Stochastic Models of Brand Choice
and the Analysis
of Brand Switching Data via Latent
Class Analysis
Panagiota Anastasiou and John
Panaretos
Stochastic choice models differ in
a number of ways from the other types of consumer choice behavior models.
They also differ as to how they handle customer heterogeneity. Some models
try to segment customer into homogeneous subsets and analyze them separately.
Brand choice models are usually distinguished according to how they deal
with population heterogeneity, purchase event feedback and exogenous market
factors. Much of the information used in a number of popular brand choice
models can be extracted from a brand-switching matrix. The elements in
this matrix are obtained from purchase information on two occasions separated
in time.
In this paper we present the
most popular methods that have been used until now in the analysis of brand
switching matrix, giving emphasis to latent class models. Questions that
are dealt with are the following: Which of these methods give us the best
information about the consumer’s behavior? Assuming misclassification error
is present, which technique best recovers the true market structure? Are
there situations under which one model may fit the data better than another
one?
PANAGIOTA ANASTASIOU
Department of Statistics, AUEB
Athens, Greece
Income Distribution: An Application
to Greek Data
X. Balogianni and C. Dimaki
Income arises from a number of different
sources and in general it is not equally distributed. On the contrary the
unequal distribution of income and wealth is one of the most prominent
features of our society. The study of income distribution is the focus
of the present paper which consists of two parts.
The first part covers a number of
important theoretical aspects of the subject. Presentation of certain distributions
that may be used to describe income distribution and discussion of the
criteria that they should satisfy. Presentation of income distribution
generation formulas. Identification of the factors affecting income inequality.
Suggestion of measures for income inequality. Study of the relation between
reported and true income. The second part is an application of the Pareto’s
Law to Greek income data for the period 1960-1995.
XANTHI BALOGIANNI
Department of Statistics, AUEB
Athens, Greece
Predictability and Model Selection
in the Context of ARCH Models
Stavros Degiannakis and Evdokia
Xekalaki
Autoregressive conditional heteroscedastic
(ARCH) models, introduced by Engle (1982), have been widely used in financial
time series analysis and particularly in analyzing the risk of holding
an asset, evaluating the price of an option, forecasting time varying confidence
intervals and obtaining more efficient estimators under the existence of
heteroscedasticity. In the recent literature one can find several forms
of ARCH models have been considered for the description of some of the
characteristics of financial markets. The problem of selecting the model
that describes best the movement of the series under study is therefore
of practical importance.
Most of the methods used in the
literature for selecting the appropriate model are based on evaluating
the ability of the models to describe the data. Standard model selection
criteria such as the Akaike Information Criterion (1973) and the Schwartz
Bayesian Criterion (1978) have been widely used in the ARCH literature,
despite the fact that their statistical properties in the ARCH context
are unknown.
In this paper, an alternative model
selection approach is examined based on the evaluation of the predictability
of the models. Attention is focused on the behavior of the estimators of
conditional means or of conditional variances. In both cases the Correlated
Gamma Ratio Distribution (CGR) developed by Panaretos et al. (1997) in
the context of linear models, is being considered to be applied in the
context of ARCH models.
STAVROS DEGIANNAKIS
Department of Statistics, AUEB
Athens, Greece
Multinomial Probit Model
Stefanos G. Giakoumatos, Petros
Dellaportas, and Dimitris N. Politis
We study the multivariate unobserved ARCH and the latent factor ARCH models which capture the cross-covariance changes commonly appearing in the financial time-series. We adopt Bayesian inference via Markov Chain Monte Carlo and we present easy to implement algorithms which use only Gibbs sampling steps. We illustrate our methodology with data from four exhange rate series.
STEFANOS GIAKOUMATOS
Department of Statistics, AUEB
Athens, Greece
Analysis and Comparison of Greek Parliamentary Electoral Systems of the Period 1974-1996
Aikaterini Kalogirou and John Panaretos
The most fundamental element of representative
democracy is the Electoral System, as it translates vote totals into parliamentary
seats. An important topic is the choice of the electoral system, which
will be applied in an election, because parliamentary seats allocated to
political parties differ when a different system is used.
In the present paper we focus attention
on the Greek parliamentary electoral systems of the period 1974-1996. Our
aim is to analyse statistically these electoral systems and compare them
by using measures of disproportionality. We also examine the behaviour
of electoral systems when one or more electoral parameters change. Sensitivity
analysis is performed in order to find out which percentages of total votes
give majority of parliamentary seats to a party, for the six electoral
systems used in Greek elections in the period under study.
AIKATERINI KALOGIROU
Department of Statistics, AUEB
Athens, Greece
Making Predictions of the
Consumer Price Index
and its Subdivisions with Applications
to Greek Data
Polyxeni Liva and Evdokia Xekalaki
The compilation of a price index
that truly reflects the present consumption patterns of the economy, is
necessary for a number of reasons, the most important being the need to
record changes linked to the value of money. Associated to the achievement
of price stability by member states of the European Union, the Consumer
Price Index is a statistical measure, which reflects the effects of price
variations for a given quantitative and qualitative composition of the
goods and services purchased by households during the period covered by
the Household Budget Survey. The importance of the use of the CPI in the
formulation of social and economic policies makes it obvious that predicting
its value and as well as the values of its subdivisions is an interesting
problem.
In this paper, the problem of forecasting
future values of the basic groups of the Consumer Price Index is examined.
For this purpose, 12 time series have been considered, that correspond
to the indexes of the 12 basic subdivisions of the Consumer Price Index,
for the period from January 1988 to September 1998. The forecast of the
General consumer price index is obtained as a weighted average of the forecasts
of the indexes of the basic groups. The source of the data is the National
Statistical Service of Greece. The Box and Jenkins method is applied for
the creation of appropriate forecasting models of the12 basic groups of
the Consumer Price Index, using SPSS for the analysis of the time series,
the estimation of the parameters of the models and for the diagnostic checking.
POLYXENI LIVA
Department of Statistics, AUEB
Athens, Greece
A New Process Capability Index
that is Based on the Proportion of Conformance
Michael Perakis and Evdokia Xekalaki
Process capability indices (PCI’s)
aim to quantify the capability of a process to produce according to some
given specifications. These specifications are determined through the lower
specification limit, the upper specification limit and the target value.
A variety of PCI’s can be found in the literature and the most prominent
among them are Cp, Cpk, Cpm and Cpmk. The basic drawback of these four
indices is that they have been developed under the somewhat restrictive
assumption that the distribution of the examined process is normal. For
this reason, some modifications of these indices have been proposed in
the literature. However, the construction of confidence limits for these
modifications without resorting to the method of bootstrap is extremely
cumbersome. Moreover, some of these indices do not have a direct association
to the proportion of conformance of the process, i.e. to the probability
of producing within the specification area.
In this paper, a new PCI is proposed,
which is based on the proportion of conformance of the examined process
and has several appealing features. Indeed, this index is simple in its
assessment and interpretation and is applicable to processes with normal
or non-normal distributions. Further, it can be used under either unilateral
or bilateral tolerances and the assessment of confidence limits for its
true value is not very involved. Moreover, point estimators and confidence
limits for the new index are constructed, under several distributional
assumptions.
MICHAEL PERAKIS
Department of Statistics, AUEB
Athens, Greece
Refined Asymptotics for Fluid
Queues with Multiplexed On-Off Inputs
D. Pinotsi and M.A. Zazanis
We consider queues with multiplexed On-Off fluid inputs under the assumption of exponential Off periods. Refined asymptotics based on the results of Willekens and Teugels (1992) are obtained for the workload distribution under the assumption of On periods with regularly varying tails. The analysis, based on the embedded M/G/1 queue, extends the results in Jelenkovic and Lazar (1999).
DIMITRA PINOTSI
Department of Statistics, AUEB
Athens, Greece
Some Statistical Analysis of Greek
Crime Data
Alexandra Tsiamtsiouri and John
Panaretos
The increase of criminal activity
worldwide calls for cooperation in order to combat crime. All over the
world, scientists, including Statisticians, of many seemingly unrelated
fields are cooperating in order to find methods to prevent or to reduce
crime.
Kent S. Borowick (1997) in "Analysis
of low probability count data with applications in crime analysis"(PhD
thesis, Baylor University) has performed several statistical analyses on
crime data. Using his approach, in this paper we analyze Greek crime data.
The data are reported crimes per month or per year for the periods 1987-1997
(for the monthly data) and 1982-1997 (for the annual data). Some of the
examined crimes are Commonly Dangerous Crimes, Robbery, Theft, Murder,
Rape, Arson, Personal Vengeance, Drug Violations, Smuggling of Antiquities
and Beggary. Data about the offenders are also examined.
Multivariate methods such as Principal
Component Analysis (PCA), Multidimensional Scaling and Clustering methods
are employed in order to map the crime activity in Greece and point out
the areas where certain crimes seem to be on the increase. Quality Control
is employed in order to make inferences about the months when different
crimes are more possible to occur.
ALEXANDRA TSIAMTSIOURI
Department of Statistics, AUEB
Athens, Greece
Large Analysis of a Class of Multivariate
ARCH and GARCH Models
I. D. Vrontos, P. Dellaportas,
and D. N. Politis
An analysis of a class of multivariate ARCH and GARCH models is proposed consisting of parameter estimation and model comparison. We study some models that already exist in the literature, as well as propose a new multivariate GARCH model. Bayesian and classical techniques are used for the estimation of the parameters of the models analyzed, and model comparisons are addressed via predictive distributions. We provide implementation details and illustrations using daily exchange rates of the Athens exchange market.
I. D. VRONTOS
Department of Statistics, AUEB
Athens, Greece
Discrete Time Series Models with
Application to Epileptic Seizures
Nikkie Yiokari and Evdokia Xekalaki
Daily seizure counts are a prime
tool in investigating epileptic disease and examining the effectiveness
of certain drugs. Does an epileptic fit make the patient prone to have
another fit in the near future? Do unobservable factors exist that may
contribute to an epileptic fit? These are a few of many questions that
arise in epileptic disease investigations.
In this paper questions of
this nature are investigated on the basis of data on epileptic fit occurrences.
These are viewed as time series of counts, i.e. as non-negative, integer-valued
stochastic processes in discrete time. We present models that are useful
for the modeling of discrete-time dependent counting processes and pay
special attention to the integer-valued autoregressive (INAR) and the switching
integer-valued autoregressive (SINAR) models. The SINAR model is fitted
to a series of daily counts of epileptic seizures of a patient. The EM-algorithm
is used in order to estimate the parameters of this model.
NIKKIE YIOKARI
Department of Statistics, AUEB
Athens, Greece