語系:
繁體中文
English
說明(常見問題)
回圖書館首頁
手機版館藏查詢
登入
回首頁
切換:
標籤
|
MARC模式
|
ISBD
Deep Residual Learning for Portfolio...
~
Wang, Jifei.
FindBook
Google Book
Amazon
博客來
Deep Residual Learning for Portfolio Optimization: With Attention and Switching Modules.
紀錄類型:
書目-電子資源 : Monograph/item
正題名/作者:
Deep Residual Learning for Portfolio Optimization: With Attention and Switching Modules./
作者:
Wang, Jifei.
出版者:
Ann Arbor : ProQuest Dissertations & Theses, : 2019,
面頁冊數:
99 p.
附註:
Source: Dissertation Abstracts International, Volume: 80-06(E), Section: B.
Contained By:
Dissertation Abstracts International80-06B(E).
標題:
Applied mathematics. -
電子資源:
http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=13423386
ISBN:
9780438815148
Deep Residual Learning for Portfolio Optimization: With Attention and Switching Modules.
Wang, Jifei.
Deep Residual Learning for Portfolio Optimization: With Attention and Switching Modules.
- Ann Arbor : ProQuest Dissertations & Theses, 2019 - 99 p.
Source: Dissertation Abstracts International, Volume: 80-06(E), Section: B.
Thesis (Ph.D.)--New York University Tandon School of Engineering, 2019.
Deep learning recently gained considerable attention in the financial econometric community. This thesis studies model-driven deep learning (DL) methodologies for portfolio optimization in the US equities market.
ISBN: 9780438815148Subjects--Topical Terms:
2122814
Applied mathematics.
Deep Residual Learning for Portfolio Optimization: With Attention and Switching Modules.
LDR
:04010nmm a2200361 4500
001
2201396
005
20190429062348.5
008
201008s2019 ||||||||||||||||| ||eng d
020
$a
9780438815148
035
$a
(MiAaPQ)AAI13423386
035
$a
(MiAaPQ)poly:10424
035
$a
AAI13423386
040
$a
MiAaPQ
$c
MiAaPQ
100
1
$a
Wang, Jifei.
$3
3428115
245
1 0
$a
Deep Residual Learning for Portfolio Optimization: With Attention and Switching Modules.
260
1
$a
Ann Arbor :
$b
ProQuest Dissertations & Theses,
$c
2019
300
$a
99 p.
500
$a
Source: Dissertation Abstracts International, Volume: 80-06(E), Section: B.
500
$a
Adviser: Deane Yang.
502
$a
Thesis (Ph.D.)--New York University Tandon School of Engineering, 2019.
520
$a
Deep learning recently gained considerable attention in the financial econometric community. This thesis studies model-driven deep learning (DL) methodologies for portfolio optimization in the US equities market.
520
$a
In the first part, we develop a novel DL paradigm that integrates deep residual network with attention mechanism for portfolio construction. Our residual learning framework exhibits substantial depth at tens of hidden layers. When applying complex non-linear models on noisy financial data, model performance is often confronted by the over-fitting obstacle. We demonstrate that over-fitting can be controlled with deep residual learning. We further incorporate attention mechanisms to complete our attention-enhanced residual network (attention ResNet) with powerful predictive properties.
520
$a
We validate the attention enhanced residual network empirically. For each ordinary share over the broad universe of US equities, we estimate the probability of next month's return being either high or low. The portfolio is constructed by long the "winners" with high predicted probability and short the "losers" with low predicted probability. Over the period of 2008--H12017, the Attention ResNet strategy verified superior out-of-sample performance with an average annual Sharpe ratio of 1.77, compared with average annual Sharpe ratio of 0.81 for the ANN-based strategy and 0.69 for the linear model. Experimental results demonstrate that the deep plain ANN over-fitted the financial data, the linear model under-fitted the financial data, whereas our proposed network is robust to over-fitting and able to optimize the degree of non-linearity in the model.
520
$a
The second part of the thesis presents a novel residual switching network architecture which combines two separate residual networks, namely a switching module that computes a condition weight mask, and another residual network that learns latent representation of momentum and reversal features. Input vectors for the switching module are stock market condition features such as squared VIX, realized volatility and variance risk premium for Standard and Poors 500 index. The switching module can automatically sense changes in stock market regimes and guide the proposed DL framework to switch between the two prominent stock market anomalies, i.e. momentum and reversal.
520
$a
Overall, momentum features are better predictors when market is bullish, and reversal features are more appropriate when market regime is bearish (Cooper, Gutierrez and Hameed 2004). Our main result shows that the behavior of the switching module is in excellent agreement with changes in market regimes. During periods of higher market volatility (bearish market), the switching module's condition weight mask concentrates on reversal latent representations. Whereas for periods of lower market volatility (bullish market), the condition weight mask switches concentration to momentum latent representations.
590
$a
School code: 1988.
650
4
$a
Applied mathematics.
$3
2122814
650
4
$a
Artificial intelligence.
$3
516317
650
4
$a
Finance.
$3
542899
690
$a
0364
690
$a
0800
690
$a
0508
710
2
$a
New York University Tandon School of Engineering.
$b
Mathematics.
$3
3428116
773
0
$t
Dissertation Abstracts International
$g
80-06B(E).
790
$a
1988
791
$a
Ph.D.
792
$a
2019
793
$a
English
856
4 0
$u
http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=13423386
筆 0 讀者評論
館藏地:
全部
電子資源
出版年:
卷號:
館藏
1 筆 • 頁數 1 •
1
條碼號
典藏地名稱
館藏流通類別
資料類型
索書號
使用類型
借閱狀態
預約狀態
備註欄
附件
W9377945
電子資源
11.線上閱覽_V
電子書
EB
一般使用(Normal)
在架
0
1 筆 • 頁數 1 •
1
多媒體
評論
新增評論
分享你的心得
Export
取書館
處理中
...
變更密碼
登入