Exploring a Pricing Model for Urban Rental Houses from a Geographical Perspective

Round 1

Reviewer 1 Report

Major revision

This paper illustrates an advanced engineering approach, which models rental housing prices formation. The combination of traditional OLS and GWR methods with the nonlinear-oriented neural network technique provides more accurate prediction for rental housing prices. The study presents a complete picture of the prediction results and their statistical performances for four housing submarkets by employing four different empirical procedures, which are the traditional HLM(OLS), GWR, CNN, and CNN with GWR. The list of POIs is lengthy and complete to model rental housing prices. Especially, the originality of this paper is the discovery of the role of “quantity-based” neighborhood variables in increasing the model performance. Some important hedonic variables, however, seems missing. In addition, some other improvements are necessary for this paper to be published in Land.

1. [The reason to look into the rental housing market] The authors provide some explanations why they look into the rental housing market, rather than the owner’s market, from the perspective of the Chinese context. Another reason why investigating the rental market for price prediction would be the complexity of the formation of sales prices, in that housing sales happens not only by the movement of use value, but also by the fluctuations of transactions value. I hope you mention this perspective with citing the following seminal papers in the housing economics discipline:

Henderson, J. V. and Ioannides, Y. M. 1983. A model of housing tenure choice. The American Economic Review 73, no.1: 98-113.

Ioannides, Y. M. and Rosenthal, S. S. 1994. Estimating the consumption and investment demand for housing and their effect on housing tenure status. Review of Economics and Statistics 76, no.1: 127-141.

2. [About hedonic variables] First of all, you need to control for residential density (FAR) for the residential building (not just for the community). We expect that density, all else equal, is negatively related to rental prices because dense environment is more likely to result in congestion and the invasion of privacy. Second, as for the building year variable, you need the completion date (in month) and the date sold for each housing unit (in month). Then, the age variable is calculated by subtracting the completing/construction date from the sales date and by dividing the new variable by 12 for it to be in years. You also need to include the age-squared variable (Consult the paper by Bokhari & Geltner(2018) about the inclusion of the age-squared variable, published in Real Estate Economics). I see that you have the “Year” variable. How did you calculate the variable? Third, you should add the (monthly) time trend variable. Suppose the earlies date sold in your data is February, 2010. Then, the observation takes the value of 1 for the time trend variable. It is 2 for the units sold in March, 2010. Finally, you need three seasonal dummy variables to control for the seasonality.

And, most importantly, the dependent variable should be the unit price, which is the rental price divided by the housing living area. In economics, the rental price is expenditure, not the price. The unit price is the dependent variable of our interest. I wonder if you used the unit price or not.

3. [Forecasting] As for the presentation of the results, please provide the difference in predicted rental housing prices when using the different four models. Consult the following paper:

Hagenauer, Julian, and Marco Helbich. “A geographically weighted artificial neural network.” International Journal of Geographical Information Science (2021): 1-21.

 

<End of the 1^st review>

Author Response

    Thank you very much for your comments! We believe that all of your suggestions are instructive, valuable, and very helpful for revising and improving our paper. In addition, the papers you mentioned have great guiding significance to our research. We have carefully studied the comments and have done our best to revise the manuscript based on the comments. Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Thank you for providing me with the opportunity to read “Exploring A Pricing Model for Urban Rental Houses from a Geographical Perspective”. I have the following comments:

• The paper needs English editing and fixing of grammatical and syntax errors. Some sentences are very lengthy and harder to follow. For example, in the abstract line 14-19 is a single sentence. Please get the paper proofread. What do you mean by the sentence “The real estate market value is one of the most important issues in people’s lives.” I am not sure if this is an issue. This can be a matter of concern, though but it is more financial than related to life. Please rephrase such ambiguous sentences
• In the abstract, add a line at the end to show the applicability/ practical application of the model and highlight who is the target user for the model.
• What are the “rental” variables considered in this study? This is not clear.
• The problem considered in the study is not clear. This needs to be clearly presented and justified. There is confusion in the paper and the readers are not clear if the problem is related to renting, rental price or rental models’ spatial heterogeneity.
• Why are deep learning models best for this study? What other models have been used in various studies? What type of deep learning model is used here? Is it ANN, CNN, FCNN?? Or any other model? Clarify this in the introduction.
• Why are these four cities considered as case studies? What are the defining parameters in this selection?
• Please provide references for the selection of primary and secondary categories as listed in Table 1.
• Please revise the model to use the latest data. The dataset from Feb to Mar 2020 is too small. My suggestion is to expand it to at least 12 months of data, including the months of 2021.
• Table 2 is verbose and should be moved to the appendix. The same goes for Table 6
• Provide a comparison of the results obtained by your model with the ones previously published and discuss in detail.
• Provide the limitations of the study in conclusion.
• Add more references from 2020 and 2021.

Author Response

    Thank you very much for your comments! We believe that your suggestions are valuable for the revision and improvement of our paper, and are of great guiding significance for our research. We have carefully studied the comments and have done our best to revise the manuscript based on the comments. Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The study proposes the coupling of two models to make predictions of house rents in four Chinese cities. One model recognises linear relationships and the other recognises both linear and non-linear relationships (based on artificial neural networks). They also build a database with different types of variables including geographical variables. Based on the results it seems that by combining these two types of models the predictions are more accurate.

In section 2. Related Works, I recommend the authors to review and include some recent work on hedonic models and ML techniques in general, e.g.,

[1]Rico-Juan, J. R. and Taltavull de La Paz, P. (2021). Machine learning with explainability or spatial hedonics tools? An analysis of the asking prices in the housing market in Alicante, Spain. Expert Systems with Applications, 171:114590,

[2]Lirong Hu, Shenjing He, Zixuan Han, He Xiao, Shiliang Su, Min Weng, Zhongliang Cai (2019). Monitoring housing rental prices based on social media:An integrated approach of machine-learning algorithms and hedonic modeling to inform equitable housing policies, Land Use Policy, Volume 82, Pages 657-673

[3]Sanglim Yoo, Jungho Im., John E. Wagner (2012). Variable selection for hedonic model using machine learning approaches: A case study in Onondaga County, NY, Landscape and Urban Planning, Volume 107, Issue 3, Pages 293-306

One question I have asked myself is whether it would be possible to use only a non-linear model, for example based on neural networks, which is also trained with the geographical variables.

Author Response

    Thank you very much for your comments! We believe that your suggestions are very instructive and valuable for revising and improving our paper. The papers you mentioned are profound and have great guiding significance to our research. We have carefully studied the comments and have done our best to revise the manuscript based on the comments. Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

I believe the current version now can be published. Please double check spelling or typos.

Reviewer 2 Report

I am satisfied with the revised version. All the best