AI Empowerment, Optimizing Predictions
As construction costs increasingly dominate Hong Kong’s development budgets, accurately predicting market trends has become crucial. The Tender Price Index (TPI), a key quarterly benchmark, provides valuable insights into evolving cost trends. However, traditional forecasting methods rely heavily on expert judgment, often leading to inconsistent results due to human factors, limited scalability, and a lack of transparency.
To tackle this challenge, RLB has developed an innovative AI-driven TPI Forecasting Model specifically designed for the Hong Kong construction market. Utilizing machine learning techniques, the model extracts patterns from vast datasets, offering objective, timely, and evidence-based industry insights.
This is not just a technological advancement; it sets a new industry standard. By merging advanced data analytics with extensive industry experience, RLB empowers developers and consultants to make informed predictions with greater confidence and flexibility in an increasingly complex market environment.
RLB’s Advanced Tender Price Index Forecasting Model
Our model employs a Generalized Linear Model (GLM) and correlation matrix analysis to identify key leading economic indicators from the domestic economy, banking sector, construction industry, and stock market. It integrates three primary indicators with strong directional influence: Money Supply (M3), Seasonally Adjusted Gross Domestic Product (GDP), and the Hang Seng Index (HSI). Both M3 and GDP demonstrate a positive correlation with the TPI, while the HSI shows a mixed relationship.
Powered by Random Forest—a widely used machine learning method—our model identifies relationships between input features and target outcomes. Leveraging two decades of historical data, it achieves an impressive predictive accuracy of 98% (99% in the training set and 98% in the testing set), setting a new benchmark for reliability in tender price forecasting. By combining expert judgment with empirical analysis, our model enables stakeholders to navigate market uncertainties with greater precision.
Figure 1: Comparison of Historical Data and Model Predictions
Transparent and Accurate AI
The prediction model is designed to ensure transparency and clarity in its decisions. It utilizes both intrinsic explanations (Gini importance) and post hoc explanations (SHAP) to validate its interpretability. Intrinsic explanations derive from the model’s inherent structure, while post hoc explanations analyze predictions using external techniques. Together, these approaches rigorously assess the model’s explanatory power.
To validate its accuracy, the model was used to forecast the RLB Tender Price Index (TPI) for the 1st Quarter of 2025, predicting a value of 2675. This forecast was retrospectively validated against actual TPI data compiled from returned tenders during the same period, demonstrating a high level of consistency and reinforcing the model’s reliability in practical applications.
Conclusion
The successful deployment of our forecasting model highlights the transformative potential of data science in modern quantity surveying. However, this is just the beginning. As market complexities grow, we remain committed to continuously innovating and refining our tools, delivering smarter solutions that empower the industry to navigate change with clarity and confidence.
Authors:
Dr Monica Zhan Research Manager
Dr K.C. Lo Quantity Surveyor
Explore the latest cost trends in RLB’s publication: Construction Cost Update for Hong Kong Q2 2025
FURTHER INFORMATION:
