Course Name:
课程名称：
GSOE9011 Engineering PGCW Research Skills
GSOE9011 工程专业研究生课程研究技能
Proposal Name:
提案名称：
Application of AI in investment and asset management
人工智能在投资和资产管理中的应用
Topic Number：ERC, 9
主题编号：ERC, 9
Group Number：175
小组编号：175
Team Members:
团队成员：
Yifei Wang z5474815
王一飞 z5474815
Eden Wang z5645864
伊甸·王 z5645864
Jiongqi Wang z5576929
王景琪 z5576929
Fei Teng z5615053
滕飞 z5615053
Yuzhou Zhu z5598414
朱宇舟 z5598414
Date of submission: 25/07/2025
提交日期：25/07/2025
Abstract
摘要
This article examines the transition to the application of artificial intelligence in investment and asset management and aims to address the increasing weaknesses of traditional investment methods and to address the complexity and volatility of modern financial markets. With the increasing complexity of financial markets and increased data density, traditional approaches, based mainly on manual analysis and historical data, are no longer able to meet the needs of optimal policy formulation. The aim of the study is to use robots to find out how to reallocate resources and to modify market forecasting and risk management procedures in order to increase the efficiency and profitability of investments and reduce the associated risks.
本文探讨了人工智能在投资和资产管理领域的应用转型，旨在解决传统投资方法的日益凸显的弱点，并应对现代金融市场的复杂性和波动性。随着金融市场复杂性的增加和数据密度的提升，主要依赖人工分析和历史数据的传统方法已无法满足最优政策制定的需求。本研究旨在利用机器人技术，探索如何重新配置资源，并调整市场预测和风险管理流程，以提高投资效率和盈利能力，并降低相关风险。
Three key questions: what are the effects of artificial intelligence in asset management? How can artificial intelligence help investors optimise the distribution of investments and minimise losses? What are the real challenges for artificial intelligence in its financial applications? The study examines the legal application of an aig strategy with respect to conventional methods of risk measurement and management.
三个关键问题：人工智能在资产管理中的影响是什么？人工智能如何帮助投资者优化投资分配并最小化损失？人工智能在金融应用中的真实挑战是什么？本研究考察了人工智能策略在传统风险测量和管理方法中的法律应用。
Product trials have shown that significant progress has been made in optimising artificial intelligence combinations and that more advanced and better learning techniques have demonstrated the validity of traditional approaches. Promising developments include the integration of forecasting models with modern portfolio theories, multi-regional optimisation methods and modern risk management systems using risk assessment and market forecasting machines. However, recent studies have shown that implementation of the models, transparency and real-time compliance remain problems, especially under extreme market conditions.
产品试用表明，在优化人工智能组合方面已取得显著进展，更先进和更好的学习技术也证明了传统方法的可行性。有前景的发展包括将预测模型与现代投资组合理论相结合、多区域优化方法和使用风险评估与市场预测机器的现代风险管理系统。然而，最近的研究表明，模型实施、透明度和实时合规性仍然是问题，尤其是在极端市场条件下。
The study, which is important for financial techniques and engineering, aims to provide practical and practical guidance for a more effective application of asset management. Innovative research aims to develop an integrated approach that integrates traditional investment theory with advanced technologies and can lead to the restoration of asset management methods and a new phase in the development of financial techniques. Expected results include the improvement of the investment policy framework, the improvement of the agreements on risk management policies and the contribution to the regulatory debate in relation to financial applications, which will ultimately benefit investors, financial institutions and the economic structure in general.
这项对金融技术和工程领域具有重要意义的研究，旨在为资产管理提供更有效的应用实践指导。创新性研究致力于开发一种整合传统投资理论与先进技术的综合方法，从而推动资产管理方法的恢复，并在金融技术发展进入新阶段。预期成果包括改进投资政策框架、完善风险管理政策协议，以及为金融应用相关的监管辩论做出贡献，最终将使投资者、金融机构及整体经济结构受益。
Introduction
引言
With the increasing complexity of financial markets, traditional investment methods are gradually unable to cope with information explosion and market fluctuations. Artificial intelligence (AI) technology has become an important force in promoting changes in the financial industry with its powerful data processing and pattern recognition capabilities. This study will focus on the practical application of artificial intelligence in investment and asset management, especially how to use AI to help investors optimize asset allocation, predict market trends and manage risks. Traditional investment methods often rely on manual analysis and historical data, but with the sharp increase in market information, the introduction of AI provides a more efficient way to deal with it.
随着金融市场日益复杂，传统投资方法逐渐无法应对信息爆炸和市场波动。人工智能（AI）技术凭借其强大的数据处理和模式识别能力，已成为推动金融行业变革的重要力量。本研究将聚焦人工智能在投资和资产管理中的实际应用，特别是如何利用 AI 帮助投资者优化资产配置、预测市场趋势和管理风险。传统投资方法通常依赖人工分析和历史数据，但随着市场信息的急剧增加，引入 AI 提供了一种更高效的处理方式。
In this research, the questions will mainly focus on: What practical effects can artificial intelligence play in asset management? How can AI help investors optimize asset allocation and reduce investment losses? What practical challenges does AI currently face in financial applications? The discussion of these issues will not only help improve investment efficiency and increase investment returns but also change the operating logic of the entire asset management industry. In addition, it will promote financial technology (FinTech) to move into a new stage of development and thus affect the stability of the economic structure and capital market.
在本项研究中，问题将主要聚焦于：人工智能在资产管理中能发挥哪些实际作用？人工智能如何帮助投资者优化资产配置并减少投资损失？人工智能目前在金融应用中面临哪些实际挑战？对这些问题的讨论不仅有助于提高投资效率、增加投资回报，还将改变整个资产管理行业的运营逻辑。此外，它还将推动金融科技（FinTech）进入新的发展阶段，从而影响经济结构和资本市场的稳定性。
In conclusion, this study will focus on analyzing the representative applications of artificial intelligence in asset management in major financial markets through a combination of literature review and case analysis. At the same time, the existing data is used to compare the performance of AI-assisted investment strategies and traditional investment strategies in terms of return rate and risk control, to evaluate the actual benefits and feasibility of AI technology in asset management.
总之，本研究将通过文献综述和案例分析相结合的方式，重点分析人工智能在主要金融市场中的典型应用。同时，利用现有数据比较人工智能辅助投资策略与传统投资策略在回报率和风险控制方面的表现，以评估人工智能技术在资产管理中的实际效益和可行性。
However, artificial intelligence has broad application prospects in investment and asset management, its promotion and application still need to solve problems such as algorithm transparency and data privacy. This study aims to provide a theoretical basis and practical reference for the application of artificial intelligence technology, so as to help it serve the financial industry more reasonably and efficiently in the future.
然而，人工智能在投资和资产管理领域具有广阔的应用前景，但其推广和应用仍需解决算法透明度和数据隐私等问题。本研究旨在为人工智能技术的应用提供理论基础和实践参考，以帮助其在未来更合理、高效地服务于金融行业。
Literature Review
文献综述
In recent years, there has been a significant rise in data-driven portfolio optimization, that can dynamically allocate assets and frequently surpass the performance of conventional methods. Many of these methods integrate traditional frameworks and predictive analytics with models, while others use deep reinforcement learning to achieve end-to-end allocation strategies.
近年来，数据驱动的投资组合优化显著增长，能够动态配置资产并频繁超越传统方法的性能。这些方法中，许多将传统框架和预测分析与传统模型相结合，而另一些则使用深度强化学习来实现端到端的配置策略。
Portfolio Optimization with AI
人工智能投资组合优化
For example, Chen et al. (2021) proposed a hybrid framework that uses an XGBoost-based predictive model to estimate stock returns, which has been used within a mean-variance optimization scheme. When applied to Korean equity markets, this method outperformed conventional mean-variance portfolio strategies. Additionally, Yan et al. (2024) introduced a Deep Portfolio Optimization (DPO) framework that integrates deep learning method and reinforcement learning into the modern portfolio theory. By extracting the key data and time-series features and incorporating transaction costs and risk into the reward function, the DPO method can get the highest cumulative portfolio value, which exceeds the original method in both profitability and risk-adjusted returns. In the cryptocurrency market, Lucarelli et al. (2020) designed a multi-agent system based on Q-learning to manage a portfolio Bitcoin, and other digital assets. The most effective configuration, utilizing a Sharpe ratio-based reward, achieved an average daily return of almost 4,67% and the highest Sharpe ratio (0.20) among these tested approaches, most of them outperforming equal-weight and genetic algorithms.
例如，陈等人（2021）提出了一种混合框架，该框架使用基于 XGBoost 的预测模型来估计股票收益，并在均值-方差优化方案中应用。当应用于韩国股票市场时，这种方法优于传统的均值-方差投资组合策略。此外，严等人（2024）引入了一种深度投资组合优化（DPO）框架，将深度学习和强化学习整合到现代投资组合理论中。通过提取关键数据和时序特征，并将交易成本和风险纳入奖励函数，DPO 方法可以获得最高的累积投资组合价值，在盈利能力和风险调整回报率方面均优于原始方法。在加密货币市场，Lucarelli 等人（2020）设计了一个基于 Q 学习的多智能体系统来管理比特币及其他数字资产的投资组合。最有效的配置，利用基于夏普比率（Sharpe ratio）的奖励，实现了平均每日收益率近 4.67%，并在所有测试方法中获得了最高的夏普比率（0.20），大多数方法在等权重和遗传算法之上表现更优。
Researchers have also investigated multi-objective optimization frameworks. For example, Choudhary et al. came up with a theory that they combine three different reinforcement learning methods, including maximizing returns, Sharpe ratio, and minimizing drawdowns. Additionally, researchers tested the daily stock index data, the approach that can adjust the risk outperformed both standard single-objective reinforcement learning methods and traditional approaches.
研究人员还研究了多目标优化框架。例如，Choudhary 等人提出了一种理论，该理论结合了三种不同的强化学习方法，包括最大化收益、夏普比率以及最小化回撤。此外，研究人员测试了每日股票指数数据，该方法能够调整风险，其表现优于标准的单目标强化学习方法和传统方法。
Risk Management and AI
风险管理与人工智能
For financial risk management, many companies implement advanced models to improve the accuracy of prediction and maximum reduce the risks, ranging from credit risk management in lending areas such as Value-at-Risk (VaR). Especially in the credit field, nowadays machine learning models generally have high accuracy to predict some companies or people whether they can afford the debt. Besides they also have high accurate predictions regarding whether the company will go bankrupt and flee, and they will issue corresponding warnings.
在金融风险管理方面，许多公司采用先进的模型来提高预测的准确性并最大限度地减少风险，范围从信贷领域的风险价值（VaR）管理。特别是在信贷领域，如今机器学习模型通常具有很高的准确性，能够预测某些公司或个人是否能够偿还债务。除此之外，它们还能准确预测公司是否会破产逃逸，并发出相应的警告。
For instance, Chang et al. (2024) used a wide range of algorithms to predict the credit card original data, including Logistic regression, random forest, neural network and enhancement methods. Their research results illustrated that XGBoost can get the highest accuracy rate among these algorithms in predicting customer default, this helps to significantly enhance the efficiency of its credit risk screening process.
例如，Chang 等人（2024）使用了一系列算法来预测信用卡原始数据，包括逻辑回归、随机森林、神经网络和增强方法。他们的研究结果表明，在预测客户违约方面，XGBoost 在这些算法中可以获得最高的准确率，这有助于显著提高其信用风险评估的效率。
In addition to the progress made in credit risk analysis, significant improvements have also happened in managing market risks and operational risks. For example, the Bayesian mixed frequency quantile vector autoregressive model has been proven to be capable of effectively enhancing the reliability of risk value (VaR) and expected loss estimations, especially when combined with information from different frequency sampling data (Iacopini et al., 2023). In the field of operational risk, denoising autoencoders are employed as a data-based approach to identify anomalies within structured financial datasets, enabling the detection of unusual trading patterns and early signs of potential system failures, without relying on pre-classified input data. (Sattarov et al., 2022). Furthermore, the latest developments in high quantile regression techniques, such as extreme quantile autoregressive models, have improved the ability to assess tail risks in financial time series with heavy-tailed distribution characteristics. (He,Fet al 2023). Additionally, improvement of methods is crucial to ensuring the clarity of the model design and compliance with regulatory requirements. This highlights the necessity of continuously conducting strict validation processes and establishing a clear supervision framework.
除了在信用风险分析方面取得的进展，市场风险和操作风险的管理也发生了显著改进。例如，贝叶斯混合频率分位数向量自回归模型已被证明能够有效提高风险价值（VaR）和预期损失估计的可靠性，尤其是在结合不同频率采样数据的信息时（Iacopini 等人，2023）。在操作风险领域，去噪自编码器被用作一种基于数据的方法，用于识别结构化金融数据集中的异常情况，从而能够检测到异常交易模式以及潜在系统故障的早期迹象，而无需依赖预先分类的输入数据。（Sattarov 等人，2022）。此外，高分位数回归技术的最新发展，如极端分位数自回归模型，提高了评估具有重尾分布特征金融时间序列尾部风险的能力。（He 等人，2023）。此外，方法的改进对于确保模型设计的清晰性和符合监管要求至关重要。 这突出了持续进行严格验证流程和建立清晰监管框架的必要性。
Market Forecasting with AI
人工智能市场预测
Market trend forecasting is one of the most active areas in data-driven finance, including lots of tasks such as stock price prediction, index movement forecasting and asset return estimation across diverse markets, which include equities, bonds, and cryptocurrencies. Since 2020, much research has demonstrated that advanced models frequently exceed traditional statistical approaches in predicting and exploring the market.
市场趋势预测是数据驱动金融中最活跃的领域之一，包括许多任务，如股票价格预测、指数运动预测和跨不同市场（包括股票、债券和加密货币）的资产回报估计。自 2020 年以来，大量研究表明，高级模型在预测和探索市场方面经常超越传统统计方法。
For instance, deep neural architectures such as RNNs, LSTMs, and Transformers have been successfully applied to stock market forecasting, Leippold, Wang, and Zhou (2022) come up with a comprehensive study that examined the Chinese equity market and developed a broad set of machine learning-based return predictors. Their findings revealed that, in a market that largely driven by retail investors, some models could exploit structural inefficiencies. Even after accounting for transaction costs, the machine learning-based strategies generated out-of-sample returns. Another emerging direction is the adaptation of hybrid and innovative model architecture. They combine different techniques to improve predictive performance. For example, convolutional neural networks (CNNs) are used for feature extraction, while transformer models are used to deal with sequence modeling. Xie et al. (2024) proposed a Deep Convolutional Transformer (DCT) that integrates CNN-based feature extraction with Transformer attention mechanisms to improve stock trend classification.
例如，深度神经网络架构如 RNN、LSTM 和 Transformer 已被成功应用于股票市场预测，Leippold、Wang 和 Zhou（2022）进行了一项全面研究，考察了中国股票市场并开发了一套基于机器学习的回报预测模型。他们的研究发现，在一个主要由散户投资者驱动的市场中，一些模型可以利用结构性低效。即使在考虑交易成本后，基于机器学习的策略也产生了样本外回报。另一个新兴方向是混合和创新模型架构的适应性。它们结合不同的技术来提高预测性能。例如，卷积神经网络（CNN）用于特征提取，而 Transformer 模型用于处理序列建模。Xie 等人（2024）提出了一种深度卷积 Transformer（DCT），该模型集成了基于 CNN 的特征提取与 Transformer 注意力机制，以改进股票趋势分类。
Gaps & Future Directions
差距与未来方向
Advanced quantitative techniques have demonstrated considerable promise in portfolio optimization, risk evaluation, and market forecasting; however, their implementation in real-time trading systems remains complex. Numerous models deliver satisfactory results under stable market conditions but often falter amid heightened volatility or financial crises, revealing limited resilience to extreme tail-risk events. (Stulz, 2008; Kolm et al., 2014). Currently, most approaches depend primarily on offline backtesting and do not possess genuine real-time decision-making capabilities, limiting their effectiveness in high-frequency environments or during sudden market disruptions. (Masini et al., 2023). Moreover, although techniques such as deep reinforcement learning have the potential to enhance investment performance, their opaque nature poses challenges for transparency, complicates regulatory oversight, and may diminish investor confidence. (Molnar et al., 2021). Finally, cost-aware portfolio frameworks offer potential in managing the trade-off between transaction costs and returns; however, their integration with robust optimization techniques continues to present a significant challenge. (Zhang et al., 2022).
先进的量化技术在投资组合优化、风险评估和市场预测方面已展现出巨大潜力；然而，将这些技术应用于实时交易系统仍然面临复杂挑战。许多模型在稳定市场条件下能取得令人满意的结果，但在高波动性或金融危机期间往往表现不佳，显示出对极端尾部风险事件适应力有限。（Stulz，2008；Kolm 等人，2014）。目前，大多数方法主要依赖离线回测，缺乏真正的实时决策能力，这限制了它们在高频环境或市场突然中断时的有效性。（Masini 等人，2023）。此外，尽管深度强化学习等技术有潜力提升投资表现，但其不透明性给透明度带来挑战，增加了监管监督的复杂性，并可能削弱投资者信心。（Molnar 等人，2021）。 最后，成本敏感的投资组合框架在管理交易成本与收益之间的权衡方面具有潜力；然而，将其与稳健的优化技术相结合仍然是一个重大挑战。（张等人，2022 年）。
Future research should focus on developing distributionally robust and cost-sensitive algorithms capable of mitigating the risks associated with extreme-event mischaracterization. Equally important is the design of end-to-end, low-latency learning systems that support continuous model adaptation to changes in market microstructure. Incorporating structured interpretability components can enhance oversight by risk managers and build trust among investors. Comprehensive validation across markets and asset classes—through both controlled simulations and real-time pilot implementations—is essential to ensure model reliability and effectiveness under varying economic conditions. Addressing these areas will support the creation of investment strategies that are both resilient under stress and transparent to all stakeholders.
未来的研究应专注于开发分布稳健且成本敏感的算法，以减轻极端事件误判相关的风险。同样重要的是设计端到端、低延迟的学习系统，支持持续适应市场微观结构的变化。引入结构化可解释性组件可以增强风险管理人员的管理监督能力，并在投资者之间建立信任。通过受控模拟和实时试点实施相结合的方式，在多个市场和资产类别上进行全面验证，对于确保模型在不同经济条件下的可靠性和有效性至关重要。解决这些领域将支持创建在压力下具有韧性且对所有利益相关者透明的投资策略。
Significance and Innovation
重要性与创新
Significance
意义
This project focuses on the application of AI in investment and asset management systems. It aims to improve the application of AI in areas such as asset allocation, market assessment and forecasting, and risk management. The increasing complexity of financial markets has made traditional methods ineffective in dealing with the flood of information and uncertainty. The data processing and analysis capabilities of artificial intelligence will drive changes in financial markets. Through this project, investors will be able to make better decisions, allocate assets more effectively, reduce investment risks, and improve investment efficiency and returns. Traditional asset management methods will be reshaped, driving financial technology to new heights and having a profound impact on the entire economic structure and the stability of capital markets. AI will play a greater role in optimising existing financial markets and addressing long-standing issues in the financial industry, such as transparency and data privacy, thereby enriching the content of this field. AI will be applied in more diverse scenarios, sparking interest in other disciplines such as computer science, statistics, and data science.
该项目聚焦于人工智能在投资和资产管理系统的应用。它旨在提升人工智能在资产配置、市场评估与预测、风险管理等领域的应用。金融市场的日益复杂使得传统方法在处理信息洪流和不确定性方面失效。人工智能的数据处理与分析能力将推动金融市场变革。通过该项目，投资者将能做出更优决策，更有效地配置资产，降低投资风险，提升投资效率与回报。传统资产管理方法将被重塑，推动金融科技迈向新高度，并对整个经济结构及资本市场稳定性产生深远影响。人工智能将在优化现有金融市场、解决金融业长期存在的透明度与数据隐私等问题中发挥更大作用，从而丰富该领域的内容。 AI 将应用于更多样化的场景，引发计算机科学、统计学和数据科学等其他学科的兴趣。
Innovation
创新
The innovative aspects of the project lie in the development of a series of new technologies and methods, which hold significant implications for leveraging AI technology to make effective investments in the highly uncertain environment of financial markets. The project will explore the integration of advanced AI technologies such as deep learning and reinforcement learning with traditional investment portfolio theories (e.g., Markowitz's portfolio theory), which will drive changes in asset allocation, market forecasting, and risk management approaches. This will enhance the effectiveness, robustness, and interpretability of financial models, facilitate the real-time and intelligent decision-making of high-frequency financial models, and help reduce risks. It will also explore how to utilise real-time AI models during market volatility or even financial crises, ensuring the adaptability and robustness of market models to address different financial market conditions. This theoretical-to-practical innovation will facilitate the development of various financial applications, including high-frequency trading, and address the shortcomings of existing models in handling extreme market conditions. Finally, further research will be conducted on the explainability of AI to establish a new model structure that is easier to regulate, thereby increasing investors' trust in machine model decisions. This innovation is not only reflected in financial models but also drives regulatory reforms across the entire financial industry. In summary, the innovation of the project should not only focus on advancements in methods and technologies but also involve systematic thinking on how to effectively address issues within the financial industry.
该项目的创新之处在于开发了一系列新技术和方法，这些技术在利用 AI 技术进行有效投资方面具有重要意义，尤其是在高度不确定的金融市场中。项目将探索深度学习和强化学习等先进 AI 技术与传统投资组合理论（例如马科维茨投资组合理论）的整合，这将推动资产配置、市场预测和风险管理方法的变革。这将提高金融模型的有效性、鲁棒性和可解释性，促进高频金融模型的实时智能决策，并有助于降低风险。此外，项目还将探索如何在市场波动甚至金融危机期间利用实时 AI 模型，确保市场模型对不同金融市场条件的适应性和鲁棒性。 这一从理论到实践的革新将促进各类金融应用的发展，包括高频交易，并解决现有模型在处理极端市场条件时的不足。最后，将针对 AI 的可解释性进行进一步研究，以建立一种更易于监管的模型结构，从而增强投资者对机器模型决策的信任。这一创新不仅体现在金融模型中，还推动整个金融行业的监管改革。总之，该项目的创新不仅应关注方法和技术的进步，还应包含系统性的思考，以有效应对金融行业内部的问题。
Proposal
提案
This study proposed a detailed exploration of the application of artificial intelligence in the fields of investment, which focused on the practical financial applications rather than the complex. Specifically, the research concentrated on four key areas, which are portfolio optimization, risk management, Future Directions, and market trend prediction. It also conducted a comprehensive practical analysis of the stock, bond, and cryptocurrency markets in China, Europe, and the United States.
本研究提出了对人工智能在投资领域应用的详细探索，重点在于实际金融应用而非复杂理论。具体而言，研究集中在四个关键领域，即投资组合优化、风险管理、未来方向和市场趋势预测。此外，研究还对中国、欧洲和美国股票、债券和加密货币市场进行了全面的实践分析。
Firstly, this study examines the practical effectiveness of recent artificial intelligence-based methods in portfolio optimization, which focus on the deep reinforcement learning (DRL) and hybrid predictive optimization frameworks. The study use the real-world dataset that obtained from existing financial databases (such as CRSP, Refinitiv, and cryptocurrency exchanges). Rigorous analyses were used on performance indicators such as Sharpe ratio, Sortino ratio, maximum drawdown, annualized return rate, and turnover rate. Through comparative analysis, the study systematically evaluated the outcomes of artificial intelligence-enhanced portfolios against traditional benchmarks (such as mean-variance optimization, equal-weighted portfolios, and industry standard indices).
首先，本研究考察了近期基于人工智能的方法在投资组合优化中的实际效果，这些方法主要关注深度强化学习（DRL）和混合预测优化框架。研究使用了从现有金融数据库（如 CRSP、Refinitiv 和加密货币交易所）获取的真实世界数据集。对夏普比率、索提诺比率、最大回撤、年化回报率和换手率等性能指标进行了严谨分析。通过比较分析，本研究系统地评估了人工智能增强型投资组合与传统基准（如均值-方差优化、等权重投资组合和行业标准指数）的成果。
Secondly, this study explores the role of artificial intelligence in risk management; in-depth case studies from leading institutions were adopted. At the same time, it will contribute to in-depth analysis of related challenges such as the transparency and interpretability of algorithms, and compliance with changing regulatory standards. Additionally, the study will explore ethical issues, which focus on biases in data-driven decision.
其次，本研究探讨了人工智能在风险管理中的作用；采用了来自领先机构的深入案例研究。同时，它将有助于深入分析算法的透明度和可解释性等相关挑战，以及遵守不断变化的监管标准。此外，研究将探讨伦理问题，这些问题聚焦于数据驱动决策中的偏见。
Thirdly, we trained long Short-Term memory (LSTM) networks and Transformer models using a vast amount of historical data covering the stock, fixed income, and commodity markets to evaluate them. We measure the prediction quality using root means square error (RMSE), mean absolute error (MAE), direction accuracy and information coefficient, and then directly compare these data with the results of the ARIMA and GARCH models. Next, we will back test the trading strategies based on these predictions to demonstrate how higher accuracy can optimize the entry and exit rules. The results demonstrate how more accurate predictions can optimize risk-adjusted portfolio allocation and provide asset managers with timely insights needed for decisive action
第三，我们使用大量涵盖股票、固定收益和商品市场的历史数据训练了长短期记忆（LSTM）网络和 Transformer 模型，以评估它们。我们使用均方根误差（RMSE）、平均绝对误差（MAE）、方向准确性和信息系数来衡量预测质量，然后将这些数据直接与 ARIMA 和 GARCH 模型的结果进行比较。接下来，我们将基于这些预测回测交易策略，以展示更高的准确度如何优化入场和出场规则。结果表明，更准确的预测可以优化风险调整后的投资组合配置，并为资产管理人提供及时洞察，以支持果断行动。
To integrate these quantitative research results with practical operations, we combine statistical work with qualitative evidence collected through structured interviews and questionnaires, which target industry experts, senior investment managers and risk analysts. We use their feedback to identify obstacles in actual operations - such as imperfect data access, heavy computing loads and constantly changing compliance requirements - and transform our conclusions into practical suggestions and best practice guidelines. Finally, we proposed a policy framework that can be adopted by regulatory authorities and stakeholders to encourage responsible innovation while controlling operational risks and systemic risks. By combining reliable data analysis with the insights of practitioners, we provide a clear roadmap for improving market forecasting and promoting lasting efficiency and innovation in the global financial services industry.
为了将这些定量研究结果与实际操作相结合，我们将统计工作与通过结构化访谈和问卷调查收集的定性证据相结合，这些访谈和问卷针对行业专家、高级投资经理和风险分析师。我们利用他们的反馈来识别实际操作中的障碍——例如不完善的数据访问、繁重的计算负载和不断变化的合规要求——并将我们的结论转化为实用建议和最佳实践指南。最后，我们提出了一个政策框架，监管机构和利益相关者可以采用该框架，以鼓励负责任的创新，同时控制操作风险和系统性风险。通过结合可靠的数据分析与从业者的见解，我们为改进市场预测和促进全球金融服务行业的持久效率与创新提供了清晰的路线图。
Reference
参考文献
[1] Chen, W, Zhang, H, Mehlawat, MK & Jia, L 2021, ‘Mean–variance portfolio optimization using machine learning-based stock price prediction’, Applied soft computing, vol. 100.
[2] Yan, R, Jin, J & Han, K 2024, ‘Reinforcement learning for deep portfolio optimization’, Electronic research archive, vol. 32, no. 9, pp. 5176–5200.
[2] Yan, R, Jin, J & Han, K 2024, ‘深度投资组合优化强化学习’, 电子研究档案, 第 32 卷, 第 9 期, pp. 5176–5200.
[3] Lucarelli, G & Borrotti, M 2020, ‘A deep Q-learning portfolio management framework for the cryptocurrency market’, Neural computing & applications, vol. 32, no. 23, pp. 17229–17244.
[3] Lucarelli, G & Borrotti, M 2020, ‘用于加密货币市场的深度 Q 学习投资组合管理框架’, 神经计算与应用, 第 32 卷, 第 23 期, pp. 17229–17244.
[4] Leippold, M, Wang, Q & Zhou, W 2022, ‘Machine learning in the Chinese stock market’, Journal of financial economics, vol. 145, no. 2, pp. 64–82.
[4] Leippold, M, Wang, Q & Zhou, W 2022, ‘中国股市中的机器学习’, 金融经济学杂志, 第 145 卷, 第 2 期, pp. 64–82.
[5] Xie, L, Chen, Z & Yu, S 2024, ‘Deep Convolutional Transformer Network for Stock Movement Prediction’, Electronics (Basel), vol. 13, no. 21, p. 4225.
[5] Xie, L, Chen, Z & Yu, S 2024, ‘用于股票运动预测的深度卷积 Transformer 网络’, 电子学(巴塞尔), 第 13 卷, 第 21 期, p. 4225.
[6] Choudhary, H, Orra, A, Sahoo, K & Thakur, M 2025, ‘Risk-Adjusted Deep Reinforcement Learning for Portfolio Optimization: A Multi-reward Approach’, International journal of computational intelligence systems, vol. 18, no. 1.
[6] Choudhary, H, Orra, A, Sahoo, K & Thakur, M 2025, ‘风险调整深度强化学习在投资组合优化中的应用：一种多奖励方法’, 国际计算智能系统杂志, 第 18 卷, 第 1 期。
[7] Chang, V, Sivakulasingam, S, Wang, H, Wong, ST, Ganatra, MA & Luo, J 2024, ‘Credit Risk Prediction Using Machine Learning and Deep Learning: A Study on Credit Card Customers’, Risks (Basel), vol. 12, no. 11, p. 174.
[7] Chang, V, Sivakulasingam, S, Wang, H, Wong, ST, Ganatra, MA & Luo, J 2024, ‘利用机器学习和深度学习进行信用风险预测：一项关于信用卡客户的研究’, 风险 (巴塞尔), 第 12 卷, 第 11 期, 第 174 页。
[8] Iacopini, M, Poon, A, Rossini, L & Zhu, D 2023, ‘Bayesian Mixed-Frequency Quantile Vector Autoregression: Eliciting tail risks of Monthly US GDP’, Journal of economic dynamics & control, vol. 157.
[8] Iacopini, M, Poon, A, Rossini, L & Zhu, D 2023, ‘贝叶斯混合频率分位数向量自回归：提取月度美国 GDP 的尾部风险’, 经济动力学与控制杂志, 第 157 卷。
[9] Sattarov, T, Herurkar, D & Hees, J 2022, ‘Explaining Anomalies using Denoising Autoencoders for Financial Tabular Data’.
[9] Sattarov, T, Herurkar, D & Hees, J 2022, ‘使用去噪自编码器解释金融表格数据的异常现象’.
[10] He, F, Wang, HJ & Zhou, Y 2022, ‘Extremal quantile autoregression for heavy-tailed time series’, Computational statistics & data analysis, vol. 176.
[10] He, F, Wang, HJ & Zhou, Y 2022, ‘极值分位数自回归模型在重尾时间序列中的应用’, 计算机统计与数据分析, 卷. 176.
[11] Stulz, R. M. (2008). Risk management failures: What are they and when do they happen? Journal of Applied Corporate Finance, 20(4), 39–48. DOI:10.1111/j.1745-6622.2008.00202.x
[11] Stulz, R. M. (2008). 风险管理失败：它们是什么以及何时发生？应用公司金融杂志, 20(4), 39–48. DOI:10.1111/j.1745-6622.2008.00202.x
[12] Kolm, PN, Tütüncü, R & Fabozzi, FJ 2014, ‘60 Years of portfolio optimization: Practical challenges and current trends’, European journal of operational research, vol. 234, no. 2, pp. 356–371.
[12] Kolm, PN, Tütüncü, R & Fabozzi, FJ 2014, ‘60 年投资组合优化：实践挑战与当前趋势’, 欧洲运筹学杂志, 卷. 234, 第 2 期, pp. 356–371.
[13] Masini, RP, Medeiros, MC & Mendes, EF 2023, ‘Machine learning advances for time series forecasting’, Journal of economic surveys, vol. 37, no. 1, pp. 76–111.
[13] Masini, RP, Medeiros, MC & Mendes, EF 2023, ‘机器学习在时间序列预测中的进展’, 经济学评论, 卷. 37, 第 1 期, pp. 76–111.
[14] Molnar, C, Casalicchio, G & Bischl, B 2020, ‘Interpretable Machine Learning -- A Brief History, State-of-the-Art and Challenges’.
[14] Molnar, C, Casalicchio, G & Bischl, B 2020, ‘可解释机器学习——简史、现状与挑战’.
[15] Zhang, Y, Zhao, P, Wu, Q, Li, B, Huang, J & Tan, M 2020, ‘Cost-Sensitive Portfolio Selection via Deep Reinforcement Learning,.
[15] 张, Y, 赵, P, 吴, Q, 李, B, 黄, J & 谭, M 2020, ‘基于深度强化学习的成本敏感投资组合选择,.