Spain | Coincident indicator model for housing prices

At BBVA we introduced a new indicator built with Big Data, based on the appraisals used in the bank's regular activity, to strengthen price monitoring in the sector. In addition, we built a set of models to improve estimates in real time.

Key points

• Key points:
• The real estate market is characterized by the coexistence of several price indicators from various sources that provide signals on pre-sale values (e.g. appraisals, initial offer price on real estate portals), during the transaction (notaries and registrars) or after the transaction (registry office). These indicators also differ in their frequency of publication and lag with respect to the reference data.
• To strengthen price monitoring in the sector, at BBVA we introduced a new indicator built with Big Data, based on the appraisals used in the bank's regular activity. In addition, to improve real-time price estimates (MITMA/Appraisals) in the short term we built a set of Dynamic Factor Models that allow us to capture the joint evolution of the available indicators.
• The results of the models are satisfactory, both in terms of the explanatory power of the variables of major interest and the ability to extract unobserved latent house price signals. The introduction of the BBVA high-frequency indicator in the models is statistically significant and is shown to increase the explained variance.
• The linear combination of models, through Bayesian algorithms, improves the predictive ability of BBVA Research. Specifically, it is shown that for different forecasting horizons different combinations of models are optimal, some of which include simpler specifications, such as univariate models, or the BBVA Research high-frequency indicator itself.

Authors

Geographies

Topics

Tags