Article Details

Abstract

By integrating the advantages of multiple individual models, combination forecasting models can effectively improve prediction accuracy and have been widely applied in fields such as economic forecasting and engineering decision-making. The combination forecasting model based on reciprocal of variance follows the core principle that "the smaller the error variance of an individual model, the greater its weight," providing an intuitive and efficient solution to weight assignment. This paper systematically presents the theoretical foundation of the model, including the principles of weight calculation and accuracy evaluation indicators. Subsequently, representative monthly commodity sales data is selected as a case study. Two individual forecasting models are constructed and then combined using the reciprocal-of-variance method. Data processing and result analysis are performed to intuitively compare the prediction performance between individual models and the combination model. The results indicate that the variance-reciprocal combination forecasting model achieves significantly higher prediction accuracy than individual models. Moreover, the model features a simple calculation process and strong practicality, providing reliable technical support for practical forecasting problems.

Keywords

reciprocal of variance; combination forecasting; prediction accuracy; weight assignment

References

1. [1] Liu Guangying, Weng Qiulei, Zhuangzi said Realized covariance matrix prediction model based on long-term memory dynamic factors [J]. Statistics and Decision, 2026, 42 (6): 154-159
2. [2] Franses H P.Model Selection for Forecast Combination [J].Applied Economics，2011，43（14）：1721-1727.
3. [3] Zhai Yuguang, Ren Jie, Nan Shenghao, etc Bayesian seepage parameter inversion and analysis based on dynamic weight coefficient multi-objective optimization combination [J]. Journal of Irrigation and Drainage Mechanical Engineering, 2024, 42 (12): 1259-1265
4. [4] Zheng Shouxiang Weight allocation optimization method for robust design of mechanical structures under multiple constraints [J]. Modern Manufacturing Technology and Equipment, 2025, 61 (9): 85-87
5. [5] Liu G H，You J H.Level Mapping Induced Loop Formulas for Weight Constraint and Aggregate Logic Programs [J].Fundamenta Informaticae，2011，106（1）：25-43.
6. [6] Lv Xinman, Yin Kedong, Li Xuemei Grey Multivariate Variable Weight Combination Prediction Model and Its Application [J]. Statistics and Decision Making, 2022, 38 (14): 25-29
7. [7] Peng Yikai, Pu Hongping Research on a Sparrow Search Algorithm Integrating Logistic Regression [J]. Modern Information Technology, 2023, 7 (8): 1-7
8. [8] Li Yilin, Su Xuefeng, Li Hui, etc Epidemic combination prediction method based on multi-source data fusion [J]. Chinese Journal of Medical Physics, 2024, 41 (2): 258-264
9. [9] Hu Yankun, Wang Ning, Liu Shu, etc Research on the Application of Time Series Models and LSTM Models in Water Quality Prediction [J]. Small Microcomputer Systems, 2021, 42 (8): 1569-1573
10. [10] Heng Hongjun, Chen Hongrong Multivariate Time Series Prediction Integrating Long term Trends and Short term Fluctuations [J/OL]. Journal of Xi'an University of Electronic Science and Technology of China (November 11, 2025) [November 12, 2025] https://doi.org/10.19665/j.issn1001-2400.20251103.
11. [11] Zhang Xingyu, Zhang Haozhe, Wang Zhuangzhuang, etc A prediction of irrigation water consumption based on ARIMA model [J]. China Science and Technology Information, 2025 (24): 97-100
12. [12] Xu Qili Setting the optimal prediction parameters for exponential smoothing model using regression method [J]. Journal of Statistics, 2021, 2 (4): 84-93
13. [13] Wang Dan, Jiang Hui, Zhang Yuqi, etc Economic dispatch optimization based on covariance matrix adaptive evolution strategy using dynamic reverse learning [J]. Journal of Tianjin University of Science and Technology, 2023, 38 (2): 63-69

Funding

Research Foundation Ability Enhancement Project for Young and Middle aged Teachers in Guangxi Universities (2024KY1409)

About the authors

Chen Chunyan (1996-), female, Han nationality, from Fangchenggang, Guangxi, intermediate economist, master's degree, research direction: machine learning, prediction and decision analysis.