Related Experiment Video
Updated: Sep 7, 2025
05:55
Modeling the Functional Network for Spatial Navigation in the Human Brain
Published on: October 13, 2023
1.2K
China's GDP forecasting using Long Short Term Memory Recurrent Neural Network and Hidden Markov Model
Junhuan Zhang1,2, Jiaqi Wen3, Zhen Yang1
1School of Economics and Management, Beihang University, Beijing, China.
Plos One
|June 17, 2022
Summary
This study introduces a Long Short-Term Memory Recurrent Neural Network and Hidden Markov Model (LSTM-HMM) for predicting China's Gross Domestic Product (GDP) fluctuations. LSTM-HMM demonstrates superior accuracy and consistency, especially with monthly Consumer Price Index (CPI) data over a 10-year window.
Area of Science:
- Econometrics
- Computational Economics
- Time Series Analysis
Background:
- Accurate forecasting of Gross Domestic Product (GDP) is crucial for economic policy and planning.
- Traditional dynamic forecast systems face challenges in capturing complex economic fluctuations.
- The integration of advanced machine learning with statistical models offers potential improvements in economic prediction.
Purpose of the Study:
- To develop and evaluate a Long Short-Term Memory Recurrent Neural Network and Hidden Markov Model (LSTM-HMM) for predicting China's GDP fluctuation states.
- To compare the predictive performance of LSTM-HMM against other dynamic forecast systems, including Hidden Markov Model (HMM) and Gaussian Mixture Model-Hidden Markov Model (GMM-HMM).
- To assess the impact of different time windows (4, 6, 8, 10 years) and input data frequencies (monthly vs. quarterly Consumer Price Index - CPI) on forecasting accuracy.
Main Methods:
- Implementation of a hybrid LSTM-HMM model incorporating recurrent neural networks for state prediction and HMM for sequence modeling.
- Comparative analysis using HMM, GMM-HMM, and LSTM-HMM with varying rolling time window lengths.
- Utilizing both monthly and quarterly CPI data as input features to evaluate data granularity's effect on model performance.
Main Results:
- The LSTM-HMM model consistently outperformed other benchmark models in predicting GDP fluctuation states.
- Utilizing monthly CPI data resulted in improved model performance compared to quarterly CPI data.
- Forecasting models demonstrated better overall performance within a 10-year rolling time window.
- Within the 10-year window, LSTM-HMM with either monthly or quarterly CPI input achieved the highest accuracy and consistency.
Conclusions:
- The LSTM-HMM framework provides a robust and accurate method for forecasting economic fluctuations, specifically China's GDP.
- The choice of input data frequency (monthly CPI) and the length of the time window (10-year) significantly enhance predictive capabilities.
- This research highlights the potential of advanced hybrid models in improving the reliability of macroeconomic forecasting.
