Redes neuronales artificiales para la predicción de precios inmobiliarios

Abstract

Los modelos econométricos en la valoración de precios inmobiliarios constituyen una herramienta útil tanto para los compradores como para las autoridades locales y fiscales. Desde los modelos hedónicos clásicos hasta los planteamientos actuales a través de redes neuronales artificiales (RNA), han tenido lugar numerosas aportaciones en la literatura económica que tratan de comparar los resultados de ambos métodos. Insistimos en el empleo de RNA en el caso de disponer de suficiente información estadística. En este trabajo se aplica dicha metodología en una ciudad de tamaño medio situada en el sur de España, utilizando una extensa muestra de datos que comprende varios años precedentes a la crisis actual. Las variables utilizadas -tanto cuantitativas como cualitativas- incluyen datos externos e internos de la vivienda, del edificio en el que está localizada, así como de su entorno. Se construyen varios modelos alternativos para distintos intervalos de tiempo, siendo capaces de estimar los efectos de los precios crecientes del mercado alcista durante la década pasada.

Modelling property values in Nigeria using artificial neural network

“…The actual address of each property used in this study could not be retrieved from the real estate firms. This is not uncommon in the literature (for instance Selim, 2009;Kontrimas & Verikas, 2011;Tabales et al, 2013). (1)…”

Office property price index forecasting using neural networks

“…the location, number of stories, lot size, house type, exterior composition, house age, number of baths and number of units) for valuations of properties (Peterson and Flanagan, 2009; Selim, 2009; Wu et al , 2009; Igbinosa, 2011; Morano and Tajani, 2013; Chiarazzo et al , 2014; Morano et al , 2015; Rafiei and Adeli, 2016; Abidoye and Chan, 2017, 2018; Kang et al , 2020; Ho et al , 2021; Xu and Li, 2021), from different macroeconomic variables (e.g. the unemployment rate, default rate, stock market index, gross national product, interest rate, consumer price index and gross domestic product) for valuations (Rafiei and Adeli, 2016; Kang et al , 2020), from housing prices themselves for technical forecasts (Xin et al , 2004; Li et al , 2009; Xiaolong and Ming, 2010; Gu et al , 2011; Wang et al , 2014; Ma et al , 2015; Ge et al , 2019; Li et al , 2020; Liu and Wu, 2020), from house characteristics for technical forecasts (Nghiep and Al, 2001; Limsombunchai, 2004; Khalafallah, 2008; Lam et al , 2008; Igbinosa, 2011; Tabales et al , 2013; Morano and Tajani, 2013; Park and Bae, 2015; Sarip et al , 2016; Chen et al , 2017; Kitapci et al , 2017; Fu, 2018; Yu et al , 2018; Li et al , 2018; Rahman et al , 2019; Liu and Liu, 2019; Piao et al , 2019; Ge et al , 2019; Shahhosseini et al , 2019; Huang, 2019; Yan and Zong, 2020; Pai and Wang, 2020; Kang et al , 2020; Rico-Juan and de La Paz, 2021; Embaye et al , 2021; Terregrossa and Ibadi, 2021; Yasnitsky et al , 2021) and from macroeconomic variables for technical forecasts (Wilson et al , 2002; Xin et al , 2004; Khalafallah, 2008; Lam et al , 2008; Li et al , 2009; Tabales et al , 2013; Azadeh et al , 2014; Plakandaras et al , 2015; …”

“…Some previous research focuses on one specific machine learning model to forecast housing prices (Wilson et al , 2002; Xin et al , 2004; Lam et al , 2008; Khalafallah, 2008; Xiaolong and Ming, 2010; Igbinosa, 2011; Gu et al , 2011; Azadeh et al , 2014; Wang et al , 2014; Chiarazzo et al , 2014; Ma et al , 2015; Morano et al , 2015; Rafiei and Adeli, 2016; Wang et al , 2016; Kitapci et al , 2017; Chen et al , 2017; Abidoye and Chan, 2017; Ćetković et al , 2018; Li et al , 2018; Piao et al , 2019; Shahhosseini et al , 2019; Rahman et al , 2019; Li et al , 2020; Kang et al , 2020; Yasnitsky et al , 2021), some on comparisons of different machine learning models (Li et al , 2009; Wu et al , 2009; Park and Bae, 2015; Sarip et al , 2016; Li et al , 2017; Fu, 2018; Yu et al , 2018; Liu and Liu, 2019; Ge et al , 2019; Huang, 2019; Pai and Wang, 2020; Yan and Zong, 2020; Ho et al , 2021; Rico-Juan and de La Paz, 2021; Xu and Li, 2021; Embaye et al , 2021), some on comparisons between machine learning models and traditional econometric models (Nghiep and Al, 2001; Limsombunchai, 2004; Selim, 2009; Peterson and Flanagan, 2009; Tabales et al , 2013; Morano and Tajani, 2013; Plakandaras et al , 2015; Lim et al , 2016; Li et al , 2017; Abidoye and Chan, 2018; Yu et al , 2018; Ge et al , 2019; Liu and Wu, 2020; Milunovich, 2020) and some on model combinations (Taffese, 2007; Wei and Cao, 2017; Terregrossa and Ibadi, 2021). More specifically, for research focusing on one particular machine learning model, our reviews here suggest that the NN (Wilson et al , 2002; Xin et al , 2004; Lam et al , 2008; …”

“…On the other hand, the random forest is determined to beat the AdaBoost, CatBoost, XGBRegressor, decision tree, nearest neighbors, linear LASSO, linear Ridge and liner regression (Rico-Juan and de La Paz, 2021). And the linear regression is found to lead to lower forecast accuracy as compared to the Ridge, LASSO, bagging, boosting and random forest (Embaye et al , 2021), as well as the NN (Nghiep and Al, 2001; Selim, 2009; Peterson and Flanagan, 2009; Tabales et al , 2013; Morano and Tajani, 2013; Abidoye and Chan, 2018; Limsombunchai, 2004). The support vector regression based upon the ensemble empirical mode decomposition is determined to beat the random walk model and Bayesian (vector) autoregression models (Plakandaras et al , 2015).…”

“…Recently, many machine learning approaches and algorithms, such as the neural network (NN), regression tree, nearest neighbor, support vector regression, random forest, bagging, ensemble learning, boosting and deep learning, have been found to be useful and promising tools to various forecasting problems regarding housing prices (Li et al , 2009; Wu et al , 2009; Gu et al , 2011; Wang et al , 2014; Park and Bae, 2015; Plakandaras et al , 2015; Rafiei and Adeli, 2016; Chen et al , 2017; Fu, 2018; Yu et al , 2018; Liu and Liu, 2019; Shahhosseini et al , 2019; Huang, 2019; Li et al , 2020; Yan and Zong, 2020; Milunovich, 2020; Pai and Wang, 2020; Ho et al , 2021; Rico-Juan and de La Paz, 2021; Xu and Li, 2021; Embaye et al , 2021). In particular, the literature has shown that NNs have great potential to forecast (economic/financial) time series that tend to be highly noised and chaotic (Xu and Zhang, 2022o; Yang et al , 2008; Wang and Yang, 2010; Xu and Zhang, 2021d; Yang et al , 2010; Xu and Zhang, 2023i; Wegener et al , 2016; Xu and Zhang, 2022j), including many different housing price series (Nghiep and Al, 2001; Wilson et al , 2002; Limsombunchai, 2004; Xin et al , 2004; Taffese, 2007; Lam et al , 2008; Khalafallah, 2008; Xu and Zhang, 2021b; Selim, 2009; Peterson and Flanagan, 2009; Xiaolong and Ming, 2010; Igbinosa, 2011; Tabales et al , 2013; Xu and Zhang, 2022h; Morano and Tajani, 2013; Chiarazzo et al , 2014; Azadeh et al , 2014; Ma et al , 2015; Morano et al , 2015; Lim et al , 2016; Sarip et al , 2016; Wang et al , 2016; Abidoye and Chan, 2017; Kitapci et al , 2017; Li et al , 2017; Fu, 2018; Yu et al , 2018; Abidoye and Chan, 2018; Ćetković et al , 2018; Li et al , 2018; …”