Sciencestatistics: Journal of Statistics, Probability, and Its Application

Single Exponential Smoothing for Forecasting Medium Rice Retail Prices in Lampung Province

2026-07-01T09:28:32+07:00

Forecasting the price of medium grade rice is a strategic effort to support decision-making in maintaining food price stability in Lampung Province. This study aims to apply the Single Exponential Smoothing (SES) method in forecasting medium grade rice’s retail price in 2023 by evaluating the performance of the model using Mean Absolute Error (MAE). The data used is monthly retail price data for medium grade rice obtained from Dinas Ketahanan Pangan, Tanaman Pangan, dan Horticultura of Lampung Province. To obtain optimal forecasting results, the forecasting process involves determining the smoothing factor (α) parameters. The results show that the SES method can provide accurate forecasting with low MAE values. These findings suggest that the Single Exponential Smoothing method is feasible to be applied as a tool in food price control and policy planning in Lampung province.

Bayesian Prior Sensitivity in Psychological Decision Modeling: Evidence from Loss Aversion Estimation Under Prospect Theory

2026-07-01T10:23:38+07:00

Prior specification is a critical yet frequently neglected decision in Bayesian inference, with potentially severe consequences for behavioral research conclusions, particularly in nonlinear psychological decision models where likelihood surfaces are often flat and parameters are weakly identified. This study presents a simulation-based framework for assessing prior sensitivity in Bayesian psychological decision modeling, using loss aversion estimation under Prospect Theory as a case study. Synthetic binary choice data were generated from the Tversky-Kahneman utility function across four true loss aversion values (λ ∈ {1.5, 2.0, 2.5, 3.0}) and three sample sizes (n ∈ {100, 200, 500}), fitted under three prior specifications: weakly informative diffuse prior, moderate informative, and strongly informative, yielding 1,080 total model fittings from 360 synthetic datasets via Laplace approximation with importance-weighted resampling. Performance was evaluated via posterior mean bias, RMSE, credible interval width, and directional probability P(λ > 2). Three findings emerged. First, diffuse default priors failed to recover the loss aversion parameter when the likelihood was insufficiently informative, regardless of sample size. Second, strongly informative priors introduced systematic bias that persisted independently of sample size when the true parameter deviated from the prior mean. Third, prior choice produced meaningful disagreements in directional behavioral conclusions that larger samples could not eliminate. These findings demonstrate that prior sensitivity is a substantive methodological concern in Bayesian psychological decision modeling that cannot be resolved by increasing sample size alone, and researchers are encouraged to treat prior specification as an explicit analytical choice supported by routine sensitivity analysis.

Gaussian Mixture Models for Human Development-Based Regional Clustering of East Java

2026-07-14T14:15:36+07:00

Regional disparities in human development require an analytical approach that can identify latent patterns in development achievements. This study applies the Gaussian Mixture Model (GMM) to cluster regencies and cities in East Java based on the Human Development Index (HDI). GMM was chosen because it offers a probabilistic and distribution-based clustering framework, assigns regions using posterior membership probabilities, and provides interpretable parameters such as mixing proportions, component means, and variances. Univariate GMMs with two, three, and four components were fitted to HDI data from 38 regencies/cities in East Java at two time points, 2015 and 2025, which represent a ten-year interval. Model selection was conducted using internal cluster-validity measures, namely the Silhouette Coefficient and the Davies–Bouldin Index. The results show that the two-component GMM is selected as the best model for both years. The selected model produces the same membership structure in 2015 and 2025, forming a larger cluster of 31 regencies/cities with relatively lower and more variable HDI values and a smaller cluster of 7 regencies/cities with higher and more homogeneous HDI values. Over the ten-year interval, HDI increased in all regions, while the two-cluster structure remained evident. These findings can support regional development planning, policy evaluation, and the formulation of more targeted development strategies across regencies and cities.

Gold Price Forecasting Using Hybrid ARIMA-IGARCH

2026-07-29T22:26:54+07:00

Gold is an important investment and hedging instrument, with highly volatile price movements that are difficult to accurately predict. Previous studies have generally used ARIMA-GARCH models to forecast financial time series, but these models have not fully captured the persistent volatility of gold prices. Therefore, this study proposes an ARIMA-IGARCH approach to simultaneously model the mean and persistent volatility patterns of gold price movements. Daily gold closing price data from November 2022 to August 2025 was analyzed using Python, with 90% used for training and 10% for testing. The ARIMA model was used to capture the mean structure, while the IGARCH model was used to represent the long-term volatility persistence. The results showed that the proposed model achieved high forecasting accuracy, with a MAPE of 9.38%, indicating strong predictive performance. These findings indicate that the ARIMA-IGARCH model can serve as an alternative approach for gold price forecasting and financial market volatility analysis

Empirical Mode Decomposition with Swarm-Optimized Support Vector Regression for Natural Gas Price Forecasting

2026-08-09T00:12:41+07:00

Natural gas is a strategic energy commodity exhibiting nonlinear and highly volatile price movements due to supply-demand fluctuations, market dynamics, and geopolitical influences. These factors complicate accurate forecasting and necessitate advanced methods capable of modeling complex data patterns. This study proposes a hybrid forecasting model that integrates Empirical Mode Decomposition (EMD), Support Vector Regression (SVR), and Particle Swarm Optimization (PSO) to predict natural gas prices and assess predictive performance. The analysis utilizes a dataset of 1,575 daily closing prices from January 2020 to December 2025. EMD decomposes the original time series into seven Intrinsic Mode Functions (IMFs) and one residual component. Each component is modeled using SVR with a Radial Basis Function (RBF) kernel, and PSO is used to optimize model parameters. Forecasting performance is evaluated using Mean Absolute Percentage Error (MAPE) across three data partitioning schemes. Results indicate that the 70:15:15 partition yields the most accurate model, achieving a MAPE of 2.2641%. The 90-day forecast projects a gradual decline in natural gas prices after a peak in mid-January 2026, followed by relative price stability through March 2026. These findings suggest that the hybrid EMD-SVR-PSO model effectively captures the nonlinear dynamics of natural gas price data and delivers accurate forecasts, positioning it as a valuable decision-support tool for policymakers, industry stakeholders, and investors.