1

Report #1, April 1997, (Ini 5)

Alfred Taudes, Martin Natter, Michael Trcka

Real Option Valuation with Neural Networks

We propose to use Neural Networks to value options when analytical solutions do not exist. The basic idea of this approach is to approximate the value function of a dynamic program by a Neural Net, where the selection of the network weights is done via Simulated Annealing. The main benefits of this method as compared to traditional approximation techniques are that there are no restrictions on the type of th underlying stochastic process and no limitations on the set of possible actions. This makes our approach especially attractive for valuing Real Options in flexible investments. We, therefore, demonstrate the method proposed by valuing flexibility for costly switch production between several products under various conditions.
(accepted for publication in: International Journal of Intelligent Systems in Accounting, Finance and Management)

Up

2

Report #2, April 1997, (Ini 3)

April 1997

Josef A. Mazanec

Simultaneous Positioning and Segmentation Analysis with Topologically Ordered Feature Maps: A Tour Operator Example

Positioning analysis seeks spatial representations of brands. Segmentation analysis searches for perceptually or preferentially homogeneous consumer segments. While these analytical steps often are taken sequentially, the positioning and segmentation problems are interrelated and need to be treated simultaneously. Topologically ordered feature maps (Self-Organizing Maps) are neural network models for feature extraction and classification. Extracting prototypes corresponds with finding market segments; ordering them topologically resembles a perceptual mapping exercise. SOM modeling may thus be relevant for deriving low-dimensional and parsimonious representations of multidimensional profile data. The application of SOMs is explored in a case study on tour operator images. The results simultaneously inform about the firms’ image positions and their perceptually homogeneous customer segments.
(under review for publication in: Journal of Retailing and Consumer Services)

Up

3

Report #3, May 1997 (Ini 3, Ini 5)

Martin Natter, Harald Hruschka

Ankerpreise als Erwartungen oder dynamische latente Variablen in Marktreaktionsmodellen

Ankerpreistheorien liefern verhaltenswissenschaftliche Erklärungen für Effekte von Preisvariationen auf Absatzmengen oder Marktanteile, die nicht nur mit dem jeweils aktuellen Preis zusammenhängen. Man geht davon aus, daß Nachfrager bei Kaufentscheidungen Preisinformationen aus der Vergangenheit im Gedächtnis behalten. Abnehmer vergleichen (mindestens) einen derartigen internen Preis mit den von ihnen beobachteten aktuellen Preisen und ziehen ihn als Bestimmungsfaktor für neue Kaufentscheidungen heran. Falls der Ankerpreis über dem aktuellen Preis liegt, bewerten dies die Abnehmer als Gewinn. Zu einer Bewertung als Verlust kommt es dagegen, wenn der Ankerpreis geringer als der aktuelle Preis ist. Die günstige (ungünstige) Bewertung des beobachteten Preises einer Marke erhöht (senkt) den Marktanteil. Die praktische Relevanz von Ankerpreisen zeigt sich an deren Einfluß auf die Form optimaler dynamischer Preisstrategien.  Im Mittelpunkt dieses Beitrags stehen die Fragen der Spezifikation und Messung von Ankerpreisen. Hypothetisch wirken sich bestimmte Prädiktoren auf Ankerpreise aus, diese ihrerseits beeinflussen Marktanteile. Es werden drei Formen von Ankerpreismechanismen, nämlich
extrapolative Erwartungen, rationale Erwartungen und dynamische latente Variablen, untersucht. Abschließend werden die Implikationen von Ankerpreiseffekten für die Preispolitik erörtert.
(accepted for publication in: Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung)

Up

4

Report #4, July 1997 (Ini 1)

Friedrich Leisch, Kurt Hornik

Combining Neural Network Voting Classifiers and Error Correcting Output Codes

We show that error correcting output codes (ECOC) can further improve the effects of error dependent adaptive resampling methods such as arc-Ih. In traditional one-in-n coding, the distance between two binary class labels is rather small, whereas ECOC are chosen to maximize this distance. We compare one-in-n and ECOC on a multiclass data set using standard MLPs and bagging and arcing voting committees.
(accepted for publication in: Proceedings of Measurement 97, May 29-31-1997, Smolenice, Slovakia)

Up

5

Report #5, July 1997 (Ini 3)

Adrian Trapletti, Manfred M. Fischer

Bayesian Modelling of High Frequency Data in Finance

The objective of stochastic modelling is not to find an exact representation of the observed data itself, but rather to build a statistical model of the process which generates the data. In contrast to the frequentist approach, the Bayesian approach provides a different framework for dealing with the issues of model complexity and, thus, avoiding the overfitting problem. The objective of this paper is to adopt the Bayesian framework for modelling the long memory of foreign currency exchange rate volatilities on an operational time scale in the context of the ARCH methodology. In addition, it is illustrated that the choice of a prior distribution for the model parameters affects the overall model quality. Eventhough attention is focused on ARCH models, the suggested approach can be applied to other model types too.
(accepted for publication in: Neural Network World, International Journal on Neural and Mass-Parallel Computing and Information Systems, Special Issue on PASE ’97)

Up

6

Report #6, July 1997 (Ini 5)

Martin Natter

Conditional Market Segmentation by Neural Networks: A Monte Carlo Study

An artificial neural network (ANN) algorithm is proposed that incorporates both cluster and discriminant (or regression) analysis of the segments. The method simultaneously estimates the models relating consumer characteristics to market segments, i.e., subjects are assigned to (unique) segments so that subjects within a class show similar purchase behaviour and share the same characteristics (psychographics/sociodemographics). Parameters of all models are estimated by the backpropagation algorithm. The performance of the ANN methodology is assessed in a Monte Carlo study. In contrast to the usual stepwise approach adopted in segmentation studies, our study found that simultaneous segmentation and discrimination are preferable for finding an overall optimum in that this way clusters are formed not only to create homogeneous submarkets but also to show a good discriminatory behaviour.
(accepted for publication in: Journal of Retailing and Consumer Services)

Up

7

Report #7, July 1997 (Ini 3, Ini 5)

Jörg Peter Heimel, Harald Hruschka, Martin Natter, Alfred Taudes

Konnexionistische Kaufakt- und Markenwahlmodelle

Verschiedene konnexionistische Modelle (künstliche neurale Netzwerke) können als nichtlineare Verallgemeinerungen bestimmter in der Marketingforschung verbreiteter multivariater Datenanalyseverfahren interpretiert werden. Im vorliegenden Beitrag wird untersucht, ob konnexionistische Modelle herkömmlichen Mikromodellen des Kaufverhaltens bei kurzlebigen Konsumgütern bezüglich Analyse und Prognose überlegen sind. Für diesen Problembereich wurden im Laufe der Zeit eine Reihe stochastischer und ökonometrischer Verfahren entwickelt, die eine fundierte Evaluierung alternativer Ansätze erlauben. Empirisch geschätzte neurale Netzwerke werden mittels statistischer Kennzahlen von Elastizitäten und Änderungsraten der Kriteriumsvariablen interpretiert.

Up

8

Report #8, October 1997 (Ini 5)

Bernd Bullnheimer, Gabriele Kotsis, Christine Strauss

Parallelization Strategies for the Ant System

The Ant System is a new meta-heuristic method particularly appropriate to solve hard combinatorial optimization problems. It is a population-based, nature-inspired approach exploiting positive feedback as well as local information and has been applied successfully to a variety of combinatorial optimization problem classes. The Ant System consists of a set of cooperating agents (artificial ants) and a set of rules that determine the generation, update and usage of local and global information in order to find good solutions. As the structure of the Ant System highly suggests a parallel implementation of the algorithm, in this paper two parallelization strategies for an Ant System implementation are developed and evaluated: the synchronous parallel algorithm and the partially asynchronous parallel algorithm. Using the Traveling Salesman Problem a discrete event simulation is performed, and both strategies are evaluated on the criteria „speedup“, „efficiency“ and „efficacy“. Finally further improvements for an advanced parallel implementation are discussed.
(accepted for publication in: Kluwer Series of Applied Optimization)

Download / Up

9

Report #9, November 1997 (Ini 2)

Helmut Strasser

Perturbation Invariant Estimates and Incidental Nuisance Parameters

It is shown that the asymptotic information bound which is valid for the estimation of a parameter in the structure (mixture) model remains valid in the functional model (incidental nuisance parameters) if only perturbation symmetric estimators are admitted. Perturbation symmetry is a property which is closely related to permutation symmetry. In particular, equicontinuous functions of empirical processes are perturbation symmetric. Thus, the results of this paper continue a discussion initiated by Bickel and Klaassen, Pfanzagl and Strasser on permutation symmetry of estimators and the exclusion of superefficiency in the functional model.
(accepted for publication in:Mathematical Methods of Statistics)

Up

10

Report #10, December 1997 (Ini 2)

Klaus Pötzelberger, Helmut Strasser

Data compression by unsupervised classification

This paper deals with a general class of classification methods which are related both to vector quantization in the sense of Pollard, as well as to competitive learning in the sense of Kohonen. The basic duality of minimum variance partitioning and vector quantization known from statistical cluster analysis is shown to be true for this whole class of classification problems. The paper contains theoretical results like existence of optima, consistency of approximate optima and characterization of local optima as fixpoints of a fix point algorithm. A fix point algorithm is proposed and its termination after finite time is proved for empirical distributions. The construction of a particular classification method is based on a statistical information measure specified by a convex function. Modifying this convex function gives room for suggesting a large variety of new classification procedures, e.g. of robust quantifiers.

Up

111

Report #11, January 1998 (Ini 3)

January 1998

Petra Staufer, Manfred Fischer

A Neural Network Classifier for Spectral Pattern Recognition. On-line versus Off-Line Backpropagation Training.

In this contributon we evaluate on-line and off-line techniques to train a single hidden layer neural network classifier with logistic hidden and softmax output transfer functions on a multispectral pixel-by-pixel classification problem. In contrast to current practice a multiple class cross-entropy error function has been chosen as the function to be minimized. The non-linear differential equations cannot be solved in closed form. To solve for a set of locally minimizing parameters we use the gradient descent technique for parameter updating based upon the backpropagation technique for evaluating the partial derivatives of the error function with respect to the parameter weights. Empirical evidence shows that on-line and epoch-based gradient descent backpropagation fail to converge within 100,000 iterations, due to the fixed step size. Batch gradient descent backpropagation training is superior in terms of learning speed and convergence behaviour. Stochastic epoch-based training tends to be slightly more effective than on-line and batch training in terms of generalization performance, especially when the number of training examples is larger. Moreover, it is less prone to fall into local minima than on-line and batch modes of operation.
(This paper is under review for publication in: Neurocomputing)

Up

12

Report #12, January 1998 (Ini 3)

Petra Staufer, Manfred Fischer

Optimization in an Error Backpropagation Neural Network Environment with a Performance Test on a Real World Pattern Classification Problem

Various techniques of optimizing the multiple cross-entropy error function to train single hidden layer neural network classifiers with softmax output transfer functions are investigated on a real-world multispectral pixel-by-pixel classification problem that is of fundamental importance in remote sensing. These techniques include epoch-based and batch versions of backpropagation of gradient descent, PR-conjugate gradient and BFGS quasi-Newton errors. The method of choice depends upon the nature of the learning task and whether one wants to optimize learning for speed or generalization performance. It was found that, comparatively considered, gradient descent error backpropagation provided the best and most stable out-of-sample performance results across batch and epoch-based modes of operation. If the goal is to maximize learning speed and a sacrifice in generalisation is acceptable, then PR-conjugate gradient error backpropagation tends to be superior. If the training set is very large, stochastic epoch-based versions of local optimizers should be chosen utilizing a larger rather than a smaller epoch size to avoid inacceptable instabilities in the generalization results.
(This paper is under review for publication in: IEEE Transactions on Neural Networks)

Up

13

Report #13, February 1998 (Ini 2)

Friedrich Hubalek, Walter Schachermayer

When does Convergence of Asset Price Processes Imply Convergence of Option Prices?

We consider weak convergence of a sequence of asset price models (Sn) to a limiting asset price model S. A typical case for this situation is the convergence of a sequence of binomial models to the Black-Scholes model, as studied by Cox, Ross and Rubinstein. We put emphasis on two different aspects of this convergence: firstly we consider convergence with respect to the given "physical" probability measures (Pn) and secondly with respect to the "risk-neutral" measures (Qn) for the asset price processes (Sn). (In the case of non-uniqueness of the risk-neutral measures also the question of the "good choice" of (Qn) arises.) In particular we investigate under which conditions the weak convergence of (Pn) to P implies the weak convergence of (Qn) to Q and thus the convergence of prices of derivative securities. The main theorem of the present paper exhibits an intimate relation of this question with contiguity properties of the sequences of measures (Pn) with respect to (Qn) which in turn is closely connected to asymptotic arbitrage properties of the sequence (Sn) of security price processes. We illustrate these results with general homogeneous binomial and some special trinomial models.
(This paper was accepted for publication in: Mathematical Finance)

Download / Up

14

Report #14, May 1998 (Ini 6)

Neal M. Stoughton, Josef Zechner

IPO-Mechanisms, Monitoring and Ownership Strucutre

This paper analyzes the effect of different IPO mechanisms on the structure of share ownership and explores the role of underpricing and rationing in determining investors’ shareholdings. We focus on the agency problem that results when large institutions are the only investors capable of monitoring the firm whereas small shareholders free-ride on these activities. The major conclusion of this paper is that some well-known aspects of IPOs may be explained as rational responses by the issuer to the existence of regulatory constraints in public capital markets. We find that there is a two-stage offering mechanism in which an investment banker acting in the interests of the issuer optimally rations the allotment of shares to small investors in order to capture the benefits associated with better monitoring by institutions. Importantly, in our model, the existence of underpricing (and oversubscription) is an indication that the issuer has received a higher ex ante price than would have been obtained through a competitive Walrasian-type offering process.
(This paper was accepted for publication in: Journal of Financial Economics)

Up

15

Report #15, May 1998 (Ini 1, Ini 6)

Christian Schittenkopf, Georg Dorffner, Engelbert J. Dockner

Volatility Prediction with Mixture Density Networks

Despite the lack of a precise definition of volatility in finance, the estimation of volatility and its prediction is an important problem. In this paper we compare the performance of standard volatility models and the performance of a class of neural models, i.e. mixture density networks (MDNs). First experimental results indicate the importance of long-term memory of the models as well as the benefit of using non-gaussian probability densities for practical applications.
(This paper was accepted for publication in: Proceedings of the International Conference on Artificial Neural Networks, September 2-4 1998, Skövde, Schweden)

Download / Up

16

Report #16, May 1998 (Ini 3)

Thomas Reutterer

Combined Market Structure and Segmentation Analysis Based on Brand Choice Data: Overcoming the Limitations of Conventional Techniques with Topologically Ordered Feature Maps

The objective and related issues of simultaneously performing competitive market structure (CMS) and market segmentation analysis is well-documented in the marketing literature. In this paper, an artificial neural network based approach of Kohonen (1982) for the formation of topological ordered feature maps in introduced into the context of brand choice data based combined CMS/segmentation analysis. Selected aspects of the methodological basis are discussed and in a demonstration study using household-level brand choice probabilities derived from diary household panel data the failure of conventional methodology is opposed to the promising properties of the suggested neurocomputing approach.
(This paper was accepted for publication in: Proceedings of the 1998 AMA Exchange Colloquium July 23-25, Vienna)

Download / Up

17

Report #17, June 1998 (Ini 1)

Friedrich Leisch, Andreas Weingessel, Evgenia Dimitriadou

Competitive Learning for Binary Valued Data

We propose a new approach for using online competitive learning on binary data. The usual Euclidean distance is replaced by binary distance measures, which take possible asymmetries of binary data into account and therefore provide a ``different point of view'' for looking at the data. The method is demonstrated on two artificial examples and applied on tourist marketing research data.
(This paper was accepted for publication in: Proceedings of the International Conference on Artificial Neural Networks, September 2-4 1998, Skoevde, Schweden)

Download / Up

18

Report #18, 1998 (Ini 1)

Peter Tino and Georg Dorffner

Recurrent neural networks with Iterated Function Systems dynamics

We suggest a recurrent neural network (RNN) model with a recurrent part corresponding to iterative function systems (IFS) introduced by Barnsley (1988) as a fractal image compression mechanism. The key idea is that 1) in our model we avoid learning the RNN state part by having non-trainable connections between the context and recurrent layers (this makes the training process less problematic and faster), 2) the RNN state part codes the information processing states in the symbolic input stream in a well-organized and intuitively appealing way. We show that there is a direct correspondence between the R\' enyi entropy spectra characterizing the input stream and the spectra of R\' enyi generalized dimensions of activations inside the RNN state space. We test both the new RNN model with IFS dynamics and its conventional counterpart with trainable recurrent part on two chaotic symbolic sequences. In our experiments, RNNs with IFS dynamics outperform the conventional RNNs with respect to information theoretic   measures computed on the training and model generated sequences.
(Accepted for: International ICSC/IFAC Symposium on Neural Computation (NC'98), Technical University in Vienna, Austria, September 23-25, 1998)

Download / Up

19

Report #19, 1998 (Ini 5)

Bernd Bullnheimer, Herbert Dawid and Ralph Zeller

Learning from Own and Foreign Experience: Technological Adaptation by Imitating Firms

In this paper we study the adaptive behavior of firms which repeatedly have to make a production decision. In a single good market the firms use own experience as well as information gathered by observing competitors to iteratively choose a production technology out of a given set. The adaptive learning of the firms is described in a dynamic model and analyzed in a simulation framework. We show that a small but positive propensity to imitate is optimal for the firms and yields production efficiencies above 95 % of the maximal value. Furthermore, we observe that in a competitive situation firms using optimal propensities to imitate outmatch pure imitators and non-imitators in production efficiency as well as in profits.
(Accepted for: Computational and Mathematical Organization Theory)

Up

20

Report #20,  1998 (Ini 5)

Herbert Dawid

Learning to Trade and Mediate

In this paper we study the behayior of boundedly rational agents in a two good economy where trading is costly with respect to time. All individuals have a fixed time budget and may spend time for the production of good one, the production of good two and trading. They update their strategies, which determine their time allocation, according to a simple imitation type learning rule with noise. In a setup with two different type of agents with different production technologies we show by the means of simulations that both direct trade and trade via mediators who specialize in trading can emerge. We can also observe the transition from a pure production economy via direct trade to an economy with mediated trade.
(Accepted for: Market Structure, Aggregation and Heterogeneity)

Download / Up

21

Report #21, 1998 (Ini 1)

Friedrich Leisch, Adrian Trapletti, Kurt Hornik

On the Stationarity of Autoregressive Neural Network Models

We analyze the asymptotic behavior of autoregressive neural network (AR-NN) processes using techniques from Markov chains and  non-linear time series analysis. It is shown that standard AR-NNs without shortcut connections are asymptotically stationary. If  linear shortcut connections are allowed, only the shortcut weights determine whether the overall system is stationary, hence standard   conditions for linear AR processes can be used.
(This paper was accepted for publication in: Proceedings of CoWAN 98: Cottbuser Workshop Aspekte Neuronalen Lernens, October 5-7, 1998, Cottbus, Germany. Shaker Verlag.)

Download / Up

Report #22, October 1998 (Ini 6)

Andrea Gaunersdorfer

Endogenous Fluctuations in a Simple Asset Pricing Model with Heterogenous Agents

In this paper we study the adaptive rational equilibrium dynamics in a simple asset pricing model introduced by Brock and Hommes (1997b, 1998). Traders have heterogeneous expectations concerning future prices and update their beliefs according to a risk adjusted performance measure and to market conditions. Further, also their expectations about conditional variances of returns vary over time. We show that even for the simple case where agents can only choose between two different predictors complicated dynamics arise and we analyse the bifurcation routes to chaos.
(This paper was accepted for publication in: Journal of Economic Dynamics and Control)

Up

23

Report #23,  January 1999 (Ini 2)

Freddy Delbaen, Walter Schachermayer

A Compactness Principle for Bounded Sequences of Martingales with Applications

For $\Cal H^1$ bounded sequences, we introduce a technique, related to the Kadec-Pelczynski-decomposition for $L^1$ sequences, that allows us to prove compactness theorems. Roughly speaking, a bounded sequence in $\Cal H^1$ can be split into two sequences, one of which is weakly compact, the other forms the singular part. If the martingales are continuous then the singular part tends to zero in the semi-martingale topology. In the general case the singular parts give rise to a process of bounded variation. The technique allows to give a new proof of the optional decomposition theorem in Mathematical Finance.
(This paper was accepted for publication in: Proceedings of the Seminar on Stochastic Analysis, Random, Fields and Application)

Download / Up

24

Report #24, January 1999 (Ini 2)

Freddy Delbaen, Walter Schachermayer

The Fundamental Theorem of Asset Pricing for Unbounded Stochastic Processes

The Fundamental Theorem of Asset Pricing states - roughly speaking - that the absence of arbitrage possibilities for a stochastic process $S$ is equivalent to the existence of an equivalent martingale measure for $S$. It turnsout that it is quite hard to give precise and sharp versions of this theorem in proper generality, if one insists on modifying the concept of ``no arbitrage" as little as possible. It was shown in [DS94] that for a locally bounded $\R^d$-valued semi-martingale $S$ the condition of No Free Lunch with Vanishing Risk is equivalent to the existence of an equivalent local martingale measure for the process $S$. It was asked whether the local boundedness assumption on $S$ may be dropped.  In the present paper we show that if we drop in this theorem the local boundedness assumption on $S$ the theorem remains true if we replace the term equivalent local martingale measure by the term equivalent sigma-martingale measure. The concept of sigma-martingales was introduced by Chou and Emery --- under the name of ``semimartingales de la classe $(\Sigma _m)$".  We provide an example which shows that for the validity of the theorem in the non locally bounded case it is indeed necessary to pass to the concept of sigma-martingales. On the other hand, we also observe that for the applications in Mathematical Finance the notion of
sigma-martingales provides a natural framework when working with non locally bounded processes $S$. The duality results which we obtained earlier are also extended to the non locally bounded case. As an application we characterize the hedgeable elements.
(This paper was accepted for publication in: Math. Annalen)

Download / Up

Report #25, January 1999 (Ini 2)

Dimitri Kramkov, Walter Schachermayer

A Condition on the Asymptotic Elasticity of Utility Functions and Optimal Investment in Incomplete Markets

The paper studies the problem of maximizing the expected utility of terminal wealth in the framework of a general incomplete semimartingale model of a financial market. We show that the necessary and sufficient condition on a utility function for the validity of several key assertions of the theory to hold true is the requirement that the asymptotic elasticity of the utility function is strictly less then one.
(This paper was accepted for publication in: Math. Annalen)

Download / Up

26

Report #26, January 1999 (Ini 2)

Peter Grandits, Werner Schachinger

Leland's Approach to Option Pricing: The Evolution of a Discontinuity

A claim of Leland (1985) states that in the presence of transaction costs a call option on a stock S, described by geometric Brownian motion, can be perfectly hedged using Black-Scholes delta hedging with a modified volatility. Recently Kabanov and Safarian (1997) disproved this claim, giving an explicit (up to an integral) expression of the limiting hedging error, which appears to be strictly negative and depends on the path of the stock price only via the stock price at expiry ST. We prove in this paper that the limiting hedging error, considered als a function of ST, exhibits a removable discontinuity at the exercise price. Furthermore, we provide a quantitative result describing the evolution of the discontinuity, which shows that ists precursors can very well be observed also in cases of reasonable length of revision intervals.
(This paper is under review for publication in: Mathematical Finance)

Download / Up

27

Report #27, January 1999 (Ini 2)

Helmut Strasser, Christian Weber

On the Asymptotic Theory of Permutation Statistics

In this paper limit theorems for the conditional distributions of linear test statistics are proved. The assertions are conditioned by the -field of permutation symmetric sets. Limit theorems are proved both for the conditional distributions under the hypothesis of randomness and under general contiguous alternatives with independent but not identically distributed observations. The proofs are based on results on limit theorems for exchangeable random variables by Strasser and Weber. The limit theorems under contiguous alternatives are consequences of an LAN-result for likelihood ratios of symmetrized product measures. The results of the paper have implications for statistical applications. By example it is shown that minimum variance partitions which are defined by observed data (e.g. by LVQ) lead to asymptotically optimal adaptive tests for the k-sample problem. As another application it is shown that conditional k-sample tests which are based on data-driven partitions lead to simple confidence sets which can be used for the simultaneous analysis of linear contrasts.
(This paper was accepted for publication in: Sonderband der Mathematical Methods of Statistics)

Up

28

Report #28, January 1999 (Ini 3)

Sara Dolnicar, Klaus Grabler, Josef Mazanec

A Tale of Three Cities: Perceptual Charting for Analyzing Destination Images

Heterogeneity of perceptions is a neglected issue in market segmentation studies. Only recently parametric approaches toward modeling segmented perception-preference structures such as combined MDS and Latent Class procedures have been introduced. A completely different nonparametric method is based on topology-sensitive vector quantization (VQ) for consumers-by-brands-by-atrributes data. It maps the segment-specific perceptual structures into bar charts with multiple brand positions exhibiting perceptual distinctiveness or similarity. An extensive literatur review is followed by an introduction into the VQ methodology and a sample study on three urban destinations competing on the world travel markets. City images serve as the underlying behavioral constructs. Preferential data are based on respondents' comes-closet-to-ideal-city jedgments and incorporated into the perceptual positions of city profiles. Perceptual charting works on two levels of aggregation named prototypes and perceptual sub-structures. The results demonstrate how this method prevents the analyst from drawing erroneous conclusions due to uncontrolled aggregation.

(Published in: Woodside (ed.), Consumer Psychology of Tourism, Hospitality and Leisure, London: CAB International, A.G. Woodside, G.I. Crouch, J.A. mazanec, M. Oppermann and M.Y. Sakai)

Up

29

Report #29, January 1999 (Ini 1)

Evgenia Dimitriadou, Andreas Weingessel, Kurt Hornik

Fuzzy Voting in Clustering

In this paper we present a fuzzy voting scheme for cluster algorithms. This fuzzy voting method allows us to combine several
runs of cluster algorithms resulting in a common fuzzy partition. This helps us to overcome instabilities of the cluster algorithms
and results in a better clustering.
(This paper was accapted for publication in: Proceedings of the "6th International Workshop Fuzzy-Neuro
Systems '99", Leipzig, Germany, March 18-19, 1999.)

Download / Up

30

Report #30, March 1999 (Ini 1)

Evgenia Dimitriadou, Andreas Weingessel, Kurt Hornik

Voting in Clustering and Finding the Number of Clusters

In this paper we present an unsupervised algorithm which performs clustering given a data set and which can also find the number of clusters existing in it. This algorithm consists of two techniques. The first, the voting technique, allows us to combine several runs of clustering algorithms, with the number of clusters predefined, resulting in a common partition. We introduce the idea that there are cases where an input point has a structure with a certain degree of confidence and may belong to more than one cluster with a certain degree of ``belongingness''. The second part consists of an index measure which receives the results of every voting process for different number of clusters and makes the decision in favor of one. This algorithm is a complete clustering scheme which can be applied to any clustering method and to any type of data set. Moreover, it helps us to overcome instabilities of the clustering algorithms and to improve the ability of a clustering algorithm to find structures in a data set.
This paper was accepted for publication in: Proceedings of the "International ICSC Symposium on Advances in Intelligent Data Analysis
(AIDA 99)'' (("International Congress on Computational Intelligence: Methods and Applications (CIMA 99)''), ICSC Academic Press

Download / Up

Report #31, April 1999 (Ini 3)

Josef A. Mazanec

Exploratory Market Structure Analysis: Topology-Sensitive Methodology

Given the recent abundance of brand choice data from scanner panels market researchers have neglected the measurement and analysis of perceptions. Heterogeneity of perceptions is still a largely unexplored issue in market structure and segmentation studies. Over the last decade various parametric approaches toward modelling segmented perception-preference structures such as combined MDS and Latent Class procedures have been introduced. These methods, however, are not taylored for qualitative data describing consumers' redundant and fuzzy perceptions of brand images. A completely different method is based on topology-sensitive vector quantization (VQ) for consumers-by-brandsby-attributes data. It maps the segment specific perceptual structures into bubble-pie-bar charts with multiple brand positions demonstrating perceptual distinctiveness or similarity. Though the analysis proceeds without any distributional assumptions it allows for significance testing. The application of exploratory and inferential data processing steps to the same data base is statistically sound and particularly attractive for market structure analysts. A brief outline of the VQ method is followed by a sample study with travel market data which
proved to be particularly troublesome for conventional processing tools.

Download / Up

32

Report #32, April 1999 (Ini 6)

Thomas Dangl

Investment and capacity choice under uncertain demand

This paper extends the real options literature by discussing an investment problem, where a firm has to determine optimal investment timing and optimal capacity choice at the same time under conditions of irreversible investment expenditures and uncertainty in future demand. After the project is installed with a certain maximum capacity, this capacity is fixed as an upper boundary to the output and cannot be adjusted later on. It turns out that, in the framework of this once and for all decision, uncertainty in future demand leads to an increase in optimal installed capacity. But on the other hand it causes investment to be delayed to an extent that even small uncertainty makes waiting and accumulation of further information the optimal decision for large ranges of demand. Limiting the capacity which may be installed weakens this extreme effect of uncertainty.

Download / Up

33

Report #33, May 1999 (Ini 3, Ini 5)

Thomas Reutterer, Martin Natter

Segmentation Based Competitive Analysis with MULTICLUS and Topology Preserving Networks

Two neural network approaches, Kohonen's Self-Organizing (Feature) Map (SOM) and the Topology Representing Network (TRN) of Martinetz and Schulten are employed in the context of competitive market structuring and segmentation analysis. In an empirical study using brands preferences derived from household panel data, we compare the SOM and TRN approach to MULTICLUS, a parametric approach which also imultaneously solves the market structuring and segmentation problem.

Download / Up

33

Report #34, July 1999 (Ini 6)

Franz Wirl, Thomas Dangl

Was Dixit und Pindyck bei der Analyse von Managementproblemen unter Unsicherheit verschweigen an Hand des Beispiels der optimalen Wartung und Ausmusterung einer Maschine

Mit ihrem Buch ‘Investment Under Uncertainty’ erwecken A. Dixit und R. Pindyck den Eindruck, daß dynamische Managementprobleme unter Unsicherheit mit Hilfe von Standardmethoden zur Lösung von Differentialgleichungen analysiert werden können. Der vorliegende Aufsatz zeigt, warum dieser Eindruck falsch ist und warum Standarddifferentialgeichungsverfahren schon bei sehr einfachen Kontrollproblemen versagen müssen. Darüber hinaus wird ein Projektionsalgorithmus präsentiert, der zur Approximation der Wertefunktion geeignet ist. Damit ist die Möglichkeit für eine betriebswirtschaftliche Anwendungen geschaffen. Demonstriert werden die einzelnen Punkte an Hand eines Modells zur optimalen Wartung und Ausmusterung einer Anlage.

Download / Up

35

Report #35, August 1999 (Ini 2)

Jörg Rahnenführer

Multivariate permutation tests for the k-sample problem with clustered data

The present paper deals with the choice of clustering algorithms before treating a k-sample problem. We investigate multivariate data sets that are quantized by algorithms that define partitions by maximal support planes (MSP) of a convex function. These algorithms belong to a wide range class containing as special cases both the well known k-means algorithm and the Kohonen (1985) algorithm and have profundly investigted by Pötzelberger and Strasser (1999). For computing the test statistics for the k-sample problem we replace the data points by their conditional expections with respect to the MSP-partition. We present Monte Carlo simulations of power functions of different tests for the k-sample problem whereas the tests are carried out as multivariate permutation tests to ensure that they hold the level. The results presented show that there seems to be a vital and decisive connection between the optimal choice of the clustering algorithm and the tails of the probability distribution of the data. Especially for distributions with heavy tails like the exponential distribution the performance of tests based on a quadratic function with k-means type partitions totally breaks down.

Download / Up

36

Report #36, September 1999 (Ini 1, Ini 6)

Christian Schittenkopf, Georg Dorffner, Engelbert J. Dockner

Forecasting Time-dependent Conditional Densities: A Neural Network Approach

In financial econometrics the modeling of asset return series is closely related to the estimation of the corresponding conditional densities. One reason why one is interested in the whole conditional density and not only in the conditional mean, is that the conditional variance can be interpreted as a measure of time-dependent volatility of the return series. In fact, the modeling and the prediction of volatility is one of the central topics in asset pricing.
In this paper we propose to estimate conditional densities semi-nonparametrically in a neural network framework. Our recurrent mixture density networks realize the basic ideas of prominent GARCH approaches but they are capable of modeling any continuous conditional density also allowing for time-dependent higher-order moments. Our empirical analysis on daily DAX data shows that out-of-sample volatility predictions of the neural network model are superior to predictions of GARCH models in that they have a higher correlation with implied volatilities.

(Submitted to: Journal of Forecasting)

Download / Up

37

Report #37, September 1999 (Ini 1, Ini 6)

Artur Baldauf, Engelbert J. Dockner, Heribert Reisinger

The Effects of Long-term Debt on a Firm's New Product Pricing Ploicy in Duopolistic Markets

While many marketing models ignore the influence of financial variables on a firm's marketing strategy, this paper explores the
effect of debt on the profit maximizing price for a new product. We assume a duopolistic market structure in which two firms produce a heterogeneous new consumer durable that is sold over two different periods.  Firms know market demand in the first period with certainty, while demand in the second period is uncertain. Moreover, firms have free access to the capital market and finance part of their operating costs by issuing long-term debt. In this setting, we study the influence of long-term debt on firms' pricing policies. It turns out that leveraged firms compared to unleveraged ones have different pricing strategies. In particular, first-period prices are lower and second-period prices are higher in case of long-term debt than in case of no leverage. Finally we find that prices for firms that take on debt are less volatile than prices for   purely equity-financed firms.

Download / Up

37

Report #38, September 1999 (Ini5)

Alfred Taudes, Michael Trcka, Lukanwicz Martin

Organizational Learning in Production Networks

If one accepts that a firm's behavior is determined by history-dependent capabilities that adapt in a goal-directed way one would like to know how a firm's organizational structure influences the way in which this distributed and partially tacit organizational memory evolves over time. In this paper, we study the impact that alternative information systems, incentive systems and modes of learning co-ordination have on the efficiency and generality of priority rules for job shop scheduling which are learnt by a network of production agents modeled by neural networks. When modeling the alternative organizational structures by different input layers, feedback and training methods, we find that efficient rules evolve when global incentives and synchronized learning are employed even if the system state is only partially known to an agent. However, organizational learning fails when it is performed asynchronously with local goals.

Download / Up

39

Report #39, September 1999 (Ini 1 und Ini 6 )

Schittenkopf, Dorffner, Dockner

Non-linear versus non-gaussian volatility models

One of the most challenging topics in financial time series analysis is the modeling of conditional variances of asset returns. Although conditional variances are not directly observable there are numerous approaches in the literature to overcome this problem and to predict volatilities on the basis of historical asset returns. The most prominent approach is the class of GARCH models where conditional variances are governed by a linear autoregressive process of past squared returns and variances.  Recent research in this field, however, has focused on modeling asymmetries of conditional variances by means of non-linear models. While there is evidence that such an approach improves the fit to empirical asset returns, most non-linear specifications assume conditional normal distributions and ignore the importance of alternative models. Concentrating on the distributional assumptions is, however, essential since asset returns are characterized by excess kurtosis and hence fat tails that cannot be explained by models with sufficient heteroskedasticity.

In this paper we take up the issue of returns' distributions and contrast it with the specification of non-linear GARCH models. We use
daily returns for the Dow Jones Industrial Average over a large period of time and evaluate the predictive power of different linear and non-linear volatility specifications under alternative distributional assumptions. Our empirical analysis suggests that while non-linearities do play a role in explaining the dynamics of conditional variances, the predictive power of the models does also depend on the distributional assumptions.

Download / Up

40

Report #40, September 1999 (Ini 1und Ini 6)

Schittenkopf, Dorffner, Dockner

On non-linear, stochastic dynamics in economic and financial time series

The search for deterministic chaos in economic and financial time series has attracted much interest over the past decade. However, clear evidence of chaotic structures is usually prevented  by large random components in the time series. In the first part of this paper we show that even if a sophisticated algorithm estimating and testing the positivity of the largest Lyapunov exponent is applied to time series generated by a stochastic dynamical system or a return series of a stock index, the results are difficult to interpret. We conclude that the notion of sensitive dependence on initial conditions as it has been developed for deterministic dynamics, can hardly be transfered into a stochastic context. Therefore, in the second part of the paper our starting point for measuring dependencies for stochastic dynamics is a distributional characterization of the dynamics, e.g. by heteroskedastic models for
economic and financial time series. We adopt a sensitivity measure proposed in the literature which is an information-theoretic
measure of the distance between probability density functions. This sensitivity measure is well defined for stochastic dynamics, and it can be calculated analytically for the classes of stochastic dynamics with conditional normal distributions of constant and state-dependent variance. In particular, heteroskedastic return series models such as ARCH and GARCH models are investigated.

Download / Up

,

Report #41, December 1999 (Ini 2)

Steinberger, Zinner

Complete Controllability of Discrete-Time Recurrent Neural Networks.

This paper presents a characterisation of complete controllability for the class of discrete-time recurrent neural networks. We prove that complete controllability holds if and only if the rank of the control matrix equals the state space dimension.

Download / Up

42

Report #42, May 1999 (Ini 5)

Bauer, Bullnheimer, Hartl, Strauss

Minimizing Total Tardiness on a Single Machine Using Ant Colony Optimization

Ant Colony Optimization is a relatively new meta-heuristic that has proven its quality and versatility on various combinatorial optimization problems such as the traveling salesman problem,  the vehicle routing problem and the job shop scheduling problem. The paper introduces an Ant Colony Optimization approach to solve the problem of determining a job-sequence  that minimizes the overall  tardiness for a given set of jobs to be processed on a single, continuously available machine, the Single Machine Total Tardiness Problem. We experiment with various heuristic information as well as with variants for local search. Experiments with 250 benchmark problems with 50 and 100 jobs illustrate that Ant Colony Optimization is an adequate method to tackle the SMTTP.

Download / Up

42

Report #43, December 1999 (Ini 2)

Stefan Geiss

On quantitative approximation of stochastic integrals with respect to the geometric Brownian motion

We approximate stochastic integrals with respect to the geometric Brownian motion by stochastic integrals over discretized integrands, where deterministic, but not necessarily equidistant, time nets are used. This corresponds to the approximation of a continuously adjusted portfolio by a discretely adjusted one. We compute the approximation orders of European Options in the Black Scholes model with respect to L_2 and the approximation order of the standard European-Call and Put Option with respect to an appropriate BMO space, which gives information about the cost process of the discretely adjusted portfolio.

Download / Up

44

Report #44, March 2000 (Ini 3)

Harald Hruschka

Semi-Parametrische Marktanteilsmodellierung

In the empirical study presented market share models with semi-parametric additive brand attractions attain better fits both
according to an information criterion like AIC that penalizes a model for degrees of freedom used and according to error measures determined by cross validation or bootstrapping. Higher flexibility compared to strictly parametric models leads to more reliable measurements of the effects of marketing instruments. Moreover, marginal effects and price elasticities computed on the basis of the semi-parametric model differ qualitatively from those obtained for the parametric counterparts. The more flexible market share model also has different managerial implications than its parametric relatives as prices and higher profits determined by the solution concept of Fictitious Play show.

In der vorliegenden empirischen Untersuchung erreichen Marktanteilsmodelle mit semi-parametrischen additiven Markenattraktionen bessere Anpassungsmaáe sowohl nach einem Informationskriterium wie AIC, das ein Modell für die Anzahl verbrauchter Freiheitsgrade bestraft, als auch nach mittels Kreuzvalidierung oder Bootstrapping bestimmten Fehlermaßen. Die höhere Flexibilität gegenüber strikt parametrischen Modellen führt zu einer verläßlicheren Messung der Effekte von Marketing-Instrumenten. Außerdem unterscheiden sich marginale Effekte und Preiselastizitäten, die auf Grundlage des semi-parametrischen Modells berechnet werden, qualitativ von jenen, die man für die parametrischen Alternativen erhält. Das flexiblere Marktanteilsmodell impliziert unterschiedliche, mit Gewinnsteigerungen verbundene optimale Entscheidungen, wie mit Hilfe des Lösungskonzepts Fictitious Play bestimmte Preise und Gewinne zeigen.

Download / Up

454

Report #45, July 2000 (Ini 1)

Christian Schittenkopf, Peter Tino, Georg Dorffner

The Benefit of Information Reduction for Trading Strategies

In Motivated by previous findings that discretization of   financial time series can effectively filter the data and reduce the noise, this experimental study compares the trading performance of predictive models based on different modelling paradigms in a realistic setting. Different methods ranging from real-valued time series models to predictive models on a symbolic level are applied to predict the daily change in volatility of two major stock indices.

The  predicted volatility changes are interpreted as trading signals for  buying or selling a straddle portfolio on the underlying stock   index. Profits realized by this trading strategy are tested for  statistical significance taking into account transactions costs. The results indicate that symbolic information processing is a promising approach to financial prediction tasks undermining the hypothesis of efficient capital markets.

Download / Up

464

Report #46, July 2000 (Ini 1)

Peter Tino, Christian Schittenkopf, Georg Dorffner

Temporal Pattern Recognition in Noisy Non-stationary Time Series Based on Quantization into Symbolic Streams: Lessons Learned from Financial Volatility Trading

In this paper we investigate the potential of the analysis of noisy non-stationary time series by quantizing it into streams of discrete symbols and applying finite-memory symbolic predictors. The main argument is that careful quantization can reduce the noise in the time series to make model estimation more amenable given limited numbers of samples that can be drawn due to the non-stationarity in the time series. As a main application area we study the use of  such an analysis in a realistic setting involving financial forecasting and trading. In particular, using historical data, we simulate the trading of straddles on the financial indexes DAX and FTSE 100 on a daily basis, based on predictions of the daily volatility differences in the underlying indexes. We propose a parametric, data-driven quantization scheme which transforms temporal patterns in the series of daily volatility changes into grammatical and statistical patterns in the corresponding symbolic streams. As symbolic predictors operating on the quantized streams we use the classical fixed-order Markov models, variable memory length Markov models and a novel variation of fractal-based predictors introduced in its original form in (Tino, 2000b). The fractal-based predictors are designed to efficiently use deep memory. We compare the symbolic models with continuous techniques such as time-delay neural networks with continuous and categorical outputs, and GARCH models. Our experiments strongly suggest that the robust information reduction achieved by quantizing the real-valued time series is highly beneficial. To deal with non-stationarity in financial daily time series, we propose two techniques that combine ``sophisticated'' models fitted on the training data with a fixed set of simple-minded symbolic predictors not using older (and potentially misleading) data in the training set. Experimental results show that by quantizing the volatility differences and then using symbolic predictive models, market makers can generate a statistically significant excess profit. However, with respect to our prediction and trading techniques, the option market on the DAX does seem to be efficient for traders and non-members of the stock exchange. There is a potential for traders to make an excess profit on the FTSE 100. We also mention some interesting observations regarding the memory structure in the studied series of daily volatility differences.

Download / Up

47

Report #47, August 2000 (Ini 1)

Christian Schittenkopf, Georg Dorffner

Risk-neutral Density Extraction from Option Prices: Improved Pricing with Mixture Density Networks

One of the central goals in finance is to find better models for pricing and hedging financial derivatives such as call and put options. We present a semi-nonparametric approach to risk-neutral density extraction from option prices which is based on an extension of the concept of mixture density networks. The central idea is to model the shape of the risk-neutral density in a flexible, non-linear way as a function of the time horizon. Thereby, stylized facts such as negative skewness and excess kurtosis are captured. The approach is applied to a very large set of intraday options data on the FTSE 100 recorded at LIFFE. It is shown to yield significantly better results in terms of out-of-sample pricing in comparison to the basic Black-Scholes model and to an extended model adjusting the skewness and kurtosis terms. From the perspective of risk management, the extracted risk-neutral densities provide valuable information about market expectations.

Download / Up

Report #48, November 2000 (Ini 3)

Harald Hruschka

An Artificial Neural Net Attraction Model (ANNAM) to Analyze Market Share Effects of Marketing Instruments

Attraction models are very popular in marketing research for studying the effects of marketing instruments on market shares. However, so far the marketing literature only considers attraction models with certain functional forms that exclude threshold or saturation effects on attraction values. We can achieve greater flexibility by using the neural net based approach introduced here. This approach assesses brands' attraction values by means of a perceptron with one hidden layer. The approach uses log-ratio transformed market shares as dependent variables. Stochastic gradient descent followed by a quasi-Newton method estimates parameters. For store-level data, neural net models perform better and imply a price response that is qualitatively different from the well-known multinomial logit attraction model. Price elasticities of neural net attraction models also lead to specific managerial implications in terms of optimal prices.

Download / Up

Report #49, January 2001 (Ini 3)

Harald Hruschka, Werner Fettes, Markus Probst

Die Bewährung von Ankerpreismodellen bei der Erklärung der Markenwahl

We evaluate reference price models with regard to their ability to explain brand choices of individual households. Reference price models are of the adaptive expectations and extrapolative expectations types. Brand choice is analyzed by means of multinomial logit (MNL) models. We specify the deterministic utility component of MNL models as both conventional linear function and nonlinear function. Nonlinear utility is approximated by an appropriate neural network, a feedforward multilayer perceptron with sigmoid hidden units. Reference price models of the extrapolative expectation type formed by lagged prices and a time trend are superior to those of the adaptive expectation type for household scanner panel data. Improvements of posterior probabilities of choice models due to the inclusion of reference prices, losses and gains are greater if nonlinear utility choice models are used.

Download / Up

Report #50, April 2001 (Ini 5)

In this paper we propose a multi-colony Ant System, where the colonies solve a multiobjective optimization problem concerned with goods transportation. The colonies differ from each other by the heuristic information, which guides their search through the solution space. Information exchange occurs as ants from one population observe the pheromone trails of other populations and decide whether or not to utilize this information. Furthermore, population sizes are adapted according to the relative firness of the populations. The results show the advantages of this approach over common Ant System approaches.

Download / Up

Report #51, April 2001 (Ini 5)

Karl Doerner, Richard F. Hartl, Marc Reimann

In this paper we propose a hybrid ACO approach to solve a full truckload transportation problem. Hybridization is achieved through the use of a problem specific heuristic. This heuristic is utilized both, to initialize the pheromone information and to construct solutions in the ACO procedure. The main idea is to use information about the required fleetsize, by initializing the system with a number of vehicles rather than opening vehicles one at a time as needed. Our results show the advantages of this new approach over more traditional, i.e. sequential, approaches.

Download / Up

Report #52, March 2001 (Ini 1)

Alfred Lehar, Martin Scheicher, Christian Schittenkopf

GARCH vs Stochastic Volatility: Option Pricing and Risk Management

This paper examines the out-of-sample performance of two common extensions of the Black-Scholes framework, namely a GARCH and a stochastic volatility option pricing model. The models are calibrated to intraday FTSE 100 option prices. We apply two sets of performance criteria, namely out-of-sample valuation errors and Value-at-Risk oriented measures. When we analyze the fit to observed prices, GARCH clearly dominates both stochastic volatility and the benchmark Black Scholes model. However, the predictions of the market risk from hypothetical derivative positions show sizable errors. The fit to the realized profits and losses is poor and there are no notable differences between the models.
Overall we therefore observe that the more complex option pricing models can improve on the Black Scholes methodology only for the purpose of pricing, but not for the Value-at-Risk forecasts.

Download / Up

Report #53, May 2001 (Ini 1 + 3)

Binary survey data from the Austrian National Guest Survey conducted in the summer season of 1997 was used to identify behavioral market segments on the basis of vacation activity information. The bagged clustering methodology applied overcomes a number of difficulties typically encountered when partitioning clustering algorithms are applied to large binary data sets. Besides rendering more stable results in the sense of reproducibility and making the yet unsolved question of the correct number of clusters to choose less important by a hierarchical step of analysis at the end of the procedure, the bagged clustering approach eases interpretation of segment profiles as classically given by the mean variable values per segment and thus markedly improves the investigation and visualization of unobserved heterogeneity within the field of exploratory market segmentation.

Up

Report #54, May 2001 (Ini 3)

Sara Dolnicar

Perceptions based market segmentation (PBMS, Mazanec & Strasser, 2000) is a simple framework for market structure analysis integrating the issues of segmentation and positioning. The only requirement is the availability of three-way data (numerous respondents evaluate numerous brands according to numerous attributes). The implicit consideration of interrelations between positioning and segmentation prevents unharmonized strategic marketing decisions and enables managers with clear strategic goals to analyze market information in depth and arrive at a profound basis for segmentation and positioning decisions. In this study, PBMS is applied to deodorant data. The simultaneous treatment of all three data dimensions enables insights into deodorant brand images (among men) that go far beyond analysis of average perceptions for each brand over all respondents.

Up

Report #55, May 2001 (Ini 1)

This paper introduces ideas and methods for testing for structural change in linear regression models and presents how these have been realized in an R package called strucchange. It features tests from the generalized fluctuation test framework as well as from the F test (Chow test) framework. Extending standard significance tests it contains methods to fit, plot and test empirical fluctuation processes (like CUSUM, MOSUM and estimates-based processes) on the one hand and to compute, plot and test sequences of F statistics with the supF, aveF and expF test on the other. Thus, it makes powerful tools available to display information about structural changes in regression relationships and to assess their significance. Furthermore it is described how incoming data can be monitored online.

Download / Up

Report #56, June 2001 (Ini 1 + 5)

The Effect of Incentive Schemes and Organizational Arrangements on the New Product Development Process

This paper proposes a new model for studying the new product development process in an artificial environment. We show how connectionist models can be used to simulate the adaptive nature of agents‘ learning exhibiting similar behavior as practically experienced learning curves. We study the impact of incentive schemes (local, hybrid and global) on the new product development process for different types of organizations. Sequential organizational structures are compared to two different types of team-based organizations, incorporating methods of Quality Function Deplyment such as the House of Quality. A key finding of this analysis is that hte firms‘ organizational structure and agents‘ incentive system signigicantly interact. We show that the House of Quality is less affected by the incentive scheme than firms using a Trial & Error approach. This becomes an important factor for new product success when the agents‘ performance measures are conflicting.

Download / Up

Report #57, June 2001 (Ini 5)

Choice-Based Conjoint (CBC) models are often used for pricing decisions, especially when scanner data models cannot be applied. Up to date, it is unclear how Choice-Based Conjoint (CBC) models perform in terms of forecasting real-world shop data. In this contribution, we measure the performance of a Latent Class CBC model not by means of an experimental hold-out sample but via aggregate scanner data. We find that the CBC model does not accurately predict real-world market shares, thus leading to wrong pricing decisions. In order to improve its forecasting performance, we propose a correction scheme based on scanner data. Our empirical analysis shows that the hybrid method improves the performance measures considerably.

Download / Up

Report #58, June 2001 (Ini 5)

Conjoint analysis is one of the most important tools to support product development, pricing and positioning decisions in management practice. For this purpose various models have been developed. It is widely accepted that models that take consumer heterogeneity into account, outperform aggregate models in terms of hold-out tasks. The aim of our study is to investigate empirically whether predictions of choice-based conjoint models which incorporate heterogeneity can successfully be generalized to a whole market. To date no studies exist that examine the real world performance of choice-based conjoint models by use of aggregate scanner panel data. Our analysis is based on four commercial choice-based conjoint pricing studies including a total of 43 stock keeping units (SKU) and the corresponding weekly scanning data for approximately two years. An aggregate model serves as a benchmark for the performance of two models that take heterogeneity into account, hierarchical Bayes and latent class. Our empirical analysis demonstrates that, in contrast to the performance using hold-out tasks, the real world performance of hierarchical Bayes and latent class is similar to the performance of the aggregate model. Our results indicate that heterogeneity cannot be generalized to a whole market and suggest that aggregate models are sufficient to predict market shares.

Download / Up

Report #59, June 2001 (Ini 2 + 6)

This article considers three standard asset pricing models with adaptive agents and stochastic non-stationary dividends. We assume that the parameters are estimated by exponential smoothing, such that prices and returns remain random variables. This paper provides sufficient conditions for the ergodicity of the return process and checks whether the perceived law assumed by the bounded rational agents can be considered to be sound with the returns observed.

Download / Up

Report #60, June 2001 (Ini 6)

We introduce a simple asset pricing model with two types of adaptively learning traders, fundamentalists and technical traders. Traders update their beliefs according to past performance and to market conditions. The model generates endogenous price fluctuations and captures some stylized facts observed in real returns data, such as excess volatility, fat tails of returns distributions, volatility clustering, and long memory. We show that the results are quite robust w.r.t.\ to different choices for the performance measure.

Download / Up

Report #61, August 2001 (Ini 1 + 3)

Sara Dolnicar, Friedrich Leisch

Binary survey data from the Austrian National Guest Survey conducted in the summer season of 1997 were used to identify behavioral market segments on the basis of vacation activity information. Bagged clustering overcomes a number of difficulties typically encountered when partitioning large binary data sets: The partitions have greater structural stability over repetitions of the algorithm and the question of the ``correct'' number of clusters is less important because of the hierarchical step of the cluster analysis. Finally, the bootstrap part of the algorithm provides means for assessing and visualizing segment stability for each input variable.

Download / Up

Report #62, September 2001 (Ini 5)

Doerner, K., Gutjahr, W. J., Hartl, R. F., Strauss, C., Stummer,

Investitionsentscheidungen bei mehrfachen Zielsetzungen und künstliche Ameisen

Die Auswahl des attraktivsten Portfolios von Investitionsprojekten zählt zu den kritischen Managementaufgaben. Angesichts mehrfacher Zielsetzungen und komplexer Projektabhängigkeiten bietet sich dazu ein zweistufiges Vorgehen an, das zunächst effiziente Portfolios identifiziert und den Entscheidungsträger anschließend bei der Suche in diesem Lösungsraum unterstützt. Bei einer großen Zahl an Vorschlägen können die möglichen Projektkombinationen aber nicht mehr in akzeptabler Zeit vollständig enumeriert werden. Adaptierte Meta-Heuristiken bieten hier einen Kompromiß zwischen dem Wunsch nach exakter Bestimmung aller Pareto-optimalen Investitionsprogramme und dem dazu nötigen Rechenaufwand. Dieser Beitrag beschreibt den entsprechenden Einsatz künstlicher Ameisen und diskutiert erste numerische Ergebnisse.

Download / Up

Report #63, December 2001 (Ini 5)

K. Doerner, M. Gronalt, R. F. Hartl, M. Reimann, C. Strauss, M. Stummer

SavingsAnts for the Vehicle Routing Problem

In this paper we propose a hybrid approach for solving vehicle routing problems. The main idea is to combine an Ant System (AS) with a problem specific constructive heuristic, namely the well known Savings algorithm. This differs from previous approaches, where the subordinate heuristic was the Nearest Neighbor algorithm initially proposed for the TSP. We compare our approach with some other classic, powerful meta-heuristics and show that our results are competitive.

Download / Up

Report #64, March 2002 (Ini 1)

Achim Zeileis, Friedrich Leisch, Christian Kleiber, Kurt Hornik

Monitoring Structural Change in Dynamic Econometric Models

The classical approach to testing for structural change employs retrospective tests using a historical data set of a given length. Here we consider a wide array of fluctuation-type tests in a monitoring situation - given a history period for which a regression relationship is known to be stable, we test whether incoming data are consistent with the previously established relationship. Procedures based on estimates of the regression coefficients are extended in three directions: we introduce (a) procedures based on OLS residuals, (b) rescaled statistics and (c) alternative asymptotic boundaries. Compared to the existing tests our extensions offer better power against certain alternatives, improved size in finite samples for dynamic models and ease of computation respectively. We apply our methods to two data sets, German M1 money demand and U.S. labor productivity.

Download / Up

Report #65, February 2002 (Ini 2)

Mark Davis, Walter Schachermayer, Robert Tompkins

Pricing, No-arbitrage Bounds and Robust Hedging of Installment Options

An installment option is a European option in which the premium, instead of being paid up-front, is paid in a series of installments. If all installments are paid the holder receives the exercise value, but the holder has the right to terminate payments on any payment date, in which case the option lapses with no further payments on either side. We discuss pricing and risk management for these options, in particular the use of static hedges, and also study a continuous-time limit in which premium is paid at a certain rate per unit time.

Download / Up

Report #66, February 2002 (Ini 2)

Walter Schachermayer

Optimal Investment in Incomplete Financial Markets

We give a review of classical and recent results on maximization of expected utility for an investor who has the possibility of trading in a financial market. Emphasis will be given to the duality theory related to this convex optimization problem.

Download / Up

Report #67, February 2002 (Ini 2)

Mark Davis, Walter Schachermayer, Robert Tompkins

Installment Options and Static Hedging

An installment option is an European option in which the premium, instead of being paid up-front, is paid in a series of installments. If all installments are paid the holder receives the exercise value, but the holder has the right to terminate payments on any payment date, in which case the option lapses with no further payments on either side. We discuss pricing and risk management for these options, in particular the use of static hedges to obtain both no-arbitrage pricing bounds and very effective hedging strategies with almost no vega risk.

Download / Up

Report #68, October 2002 (Ini 5)

Marc Reimann, Karl Doerner, Richard F. Hartl

Insertion based Ants for Vehicle Routing Problems with Backhauls and Time Windows

In this paper we present and analyze the application of an Ant System to the Vehicle Routing Problem with Backhauls and Time Windows (VRPBTW). At the core of the algorithm we u se an Insertion procedure to construct solutions. We provide results on the learning and runtime behavior of the algorithm as well as a comparison with a custom made heuristic for the problem.

Download / Up

Report #69, March 2002 (Ini 1)

Friedrich Leisch

Sweave: Dynamic Generation of Statistical Reports Using Literate Data Analysis

Sweave combines typesetting with LaTeX and data anlysis with S into integrated statistical documents. When run through R or Splus, all data analysis output (tables, graphs, ...) is created on the fly and inserted into a final LaTeX document.  Options control which parts of the original S code are shown to or hidden from the reader, respectively.   Many S users are also LaTeX users, hence no new software has to be learned.   The report can be automatically updated if data or analysis change, which allows for truly reproducible research.

Download / Up

Report #70, June 2002 (Ini 2)

Sylvia Frühwirth-Schnatter

In the present paper we discuss the problem of estimating model likelihoods from the MCMC output for a general mixture and switching model. Estimation is based on the method of bridge sampling (Meng and Wong, 1996), where the MCMC sample is combined with an iid sample from an importance density. The importance density is constructed in an unsupervised manner from the MCMC output using a mixture of complete data posteriors. Whereas the importance sampling estimator as well as the reciprocal importance sampling estimator are sensitive to the tail behaviour of the importance density, we demonstrate that the bridge sampling estimator is far more robust in this concern. Our case studies range from computing marginal likelihoods for a mixture of multivariate normal distributions, testing for the inhomogeneity of a discrete time Poisson process, to testing for the presence of Markov switching and order selection in the MSAR model.

Download / Up

Report #71, June 2002 (Ini 2 + 3)

Thomas Otter, Regina Tüchler, Sylvia Frühwirth-Schnatter

This paper presents the fully Bayesian analysis of the latent class model using a new approach towards MCMC estimation in the context of mixture models. The approach starts with estimating unidentified models for various numbers of classes. Exact Bayes factors are computed by the bridge sampling estimator to compare different models and select the number of classes. Estimation of the unidentified model is carried out using the random permutation sampler. From the unidentified model estimates for model parameters that are not class specific are derived. Then, the exploration of the MCMC output from the unconstrained model yields suitable identifiability constraints. Finally, the constrained version of the permutation sampler is used to estimate group specific parameters. Conjoint data from the Austrian mineral water market serve to illustrate the method.

Download / Up

Report #72, June 2002 (Ini 2 + 3)

Sylvia Frühwirth-Schnatter, Regina Tüchler, Thomas Otter

In the present paper we consider Bayesian estimation of a finite mixture of models with random effects which is also known as the heterogeneity model. First, we discuss the properties of various MCMC samplers that are obtained from full conditional Gibbs sampling by grouping and collapsing. Whereas full conditional Gibbs sampling turns out to be sensitive to the parameterization chosen for the mean structure of the model, the alternative sampler is robust in this respect. However, the logical extension of the approach to the sampling of the group variances does not further increase the efficiency of the sampler. Second, we deal with the identifiability problem due to the arbitrary labeling within the model. Finally, a case study involving metric Conjoint analysis serves as a practical illustration.

Download / Up

Report #73, June 2002 (Ini 2 + 3)

Regina Tüchler, Sylvia Frühwirth-Schnatter, Thomas Otter

In this paper we carry out fully Bayesian analysis of the general heterogeneity model, which is a mixture of random effects model, and its special cases, the random coefficient model and the latent class model. Our application comes from Conjoint analysis and we are especially interested in what is gained by the general heterogeneity model in comparison to the other two when modeling consumers' heterogeneous preferences.

Download / Up

Report #74, October 2002 (Ini 1 + 3)

David Meyer, Christian Buchta, Alexandros Karatzoglou, Friedrich Leisch, Kurt Hornik

A Simulation Framework for Heterogeneous Agents

We introduce a generic simulation framework suitable for agent-based simulations featuring the support of heterogeneous agents, hierarchical scheduling and flexible specification of design parameters. One key aspect of this framework is the design specification we use an XML-based format which is simple-structured yet still enables the design of flexible models. Another issue in agent-based simulations, especially when ready-made components are used, is the heterogeneity arising from both the agents' implementations and the underlying platforms. To tackle these kind of obstacles, we introduce a wrapper technique for mapping the functionality of agents living in an interpreter-based environment to a standardized JAVA interface, thus facilitating the task for any control mechanism (like a simulation manager) because it has to handle only one set of commands for all agents involved. Again, this mapping is made by an XML-based definition format. We demonstrate the technique by applying it to a simple sample simulation of two mass marketing firms operating in an artificial consumer environment.

Download / Up

Report #75, October 2002 (Ini 1)

David Meyer,   Friedrich Leisch, Torsten Hothorn, Kurt Hornik

StatDataML An XML Format for Statistical Data

In order to circumvent common difficulties in exchanging statisticaldata between heterogeneous applications (format incompatibilities, technocentric data representation), we introduce an XML-based markup language for statistical data, called StatDataML. After comparing StatDataML to other data concepts, we detail the design which borrows from the language S, such that data objects are basically organized as recursive and non-recursive structures, and may also be supplemented with meta-information.

Download / Up

Report #76, October 2002 (Ini 3 + 5)

Andreas Mild, Thomas Retterer

An improved collaborative filtering approach for predicting cross-category purchases
based on binary market basket data

Retail managers have been interested in learning about cross-category purchase behavior of their customers for a fairly long time. More recently, the task of inferring cross-category relationship patterns among retail assortments is gaining attraction due to its promotional potential within recommender systems used in online environments. Collaborative filtering algorithms are frequently used in such settings for the prediction of choices, preferences and/or ratings of online users. This paper investigates the suitability of such methods for situations when only binary pick-any customer information (i.e., choice/nonchoice of items, such as shopping basket data) is available. We present an extension of collaborative filtering algorithms for such data situations and apply it to a real-world retail transaction dataset. The new method is benchmarked against more conventional algorithms and can be shown to deliver superior   results in terms of predictive accuracy.

Download / Up

Report #77, November 2002 (Ini 1)

Evgenia Dimitriadou, Andreas Weingessel, Kurt Hornik

A Mixed Ensemble Approach for the Semi-Supervised Problem

In this paper we introduce a mixed approach for the semi-supervised data problem. Our approach consists of an ensemble unsupervised learning part where the labeled and unlabeled points are segmented into clusters. Continuing, we take advantage of the a priori information of the labeled points to assign classes to clusters and proceed to predicting with the ensemble method new incoming ones. Thus, we can finally conclude classifying new data points according to the segmentation of the whole set and the association of its clusters to the classes.

Download / Up

Report #78, November 2002 (Ini 1)

David Meyer,   Friedrich Leisch, Kurt Hornik

Benchmarking Support Vector Machines

Support Vector Machines (SVMs) are rarely benchmarked against other classification or regression methods. We compare a popular SVM implementation (libsvm) to 16 classification methods and 9 regression methods—all accessible through the software R—by the means of standard performance measures (classification error and mean squared error) which are also analyzed by the means of bias-variance decompositions. SVMs showed mostly good performances both on classification and regression tasks, but other methods proved to be very competitive.

Download / Up

Report #79, November 2002 (Ini 5)

Roland Bauer, Sabine T. Köszegi, Michaela Wolkerstorfer

Measuring the Degree of Virtualization - An Empirical Analysis in two Austrian Industries

Strategic management literature suggests that especially in young and dynamic industries Virtual Corporations are more likely to emerge, as this type of organization is flexible enough to deal with rapidly changing environments. This paper challenges the proposition that environ-mental uncertainty and technological change lead to organizational adaptation towards virtual structures. We analyzed companies of two Austrian industries, data processing and engineering, which are characterized by different rates of innovation and environmental uncertainty, and compare their strategic, structural, and process characteristics by measuring their Degree of Virtualization. Results indicate almost no difference in the Degree of Virtualization. From these findings, we draw implications for the theoretical concept of Virtual Corpora-tions as well as for empirical research.

Download / Up

Report #80, June 2003 (Ini 1)

Achim Zeileis, Kurt Hornik

Generalized M-Fluctuation Tests for Parameter Instability

A general class of fluctuation tests for parameter instability in an M-estimation framework is suggested. The tests are based on partial sum processes of M-estimation scores for which functional central limit theorems are derived under the null hypothesis of parameter stability and local alternatives. Special emphasis is given to parameter instability in (generalized) linear regression models and it is shown that the introduced M-fluctuation tests contain a large number of parameter instability or structural change tests known from the statistics and econometrics literature. The usefulness of the procedures is illustrated using artificial data and data for the German M1 money demand, historical demographic time series from Großarl, Austria, and youth homicides in Boston.

Download / Up

Report #82, Nov 2003 (Ini 1)

Thorsten Hothorn, Friedrich Leisch, Achim Zeileis, Kurt Hornik

The Design and Analysis of Benchmark Experiments

The assessment of the performance of learners by means of benchmark experiments is established exercise. In practice, benchmark studies are a tool to compare the performance of several competing algorithms for a certain learning problem. Cross-validation or resampling techniques are commonly used to derive point estimates of the performances which are compared to identify algorithms with good properties. For several benchmarking problems, test procedures taking the variability of those point estimates into account have been suggested. Most of the recently proposed inference procedures are based on special variance estimators for the cross-validated performance.
We introduce a theoretical framework for inference problems in benchmark experiments and show that standard statistical test procedures can be used to test for differences in the performances. The theory is based on well defined distributions of performance measures which can be compared with established tests. To demonstrate the usefulness in practice, the theoretical results are applied to benchmark studies in a supervised learning situation based on artificial and real-world data.

Download / Up

Report #83, Nov 2003 (Ini 2)

Tatiana Miazhynskaia, Sylvia Frühwirth-Schnatter, Georg Dorffner

A Comparison of Bayesian Model Selection based on MCMC with an application to GARCH-Type Models

This paper presents a comprehensive review and comparison of five computational
methods for Bayesian model selection, based on MCMC simulations from posterior
model parameter distributions. We apply these methods to a well-known and important
class of models in financial time series analysis, namely GARCH and GARCH-t models
for conditional return distributions (assuming normal and t-distributions). We compare
their performance vis--vis the more common maximum likelihood-based model
selection on both simulated and real market data. All five MCMC methods proved feasible
in both cases, although differing in their computational demands. Results on simulated
data show that for large degrees of freedom (where the t-distribution becomes
more similar to a normal one), Bayesian model selection results in better decisions in
favour of the true model than maximum likelihood. Results on market data show the
feasibility of all model selection methods, mainly because the distributions appear to
be decisively non-Gaussian.

Download / Up

Report #84, Nov 2003 (Ini 6)

Tatiana Miazhynskaia, Georg Dorffner, Engelbert J. Dockner

Non-linear versus Non-gaussian Volatility Models in Application to Different Financial Markets

Download / Up

Report #86, Nov 2003 (Ini 1)

Friedrich Leisch

FlexMix: A general framework for finite mixture models and latent class regression in R

Download/ Up

Funded by the Austrian Science Foundation