東華大學圖書館 |

Deep Learning for Predicting Real-World Outcomes /

Record Type:	Electronic resources : Monograph/item
Title/Author:	Deep Learning for Predicting Real-World Outcomes // Zikai Wei.
Author:	Wei, Zikai,
Description:	1 electronic resource (181 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 86-08, Section: B.
Contained By:	Dissertations Abstracts International86-08B.
Subject:	Deep learning. -
Online resource:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31869222
ISBN:	9798304975216

Deep Learning for Predicting Real-World Outcomes /
Wei, Zikai,

Deep Learning for Predicting Real-World Outcomes /Zikai Wei. - 1 electronic resource (181 pages)

Source: Dissertations Abstracts International, Volume: 86-08, Section: B.

This thesis addresses various real-world prediction problems, including predicting human trajectories and financial market trends in the domains of team sports, pedestrian prediction, and intelligent investment. Given that these topics are intrinsically linked to human decision-making, we first study how humans make decisions in collaborative and competitive environments. This led us to explore a new approach for human trajectory prediction. We then apply this approach to financial markets, which involve a larger number of human participants, and develop specific models to solve classical investment questions that are closely related to prediction problems. Our ultimate goal is to provide useful insights into human trajectory prediction and intelligent investment, and our research strives to achieve this objective. In this thesis, we first propose to infer the future trajectories of multiple agents by characterizing the game process between different interest groups. To this end, a variational-based sequential modeling method is pro-posed to solve the human trajectory prediction problem. For accurate trajectory prediction in multi-agent settings, e.g., team games, it is important to effectively model the interactions among agents. Whereas a number of methods have been developed for this purpose, existing methods implicitly model these interactions as part of the deep net architecture. However, in the real world, interactions often exist at multiple levels, e.g., individuals may form groups, where interactions among groups and those among the individuals in the same group often follow significantly different patterns. In this part, we present a novel formulation for multiagent trajectory prediction, which explicitly introduces the concept of interactive group consensus over an interactive hierarchical latent space. This formulation allows group-level and individual-level interactions to be jointly captured, thus substantially improving the capability of modeling complex dynamics. In two multi-agent settings, i.e., team sports and pedestrians, the proposed framework consistently outperforms existing methods.The second part delivers a deep multi-factor model for factor investing. Modeling and characterizing multiple factors is perhaps the most important step in achieving excess returns over market benchmarks. Both academia and industry are striving to find new factors that have good explanatory power for future stock returns and good stability of their predictive power. In practice, factor investing is still largely based on linear multifactor models, although many deep learning methods show promising results in predicting stock trends and managing portfolio risks compared to traditional methods. However, the existing non-linear methods have two drawbacks: 1) there is a lack of interpretation of the newly discovered factors, 2) the financial in-sights behind the mining process are unclear, making practitioners reluctant to apply the existing methods to factor investing. To address these two short-comings, we develop a novel deep multi-factor model using industry neutralization and market neutralization modules with clear financial insights, which help us casely build a dynamic and multi-related stock graph in a hierarchical structure to learn the graph representation of stock relationships at different levels, e.g., industry level and universal level. Then, graph-attention modules are used to estimate a series of deep factors that maximize cumulative factor returns. Further, a factor-attention module is developed to approximate the estimated deep factors from the input factors so that the deep factors can be interpreted explicitly. Extensive experiments on real stock market data demonstrate the effectiveness of our deep multi-factor model in the factor investing task. In the third part of this thesis, we present an end-to-end and full-stack intelligent investment framework. Active investing aims to construct a port-folio of assets that are believed to be relatively profitable in the markets. One popular method is to construct a portfolio using factor-based strategies. In recent years, there have been increasing efforts to apply deep learning to find "deep factors" with more active returns or promising pipelines for predicting asset trends. However, the question of how to construct an active investment portfolio using an end-to-end deep learning (E2E) framework remains open and is rarely addressed in existing work. In this work, we are the first to pro-pose an E2E that covers almost the entire factor investment process through factor selection, factor combination, and portfolio construction. Extensive experiments on real stock market data demonstrate the effectiveness of our end-to-end deep learning framework for active investing. In the fourth part of this thesis, we explore the regime switching problem and how it can help build a regime-specific factor model. Factor model is a fundamental investment tool in quantitative investment, which can be empowered by deep learning to become more flexible and efficient in practical complicated investment situations. However, it is still an open question to build a factor model that can conduct stock prediction in an online and adaptive setting, where the model can adapt itself to match current market regime identified based on only point-in-time market information. To address this problem, we propose the first deep learning based online and adaptive factor model, HireVAE, at the core of which is a hierarchical latent space that embeds the underlying relationship between the market global situation and stock-specific latent factors, so that HireVAE can effectively estimate useful latent factors and then predict accurate stock returns when historical market information is available. On four commonly used real stock market benchmarks, the proposed HireVAE shows superior performance in terms of active returns compared to previous methods, indicating the potential usefulness of such online and adaptive factor models. Our research achieves the following goals: First, we introduce the concept of team games into human trajectory prediction, enabling the consideration of high-level interactions between human agents. Second, we apply deep learning to solve prediction problems in financial markets by providing a deep and interpretable multifactor model for factor investing that goes be-yond classical linear models. Our model incorporates interactions among different stocks at multiple levels, improving its capability to model complex dynamics. Third, we introduce the first end-to-end deep learning framework for active investing that covers the entire process from factor selection to port-folio construction. We develop a directional loss to ensure the consistency of the deep factor learned during the multi-task learning process with different objectives and procedures. Finally, we design an online regime switching algorithm and develop a regime-specific model for each market regime. Over-all, our research provides valuable insights into human trajectory prediction and intelligent investment, showcasing the potential of deep learning to solve prediction problems related to human decision-making.

English

ISBN: 9798304975216Subjects--Topical Terms:

3554982
Deep learning.
Subjects--Index Terms:

Information engineering

Deep Learning for Predicting Real-World Outcomes /
LDR:14793nmm a22003973i 4500 001 2437710
005 20251124055001.5
006 m o d
007 cr|nu||||||||
008 261011s2023 miu||||||m |||||||eng d
020 $a 9798304975216
035 $a (MiAaPQD)AAI31869222
035 $a AAI31869222
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Wei, Zikai, $e author. $3 3828784
245 1 0 $a Deep Learning for Predicting Real-World Outcomes / $c Zikai Wei.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2023
300 $a 1 electronic resource (181 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 86-08, Section: B.
500 $a Advisors: Tang, Xiaoou.
502 $b Ph.D. $c The Chinese University of Hong Kong (Hong Kong) $d 2023.
520 $a This thesis addresses various real-world prediction problems, including predicting human trajectories and financial market trends in the domains of team sports, pedestrian prediction, and intelligent investment. Given that these topics are intrinsically linked to human decision-making, we first study how humans make decisions in collaborative and competitive environments. This led us to explore a new approach for human trajectory prediction. We then apply this approach to financial markets, which involve a larger number of human participants, and develop specific models to solve classical investment questions that are closely related to prediction problems. Our ultimate goal is to provide useful insights into human trajectory prediction and intelligent investment, and our research strives to achieve this objective. In this thesis, we first propose to infer the future trajectories of multiple agents by characterizing the game process between different interest groups. To this end, a variational-based sequential modeling method is pro-posed to solve the human trajectory prediction problem. For accurate trajectory prediction in multi-agent settings, e.g., team games, it is important to effectively model the interactions among agents. Whereas a number of methods have been developed for this purpose, existing methods implicitly model these interactions as part of the deep net architecture. However, in the real world, interactions often exist at multiple levels, e.g., individuals may form groups, where interactions among groups and those among the individuals in the same group often follow significantly different patterns. In this part, we present a novel formulation for multiagent trajectory prediction, which explicitly introduces the concept of interactive group consensus over an interactive hierarchical latent space. This formulation allows group-level and individual-level interactions to be jointly captured, thus substantially improving the capability of modeling complex dynamics. In two multi-agent settings, i.e., team sports and pedestrians, the proposed framework consistently outperforms existing methods.The second part delivers a deep multi-factor model for factor investing. Modeling and characterizing multiple factors is perhaps the most important step in achieving excess returns over market benchmarks. Both academia and industry are striving to find new factors that have good explanatory power for future stock returns and good stability of their predictive power. In practice, factor investing is still largely based on linear multifactor models, although many deep learning methods show promising results in predicting stock trends and managing portfolio risks compared to traditional methods. However, the existing non-linear methods have two drawbacks: 1) there is a lack of interpretation of the newly discovered factors, 2) the financial in-sights behind the mining process are unclear, making practitioners reluctant to apply the existing methods to factor investing. To address these two short-comings, we develop a novel deep multi-factor model using industry neutralization and market neutralization modules with clear financial insights, which help us casely build a dynamic and multi-related stock graph in a hierarchical structure to learn the graph representation of stock relationships at different levels, e.g., industry level and universal level. Then, graph-attention modules are used to estimate a series of deep factors that maximize cumulative factor returns. Further, a factor-attention module is developed to approximate the estimated deep factors from the input factors so that the deep factors can be interpreted explicitly. Extensive experiments on real stock market data demonstrate the effectiveness of our deep multi-factor model in the factor investing task. In the third part of this thesis, we present an end-to-end and full-stack intelligent investment framework. Active investing aims to construct a port-folio of assets that are believed to be relatively profitable in the markets. One popular method is to construct a portfolio using factor-based strategies. In recent years, there have been increasing efforts to apply deep learning to find "deep factors" with more active returns or promising pipelines for predicting asset trends. However, the question of how to construct an active investment portfolio using an end-to-end deep learning (E2E) framework remains open and is rarely addressed in existing work. In this work, we are the first to pro-pose an E2E that covers almost the entire factor investment process through factor selection, factor combination, and portfolio construction. Extensive experiments on real stock market data demonstrate the effectiveness of our end-to-end deep learning framework for active investing. In the fourth part of this thesis, we explore the regime switching problem and how it can help build a regime-specific factor model. Factor model is a fundamental investment tool in quantitative investment, which can be empowered by deep learning to become more flexible and efficient in practical complicated investment situations. However, it is still an open question to build a factor model that can conduct stock prediction in an online and adaptive setting, where the model can adapt itself to match current market regime identified based on only point-in-time market information. To address this problem, we propose the first deep learning based online and adaptive factor model, HireVAE, at the core of which is a hierarchical latent space that embeds the underlying relationship between the market global situation and stock-specific latent factors, so that HireVAE can effectively estimate useful latent factors and then predict accurate stock returns when historical market information is available. On four commonly used real stock market benchmarks, the proposed HireVAE shows superior performance in terms of active returns compared to previous methods, indicating the potential usefulness of such online and adaptive factor models. Our research achieves the following goals: First, we introduce the concept of team games into human trajectory prediction, enabling the consideration of high-level interactions between human agents. Second, we apply deep learning to solve prediction problems in financial markets by providing a deep and interpretable multifactor model for factor investing that goes be-yond classical linear models. Our model incorporates interactions among different stocks at multiple levels, improving its capability to model complex dynamics. Third, we introduce the first end-to-end deep learning framework for active investing that covers the entire process from factor selection to port-folio construction. We develop a directional loss to ensure the consistency of the deep factor learned during the multi-task learning process with different objectives and procedures. Finally, we design an online regime switching algorithm and develop a regime-specific model for each market regime. Over-all, our research provides valuable insights into human trajectory prediction and intelligent investment, showcasing the potential of deep learning to solve prediction problems related to human decision-making.
520 $a 本論文探討了各種真實世界的預測問題,包括在團隊運動、行人軌跡的預測問題以及智能投資領域的各類預測問題。由於這些主題與人類決策密切相關,我們首先研究了人類在合作和競爭環境中如何做出決策。這促使我們先從人類軌跡預測出發。探索新的軌跡預測方法。然後,我們將這種方法應用於涉及更多人類參與者的系統裡,即金融市場,並針對與預測問題相關的經典投資問題的開發了特定的模型。我們的最終目標是為人類軌跡預測和智能投資提供有用的見解,並解決現實場景下遇到的實際問題。在本論文中,我們首先提出通過描述不同利益團體之間的博弈過程來推斷多個代理人的未來軌跡。為此,提出了一種基於變分連續序列的建模方法來解決人類軌跡預測問題。為了能在在多個代理人的環境中準確預測他們的未來軌跡,我們認為有效建模代理人之間的博弈過程來刻畫代理人之間的相互影響至圖重要。儘管現有研究已經開發了許多方法來建模代理人之間的相互影響,但現有方法通常將這些互動作為深度網絡架構的一部分隱式表達,直接基於他們的互動建模。然而,在現實世界中,互動往往存在於多個層次,例如個體可能形成團體,團體間的互動和團體內的個體間的互動往往呈現出顯著不同的模式。為了更好的刻畫不同模式,我們提出了一種新的多代理人軌跡預測方法:通過在個體及團體多個層次建模,並用博弈來刻畫團體之間的互動。這種方法允許同時捕捉團體級和個體級互動,從而大大提高了建模複雜動態的能力。在兩個多代理設置中,即團隊運動和行人預測場景下,我們所提出的框架與現有方法相比始終具有卓越的性能。在第二部分,我們提出了深度多因子模型,用於因子投資。通過對多個因子對股票未来收益進行建模是實現超過市場基準的超额收益的重要步驟。學術界和業界都在努力尋找具有良好解釋力同時具有穩定預測股票未來回報能力的新因子。在實踐中,因子投資仍然主要基於線性多因子模型,儘管許多深度學習方法在預測股票趨勢和管理組合風險方面相對於傳統方法表現出更有前景的結果。然而,現有非線性方法存在兩個缺陷: 其一,對新發現因子的可解釋性不足;其二,為挖掘過程背後的金融含義不清晰,使得實踐者不願將現有方法應用於因子投資。為解決這兩個不足,我們開發了一種新穎的深度多因子模型,該模型使用行業中性和市場中性模塊並具有明確的金融含義,這有助於我們在層次結構中輕鬆構建動態和多關聯股票圖以學習不同層次的股票關係的圖表示,例如,行業層次和普遍層次。然後,使用圖注意力模塊來估計一系列最大化累積因子收益的深度因子。並開發了一個因子關注模塊,用於從輸入因子中近似估計的深度因子,以便明確解釋深度因子。在真實股票市場數據上的廣泛實驗證明了我們的深度多因子模型在因子投資任務中的有效性。在本論文的第三部分,我們提出了一個端到端的全機智能投資框架。積極投資旨在構建一個相對市場有益的資產組合。一種流行的方法是使用基於因子的策略構建投資組合。近年來, 越來越多的努力應用深度學習來找到具有更多積極收益的「深度因子」或有前景的資產趨勢預測管道。然而,如何使用端到端深度學習(E2E)框架構建積極投資組合的問題仍然懸而未決,並且在現有的主動投資研究中很少被提及。在這項工作中, 我們首次提出了一個端深度學習投資框架,幾乎涵蓋了整個因子投資過程,包括因子選擇,因子組合和投資組合構建。在真實股票市場數據上,廣泛的實驗證明了我們的端到端深度學習框架在積極投資中的有效性。在本論文的第四部分,我們探討機制轉換問題以及如何透過它建立一個適應各類市場機制的因子模型。因子模型是量化投資中的一個基本投資工具,可以通過深度學習為其賦能,使其在實際復雜的投資情況下變得更靈活和高效。然而,建立一個可以同時滿足在線學習並能根據市場機制切換自適應地進行股票預測的因子模型,仍然是一個懸而未決的問題。其實就是, 如何讓模型可以根據僅當時市場信息確定的當前市場機制並能自我調整。為了解決這個問題,我們提出了首個基於深度學習框架的在線及自適應因子模型(簡稱HireVAE) HireVAE的核心是在多層隱變量的框架下,構建了市場隱式因子(全局信息) 和股票隱式因子之間的基本關係。這使得它僅根據歷史市場信总就能有效地估計有用的隱式因子,並能更準確地預測股票收益。在四個常用的股票市場基準上,我們的方法在主動回報方而表現出色,即其表現均優於之前的方法。這凸顯了這種在線自適應因子模型的潛力。綜上所述,我們的研究實現了以下目標:首先,我們將博弈的概念引入人類軌跡預測,使得能夠考慮人類問的高層次的博弈互動。其次,我們應用深度學習解決金融市場的預測問題。具體就是為因子投資提供一個能夠超越經典線型模型且具有可解釋性的深度多因子模型。我們的模型在多層關係上考慮了股票間的相互作用,提高了該模型處理復雜關係的能力。再次, 我們提出了首個涵蓋從因子選擇到組合建構的整個過程的端到端深度學習投資框架。我們開發了一個方向性損失函數,以確保在多任務學習過程中保證深度因子與股票未來收益的關係的一致性。最後,我們設計了一個在線學習的機制轉換算法。該算法能識別不同的市場機制,並為每個市場機制學習特定的因子模型。總之,我們的研究為人類軌跡預測和智能投资提供了寶貴的見解,展示了深度學習在解決與人類決策相關的預測問題的潛力。.
546 $a English
590 $a School code: 1307
650 4 $a Deep learning. $3 3554982
650 4 $a Investments. $3 566987
650 4 $a Team sports. $3 3311497
650 4 $a Graph representations. $3 3560730
650 4 $a Computer engineering. $3 621879
653 $a Information engineering
653 $a Deep learning
653 $a Real-world outcomes
690 $a 0464
710 2 $a The Chinese University of Hong Kong (Hong Kong). $e degree granting institution. $3 3765695
720 1 $a Tang, Xiaoou $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-08B.
790 $a 1307
791 $a Ph.D.
792 $a 2023
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31869222