Topological data analysis in investment decisions

•We explore the applications of topological data analysis in investment decision making.•A new strategy based on L1 norms of persistence landscape is proposed to filter assets.•Extensive empirical analysis is performed on assets from economies across the world.•Enhanced indexing models are solved to...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Expert systems with applications Jg. 147; S. 113222
Hauptverfasser: Goel, Anubha, Pasricha, Puneet, Mehra, Aparna
Format: Journal Article
Sprache:Englisch
Veröffentlicht: New York Elsevier Ltd 01.06.2020
Elsevier BV
Schlagworte:
Algorithms
Asset allocation
Data analysis
Datasets
Decision analysis
Empirical analysis
Enhanced indexing
Financing
Homology
Investment
Mathematical functions
Norms
Optimization
Persistence landscape
Portfolio selection
Risk
Standard deviation
Takens’ embedding
Three dimensional models
Time dependence
Time series
Topological data analysis
Topology
Portfolio selection
Enhanced indexing
Topological data analysis
Takens’ embedding
Persistence landscape
ISSN:0957-4174, 1873-6793
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:•We explore the applications of topological data analysis in investment decision making.•A new strategy based on L1 norms of persistence landscape is proposed to filter assets.•Extensive empirical analysis is performed on assets from economies across the world.•Enhanced indexing models are solved to obtain optimal portfolios from the filtered assets.•Comparative studies show that the proposed strategy leads to more robust portfolios. This article explores the applications of Topological Data Analysis (TDA) in the finance field, especially addressing the primordial problem of asset allocation. Firstly, we build a rationale on why TDA can be a better alternative to traditional risk indicators such as standard deviation using real data sets. We apply Takens embedding theorem to reconstruct the time series of returns in a high dimensional space. We adopt the sliding window approach to draw the time-dependent point cloud data sets and associate a topological space with them. We then apply the persistent homology to discover the topological patterns that appear in the multidimensional time series. The temporal changes in the persistence landscapes, which are the real-valued functions that encode the persistence of topological patterns, are captured via Lp norm. The time series of the Lp norms shows that it is better at measuring the dynamics of returns than the standard deviation. Inspired by our findings, we explore an application of TDA in Enhanced Indexing (EI) that aims to build a portfolio of fewer assets than that in the index to outperform the latter. We propose a two-step procedure to accomplish this task. In step one, we utilize the Lp norms of the assets to propose a filtration technique of selecting a few assets from a larger pool of assets. In step two, we propose an optimization model to construct an optimal portfolio from the class of filtered assets for EI. To test the efficiency of this enhanced algorithm, experiments are carried out on ten data sets from financial markets across the globe. Our extensive empirical analysis exhibits that the proposed strategy delivers superior performance on several measures, including excess mean returns from the benchmark index and tail reward-risk ratios than some of the existing models of EI in the literature. The proposed filtering strategy is also noted to be beneficial for both risk-seeking and risk-averse investors.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0957-4174
1873-6793
DOI:10.1016/j.eswa.2020.113222