Generative adversarial networks-based system for synthetic data generation and predictive analytics in asset management

Abstract

Disclosed is a Generative Adversarial Networks-based system for synthetic data generation and predictive analytics in asset management. The system comprises a noise input unit supplies random noise data as input. A generator unit processes the received random noise data and generates synthetic datasets resembling real financial data. A discriminator unit evaluates and distinguishes between the synthetic datasets and the real datasets. A feedback unit interconnects the generator unit and the discriminator unit and iteratively refines the synthetic datasets based on evaluation results from the discriminator unit until the synthetic datasets closely mirror real financial conditions. A data analysis unit utilizes the refined synthetic datasets to perform scenario simulations and generate predictive analytics for optimizing asset management.

Description

GENERATIVE ADVERSARIAL NETWORKS-BASED SYSTEM FOR SYNTHETIC DATA GENERATION AND PREDICTIVE ANALYTICS IN ASSET MANAGEMENT TECHNICAL FIELD

[0001] The present disclosure generally relates to artificial intelligence and machine learning systems. Further, the present disclosure particularly relates to Generative Adversarial Networks-based systems for synthetic data generation and predictive analytics in asset management. BACKGROUND

[0002] Asset management has become increasingly complex due to the emergence of multiple asset classes and the growing regulatory requirements in financial markets. The expansion of financial markets has led to increased competitiveness and stricter regulations, necessitating effective and efficient methods to handle assets. Small businesses, which account for a significant proportion of market participants, often lack proper asset management practices. Reports indicate that nearly half of small businesses are unable to implement structured asset management, leading to inefficiencies. Additionally, approximately 40% of businesses rely on manual tracking or do not manage their assets at all, resulting in delays, overpayments, and customer dissatisfaction.Docket No. RP-2024.98.007 2

[0003] Existing asset management systems utilise traditional methods of data tracking and financial forecasting. Such systems, comprising spreadsheet-based methods and basic database management techniques, are associated with several limitations. Spreadsheets are often used by financial executives to perform fixed asset accounting. However, such tools are prone to errors, require significant manual effort, and lack scalability. On average, nearly 46.8% of finance executives spend a week per month managing databases, resulting in wasted resources and suboptimal outcomes.

[0004] Furthermore, asset management faces challenges due to large, complex, and diverse data sources. Financial markets generate substantial data, with approximately 73% of the available data remaining unused for analysis. The lack of analytical tools capable of processing such diverse datasets further exacerbate the issue, leaving organisations with low analytics maturity. Over 87% of organisations exhibit low levels of business intelligence (BI) and analytics maturity, restricting their ability to derive meaningful insights for decision-making.

[0005] Another challenge associated with conventional asset management systems is the presence of incomplete or erroneous data. Investment policies often rely on accurate and complete datasets. However, data inaccuracies and incomplete information lead to flawed policies and suboptimal decisions. Keeping track of vast datasets from multiple sources becomes an arduous task for organisations. Research indicates that nearlyDocket No. RP-2024.98.007 3 86% of global financial services institutions lack confidence in their datasets for decision-making purposes, highlighting the need for improved data integration and validation mechanisms.

[0006] Regulatory changes and market volatility further complicate asset management processes. The financial sector is subject to frequent regulatory updates, geopolitical influences, and shifts in investor sentiment. Financial firms often struggle to access up-to-date regulatory information. Market research has shown that approximately 95% of the data used in financial decision-making is more than one hour old, 63% is more than 24 hours old, and 38% is over four days old. Such delays lead to outdated analyses and reduced accuracy in financial forecasting.

[0007] Current systems also fail to address the growing demand for personalised investment guidance. High net-worth individuals and retail investors require customised plans aligned with their specific financial objectives. However, the platforms available to such investors do not incorporate improved analytical capabilities or adaptive scenario simulations, thereby limiting the ability of investors to optimise financial outcomes. Research highlights that in 2023, 71% of Canadians aged between 18 to 34 found artificial intelligence useful in automating savings. However, existing systems lack effective tools to meet such expectations.

[0008] Moreover, global asset management markets are projected to grow significantly in the coming years. For instance, in 2024, assets underDocket No. RP-2024.98.007 4 management (AUM) in Canada are estimated to reach CAD 2,633.99 billion, with CAD 37.27 billion allocated to digital investments. Despite such growth, Canadian robo-advisors hold less than 1% of the market share in investment assets, indicating a gap in adopting effective technological tools.

[0009] The global asset management market was valued at CAD 507.9 billion in 2022 and is anticipated to grow at a compound annual growth rate (CAGR) of 35.2% from 2023 to 2032. Similarly, artificial intelligence in asset management is projected to grow at a CAGR of 24.5% from 2023 to 2030, with machine learning techniques dominating over 65% of the revenue share. Despite such improvements, process automation segments account for only 26% of the market share, and challenges in integrating diverse datasets persist.

[0010] In light of the above discussion, there exists an urgent need for solutions that overcome the problems associated with conventional systems and techniques for asset management, comprising challenges, for example lack of structured practices, unused and erroneous data, outdated decisionmaking frameworks, and the inability to meet personalised financial objectives. SUMMARY

[0011] An objective of the present disclosure is to enable synthetic data generation and predictive analytics for asset management by Generative Adversarial Networks. The system aims to address challenges associated withDocket No. RP-2024.98.007 5 generating reliable financial data, simulating economic conditions, and optimising financial decision-making.

[0012] In an aspect, the present disclosure provides a Generative Adversarial Networks (GANs)-based system for synthetic data generation and predictive analytics in asset management. The GANs-based system comprises a noise input unit supplies random noise data as input. A generator unit processes random noise data to generate synthetic datasets resembling real financial data. A discriminator unit evaluates the synthetic datasets against real datasets and iteratively refines the synthetic datasets through a feedback unit until the synthetic datasets closely mirror real financial conditions. A data analysis unit utilises refined synthetic datasets to perform scenario simulations and generate predictive analytics for asset management.

[0013] The GANs-based system integrates improved neural network architectures to simulate economic conditions and generate synthetic datasets mimicking real financial trends. The discriminator unit employs a layered evaluation process to validate synthetic datasets against real datasets, thereby improving the reliability of analytics. The GANs-based system incorporates a personalised investment guidance unit to adapt scenario simulations based on specific financial objectives of investors. Macroeconomic indicators are utilised to refine synthetic datasets and align the datasets with global financial patterns. A multi-stream evaluation framework is incorporated to validate diverse data inputs, comprising structured financial data, socialDocket No. RP-2024.98.007 6 media feedback, and news reports. Furthermore, the GANs-based system comprises a risk management tool integrated with the data analysis unit to provide risk mitigation approaches for financial portfolios. Machine learning techniques are incorporated to recommend optimal asset allocation plans for long-term investment objectives. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The summary above, as well as the following detailed description of illustrative embodiments, is better understood when read in conjunction with the appended drawings. For the purpose of illustrating the present disclosure, exemplary constructions of the disclosure are shown in the drawings. However, the present disclosure is not limited to specific methods and instrumentalities disclosed herein.

[0015] Embodiments of the present disclosure will now be described, by way of example only, with reference to the following diagrams.

[0016] FIG. 1 illustrates a Generative Adversarial Networks (GANs)- based system 100 for synthetic data generation and predictive analytics in asset management, in accordance with various implementations of the present disclosure;

[0017] FIG. 2 illustrates a sequential diagram of the GANs-based system 100 for synthetic data generation and predictive analytics in asset management, in accordance with embodiments of the present disclosure; andDocket No. RP-2024.98.007 7

[0018] FIG. 3 illustrates an architecture of the GANs-based system 100 for synthetic data generation and predictive analytics in asset management, in accordance with embodiments of the present disclosure. DETAILED DESCRIPTION OF EMBODIMENTS

[0019] The following detailed description illustrates embodiments of the present disclosure and ways in which they can be implemented. Although some modes of carrying out the present disclosure have been disclosed, those skilled in the art would recognize that other embodiments for carrying out or practicing the present disclosure are also possible.

[0020] FIG. 1 illustrates a Generative Adversarial Networks (GANs)- based system 100 for synthetic data generation and predictive analytics in asset management, in accordance with various implementations of the present disclosure. The GANs-based system 100 comprises a noise input unit 102 to supply random noise data as input. The noise input unit 102 serves as the initial component of the GANs-based system 100, providing a stream of random noise data. Said random noise data acts as a foundational input for subsequent components in the GANs-based system 100. The noise input unit 102 generates random values or patterns, which may be derived from a stochastic process, pseudo-random number generators, or other suitable mechanisms for creating non-deterministic data. Said random noise data represents an important parameter for initiating data generation processes within the GANs-based system 100. The structure of the noise input unit 102Docket No. RP-2024.98.007 8 may comprise a combination of hardware, software, or both, depending on implementation requirements. The noise input unit 102 may operate in realtime or batch-processing modes, assuring that random noise data is consistently supplied to the generator unit 104. For instance, the noise input unit 102 could use a stochastic process, for example Gaussian noise or uniform random numbers, to create unpredictable input data patterns. Such patterns might be used to simulate financial market fluctuations as a baseline for generating synthetic datasets. The noise input unit 102 might produce inputs which simulate varying degrees of market volatility or random economic shocks.

[0021] In an embodiment, the GANs-based system 100 comprises a generator unit 104 that is operatively connected to the noise input unit 102, processes the received random noise data. The generator unit 104 applies computational techniques to transform said random noise data into synthetic datasets resembling real financial data. Such synthetic datasets are generated based on pre-determined patterns or distributions learned from real-world financial datasets. The generator unit 104 may comprise one or more computational models, for example deep learning architectures, trained on historical financial data. The generator unit 104 iteratively processes random noise data to produce outputs which are statistically and contextually aligned with financial trends, transactions, or market activities. Said synthetic datasets generated by the generator unit 104 are then made available forDocket No. RP-2024.98.007 9 subsequent evaluation and refinement by downstream components. For example, the generator unit 104 could simulate datasets representing market behaviors (daily stock prices, interest rates, commodity performance, etc.). The generator unit 104 may utilize computational models to transform noise input into data patterns which resemble real-world trends, for example the rise and fall of financial indices during different market cycles. Such synthetic datasets might be applied to model the impact of macroeconomic events on portfolio performance.

[0022] In an embodiment, the GANs-based system 100 comprises a discriminator unit 106 receives synthetic datasets generated by the generator unit 104 and real datasets. Said discriminator unit 106 evaluates and distinguishes between the synthetic datasets and the real datasets based on various statistical, contextual, and structural parameters. The discriminator unit 106 performs said evaluation using computational models trained on both datasets (synthetic and real) and by comparative analysis to identify distinguishing aspects. Such evaluation by the discriminator unit 106 enables the identification of deviations or inconsistencies within the synthetic datasets, providing actionable feedback for iterative refinement. The discriminator unit 106 is capable of processing diverse data formats, comprising numerical, categorical, and temporal datasets, improving diverse data formats applicability in financial data scenarios. The discriminator unit 106 could evaluate the synthetic datasets by comparing statistical aspects, for exampleDocket No. RP-2024.98.007 10 mean and variance, with real datasets. For instance, the discriminator unit 106 might identify deviations between synthetic and real datasets, for example an unrealistic pattern of price movements or market trends. The discriminator unit 106 helps assure the generated datasets align with the expected characteristics of real financial conditions.

[0023] In an embodiment, the GANs-based system 100 comprises a feedback unit 108 that interconnects the generator unit 104 and the discriminator unit 106. The feedback unit 108 iteratively refines the synthetic datasets based on evaluation results received from the discriminator unit 106. Such refinement involves modifying parameters or inputs within the generator unit 104 to reduce discrepancies between synthetic datasets and real financial datasets. The feedback unit 108 operates through a dynamic loop wherein evaluation metrics from the discriminator unit 106 are utilised to improve the accuracy and quality of synthetic datasets. The feedback unit 108 may employ gradient-based optimisation or other iterative refinement techniques to assure that the synthetic datasets progressively converge toward real-world financial conditions. For instance, the feedback unit 108 might adjust processing parameters of the generator unit 104 to correct identified inconsistencies, for example unrealistic volatility patterns or missing correlations in asset performance. By iteratively refining the synthetic datasets, the feedback unit 108 assures that the generated data more accurately reflects real financial market behavior.Docket No. RP-2024.98.007 11

[0024] In an embodiment, the GANs-based system 100 comprises a data analysis unit 110 utilises refined synthetic datasets for performing scenario simulations and generating predictive analytics for optimising asset management. Said data analysis unit 110 applies computational techniques to analyse refined synthetic datasets and extract actionable insights. Such insights may comprise forecasting market trends, evaluating risk factors, or determining optimal investment plans. The data analysis unit 110 integrates various analytical approaches, for example statistical analysis, machine learning, and predictive modelling, to process the refined synthetic datasets. Scenario simulations performed by the data analysis unit 110 enable the evaluation of market scenarios and their implications for asset management. Said data analysis unit 110 facilitates informed decision-making in asset management by providing predictive outputs aligned with financial objectives. For example, the data analysis unit 110 might simulate a market downturn scenario to evaluate the impact on portfolio value or analyze a recovery phase to identify growth opportunities. The data analysis unit 110 might also predict asset allocation strategies, for example recommending an increased exposure to equities during bullish market conditions, or a shift to safer assets during periods of high volatility. Such insights support asset managers in making informed decisions based on realistic market scenarios.

[0025] In an exemplary aspect, a financial institution managing a diversified investment portfolio seeks to generate synthetic datasets toDocket No. RP-2024.98.007 12 simulate market conditions from 2025 to 2030, evaluate risk exposure, and optimise asset allocation plans. The process begins with the noise input unit 102 generating random noise data in the range of -1 to 1, representing stochastic variations in market parameters, for example interest rates, GDP growth, and inflation. The generator unit 104 processes said random noise data using pre-trained computational models to create synthetic datasets resembling real financial data, for example daily equity price movements, bond yield curves, and alternative asset performance. The discriminator unit 106 evaluates said synthetic datasets against historical datasets, identifying deviations in statistical properties, for example volatility clustering and autocorrelation, and provides evaluation metrics indicating alignment with real data. Based on said evaluation results, the feedback unit 108 refines synthetic datasets by modifying parameters in the generator unit 104, iteratively improving accuracy until synthetic datasets closely reflect real-world financial patterns. The data analysis unit 110 utilises said refined synthetic datasets to perform scenario simulations, for example modeling a global recession in 2026 and recovery in subsequent years, and generates predictive analytics suggesting optimal asset allocation plans, for example increasing equity exposure in emerging markets by 15% while reducing exposure to long-term bonds by 10%, to maximise returns under projected growth scenarios. Such a process enables the institution to enhance decision-making and optimise portfolio performance effectively.Docket No. RP-2024.98.007 13

[0026] In an embodiment, the generator unit 104 may utilize neural network architectures to simulate varying economic conditions and generate synthetic datasets mimicking real financial trends. The generator unit 104 employs computational models trained on historical financial datasets which incorporate key parameters, for example equity prices, bond yields, interest rate fluctuations, and inflation patterns. The neural network architectures within the generator unit 104 may comprise multi-layer perceptrons, recurrent neural networks, or convolutional neural networks, which enable the system to capture complex relationships and temporal dependencies in financial data. The generator unit 104 processes random noise data supplied by the noise input unit 102 and transforms said noise into synthetic datasets which represent realistic financial scenarios. The generator unit 104 dynamically adjusts its output by adapting to changing market trends and conditions through continuous retraining on updated financial data. Synthetic datasets produced by the generator unit 104 enable the representation of market events, for example recessions, recoveries, or periods of sustained growth, facilitating predictive analyses and scenario simulations. The generator unit 104 also incorporates mechanisms for fine-tuning its outputs, assuring that the generated datasets align with known financial patterns and statistical distributions observed in real markets.

[0027] In an embodiment, the discriminator unit 106 can employ a layered evaluation process to validate synthetic datasets against real datasets,Docket No. RP-2024.98.007 14 assuring the reliability of data for downstream processing. The discriminator unit 106 analyzes synthetic datasets generated by the generator unit 104 and compares them with historical financial datasets to identify discrepancies in key metrics, for example distributional characteristics, temporal correlations, and structural patterns. The layered evaluation process involves multiple stages of analysis, where each stage focuses on specific aspects of the datasets, for example statistical validity, contextual relevance, and consistency with market trends. The discriminator unit 106 employs computational techniques, for example hypothesis testing, feature extraction, and pattern recognition to evaluate the quality of synthetic datasets. Insights obtained from the evaluation process are transmitted to the feedback unit 108, which enables iterative refinement of synthetic datasets by the generator unit 104. The discriminator unit 106 handles a wide range of data formats, for example time-series data, categorical variables, and numerical inputs, assuring thorough evaluation across different dimensions of financial data.

[0028] In an embodiment, the GANs-based system 100 may comprise a personalized investment guidance unit which adapts scenario simulations based on specific financial objectives provided by institutional and retail investors. The personalized investment guidance unit processes refined synthetic datasets and incorporate user-defined parameters, for example risk tolerance, investment horizon, and target return objectives. For institutional investors, the personalized investment guidance unit simulates the impact ofDocket No. RP-2024.98.007 15 macroeconomic events, for example monetary policy changes or geopolitical disruptions, on large-scale portfolios. For retail investors, the personalized investment guidance unit provides customized recommendations for achieving individual financial goals, for example retirement planning or saving for significant life events. The unit evaluates multiple market scenarios, assessing how varying economic conditions influence investment plans, and generates actionable insights to align portfolio adjustments with specific objectives. The personalized investment guidance unit integrates data from the analysis unit 110 to assure that the generated insights reflect current market dynamics and investor priorities.

[0029] In an embodiment, the generator unit 104 may integrate macroeconomic indicators into the generation of synthetic datasets to assure alignment with global financial patterns. Key macroeconomic indicators, for example GDP growth, inflation rates, employment statistics, and trade balances, are incorporated into the data generation process to capture systemic economic trends. The generator unit 104 maps relationships between micro-level financial metrics, for example stock price fluctuations, and macroeconomic indicators, enabling the creation of datasets which reflect broader economic contexts. The generator unit 104 processes macroeconomic data in conjunction with random noise inputs, assuring that synthetic datasets reflect the interplay between macroeconomic forces and market behaviors. By integrating said indicators, the generator unit 104 improves said indicatorsDocket No. RP-2024.98.007 16 ability to simulate realistic financial scenarios, providing datasets which are valuable for scenario-specific analyses and forecasting applications.

[0030] In an embodiment, the discriminator unit 106 may incorporate a multi-stream evaluation framework which integrates diverse data inputs, comprising structured financial data, unstructured text from news reports, and sentiment data derived from social media. The multi-stream framework enables the discriminator unit 106 to validate synthetic datasets against multiple dimensions of real-world information. For structured financial data, the discriminator unit 106 analyzes time-series trends, pricing structures, and key performance indicators to verify statistical consistency. For news reports, the unit employs text-mining techniques to extract contextually relevant events or trends which may influence market behaviors. Sentiment data from social media platforms is evaluated for correlations between public sentiment and synthetic data outputs, assuring alignment with real-time market perceptions. The multi-stream framework enables cross-validation across different data channels, enhancing the overall reliability and applicability of synthetic datasets for predictive and analytical tasks.

[0031] In an embodiment, the GANs-based system 100 may comprise a risk management tool integrated with the data analysis unit 110 to provide customized risk mitigation strategies for financial portfolios. The risk management tool processes refined synthetic datasets to identify risks, for example exposure to volatile asset classes, sector-specific downturns, orDocket No. RP-2024.98.007 17 geopolitical disruptions. The tool quantifies risk metrics, comprising value-atrisk, conditional value-at-risk, and stress test results, based on scenario simulations generated by the data analysis unit 110. The risk management tool evaluates multiple risk factors simultaneously, generating recommendations for portfolio diversification, hedging strategies, and dynamic rebalancing to mitigate exposure to identified risks. Said recommendations are customized to the specific characteristics of the portfolio under evaluation, assuring that the plans align with investor-defined objectives and market conditions.

[0032] In an embodiment, the data analysis unit 110 may incorporate machine learning techniques to evaluate refined synthetic datasets and recommend optimal asset allocation plans for long-term financial objectives. Said machine learning techniques comprise supervised learning models for predicting asset performance, unsupervised learning methods for clustering similar market conditions, and reinforcement learning approaches for dynamic portfolio adjustments. The data analysis unit 110 processes both historical financial data and refined synthetic datasets to identify trends and correlations which influence asset performance. The unit forecasts market behaviors under varying economic scenarios, providing insights into the expected returns, volatility, and diversification benefits of different asset classes. Based on said forecasts, the data analysis unit 110 generates recommendations for asset allocation plans customized to specific investor objectives, for exampleDocket No. RP-2024.98.007 18 growth-focused investments, income generation, or capital preservation. The unit continuously updates its models to reflect evolving market conditions, assuring that the recommendations remain relevant and actionable.

[0033] FIG. 2 illustrates a sequential diagram of the GANs-based system 100 for synthetic data generation and predictive analytics in asset management, in accordance with embodiments of the present disclosure. The process begins with the user interacting with an input panel, which collects initial input data, for example random noise, to simulate financial scenarios. The input panel transmits the random noise data to the generator unit 104, which creates synthetic datasets resembling real financial data. The synthetic datasets are forwarded to the discriminator unit 106, where the synthetic datasets are evaluated against real datasets to identify differences or inconsistencies. The discriminator unit 106 provides feedback to the generator unit 104 through an iterative feedback loop, allowing the generator to refine and improve the synthetic datasets. Once validated, the refined datasets are sent to the data analysis unit 110, which performs scenario simulations and generates predictive analytics. Insights and recommendations, for example asset allocation plans or risk assessments, are then presented to the user through a dashboard interface, enabling informed decision-making in asset management.

[0034] FIG. 3 illustrates an architecture of the GANs-based system 100 for synthetic data generation and predictive analytics in asset management,Docket No. RP-2024.98.007 19 in accordance with embodiments of the present disclosure. The process begins with the noise input unit 102, denoted as Z, which supplies random noise data as input to the generator unit 104. The generator unit 104 processes the noise input to create synthetic datasets, denoted as X fake, which resemble real financial datasets. Concurrently, real datasets, denoted as X real, are provided to the discriminator unit 106 as a baseline for comparison. Both the synthetic datasets (X fake) and the real datasets (X real) are evaluated by the discriminator unit 106, which distinguishes between synthetic datasets and real datasets. The discriminator unit 106 outputs a result, denoted as y, indicating whether the input data is classified as real or synthetic (fake). Feedback from the discriminator unit 106 is iteratively processed through the feedback unit 108 to refine the generator unit 104, enabling the synthetic datasets to progressively align more closely with real financial conditions. Through said iterative process, the GANs-based system 100 generates highquality synthetic datasets for further use in scenario simulations and predictive analytics by the data analysis unit 110.

Claims

CLAIMS What is claimed is:

1. A Generative Adversarial Networks (GANs)-based system for synthetic data generation and a predictive analytics in an asset management, comprising: a noise input unit, supplies a random noise data as input; a generator unit, operatively connected to the noise input unit, wherein the generator unit is configured to process the received random noise data and generate the synthetic datasets resembling a real financial data; a discriminator unit, receives the synthetic datasets, to evaluate and distinguish between the synthetic datasets and the real datasets; a feedback unit, interconnecting the generator unit and the discriminator unit, wherein the feedback unit iteratively refines the synthetic datasets based on the evaluation results from the discriminator unit until the synthetic datasets closely mirror the real financial conditions; andDocket No. RP-2024.98.007 21 a data analysis unit, utilizes the refined synthetic datasets to perform the scenario simulations and generate the predictive analytics for optimizing the asset management.

2. The system of Claim 1, wherein the generator unit utilizes the neural network architectures to simulate the varying economic conditions and generate the synthetic datasets mimicking the real financial trends.

3. The system of Claim 1, wherein the discriminator unit employs a layered evaluation process to validate the synthetic datasets against the real datasets, enhancing reliability of the predictive analytics.

4. The system of Claim 1, further comprising a personalized investment guidance unit adapts the scenario simulations based on the specific financial objectives of the institutional investors and the retail investors.

5. The system of Claim 1, wherein the generator unit integrates the macroeconomic indicators to refine the synthetic datasets and align the datasets with the global financial patterns by applying a generative artificial intelligence (AI) technique.

6. The system of Claim 1, wherein the discriminator unit comprises a multistream evaluation framework for incorporating the diverse data input, including a financial data, a social media feedback, and the news reports.Docket No. RP-2024.98.007 22 7. The system of Claim 1, further comprising a risk management tool integrated with the data analysis unit, wherein the risk management tool provides the customized risk mitigation strategies for the financial portfolios.

8. The system of Claim 1, wherein the data analysis unit incorporates a machine learning technique to recommend the optimal asset allocation strategies for the long-term investment objectives.