Book Releases

Conditional density estimation is an important problem in a variety of areas such as system identification, machine learning, artificial intelligence, empirical economics, macroeconomic analysis, quantitative finance and risk management.
This work considers the general problem of conditional density estimation, i.e., estimating and predicting the density of a response variable as a function of covariates. The semi-parametric models proposed and developed in this work combine fuzzy and probabilistic representations of uncertainty, while making very few assumptions regarding the functional form of the response variable's density or changes of the functional form across the space of covariates. These models possess sufficient generalization power to approximate a non-standard density and the ability to describe the underlying process using simple linguistic descriptors despite the complexity and possible non-linearity of this process.
These novel models are applied to real world quantitative finance and risk management problems by analysing financial time-series data containing non-trivial statistical properties, such as fat tails, asymmetric distributions and changing variation over time.

More information: Download PhD thesis

25 June 2014

This PhD thesis contributes to the newly emerged, growing body of scientific work on the use of News Analytics in Finance. Regarded as the next significant development in Automated Trading, News Analytics extends trading algorithms to incorporate information extracted from textual messages, by translating it into actionable, valuable knowledge. The thesis addresses one main theme: the incorporation of news into trading algorithms. This relates to three main tasks: i) the extraction of the information contained in news, ii) the representation of the information contained in news, and iii) the aggregation of this information into actionable knowledge. We validate our approach by designing and implementing three semantic systems: a system for the computational content analysis of European Central Bank statements, a system for incorporating news in stock trading strategies, and a time-aware system for trading based on analyst recommendations. The approach we choose for addressing these tasks is an interdisciplinary one. For the extraction of information from news we rely on approaches borrowed from Computer Science and Linguistics. The representation of the information contained in news is realized by using, and extending, the state-of-the-art in Semantic Web technology. We do this by bringing together insights from Logics, Metaphysics, and Computational Semantics. The aggregation of information is done by using techniques and results from Computational Intelligence and Finance.

More Information: Download PhD thesis

9 February 2013

This thesis studies various computational and game-theoretic approaches to economic modeling. Unlike traditional approaches to economic modeling, the approaches studied in this thesis do not rely on the assumption that economic agents behave in a fully rational way. Instead, economic agents are assumed to be boundedly rational. Abandoning the assumption of full rationality has a number of consequences for the way in which economic reality is being modeled. Traditionally, economic models are mostly of a static nature and have a strong focus on deriving equilibria. Also, models are usually analyzed mathematically. In models of boundedly rational behavior, dynamic elements play a much more prominent role and there is less emphasis on equilibrium behavior. Also, to analyze models of boundedly rational behavior, researchers not only use mathematical techniques but they also rely heavily on computer simulations. This thesis presents four studies into the modeling of boundedly rational behavior of economic agents. Two studies are concerned with investigating the emergence of cooperation among boundedly rational agents. One study focuses on cooperation among firms in a Cournot oligopoly market, while the other study examines cooperation in a spatial model of price-competing firms. The other two studies in this thesis are concerned with methodological issues in the use of evolutionary algorithms for economic modeling purposes. One study shows how evolutionary algorithms can be analyzed mathematically rather than using computer simulations. The other study criticizes the use of a so-called binary encoding in evolutionary algorithms.

More Information: Download PhD thesis

13 October 2011

Bibliometric mapping of science is concerned with quantitative methods for visually representing scientific literature based on bibliographic data. Since the first pioneering efforts in the 1970s, a large number of methods and techniques for bibliometric mapping have been proposed and tested. Although this has not resulted in a single generally accepted methodological standard, it did result in a limited set of commonly used methods and techniques. In this thesis, a new methodology for bibliometric mapping is presented. It is argued that some well-known methods and techniques for bibliometric mapping have serious shortcomings. For instance, the mathematical justification of a number of commonly used normalization methods is criticized, and popular multidimensional-scaling-based approaches for constructing bibliometric maps are shown to suffer from artifacts, especially when working with larger data sets. The methodology introduced in this thesis aims to provide improved methods and techniques for bibliometric mapping. The thesis contains an extensive mathematical analysis of normalization methods, indicating that the so-called association strength measure has the most satisfactory mathematical properties. The thesis also introduces the VOS technique for constructing bibliometric maps, where VOS stands for visualization of similarities. Compared with well-known multidimensional-scaling-based approaches, the VOS technique is shown to produce more satisfactory maps. In addition to the VOS mapping technique, the thesis also presents the VOS clustering technique. Together, these two techniques provide a unified framework for mapping and clustering. Finally, the VOSviewer software for constructing, displaying, and exploring bibliometric maps is introduced.

More Information: Download PhD thesis

13 October 2011

Studying the behavior of market participants is important due to its potential impact on asset prices and the dynamics of financial markets. The idea of individual investors who are prone to biases in judgment and who use various heuristics, which might lead to anomalies on the market level, has been explored within the field of behavioral finance. In this dissertation, we analyze market-wise implications of investor behavior and their irrationalities by means of agent-based simulations of financial markets. The usefulness of agent-based artificial markets for studying the behavioral finance topics stems from their ability to relate the micro-level behavior of individual market participants (represented as agents) and the macro-level behavior of the market (artificial time-series). This micro-macro mapping of agent-based methodology is particularly useful for behavioral finance, because that link is often broken when using other methodological approaches. In this thesis, we study various biases commented in the behavioral finance literature and propose novel models for some of the behavioral phenomena. We provide mathematical definitions and computational implementations for overconfidence (miscalibration and better-than-average effect), investor sentiment (optimism and pessimism), biased self-attribution, loss aversion, and recency and primacy effects. The levels of these behavioral biases are related to the features of the market dynamics, such as the bubbles and crashes, and the excess volatility of the market price. The impact of behavioral biases on investor performance is also studied.

More information: Download PhD thesis

25 March 2011

Over the past two decades the internet has rapidly become an important medium to retrieve information, maintain social contacts, and to do online shopping. The latter has some important advantages over traditional shopping. Products are often cheaper on the internet, internet companies sell a wider collection of products and consumers can buy items whenever they like without leaving their homes. On the other hand, the current state of online shops still has two major disadvantages over `real' shops: Products are often much harder to find than in traditional shops and there are no salesmen to advise the customers.

In this thesis, we address both these disadvantages. We introduce and evaluate several new user interfaces for online shops that are based on representing products in maps instead of lists to user, such that products are easier to find. In these maps similar products are located close to each other. To create these maps, statistical techniques such as multidimensional scaling are used. Furthermore, we combine these maps with recommender systems to address the second disadvantage and to help the user in finding the product best suiting her needs. Also, we introduce a recommender system that is able to explain the recommendations it gives to users. We think that the methods discussed in this thesis can form a basis for new promising online shopping interfaces both in research as in practice.

More information: Download PhD thesis

19 May 2010

The essays in this thesis are concerned with two main themes in port logistics. The first theme is the coordination of transport arrivals with the distribution processes and the use of storage facilities. We study this for both containerized and bulk chemical transport. The second theme is the uncertainty associated with the arrival time of ships with bulk chemicals and the impact on port logistics. Each essay describes a case study where quantitative methods, especially simulation, are used. The operation of container terminals and in particular the way in which containers are stacked in a yard is influenced by information about the departure of a container. We find that even inaccurate information is valuable and helps to reduce unproductive moves. Next, we present the "floating stocks'' distribution concept which uses intermodal transport to deploy inventories in a supply chain in advance of retailer demand. We demonstrate that a main drawback of intermodal transport, a longer transit time, can be mitigated using this concept. This concept also influences the choice of a port: we provide a quantitative interpretation of routing flexibility in port selection.

More information: Download PhD thesis

18 November, 2009

The dynamics of financial markets is subject of much debate among researchers and financial experts trying to understand and explain how financial markets work and traders behave. Diversified explanations result from the complexity of markets, and the hardly observable aspects of price formation mechanisms and of participants' motivation behind trading decisions. In an attempt to provide a better understanding of market dynamics, studies in the realm of agent-based computational economics represent markets from bottom-up. The aim of this thesis is to contribute to the understanding of market dynamics by extending the agent-based computational approach. In order to achieve our goal we propose a modular, continuous-time, agent-based trading environment, with individual, autonomous representation of market participants. In order to be able to develop such an environment we first analyze and compare real and artificial stock markets (ASMs). Based on this analysis we propose a conceptual framework to describe real markets. By enriching the framework with design and implementation issues we get a multi-dimensional taxonomy of artificial stock markets. ABSTRACTE, the proposed modular environment is an operational form of these frameworks. ABSTRACTE is aimed to embed the common aspects of real markets that exhibit big variations and are rarely represented in artificial stock markets. This environment provides the user with a flexible mechanism to implement many of the varying and hardly observable aspects of stock markets and traders' behavior. In this way it can contribute to the understanding of market dynamics as it can be used both as a test bed to replicate and evaluate existing market models, and to compare dynamics of multiple ASMs, as well as a tool to conduct experiments with new models and traders.

More information: Download PhD thesis

25 January 2008