Long memory analysis is one of the most active areas in econometrics and time series where various methods have been introduced to identify and estimate the long memory parameter in partially integrated time series. One of the most common models used to represent time series that have a long memory is the ARFIMA (Auto Regressive Fractional Integration Moving Average Model) which diffs are a fractional number called the fractional parameter. To analyze and determine the ARFIMA model, the fractal parameter must be estimated. There are many methods for fractional parameter estimation. In this research, the estimation methods were divided into indirect methods, where the Hurst parameter is estimated first, and then the fractional integration parameter is estimated from it by a relation between them. As for direct methods, the fractional integration parameter is estimated directly without relying on Hurst's parameter, and most of them are semi parametric methods. In this paper, some of the most common direct methods were used to estimate the fraction modulus namely (Geweke-Porter-Hudak, Smoothed Geweke-Porter-Hudak, Local Whittle, Wavelet and weighted wavelet), using simulation method with different value of (d) and different size of time series. The comparison between the methods was done using the mean squared error (MSE). It turns out that the best methods to estimate the fractional parameter is (Local Whittle).       The ARFIMA model was generated by a function programmed by the MATLAB statistical program

Recent statistical research has witnessed activity on the study of skew normal distribution (SND) due to the fact that the data set does not fit well with the normal distribution due to Skewnessand excessive Kurtosis. For the purpose of estimating the parameters of the model (SND), the maximum likelihood method (ML) was used, but the probability equations of this method do not have clear solutions in the distribution (SND), and the problem was solved using the genetic algorithm (GA) and Other repetitive techniques are Newton Raphson, Nelder Mead and Iteratively Reweighting Algorithm, using the simulation method with different sample sizes and comparing the preference of results methods used based on criteria (Mean, Mse and Def). It has been concluded that (ML) capabilities using the (GA) of parameters (SND) are best in the case of a small or medium sample size and the best (IR) algorithm at a large sample size.

One-parameter exponential regression is one of the most common and widely used models in several fields, to estimate the parameters of the one-parameter exponential regression model use the ordinary least square method but this method is not effective in the presence of outlier values, so robust methods were used to treat outlier values in the one-parameter exponential regression model are to estimate the parameters using robust method (Median-of-Means, Forward search, M-Estimation), and the simulation was used to compare between the estimation methods with different sample sizes and assuming four ratios from the outliers of the data (10%, 20%, 30%, 40%). And the mean square error (MSE) was made to reach the best estimation method for the parameters, where the results obtained using the simulation showed that the forward search is the best because it gives the lowest mean of error. On the practical side, expenditure and revenue data were used to estimate the parameters of the one-parameter exponential regression, where the data was tested, it appeared to have an exponential distribution, and the boxplot and (COOK) test were used to detect the outliers present in the real data. The Goodness of fit test was used for the one-parameter exponential model, and it was found that the data did not follow the normal distribution, and it was found that it suffers from the problem of heterogeneity of variance. The one-parameter exponential regression model for the expenditure and revenue data was estimated using the advanced search method because it was the best estimate. Paper type Research paper

Artificial Neural Networks (ANN) is one of the important statistical methods that are widely used in a range of applications in various fields, which simulates the work of the human brain in terms of receiving a signal, processing data in a human cell and sending to the next cell. It is a system consisting of a number of modules (layers) linked together (input, hidden, output). A comparison was made between three types of neural networks (Feed Forward Neural Network (FFNN), Back propagation network (BPL), Recurrent Neural Network (RNN). he study found that the lowest false prediction rate was for the recurrentt network architecture and using the Data on graduate students at the College of Administration and Economics, University of Baghdad for the period from 2014-2015 to The academic year 2017-2018. The variables are use in the research is (student's success, age, gender, job, type of study (higher diploma, master's, doctorate), specialization (statistics, economics, accounting, industry management, administrative management, and public administration) and channel acceptance). It became clear that the best variables that affect the success of graduate students are the type of study, age and job.