Reproduce “Sub-City Real Estate Price Index Forecasting at Weekly Horizons Using Satellite Radar and News Sentiment” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Baris Arat, Hasan Fehmi Ates, Emre Sefer
This study forecasts weekly sub-city real estate price indices by combining satellite radar data and news sentiment with traditional transaction records. The multimodal approach reduced Mean Absolute Error by up to 34.6% for K-Nearest Neighbors at long horizons (26-34 weeks), outperforming models relying solely on historical prices.
Key insights
- The predictive utility of multimodal data is horizon-dependent, showing a substantial advantage for forecasting sub-city real estate indices at long horizons (beyond 14 weeks).
- Nonparametric models like K-Nearest Neighbors, Random Forest, and XGBoost consistently outperform linear models and LSTMs in this specific forecasting task.
- Semantic features derived from news embeddings and Sentinel-1 SAR backscatter prove more effective than lexical sentiment or optical NDBI for capturing market narratives and urban development, respectively.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.