Abstract: With the rapid development of society and economy, the real estate industry has maintained a leading position, yet fluctuating housing prices remain a significant concern for the Chinese public. This paper predicts the trend of real estate prices in Nanning over the next 10 years. A grey model is established, and the simulation sequence is validated using the posterior difference test method. Subsequently, grey GM(1, 1) models are developed for four primary influencing factors to project their values through 2031. To address overfitting in certain factors, linear regression fitting is applied after correlation verification. Finally, the GM(0, N) estimation formula is utilized to forecast Nanning's real estate prices for the coming decade.
Introduction
Predicting real estate prices is a complex challenge due to the market's nature as a "grey system"—one influenced by a combination of known macroeconomic indicators and unknown local policy or psychological variables. Traditional statistical methods often struggle with regional economic data, which typically involves small sample sizes and non-standard distributions.
This study focuses on Nanning, the capital of the Guangxi Autonomous Region and a vital hub in Southwest China. By providing a reliable 10-year forecast, the research aims to assist developers and policymakers in navigating a market that has experienced significant volatility due to global events and shifting domestic economic drivers.
Methodology
The research employs Grey System Theory, which is specifically designed for systems with limited information. The forecasting framework consists of three main components:
- GM(1,1) Model: A univariate first-order differential equation used to forecast independent influencing factors: urban per capita disposable income, Gross Regional Product (GDP), commercial housing sales area, and registered population.
- Linear Regression Integration: For factors where the GM(1,1) model exhibited overfitting or suboptimal fit despite high correlation, linear regression was used to ensure more robust projections.
- GM(0,N) Model: A multivariate model used to synthesize the predicted values of the four influencing factors into a final real estate price estimate.
- Model Validation: The accuracy of these models was rigorously tested using the posterior difference test, calculating $p$ (probability of error) and $C$ (ratio of posterior differences) values to grade simulation quality.
Data and Variables
The study utilizes 20 years of historical annual statistics (2003–2021) for Nanning City. The target variable is the average sales price of commercial housing. The predictor variables include:
- Urban per capita disposable income
- Gross Regional Product (GRP/GDP)
- Commercial housing sales area
- Year-end registered population
Results and Analysis
The individual factor models demonstrated high reliability. For instance, the population prediction model achieved a "Level 1 (Good)" accuracy rating ($p = 0.9831$, $C = 0.3306$). As shown in the figure below, the model successfully captures the steady upward trajectory of Nanning's population, projected to reach approximately 889 million by 2031.
The final synthesized housing price model achieved a "Level 2 (Pass)" rating ($p = 0.8571$, $C = 0.4947$), confirming its suitability for long-term forecasting. While the model encountered some initial fitting challenges in the early 2000s, it closely mirrors the historical trend from 2010 onwards.
Key insights from the results include:
- Economic Coupling: Housing prices in Nanning are deeply tied to the city's fundamental economic health, particularly GDP and per capita income.
- Small Sample Efficiency: The study proves that Grey models can yield meaningful results even with limited data points (20 years of annual data), where traditional big-data methods might fail.
- Consistent Growth: Despite short-term fluctuations, the underlying economic momentum of Nanning suggests a sustained rise in real estate costs over the next decade.
Conclusion
The hybrid grey model framework provides a qualified tool for regional real estate forecasting. However, the authors note certain limitations: the model assumes stable supply-demand dynamics and does not account for psychological buyer preferences or sudden "black swan" policy shifts. Future work could improve accuracy by integrating these external shocks into the grey system framework to better simulate market reactions to sudden economic mutations.
Summary updated