software modules to perform processing according to the present disclosure, and vice versa.

[0342] The specific implementations described in this disclosure are exemplary embodiments and do not limit the scope of the present disclosure in any way. For the sake of brevity, descriptions of conventional electronic components, control systems, software, and other functional aspects of the systems may be omitted. In addition, the lines connecting or connecting members between components depicted in the drawings are merely representative of functional connections and / or physical or circuit connections, and may be replaced or represented as various additional functional connections, physical connections, or circuit connections in an actual device. In addition, unless specifically mentioned as “essential,” “important,” etc., a component may not be absolutely necessary for the application of the present disclosure.

[0343] Although the detailed description of the present disclosure has been described with reference to preferred embodiments of the present disclosure, it will be understood by those skilled in the art or having ordinary knowledge in the art that various modifications and changes can be made to the present disclosure without departing from the spirit and technical scope of the present disclosure as set forth in the claims below. Accordingly, the technical scope of the present disclosure should not be limited to the contents described in the detailed description of the specification, but should be defined by the claims.

[0344] The present disclosure relates to a method and system for predicting asset portfolio weights using a deep learning model based on a Temporal Fusion Transformer (TFT) utilizing adjustable objectives and multi-faceted data, and thus has industrial applicability as it can be used in the artificial intelligence industry.

Claims

1. A method for predicting an asset portfolio performed by a computing system, A step of determining a target prediction task, which is data that specifies a target task; A step of collecting raw data for performing the above-determined target prediction task; A step of arranging the collected raw data in a time series manner along a specific time axis; A step of training a predetermined time series prediction model based on the sorted raw data and the determined target prediction task; A step of predicting daily portfolio asset weight data for a predetermined prediction unit period based on the learned time series prediction model; A step of processing the predicted daily portfolio asset weight data at each predetermined rebalancing point to create a single portfolio asset weight data; and A step of providing the generated single portfolio asset weight data is included. How to predict asset portfolios.

2. In paragraph 1, The step of determining the above target prediction task is: A step of acquiring task setting information, which is information including at least two or more data from a given domain, current portfolio asset ratio, rebalancing period, asset pool, and investment tendency data. How to predict asset portfolios.

3. In paragraph 2, The above investment propensity data is, Contains data specifying at least one of the following tendencies: a return maximization tendency that allocates weight to assets with positive past cumulative returns, a risk minimization tendency that allocates weight to assets with low past volatility, and a Sharpe ratio optimization tendency that allocates weight in order of highest Sharpe ratio for each asset. How to predict asset portfolios.

4. In paragraph 1, The steps of collecting the above raw data are: A step of collecting at least one of predetermined price time series data, macroeconomic indicator data, market indicator data, and target impact variable data. How to predict asset portfolios.

5. In paragraph 4, The above target influence variable data is, A variable that is related to at least one of the domain and asset data corresponding to the above target prediction task by a predetermined standard or higher. How to predict asset portfolios.

6. In paragraph 2, The step of aligning the above raw data in a time series manner along a specific time axis is: A step of aligning the task setting information corresponding to the target prediction task along the time axis. How to predict asset portfolios.

7. In paragraph 6, The step of training the above time series prediction model is: A step of inputting time series data from a predetermined time point tp to a predetermined time point t into the time series prediction model in a learning data set including the time-series-sorted raw data and the task setting information, A step of predicting the daily portfolio asset weight data from a predetermined time point t+1 to a predetermined time point t+h using a time series prediction model that has received the above time series data as input, For each of the predicted daily portfolio asset weight data, a step of performing normalization post-processing so that the sum of the weights of each asset included in the portfolio asset weight data becomes '1'; A step of comparing the normalized daily portfolio asset weight data with the actual daily portfolio asset weight data from the predetermined time point t+1 to the predetermined time point t+h, A step of calculating the loss according to the results of the above comparison, A step of updating the parameters of the time series prediction model based on the above-described loss is included. How to predict asset portfolios.

8. In paragraph 1, The step of predicting the daily portfolio asset weight data for the above prediction unit period is as follows: For each of the predicted daily portfolio asset weight data, a step of performing normalization post-processing so that the sum of the weights of each asset included in the portfolio asset weight data becomes '1' is included. How to predict asset portfolios.

9. In paragraph 1, The steps for generating the above single portfolio asset weight data are: A step of applying at least one rule among a rule for applying a predetermined operation to the daily portfolio asset weight data and a rule for determining whether the daily portfolio asset weight data satisfies a predefined standard to the daily portfolio asset weight data, A step of generating the single portfolio asset weight data based on the results according to the above applied rules. How to predict asset portfolios.

10. In paragraph 9, The steps for generating the above single portfolio asset weight data are: If there are constraints on the transaction costs and asset weights, the method further includes a step of generating the single portfolio asset weight data by selectively applying at least one of the clipping and smoothing operations. How to predict asset portfolios.

11. In paragraph 1, Further comprising a step of providing basis data, which is data explaining the causal relationship and data flow on which the above single portfolio asset weight data was determined. How to predict asset portfolios.

12. In paragraph 1, A step of evaluating the performance of the time series prediction model based on actual investment performance according to the single portfolio asset weight data, Further comprising a step of updating the parameters of the time series prediction model according to the evaluation results. How to predict asset portfolios.

13. In paragraph 1, A step of obtaining a target label data set, which is data that generates daily portfolio asset weight data corresponding to the prediction period of the time series prediction model based on a predetermined existing prediction model, and Further comprising a step of training the time series prediction model based on the obtained correct answer label data set. How to predict asset portfolios.

14. In paragraph 13, The step of training the time series prediction model based on the above correct label data set is: A step of comparing 1:1 each of the daily portfolio asset weight data included in the above correct answer label data set and each of the daily portfolio asset weight data predicted by the time series prediction model; A step of calculating a loss according to the results of the above comparison, A step of updating the parameters of the time series prediction model based on the above-described loss is included. How to predict asset portfolios.

15. In paragraph 13, The steps for obtaining the above correct label data set are: It includes a step of obtaining the correct answer label data set for each purpose based on an existing prediction model optimized for prediction for each purpose, The step of training the time series prediction model based on the above correct label data set is: A step of determining a correct answer label data set corresponding to a task purpose based on at least one of a predetermined investment tendency, market situation, and domain among the correct answer label data sets acquired for the above purpose, A step of training the time series prediction model based on the determined correct answer label data set. How to predict asset portfolios.

16. In paragraph 1, The step of providing the above single portfolio asset weight data is: A step of providing personalized portfolio rebalancing recommendation information based on the single portfolio asset weight data through the user interface of the robo-advisor service. How to predict asset portfolios.

17. In paragraph 11, A step of obtaining a What-if scenario that assumes changes in specific macroeconomic indicators or market indicators through user input, Further comprising a step of generating changed single portfolio asset weight data and changed basis data through the time series prediction model according to the above-mentioned acquired scenario and providing them to the fund operation support system. How to predict asset portfolios.

18. In paragraph 1, The above target prediction task includes a predetermined investment theme or factor, The above single portfolio asset weight data is used as input to determine the constituent portfolio of a new listed index fund (ETF) product. How to predict asset portfolios.

19. In paragraph 1, The step of training the above time series prediction model is: A step of training the time series prediction model, which is a deep learning model based on a Temporal Fusion Transformer (TFT), based on the sorted raw data and the determined target prediction task. How to predict asset portfolios.

20. At least one memory; and At least one processor for predicting an asset portfolio by reading out at least one application stored in the memory; The instructions of the above processor are: A step of determining a target prediction task, which is data that specifies the target task; A step of collecting raw data for performing the above-determined target prediction task, A step of arranging the collected raw data in a time series manner along a specific time axis, A step of training a predetermined time series prediction model based on the above-mentioned sorted raw data and the above-mentioned determined target prediction task, A step of predicting daily portfolio asset weight data for a predetermined prediction unit period based on the above-mentioned learned time series prediction model, A step of processing the predicted daily portfolio asset weight data at each predetermined rebalancing point to create a single portfolio asset weight data, A command comprising a step of providing the single portfolio asset weight data generated above. Asset portfolio prediction system.

21. In paragraph 20, A plurality of neurons comprising an array including at least one register, at least one programmable logic, and at least one input interface; a plurality of synaptic circuits storing synaptic weights that adjust connection strengths between the plurality of neurons; and at least one routing network that controls data flow between the plurality of neurons; Each of the plurality of neurons further includes a field programmable gate array (FPGA) implementation for a predetermined artificial neural network that is connected to at least one other neuron through the routing network to set a transmission path of the weight. Asset portfolio prediction system.

22. In paragraph 20, A plurality of neurons organized in an array including at least one register, at least one microprocessor, and at least one input; and a plurality of synaptic circuits storing synaptic weights that adjust connection strengths between the plurality of neurons; Each of the plurality of neurons further includes an application specific integrated circuit (ASIC) for a given artificial neural network that is connected to at least one other neuron through one of the plurality of synaptic circuits. Asset portfolio prediction system.

Images