Bayesian prediction of waiting times in stochastic models

The authors show how saddlepoint techniques lead to highly accurate approximations for Bayesian predictive densities and cumulative distribution functions in stochastic model settings where the prior is tractable, but not necessarily the likelihood or the predictand distribution. They consider more specifically models involving predictions associated with waiting times for semi‐Markov processes whose distributions are indexed by an unknown parameter θ. Bayesian prediction for such processes when they are not stationary is also addressed and the inverse‐Gaussian based saddlepoint approximation of Wood, Booth & Butler (1993) is shown to accurately deal with the nonstationarity whereas the normal‐based Lugannani & Rice (1980) approximation cannot, Their methods are illustrated by predicting various waiting times associated with M/M/q and M/G/1 queues. They also discuss modifications to the matrix renewal theory needed for computing the moment generating functions that are used in the saddlepoint methods.     

Consistency of the Subsample Bootstrap empirical process

In the classical Bootstrap approach the number of distinct observation in the resample is random. To overcome this hitch Rao et al. [Bootstrap by sequential resampling, J. Statist. Plan. Inference 64 (1997), pp. 257–281] have proposed a modified resampling procedure – the so-called Sequential Bootstrap or 0.632-Bootstrap – in which each resample has exactly the same number m ～eq ?0.632 n? of distinct observations. Motivated by this idea we introduce an akin procedure, the Subsample Bootstrap, where additionally even the size of each resample is equal. It will turn out that the Subsample Bootstrap empirical process is consistent for a wide class of Donsker classes.     

Bayes sequential estimation for a Poisson process under a LINEX loss function

In this paper, within the framework of a Bayesian model, we consider the problem of sequentially estimating the intensity parameter of a homogeneous Poisson process with a linear exponential (LINEX) loss function and a fixed cost per unit time. An asymptotically pointwise optimal (APO) rule is proposed. It is shown to be asymptotically optimal for the arbitrary priors and asymptotically non-deficient for the conjugate priors in a similar sense of Bickel and Yahav [Asymptotically pointwise optimal procedures in sequential analysis, in Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, Vol. 1, University of California Press, Berkeley, CA, 1967, pp. 401–413; Asymptotically optimal Bayes and minimax procedures in sequential estimation, Ann. Math. Statist. 39 (1968), pp. 442–456] and Woodroofe [A.P.O. rules are asymptotically non-deficient for estimation with squared error loss, Z. Wahrsch. verw. Gebiete 58 (1981), pp. 331–341], respectively. The proposed APO rule is illustrated using a real data set.     

Efficient sequential estimation in a markov branching process with immigration

Sequential estimation of parameters In a continuous time Markov branching process with Immigration with split rate λ₁ Immigration rate λ₂, offspring distribution {p₁j≥O) and Immigration distribution {p₂j≥l} is considered. A sequential version of the Cramér-Rao type information inequality is derived which gives a lower bound on the variances of unbiased estimators for any function of these parameters. Attaining the lower bounds depends on whether the sampling plan or stopping rule S, the estimator f, and the parametric function g = E(f) are efficient. All efficient triples (S,f,g) are characterized; It Is shown that for i = 1,2, only linear combinations of λ_ip_{ij j}'s or their ratios are efficiently estimable. Applications to a Yule process, a linear birth and death process with immigration and an M/M/∞ queue are also considered     

A Bivariate Beta-Gamma Autoregressive Process (BVBGAR(1))

In this article, we introduce a bivariate autoregressive process with Gamma marginal distributions using the form of the BGAR(2) process (Risti?, 2005 Risti? , M. M. ( 2005 ). A Beta-Gamma autoregressive process of the second-order (BGAR(2)) . Statist. Probab. Lett. 73 : 403 – 410 . [Google Scholar]) and the Beta-Gamma transformation. Some properties of the process such as the autocovariance matrix, the autocorrelation matrix, and the spectral density matrix are derived. The unknown parameters of the process are estimated using the method of moments and the method of conditional least squares. Some numerical results of the estimators are given. We investigate nonparametric and parametric estimation of the spectral density matrix of this process.     

On residual lifetimes in sequential (n ? k?+?1)-out-of-n systems

Sequential order statistics is an extension of ordinary order statistics. They model the successive failure times in sequential k-out-of-n systems, where the failures of components possibly affect the residual lifetimes of the remaining ones. In this paper, we consider the residual lifetime of the components after the kth failure in the sequential (n − k + 1)-out-of-n system. We extend some results on the joint distribution of the residual lifetimes of the remaining components in an ordinary (n − k + 1)-out-of-n system presented in Bairamov and Arnold (Stat Probab Lett 78(8):945–952, 2008) to the case of the sequential (n − k + 1)-out-of-n system.     

Construction of fixed width confidence intervals for a Bernoulli success probability using sequential sampling: a simulation study

This article considers the construction of level 1?α fixed width 2d confidence intervals for a Bernoulli success probability p, assuming no prior knowledge about p and so p can be anywhere in the interval [0, 1]. It is shown that some fixed width 2d confidence intervals that combine sequential sampling of Hall [Asymptotic theory of triple sampling for sequential estimation of a mean, Ann. Stat. 9 (1981), pp. 1229–1238] and fixed-sample-size confidence intervals of Agresti and Coull [Approximate is better than ‘exact’ for interval estimation of binomial proportions, Am. Stat. 52 (1998), pp. 119–126], Wilson [Probable inference, the law of succession, and statistical inference, J. Am. Stat. Assoc. 22 (1927), pp. 209–212] and Brown et al. [Interval estimation for binomial proportion (with discussion), Stat. Sci. 16 (2001), pp. 101–133] have close to 1?α confidence level. These sequential confidence intervals require a much smaller sample size than a fixed-sample-size confidence interval. For the coin jamming example considered, a fixed-sample-size confidence interval requires a sample size of 9457, while a sequential confidence interval requires a sample size that rarely exceeds 2042.     

Perfect Simulation for Random Sequential Adsorption of d-Dimensional Spheres with Random Radii

An efficient simulation algorithm for random sequential adsorption of spheres with radii chosen from a (prior) probability distribution is implemented. The algorithm is based on dividing the whole domain in small subcubes of different edge length. Samples obtained by this algorithm satisfy the jamming limit propertyi.e., no further sphere can be placed in the final configuration without overlapping. Samples for both discrete and continuous radii distributions are simulated and analyzed, especially jamming coverage, pair correlation functions and posterior radii distributions of the obtained sphere configurations.     

Bayesian estimation of an AR(1) process with exponential white noise

The object of this paper is a Bayesian analysis of the autoregressive model X _t ?=?ρX _t?1?+?Y _t where 0?Y _t are independent random variables with an exponential distribution of parameter θ. Our study generalizes some results obtained by Turkmann (1990 Amaral Turkmann, M. A. (1990). Bayesian analysis of an autoregressive process with exponential white noise. Statistics, 4: 601–608.  [Google Scholar]). Our analysis is based on a more general non-informative prior which allows us to improve the estimators of ρ and θ.     

Optimal sequential estimation procedures of a function of a probability of success under LINEX loss

In this paper, we investigate the problem of estimating a function g(p), where p is the probability of success in a sequential sample of independent identically Bernoulli distributed random variables. As a loss associated with estimation we introduce a generalized LINEX loss function. We construct a sequential procedure possessing some asymptotically optimal properties in the case when p tends to zero. In this approach to the problem, the conditions are given, under which the stopping time is asymptotically efficient and normal, and the corresponding sequential estimator is asymptotically normal. The procedure constructed guarantees that its sequential risk is asymptotically equal to a prescribed constant.     

On-Line Learning for the Infinite Hidden Markov Model

We develop a sequential Monte Carlo algorithm for the infinite hidden Markov model (iHMM) that allows us to perform on-line inferences on both system states and structural (static) parameters. The algorithm described here provides a natural alternative to Markov chain Monte Carlo samplers previously developed for the iHMM, and is particularly helpful in applications where data is collected sequentially and model parameters need to be continuously updated. We illustrate our approach in the context of both a simulation study and a financial application.     

Large Deviation Results for Wave Governed Random Motions Driven by Semi-Markov Processes

In this article, we present large deviation results for a model {ξ₁ + … + ξ _n : n ≥ 1} which is close to a random walk. More precisely, we consider independent random variables {ξ _n : n ≥ 1} such that {ξ _n : n ≥ 2} are i.i.d. and a different distribution for ξ₁ is allowed. We prove large deviation estimates for P(N _x  ≤ xT) and P(N _x < ∞) as x → ∞, where N _x : = inf {n ≥ 1: ξ₁ + … + ξ _n  ≥ x}. Moreover, we provide an asymptotically efficient simulation law for the estimation of P(N _x  ≤ xT) and P(N _x < ∞) by Monte Carlo simulation based on the importance sampling technique. These results will be adapted to wave governed random motions driven by semi-Markov processes and we present some simulations. Finally, we study the convergence of some large deviation rates for standard wave governed random motions based on a scaling presented in the literature (see Kac, 1974 Kac , M. ( 1974 ). A stochastic model related to the telegrapher's equation . Rocky Mountain Journal of Mathematics 4 : 497 – 509 .[Crossref] , [Google Scholar]; Orsingher, 1990 Orsingher , E. ( 1990 ). Probability law, flow function, maximum distribution of wave governed random motions and their connections with Kirchoff's laws . Stochastic Processes and their Applications 34 ( 1 ): 49 – 66 . [Google Scholar]).     

Simulation algorithms for integrals of a class of sampling distributions arising in population genetics

Efficient stochastic algorithms are presented in order to simulate allele configurations distributed according to a family π _A , 0<A<∞, of exchangeable sampling distributions arising in population genetics. Each distribution π _A has two parameters n and k, the sample size and the number of alleles, respectively. For A→0, the distribution π _A is induced from neutral sampling, whereas for A→∞, it is induced from Maxwell–Boltzmann sampling. Three different Monte Carlo methods (independent sampling procedures) are provided, based on conditioning, sequential methods and a generalization of Pitmans ‘Chinese restaurant process’. Moreover, an efficient Markov chain Monte Carlo method is provided. The algorithms are applied to the homozygosity test and to the Ewens–Watterson–Slatkin test in order to test the hypothesis of selective neutrality.     

Empirical Likelihood for Compound Poisson Processes

Let {N(t), t > 0} be a Poisson process with rate λ > 0, independent of the independent and identically distributed random variables with mean μ and variance . The stochastic process is then called a compound Poisson process and has a wide range of applications in, for example, physics, mining, finance and risk management. Among these applications, the average number of objects, which is defined to be λμ, is an important quantity. Although many papers have been devoted to the estimation of λμ in the literature, in this paper, we use the well‐known empirical likelihood method to construct confidence intervals. The simulation results show that the empirical likelihood method often outperforms the normal approximation and Edgeworth expansion approaches in terms of coverage probabilities. A real data set concerning coal‐mining disasters is analyzed using these methods.     

Response Time Distribution in a D-MAP/PH/1 Queue with General Customer Impatience

ABSTRACT

This paper presents two methods to calculate the response time distribution of impatient customers in a discrete-time queue with Markovian arrivals and phase-type services, in which the customers’ patience is generally distributed (i.e., the D-MAP/PH/1 queue). The first approach uses a GI/M/1 type Markov chain and may be regarded as a generalization of the procedure presented in Van Houdt ^[14] Van Houdt , B. ; Lenin , R. B. ; Blondia , C. Delay distribution of (im)patient customers in a discrete time D-MAP/PH/1 queue with age dependent service times Queueing Systems and Applications 2003 , 45 1 , 59 – 73 . [CROSSREF]  [Google Scholar] for the D-MAP/PH/1 queue, where every customer has the same amount of patience. The key construction in order to obtain the response time distribution is to set up a Markov chain based on the age of the customer being served, together with the state of the D-MAP process immediately after the arrival of this customer. As a by-product, we can also easily obtain the queue length distribution from the steady state of this Markov chain.

We consider three different situations: (i) customers leave the system due to impatience regardless of whether they are being served or not, possibly wasting some service capacity, (ii) a customer is only allowed to enter the server if he is able to complete his service before reaching his critical age and (iii) customers become patient as soon as they are allowed to enter the server. In the second part of the paper, we reduce the GI/M/1 type Markov chain to a Quasi-Birth-Death (QBD) process. As a result, the time needed, in general, to calculate the response time distribution is reduced significantly, while only a relatively small amount of additional memory is needed in comparison with the GI/M/1 approach. We also include some numerical examples in which we apply the procedures being discussed.