Loading...
|
Please use this identifier to cite or link to this item:
https://nccur.lib.nccu.edu.tw/handle/140.119/158044
|
| Title: | 以貝氏網路建構財務之有限理性模型
On Modelling Bounded Rationality in Finance with Bayesian Networks |
| Authors: | 何靜嫺 |
| Contributors: | 經濟系 |
| Date: | 2017-06 |
| Issue Date: | 2025-07-16 11:12:36 (UTC+8) |
| Abstract: | 本計畫欲以『貝氏網路架構』建立一系列財務上之有限理性模型. 貝氏網路為一機率及圖像化模型, 用以描述一群變數間的相關性及因果關係. Spiegler(2014)首倡以Pearl(2000)提出的包含因果結構的貝氏網路來分析經濟學上的認知偏誤. 本計畫以三年時間, 將貝氏網路架構運用於分析『投資者有限理性行為在財務』的影響.第一年計畫首先以貝氏網路架構重新檢視現存有關『財務有限理性行為』的文章. 重新檢視的目的有二：（1）由於現有理論多數為偏好導向模型（承襲展望理論）, 而投資者異質化的主因是因為投資者間的先驗分配不同. 然而根據學習理論, 若異質性來自於先驗分配的不同, 則有可能在學習的過程中逐漸趨於一致, 但如Hirshleifer(2014)所提, 認知偏誤很少會因為學習而消失. 因此貝氏網路架構相較於傳統偏好導向模型更能處理此類頑強性的偏誤行為的影響. 我們想看到這對現有理論的改變為何. (2)第一年針對各類認知偏誤所得出的基本貝氏網路模型, 可以將其運用於分析認知偏誤對資產定價模型的影響及偏誤傳播的分析.第二年計畫將第一年所建立各類認知偏誤的基本貝氏網路模型運用於分析各類認知偏誤對『資產定價模型』的影響. 本文以貝氏網路架構重新檢視現存有關不對稱訊息下的資產定價模型（以Brunnermeier (2001)為主,佐以其他資產定價文獻). 目的在於探討當認知偏誤來自於對結構的不同看法時, 頑強性的偏誤行為對現有資產定價理論的結果之影響.第三年計畫將結合第一年所建立各類認知偏誤所得出的基本貝氏網路模型以及『社群網路』理論, 來分析認知偏誤之演變及傳播. 此分析對有關投資情緒傳播的探討（特別是恐慌行為）, 提供一可模擬的探討架構. 最後, 由於傳統數據挖掘理論以尋找單一最適結構為宗旨, 但一旦我們考慮決策者對決策結構的認知可能因為偏誤而不一致(top-down), 那就很難說服為何我們必須在資料中搜尋單一的決策結構(bottom-up）. 因此本計畫最後將探討有限理性對數據挖掘理論的影響.
This three-year project develops a series of Bayesian network models to analyze the impacts of bounded rationality in finance. A Bayesian network is a probabilistic graphical model that represents a set of random variables and their conditional dependencies via a directed acyclic graph. Spiegler (2014) illustrates that the Bayesian networks proposed by Pearl (2000), which advocates the view of Bayesian network as a causal structure and uses the graph-theoretic representation to visualize causal relations and systematize reasoning about causality, can capture several cognition biases. I will use this approach to model boundedly rational behaviors in finance. In year-one project, I will use the basic Bayesian network model to remodel the existing discussions on the impacts of cognition biases in finance, including models on neglect of strategic motives, narrow framing and heuristic learning, investor sentiment and overconfidence, category learning, reinforcement learning, and limited attention. Remodeling the existing models with Bayesian networks has two meanings. First, since Bayesian network approach assumes that agents’ heterogeneity comes from their incomplete knowledge about the environment (rather than the conventional assumption of uncommon priors on preferences), we can expect that remodeling with Bayesian networks can predict different results from the existing preference-based models. Second, remodeling the existing models gives us a uniform representation for various cognition biases, which then can be used to further discuss their impacts on asset pricing or bias transmission generally. In year-two project, I use this basic Bayesian network model to remodel the asset pricing model under asymmetric information (see Brunnermeier, 2001; Cao et al.,2011), and address the impacts of various cognition biases on asset pricing. Different from the existing models which assume that investors’ heterogeneity comes from non-common priors, our framework assumes that heterogeneity comes from investors’ different knowledge about the structure and using this incomplete knowledge to forecast prices (see Eyster and Piccione, 2013). Heterogeneity from uncommon prior beliefs could fade in the dynamic updating process, while investors can persistently neglect the relevance of certain observable variables, still believing that they are statistically correct. In year-three project, I combine this basic Bayesian network model with social networks (DeGroot, 1974) to study investors’ bias transmissions and amplification. According to Hirshleifer (2014), analysis of social interactions promises to provide greater insight into where heuristics come from and understanding how financial ideas spread from person to person may eventually suggest theories of how investment and corporate ideologies evolve. For a better picture of the transmission processes, simulations using software such as Bayes Server will be provided. Finally, to conclude this series of researches, I will address data mining using Bayesian networks equipped with investors’ bounded rationality. |
| Relation: | 科技部, MOST104-2410-H004-040, 104.08-105.07 |
| Data Type: | report |
| Appears in Collections: | [經濟學系] 國科會研究計畫
|
Files in This Item:
| File |
Description |
Size | Format | |
|---|
| index.html | | 0Kb | HTML | 2 | View/Open |
|
All items in 政大典藏 are protected by copyright, with all rights reserved.
|
