Des techniques neuronales dans l'alternatif
Year: 2009
Type: dissertation
Abstract: In this thesis, we propose to redefine widely applied financial models, in less restrictive frameworks that comply with hedge funds return peculiarities. This extended setting leads to more complex models; we propose to solve the proposed approaches using natural computation methods. In the first chapter, we answer to the general issue of the pre-processing of databases with missing values. A new method is proposed that efficiently takes into account non-linear dependences and time-series structure of the underlying observations. The second chapter of the thesis is dedicated to higher-order moments asset allocation models. The set of possible portfolios is non-convex in the first four moment space. Thus, we propose an original approach based on the so-called shortage function to solve optimization programs of a rational investor. Then, in the third chapter, we evaluate the impact of outliers on higher-order moment asset allocations. An ANN-GARCH allows us to model the volatility and to extend the methodology proposed by Charles and Darné (2005) in a four moments world. The impact of outliers is finally evaluated through an analysis of moments of returns on dynamically optimized portfolios (with and without pre-processing). The last chapter of the thesis is devoted to style analysis. We develop an original approach based on the Self-Organizing Maps of Kohonen (1995). In this framework, each neuron is considered as a style factor. A conditional analysis is conducted to establish a link between the conceptual factors and real financial benchmarks. Through a dynamic projection, we finally gauge the stability of manager investment strategies in various market conditions.
Source: HAL
Author: Paul Merlin
Language: French
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W37521535
