Journal of Statistical Sciences

Selection in Small Area Estimation under AR-GARCH Models Based on the Gradient Boosting Algorithm

Abdolreza Sayyareh

The boosting algorithm is a hybrid algorithm to reduce variance, a family of machine learning algorithms in supervised learning. This algorithm is a method to transform weak learning systems into strong systems based on the combination of different results. In this paper, mixture models with random effects are considered for small areas, where the errors follow the AR-GARCH model. To select the variable, machine learning algorithms, such as boosting algorithms, have been proposed. Using simulated and tax liability data, the boosting algorithm's performance is studied and compared with classical variable selection methods, such as the step-by-step method.

Analysis of Housing Prices in Mashhad City With a Two-Stage Heterogeneous Spatial Modeling Framework

Hosein Baghishani

Financial and economic indicators, such as housing prices, often show spatial correlation and heterogeneity. While spatial econometric models effectively address spatial dependency, they face challenges in capturing heterogeneity. Geographically weighted regression is naturally used to model this heterogeneity, but it can become too complex when data show homogeneity across subregions. In this paper, spatially homogeneous subareas are identified through spatial clustering, and Bayesian spatial econometric models are then fitted to each subregion. The integrated nested Laplace approximation method is applied to overcome the computational complexity of posterior inference and the difficulties of MCMC algorithms. The proposed methodology is assessed through a simulation study and applied to analyze housing prices in Mashhad City.

Modeling error distribution in accelerated failure time models using a Dirichlet processes mixture model with the Burr XII distribution as the kernel

Soliman Khazaei

Accelerated failure time models are used in survival analysis when the data is censored, especially when combined with auxiliary variables. When the models in question depend on an unknown parameter, one of the methods that can be applied is Bayesian methods, which consider the parameter space as infinitely dimensional. In this framework, the Dirichlet process mixture model plays an important role. In this paper, a Dirichlet process mixture model with the Burr XII distribution as the kernel is considered for modeling the survival distribution in the accelerated failure time. Then, MCMC methods were employed to generate samples from the posterior distribution. The performance of the proposed model is compared with the Polya tree mixture models based on simulated and real data. The results obtained show that the proposed model performs better.

Parameter Estimation of AR(1) Model with Change Point and Its Application in Annual Inflation Rate Modeling

Yadollah waghei

Change point detection is one of the most challenging statistical problems because the number and position of these points are unknown. In this article, we will first introduce the concept of change point and then obtain the parameter estimation of the first-order autoregressive model AR(1); in order to investigate the precision of estimated parameters, we have done a simulation study. The precision and consistency of parameters were evaluated using MSE. The simulation study shows that parameter estimation is consistent. In the sense that as the sample size increases, the MSE of different parameters converges to zero. Next, the AR(1) model with the change point was fitted to Iran's annual inflation rate data (from 1944 to 2022), and the inflation rate in 2023 and 2024 was predicted using it.

Autoregressive Modeling of Fuzzy Data Based on the Support Vector Machine

Mohamad Ghasem Akbari

In the time series analysis, we may encounter situations where some elements of the model are imprecise quantities. One of the most common situations is the inaccuracy of the underlying observations, usually due to measurement or human errors. In this paper, a new fuzzy autoregressive time series model based on the support vector machine approach is proposed. For this purpose, the kernel function has been used for the stability and flexibility of the model, and the constraints included in the model have been used to control the points. In order to examine the performance and effectiveness of the proposed fuzzy autoregressive time series model, some goodness of fit criteria are used. The results were based on one example of simulated fuzzy time series data and two real examples, which showed that the proposed method performed better than other existing methods.

On Properties of a New Generalization of the Log-Logistic Distribution and Its Application

Mohammad Shafaei Noughabi

This article introduces a new extension of the log-logistic distribution, and its properties and parameter estimation are studied and analyzed. It is shown that adding a parameter to this distribution makes its shape more symmetric and less skewed as the parameter increases. Unlike the original distribution, the moments of the new distribution and its quantile function always exist. Furthermore, it is demonstrated that the reliability measures, such as the hazard rate function, the mean residual life function, and stochastic orderings, are more flexible in the new distribution. Additionally, the parameters of the distribution are estimated using the LLP and ML methods, and the efficiency and consistency of the estimators are evaluated through simulation studies. Finally, the practical applicability of the model is demonstrated by applying the new model to real-world data from airborne equipment and lung cancer patients.

Inference for the Modified Lindley Distribution Based on Progressively Type-II Censored Data

adeleh fallah

In this paper, estimation for the modified Lindley distribution parameter is studied based on progressive Type II censored data. Maximum likelihood estimation, Pivotal estimation, and Bayesian estimation were calculated using the Lindley approximation and Markov chain Monte Carlo methods. Asymptotic, Pivotal, bootstrap, and Bayesian confidence intervals are provided. A Monte Carlo simulation study has been conducted to evaluate and compare the performance of different estimation methods. To further illustrate the introduced estimation methods, two real examples are provided.

Advanced Missing Value Imputation Techniques: Machine Learning Methods with an Emphasis on an Ensemble Method for Multiple Imputation by Chained Equations

Zahra Rezaei Ghahroodi

Researchers often face the problem of how to address missing data. Multiple imputation by chained equations is one of the most common methods for imputation. In theory, any imputation model can be used to predict the missing values. However, if the predictive models are incorrect, it can lead to biased estimates and invalid inferences. One of the latest solutions for dealing with missing data is machine learning methods and the SuperMICE method. In this paper, We present a set of simulations indicating that this approach produces final parameter estimates with lower bias and better coverage than other commonly used imputation methods. Also, implementing some machine learning methods and an ensemble algorithm, SuperMICE, on the data of the Industrial establishment survey is discussed, in which the imputation of different variables in the data co-occurs. Also, the evaluation of various methods is discussed, and the method that has better performance than the other methods is introduced.

Evaluating Systemic Risk with Conditional Value at Risk and Vine Copulas in the Iranian Banking Network

Ali M. Mosammam

Systemic risk, as one of the challenges of the financial system, has attracted special attention from policymakers, investors, and researchers. Identifying and assessing systemic risk is crucial for enhancing the financial stability of the banking system. In this regard, this article uses the Conditional Value at Risk method to evaluate the systemic risk of simulated data and Iran's banking system. In this method, the conditional mean and conditional variance are modeled using Autoregressive Moving Average and Generalized Autoregressive Conditional Heteroskedasticity models, respectively. The data studied includes the daily stock prices of 17 Iranian banks from April 8, 2019, to May 1, 2023, which contains missing values in some periods. The Kalman filter approach has been used for interpolating the missing values. Additionally, Vine copulas with a hierarchical tree structure have been employed to describe the nonlinear dependencies and hierarchical risk structure of the returns of the studied banks. The results of these calculations indicate that Bank Tejarat has the highest systemic risk, and the increase in systemic risk, in addition to causing financial crises, has adverse effects on macroeconomic performance. These results can significantly help in predicting and mitigating the effects of financial crises and managing them effectively.

Multi-class support vector machine with random inputs using non-deterministic programming problem

Hadi Jabbari

‎Support vector machine (SVM) as a supervised algorithm was initially invented for the binary case‎, ‎then due to its applications‎, ‎multi-class algorithms were also designed and are still being studied as research‎. ‎Recently‎, ‎models have been presented to improve multi-class methods‎. ‎Most of them examine the cases in which the inputs are non-random‎, ‎while in the real world‎, ‎we are faced with uncertain and imprecise data‎. ‎Therefore‎, ‎this paper examines a model in which the inputs are uncertain and the problem's constraints are also probabilistic‎. ‎Using statistical theorems and mathematical expectations‎, ‎the problem's constraints have been removed from the random state‎. ‎Then‎, ‎the moment estimation method has been used to estimate the mathematical expectation‎. ‎Using Monte Carlo simulation‎, ‎synthetic data has been generated and the bootstrap resampling method has been used to provide samples as input to the model and the accuracy of the model has been examined‎. ‎Finally‎, ‎the proposed model was trained with real data and its accuracy was evaluated with statistical indicators‎. ‎The results from simulation and real examples show the superiority of the proposed model over the model based on deterministic inputs‎.

Optimal Repetitive Acceptance Sampling Inspection Plans by Attributes Based on Type I Censoring Using Two-point and Limited Weighted Methods

mehran naghizadeh qomi

This paper investigates repetitive acceptance sampling inspection plans of lots based on type I censoring when the lifetime has a Tsallis q-exponential distribution. A repetitive acceptance sampling inspection plan is introduced, and its components, along with the optimal average sample number and the operating characteristic value of the plan, are calculated under the specified values for the parameter of distribution and consumer's and producer's risks using a nonlinear programming optimization problem. Comparing the results of the proposed repetitive acceptance sampling plan with the optimal single sampling inspection plan demonstrates the efficiency of the repetitive acceptance sampling plan over the single sampling plan. Moreover, repetitive sampling plans with a limited linear combination of risks are introduced and compared with the existing plan. Results of the introduced plan in tables and figures show that this plan has a lower ASN and, therefore, more efficiency than the existing design. A practical example in the textile industry is used to apply the proposed schemes.

The Uniformly More Powerful Tests than the Likelihood Ratio Test Using Intersection-Union Hypotheses for Variance of Independent Samples from Normal Distribution

Rahim Chinipardaz

Classical hypothesis tests for the parameters provide suitable tests when the hypotheses are not restricted. The best are the uniformly most powerful test and the uniformly most powerful unbiased test. These tests are designed for specific hypotheses, such as one-sided and two-sided for the parameter. However, in practice, we may encounter hypotheses that the parameters under test have typical restrictions in the null or alternative hypothesis. Such hypotheses are not included in the framework of classical hypothesis testing. Therefore, statisticians are looking for more powerful tests than the most powerful ones. In this article, the union-intersection test for the sign test of variances in several normal distributions is proposed and compared with the likelihood ratio test. Although the union-intersection test is more powerful, neither test is unbiased. Two rectangular and smoothed tests have been examined for a more powerful test.