National Pension Hub Publications
Real Estate Valuation:
A Deep Learning Approach
The project will help develop a better understanding of real estate values in the Canadian property markets using a sample from Montreal housing market. We also aim to work on the commercial real estate markets to improve our understanding in the commercial markets using a US dataset. The neighborhood dataset and deep learning model created in this project are aimed to be extended to other regions of Canada such as Vancouver and Toronto and to the commercial real estate markets in Canada following the proposed project. The output in this project such as the neighborhood data and the deep learning model developed will help pension funds and practitioners improve their knowledge in their real estate holdings in the Canadian real estate markets.
This project will use a hybrid deep learning model to predict real estate prices by using real estate data by combining with point of interest (POI) dataset that we aim to create. POI data will help us identify the spatiotemporal neighboring effects on real estate prices. In the macro level, trends on house prices change in a low frequency manner since the influencing factors on prices changes slowly, i.e. the main characteristics of a property remains the same. However, in the micro level, due to neighboring effects or demographical changes, trends tend to change more frequently. In other words, these changes create a high frequency anomaly on the general trend in that particular location, which is difficult for traditional models such as hedonic or repeat sales models to capture such impacts. We propose a hybrid deep learning model combining historical POI data to determine neighboring effects. Together with locational and structural data, our hybrid deep learning model will be fed.
The output is to create a dataset on neighborhood effects and a hybrid deep learning model for the Canadian/US real estate markets to value such private assets. The hybrid deep learning model will improve the predictability of the current real estate valuation models and help us make more precise real estate valuation. We will also create real estate indices based on the model.
John Molson School of Business, Concordia University, Canada
Istanbul Technical University, Turkey
Erkan Yönder & Meriç Yücel
Dr. Erkan Yonder is an assistant professor of Finance and Real Estate at the John Molson School of Business (JMSB) at Concordia University in Canada. Before joining JMSB, Erkan has worked at Ozyegin University as an assistant professor for 5 years. Erkan has received his PhD degree in Finance and Real Estate at Maastricht University in 2013. Erkan has visited the Center for Real Estate, the Massachusetts Institute of Technology (MIT), as a visiting PhD student in 2012. He also hold a second PhD degree in Economics from Middle East Technical University.
Erkan mainly work on real estate finance covering subfields such as Real Estate Investment Trusts (REITs), green buildings, commercial real estate, commercial mortgages, behavioral real estate, and sustainability. His research projects in these fields have been published in leading academic journals including Real Estate Economics, the Journal of Real Estate Finance and Economics, the Journal of Banking and Finance, the Journal of Economic Geography, and the Journal of International Money and Finance.
Erkan’s research has been recognized and funded by leading academic and private institutions. Several of his works have received the Real Estate Research Institute (RERI) grants in the United States. One of his papers on green buildings has also been awarded for the best research article in Finance and Sustainability in Europe by the Principles for Responsible Investment (PRI), a United Nations supported initiative. One of his projects has also been awarded for a grant by European Public Real Estate Association (EPRA). The same project is awarded for the Nick Tyrrell Research Prize by INREV, the IPF, and SPR in the UK.
Meriç is a computer engineer specialized in robotics, signal processing and learning algorithms. He holds BS and MS degrees in computer engineering from Istanbul Technical University. Meriç is currently working on his PhD dissertation about Self Organized Signal Pattern Encoding using Wavelet Transform Method. He has conducted academic visits including the University of Washington to work on Biomimetic Information Coding project related to his PhD dissertation. Meriç is also involved in various projects on machine learning and deep learning.