ENBIS-13 in Ankara

15 – 19 September 2013 Abstract submission: 5 February – 5 June 2013

My abstracts


The following abstracts have been accepted for this event:

  • Risk Assessment of Turkish External Debt by Using Hidden Markov Model

    Authors: Ceren Eda Can (Hacettepe University), Gül Ergün (Hacettepe University)
    Primary area of focus / application: Finance
    Keywords: Hidden Markov model, Normal distribution, Turkish external debt, External debt management, Borrowing limit, Country risk
    Submitted at 15-Apr-2013 10:45 by Ceren Eda Can
    External borrowing is an important source of income for developing countries. In the first stage, due to the inflows of foreign resources, external debt provides positive impacts on the economy of the country. However, at the time of the interest and principal payments, due to the outflow of internal resources, the external debt gives rise to negative impacts on the economy. The sustainability of external debts depends on the ability of the country to meet its current and future debt obligations and other expenses. If the liquidity and cash flow in the country is not sufficient to service its external debt, it may be required to refinance the debt, which can result in serious debt crisis in the country. The increasing of the existing external debts and payments leads to establish an effective and efficient external debt management.

    In particular, economic crisis in Turkey is mostly triggered by its external debt. Therefore, the external debt is the one of the most importand problems in Turkey. In the scope of Turkish debt management system, it is a crucial task to model the structure of Turkish external debt and then define both its borrowing limits and the country risk, which represents the credibility of the Turkey on international financial markets and represents the level for the sustainability of the debt from Turkey.

    The aim of this study is to model the return of total Turkish external debt stocks and some indicators of Turkish external debt by Hidden Markov Model (HMM). The financial time series has a dynamic structure and thus irrational rapid changes in the series can be a part of a pattern. Each pattern consists of different regime, which represents the economic situation. The long-term trend and short-term sideway movements can be defined by the change of regime. When modeling the financial time series, HMM allows for not only the change of regime but also the dependency between regimes. In this study, Normal distribution is chosen for the impacts of different regimes and then Normal-HMM is used to forecast the future level of Turkish external debt and also determine its structure in order to take necessary policies, which ensure economic stabilitiy.
  • Treatment of a Functional Input for the Optimization of a Computer-based Model

    Authors: Miguel Munoz Zuniga (IRSN), Yann Richet (IRSN), Gregory Caplin (IRSN)
    Primary area of focus / application: Design and analysis of experiments
    Keywords: Functional data, Dimension reduction, Functional optimization, Computer experiments, Kriging
    Submitted at 15-Apr-2013 11:13 by Miguel Munoz Zuniga
    16-Sep-2013 12:15 Treatment of a Functional Input for the Optimization of a Computer-based Model
    In the last decade computer experiment has become a key tool for the analysis of fundamental or industrial issues in an increasing number of fields as diverse as biology, sociology, mechanic, risk management and many others as soon as a computer simulation based model is available. Mathematics and in particular statistics are the versatile language on which design of computer experiments intrinsically rely. In this context, we propose an algorithm to solve a practical optimization problem involving functional data as the objective function argument.

    One of the main IRSN concern is the risk measurement all along the nuclear fuel cycle. The so called ”burnup” problematic is to consider an irradiated nuclear fuel assembly after its life in the core. One assembly can be characterized by its rate of combustion along its vertical axis (taken as a functional data), which might lead to an uncontrolled neutronic chain reaction. Hence, the ”burn-up” problem consists in finding the most penalizing ”burn-up” profile regarding the neutronic criticality coefficient all along the storage.

    In order to measure the storage criticality risk, a simulation chain model is used, starting from the depletion of the fuel, ending with the modeling of its neutronic behaviour. These two models are computed through the HPC platform PROMOTHÉE, driven by R. The algorithm presented involves functional data treatment and approximation, dimension reduction, Monte Carlo sampling, optimization and efficiently solves this functional optimization problem. Beyond these mathematical tools, we will also focus on the technical issues derived from launching the underlying mechanistic simulations, which are mostly not related to R, high CPU consuming and available through a remote computing cluster.
  • Bayesian Analysis and Prediction of Patients' Demands for Visits in Home Health Care

    Authors: Raffaele Argiento (CNR-IMATI), Alessandra Guglielmi (Dipartimento di Matematica, Politecnico di Milano), Ettore Lanzarone (CNR-IMATI), Inad Nawajah (Dipartimento di Matematica, Politecnico di Milano)
    Primary area of focus / application: Other: ISBA/IS Invited Session
    Keywords: Home care, Bayesian modeling and estimation, MCMC algorithm, Random effect
    Submitted at 15-Apr-2013 11:46 by Raffaele Argiento
    18-Sep-2013 10:35 Bayesian Analysis and Prediction of Patients' Demands for Visits in Home Health Care
    Home Care (HC) service consists of providing care to patients at their own home, without the necessity of bringing them to hospitals or nursing homes. This service allows a high quality of life for the assisted patients at, at the same time, a cost reduction for the health care system. Planning human resources is a difficult task and, for a good quality of planning, a knowledge of future demands for visits from patient is required.
    In the literature, several studies deal with stochastic models for representing patient conditions in the health care system but, to the best of our knowledge, few works deal with HC service and furthermore Bayesian approaches have not been considered in the HC context, yet.
    The aim of this talk is to propose a methodology for estimating and predicting the demand for care by HC patients in terms of number of visits N required in a defined time slot. Patients are characterized by a Care Profile (CP) which varies along
    with the time secondary to a periodic revision or sudden variations in
    health state. Our approach considers the joint distribution of N and CP over time as a
    conditional distribution of N given CP, times the marginal of the CPs; in addition, the transition between CP states is regulated by a homogeneous multistate Markov Chain. The proposed model is developed and validated considering the data of
    one of the largest HC providers in Italy. We obtain the posterior densities of model parameters through MCMC simulation and predict the number of visits of patients
    in future time slots. Results show the applicability of the approach in the practice
    and a good predictive fit of the model to the data.
  • On-line Process Monitoring Using Partial Correlations

    Authors: Tiago Rato (University of Coimbra), Marco Reis (University of Coimbra)
    Primary area of focus / application: Process
    Keywords: Process monitoring, Multivariate dynamical processes, Variable transformation, Partial correlations, Marginal correlations, Individual observations
    Submitted at 15-Apr-2013 14:06 by Tiago Rato
    Accepted (view paper)
    17-Sep-2013 09:40 On-line Process Monitoring Using Partial Correlations
    The monitoring of the process variability with multivariate individual observations raises a variety of challenges, mostly due to the need of a continuous estimation of the covariance matrix. Usual solutions consist in adopting a moving window approach or an updating scheme such as an EWMA recursion. This latter alternative has attracted more interest, and monitoring statistics have been proposed for the trace of the up-dated covariance matrix or regarding squared deviations from target. However, these methodologies are based only on the marginal covariance and therefore unable to discern local changes in the process structure. In a previous work, presented in the last ENBIS conference, it was shown that partial correlations are indeed useful for enhancing the detection and identification of structural changes, leading to superior results than its counterparts, when applied on non-overlapping windows. Therefore, in this work, we extend their application to on-line monitoring and establish a conceptual scheme that allows the transfer of most of the interesting monitoring statistics properties, to on-line applications. This relation is made by an equivalence equation that relates the forgetting factor of EWMA recursions (on-line case) with the number of non-overlapping observations (off-line case). Furthermore, a set of sensitivity enhancing transformations based on partial correlation are also proposed.
    The proposed methodologies were applied on multivariate systems and their performances were compared against current alternatives. The results obtained showed that the sensitivity enhancing transformations play a major role on the monitoring statistics performance, allowing them to detect faults more rapidly than with original variables.
  • A New High Order Fuzzy Time Series Method Utilizing on Fuzzy Rule Based Systems

    Authors: Murat Alper Başaran (Akdeniz University Faculty of Engineering at Alanya)
    Primary area of focus / application: Modelling
    Keywords: Fuzzy set, fuzzy rule based systems, high order fuzzy time series, forecasting, fuzzy numbers
    Submitted at 15-Apr-2013 16:03 by Murat Alper Başaran
    Fuzzy time series is an alternative forecasting method based on fuzzy set theory. It is a proven method when classic time series methods violate the assumptions. Recently, high order fuzzy times series methods are proven to be effective when forecasting accuracy is the target value due to the complexity of the real data. High order fuzzy time series are applied to forecasting issues such as financial and weather data and so on. Fuzzy rule based systems are used effectively in several disciplines in order to model data whose pattern nonlinear. The most used type of them is called Mamdani whose scope varies from engineering problems to education. In this paper, the forecasting accuracy of high order fuzzy times series are constructed utilizing fuzzy rule based systems. The results found based on this method is compared with those available in the literature.
  • Grading of Carcasses - How to Measure Lean Meat Percentage?

    Authors: Froydis Bjerke (Animalia Meat and Poultry Research Centre)
    Primary area of focus / application: Other: applied regression analysis
    Keywords: Multiple linear regression, Data analysis, Carcass grading, Applied statistics
    Submitted at 15-Apr-2013 16:31 by Froydis Bjerke
    Accepted (view paper)
    17-Sep-2013 17:30 Grading of Carcasses - How to Measure Lean Meat Percentage?
    Grading of carcasses is important in order to obtain a fair price for the farmer, and for estimating meat yield for the meat industry. Pig carcasses are graded by a probe that reads fat and muscle tissue thickness in fixed points of the pork back and belly. Lean meat percentage (LM%) is then estimated via a regression model. Over time, the conditions for this model change, e.g. due to breeding, hence the model coefficients must be updated/recalibrated occasionally, and sampling of LM% data from new carcasses is necessary.
    The presentation is a case study of applied regression analysis where data collection is expensive and tedious, and the economical impact is huge throughout the meat production value chain.
    The objective of the study is to build a model for monitoring carcass grading from regularly collected meat cutting data. The model utilises LM% estimated from CT scans (images) and corresponding meat cutting data from 144 carcasses. Model properties and sources of variance are also discussed.