This thesis develops a machine learning-based forecasting model to predict monthly housing price growth in Shanghai, with the aim of capturing short-term market dynamics through a combination of macroeconomic, financial, policy, and sentiment-based indicators. The model is built using the XGBoost algorithm and is interpreted using SHAP values to ensure transparency in feature attribution. By reformulating the prediction target as monthly price growth and aligning it with percentage-based input features, the study improves model coherence and performance. The results show that recent price momentum plays a key role in shaping expectations, while liquidity conditions—particularly money supply (M2)—emerge as strong macro-financial predictors. In contrast, other variables such as CPI and policy scores show more limited and variable impact, while consumer sentiment proxies demonstrate moderate influence, with their contribution varying depending on broader market conditions. The inclusion of financial market indicators further confirms the influence of investor sentiment and capital reallocation in short-term housing price movements. Beyond its technical contribution, the study offers a structured and explainable framework for urban housing market forecasting. It demonstrates the value of integrating data-driven models with economic reasoning and provides insights relevant to policymakers, analysts, and stakeholders seeking to understand and anticipate price changes in complex real estate environments.
Questa tesi sviluppa un modello di previsione basato su tecniche di machine learning per stimare la crescita mensile dei prezzi delle abitazioni a Shanghai, con l’obiettivo di catturare le dinamiche di mercato a breve termine attraverso una combinazione di indicatori macroeconomici, finanziari, politici e legati al sentimento dei consumatori. Il modello è costruito utilizzando l’algoritmo XGBoost ed è interpretato tramite i valori SHAP per garantire trasparenza nell’attribuzione delle caratteristiche. Riformulando l’obiettivo di previsione come crescita mensile dei prezzi e allineandolo a variabili di input espresse in percentuale, lo studio migliora la coerenza e le prestazioni del modello. I risultati mostrano che lo slancio recente dei prezzi ha un ruolo centrale nel formare le aspettative, mentre le condizioni di liquidità — in particolare l’offerta di moneta (M2) — emergono come forti predittori macro-finanziari. Al contrario, variabili come l’indice dei prezzi al consumo (CPI) e i punteggi di politica mostrano un impatto più limitato e variabile, mentre i proxy del sentimento dei consumatori hanno un’influenza moderata, che varia in funzione delle condizioni di mercato generali. L’inclusione di indicatori di mercato finanziario conferma ulteriormente l’influenza del sentimento degli investitori e della riallocazione del capitale nei movimenti a breve termine dei prezzi delle abitazioni. Oltre al contributo tecnico, lo studio offre un quadro strutturato e interpretabile per la previsione del mercato immobiliare urbano. Dimostra il valore dell’integrazione tra modelli basati sui dati e il ragionamento economico, fornendo spunti utili per decisori politici, analisti e stakeholder interessati a comprendere e anticipare i cambiamenti dei prezzi in ambienti immobiliari complessi.
Machine learning-based prediction of Shanghai housing price trends and analysis of influencing factors
CHEN, FAN
2024/2025
Abstract
This thesis develops a machine learning-based forecasting model to predict monthly housing price growth in Shanghai, with the aim of capturing short-term market dynamics through a combination of macroeconomic, financial, policy, and sentiment-based indicators. The model is built using the XGBoost algorithm and is interpreted using SHAP values to ensure transparency in feature attribution. By reformulating the prediction target as monthly price growth and aligning it with percentage-based input features, the study improves model coherence and performance. The results show that recent price momentum plays a key role in shaping expectations, while liquidity conditions—particularly money supply (M2)—emerge as strong macro-financial predictors. In contrast, other variables such as CPI and policy scores show more limited and variable impact, while consumer sentiment proxies demonstrate moderate influence, with their contribution varying depending on broader market conditions. The inclusion of financial market indicators further confirms the influence of investor sentiment and capital reallocation in short-term housing price movements. Beyond its technical contribution, the study offers a structured and explainable framework for urban housing market forecasting. It demonstrates the value of integrating data-driven models with economic reasoning and provides insights relevant to policymakers, analysts, and stakeholders seeking to understand and anticipate price changes in complex real estate environments.File | Dimensione | Formato | |
---|---|---|---|
2025_07_Chen.pdf
accessibile in internet per tutti
Descrizione: Text of the thesis
Dimensione
2 MB
Formato
Adobe PDF
|
2 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/239790