مجله علوم آماری
سال پنجم شماره 1 (پیاپی 9، بهار و تابستان 1390)

Usually، in testing hypothesis a p_value is used for making decision. Would p_value be the best measure to accept or reject the null hypothesis? Would it be possible to have a better measure than the ordinary p_value? In this paper، hypothesis testing has been considered not as a choice to make decision but as an estimating problem to possible accuracy of a given set، labeled by Θ_0 and p_value would be used as an estimator to possible accuracy of Θ_0 Real numbers as a parametric space has been usually accepted by researcher although the parametric space has been limited in many of applications. A measure named as modified p_value which functions more better than usual p_value in bounded parametric space، would be introduced in normal distribution of one-side and two-side testing.

In this paper، first a brief introduction of ranked set sampling is presented. Then، construction of confidence intervals for a quantile of the parent distribution based on ordered ranked set sample is given. Because the corresponding confidence coefficient is an step function، one may not be able to find the exact prescribed value. With this in mind، we introduce a new method and show that one can obtained an optimal confidence interval by appealing the proposed approach. We also compare the proposed scheme with the other existence methods.

In sampling، arisen data with probability proportional to its length is called Length-bised. Nonparametric density estimation in length-biased sampling is more difficult than other states. One of the famous estimators in this context is the one introduced by Jones (1991). In this paper، we calculate the bandwidth parameter of this estimator by Bayes''method. The strong consistency of this estimator have been proved with a random Bandwidth. We have compared the performance of Bayes''method with cross validation by using simulation studies.

There is a family of generalized Farlie-Gumbel-Morgenstern copulas، known as the semiparametric family، which is generated by a function called distribution-based generator. These generators have been studied typically for symmetric distributions in the literature. In this article، is proposed a method for asymmetric case which increases the flexibility of distribution-based generators and، thus، the model. In addition، a method for generalizing general generators is provided which can also be used to obtain more flexible distribution-based generators. Clearly، with more flexible generators more desirable models can be found to fit real data.

In this article the characterization of distributions is considered by using Kullback-Leibler information and records values. Then some characterizations are obtained based on Kullback-Leibler information and Shannon entropy of order statistics and record values.

In this paper a new version of skew uniform distribution is introduced which is completely different from the previous works. Some important properties of the new distribution contain the expression for the density and distribution، kth moments، moment generating and characteristic functions، variance، skewness and kurtusis، mean deviation from the mean، median and mode and parameter estimation are investigated. Also a simulation study on this distribution is carried out to show the consistency of the maximum likelihood and moments estimators. In the end، the new skew uniform distribution is compared with uniform distribution.

This paper offers a method of the estimation of the transition probability for the behaviors of financial time series by Markov Switching Autoregressive model. Using this model، the behaviors of fluctuations of exchange rate form two regimes low and high changes rate are considered. Results of prediction show that the persistence probability of regimes will be decreased. Thus، the probability of transition to other regime will be increased if process were in a specific regime.