Agent-based modelling of purchasing, renting and investing behaviour in dynamic housing markets
Introduction
Residential Location Choice(RLC) and Real Estate Price(REP) models have been developed as essential components for many urban models, such as UrbanSim [1] and ILUTE [2]. In general, the studies of RLC were focused on two research questions [3]: 1) whether a household decides to move; 2) and where to move. Two classical and most-cited studies for these two questions are Rossi [4] and Brown and Moore [5], respectively. A comprehensive review of the RLC studies can be found in the work of Dieleman [6]. Recently, agent-based modelling has gained popularity in the studies of urban micro-simulations and has also applied to model RLC. A comprehensive review of the agent-based modelling of RLC can be found in the work of Huang, et al. [7]. The studies of REP generally investigated the factors that might influence the price and then used the empirical findings to develop methods to predict the price. Among them, the hedonic model appears to one of the most-used approaches [8], [9], [10], [11], [12], [13]. Since RLC and REP are highly correlated and interact with each other, they have been jointly studied. The economic bid-rent theory is a traditional approach to simultaneously modelling RCL and REP, in general with the assumption of a static demand-supply equilibrium [14], [15]. Some attempts have been made to couple the bid-rent theory with agent-based modelling, in order to investigate disequilibrium housing markets [16], [17], [18], [19], [20].
In general, these agent-based RLC-REP models mainly differ from each other in two aspects: agent types included and the behavioural rules of agents used. Buyer and seller agents were two typical types in agent-based RLC-REP models [16], [21]. Some of the models also considered developers and land owners, in order to simulate the land market as well [19], [20]. However, the renters and landlords in the rental market, as well as investors in the housing market have received relatively scant attention in these RLC-REP models. In response to this limitation, the proposed RLC-REP model will simultaneously simulate purchasers, sellers, landlords, renters and investors, considering the interactions and competitions between them. In terms of behavioural rules of agents, the utility maximization theory has been widely applied to model the decision-making of different agent types, including buyers and sellers, with the assumption that agents always choose the alternatives with the highest utilities [16], [17], [18], [20], [22]. Further, it has been argued that the loss in utility due to choosing a new residential location might have heavier influence than an equivalent-sized gain in utility, according to the prospect theory [23], [24], [25], [26]. Therefore, some attempts have been made to incorporate the prospect theory into utility functions, in order to consider the difference between gain and loss utilities [19], [21], [27]. Since many integrated urban models incorporated RLC and REP as key components, some agent-based RLC-REP models have been developed particularly for these integrated models: Ettema [16] proposed an agent-based micro-simulation model of housing market processes as a component of PUMA (Predicting Urbanisation with Multi-Agents) [28]; Habib [27] simulated both residential mobility and location choice processes within ILUTE (Integrated Land Use, Transportation, Environment) modelling system [2]. Zhuge, et al. [21] developed an agent-based RLC-REP model as a component of SelfSim, an agent-based land use and transport model. It is worth noting that the latter two RLC-REP models incorporated the prospect theory into their utility functions that were used to simulate the decision-making of agents.
This paper attempts to extend the agent-based RLC-REP model in SelfSim by incorporating the rental market and investing behaviour, aimed at fully capturing the interactions and competitions in the dynamic housing market. Specifically, the proposed RLC-REP model will define the behavioural rules of purchasers, sellers, landlords, renters and investors, and simulate the interactions and negotiations between these agents, resulting in new residential locations and real estate prices. As a component of SelfSim, the proposed RLC-REP model is directly and indirectly linked to several SelfSim sub-models. Among them, the demographic evolution model, which is used to simulate demographic transitions, is more closely associated with RLC-REP, as the life-cycle events (e.g., death and birth) of agents, which are the outputs of the demographic evolution model, are the key inputs of the RLC-REP model. Therefore, the demographic evolution model, as well as its relationship with the RLC-REP model will also be introduced. Another focus of this paper is to fully test the RLC-REP model using parameter Sensitivity Analysis (SA), in order to better understand how RLC-REP behaves, and also how the outputs of interest may be influenced by the model parameters. This kind of full SA has not been done for the previous version of RLC-REP in SelfSim and has also seldom carried out for the other land use and transport models or their components [16], [17], [18], [20], [22], [29], [30], [31]. SA is of great importance to new proposed models [32], especially for agent-based urban models that are generally involved in many parameters, as the SA results are expected to be useful for model simplification and calibration, as well as scenario analysis.
Section snippets
Agent-based land use and transport model-SelfSim
As aforementioned, the RLC-REP model is developed as a key component of an agent-based land use and transport, SelfSim. Compared with other micro-simulation land use and transport models, such as UrbanSim [1], ILUTE [2], ILUMASS [33] and PUMA [28], SelfSim aims to be more theoretically advanced and much more easily applied. Specifically, SelfSim attempts to use agent-based models to implement each component as far as possible, thereby simulating the co-evolution of land use and transport
Framework of the RLC-REP model
Fig. 2 shows the framework of the RLC-REP model which is composed of two modules, namely screening and negotiation modules. The former is used to sequentially check the number of triggers and buying/renting capacity of each household agent, and then to select those agents passing through the screening as the potential seekers that will enter the housing market. The triggers refer to the factors that could give rise to purchasing or renting demand, and most of them are life-cycle events, such as
Framework of the demographic evolution model
As aforementioned, the demographic evolution model is a key component of SelfSim that is directly linked to RLC-REP. Specifically, the outputs of the demographic model, such as income and birth, are the main inputs of the RLC-REP model for both trigger and capacity checking in the screening module (see Section 3.2). Fig. 4 shows the framework of the demographic evolution model which is composed of ten sub-models, and also reveals the relationships between the sub-models, as well as the linkage
Introduction to SA
The uncertainty in model parameters is generally viewed as one of the five types of model uncertainty [44]. It is vital to investigate the parameter sensitivity, especially for a new model, in order to quantify the extent to which the model outputs of interest are sensitive to the model parameters. Essentially, there are two types of SA, namely global and local SAs. They differ from each other in the consideration of interactions between parameters. Specifically, the local SA only varies one
Conclusions
An attempt was made in this paper to model the behaviour of several agent types in the housing market, including purchasers, renters, investors, landlords and sellers. The model starts with a screening module which only keeps those agents passing both trigger and capacity checking. Then the negotiations between seeker and offer agents are explicitly simulated, resulting in new residential locations of household agents and real estate prices. As a new model, its sensitivity was fully tested
Acknowledgements
This research was supported by the National Natural Science Foundation of China (Grant No. 51678044 and Grant No. 51338008), the Science Fund for Creative Research Groups of the National Natural Science Foundation of China (71621001) and the Hebei Natural Science Foundation (Grant No. E2016513016). We would also thank Dr. Mike Bithell for discussing with us about modelling and testing the sensitivity of the model.
Chengxiang Zhuge is currently studying for his Ph.D. degree in Geography at the University of Cambridge, United Kingdom. He is an urban geographer with particular interest in modelling dynamic urban and economic systems. His research is focused on modelling, implementing, testing and applying an agent-based Integrated Land Use-Transport (ILUT) model – SelfSim – that is Java-based. SelfSim was firstly applied to a Chinese medium-sized city, Baoding to simulate how its land use-transport system
References (49)
Hedonic analysis of housing markets
Handb. Reg. Urban Econ.
(1999)A multi-agent model of urban processes: modelling relocation processes and price setting in housing markets, Computers
Comput. Environ. Urban Syst.
(2011)- et al.
An economic agent-based model of coupled housing and land markets (CHALMS)
Comp. Environ. Urban Syst.
(2011) - et al.
A conceptual design for a bilateral agent-based land market with heterogeneous economic agents
Comput. Environ. Urban Syst.
(2008) - et al.
Agent-based joint model of residential location choice and real estate price for land use and transport model, Computers
Comput. Environ. Urban Syst.
(2016) - et al.
Multivariate uncertainty analysis of an integrated land use and transportation model: MEPLAN
Transp. Res. Part D: Transp. Environ.
(2006) - et al.
Assessing uncertainty in urban simulations using Bayesian melding
Transp. Res. Part B: Methodol.
(2007) - et al.
Anchoring and loss aversion in the housing market: implications on price dynamics
China Econ. Rev.
(2013) - et al.
An effective screening design for sensitivity analysis of large models
Environ. Modell. Softw.
(2007) - et al.
A matlab toolbox for global sensitivity analysis
Environ. Modell. Softw.
(2015)
UrbanSim. Modeling urban development for land use, transportation, and environmental planning
J. Am. Plann. Assoc.
ILUTE. An operational prototype of a comprehensive microsimulation model of urban systems
Netw. Spat. Econo.
Residential mobility and location choice: a nested logit model with sampling of alternatives
Transportation
Why Families Move
The intra-urban migration process: a perspective
Geografiska Annaler Ser. B, Hum. Geogr.
Modelling residential mobility; a review of recent trends in research
J. Hous. Built Environ.
A review of urban residential choice models using agent-based modeling
Environ. Plann. B: Plann. Des.
The composition of hedonic pricing models
J. Real Estate Lit.
Analysis of house prices to assess economic impacts of new public transport infrastructure, transportation research record
J. Transp. Res. Board
Location, regional accessibility, and price effects
Transp. Res. Rec.: J. Transp. Res. Board
Effects of transportation accessibility on residential property values
Transp. Res. Rec.: J. Transp. Res. Board
Neural network hedonic pricing models in mass real estate appraisal
J. Real Estate Res.
MUSSA: A Behavioural Land Use Equilibrium Model with Location Externalities, Planning Regulations and Pricing Policies, In, Working Paper
MUSSA: land use model for Santiago city
Transp. Res. Rec.: J. Transp. Res. Board
Cited by (11)
Exploring the role of technology innovations in the diffusion of electric vehicle with an agent-based spatial integrated model
2021, Resources, Conservation and RecyclingCitation Excerpt :Their functions are introduced as follows: The demographic evolution model (Sub-model 1), which comprises a set of sub-models, is used to simulate some key demographic transitions, including birth, death, marriage, divorce, employment, and unemployment (see the work by Zhuge and Shao (2018b) for model specification), so as to better simulate agents’ behaviours, because many of the individual behaviours in SelfSim-EV (e.g., parking and charging behaviours of EV users) are associated with their socio-demographic characteristics. The transport facility development model (Sub-model 6) is used to simulate the development of EV-related infrastructures, including parking lots, refuelling stations, charging posts at parking lots, and fast charging stations, according to infrastructural demand calculated based on the MATSim-EV simulation: see the work by Zhuge and Shao (2018a) for model specification.
Integrated modelling of autonomous electric vehicle diffusion: From review to conceptual design
2021, Transportation Research Part D: Transport and EnvironmentCitation Excerpt :For the economic system, it connects to the demographic evolution model through the employment and income modules (Zhuge and Shao, 2018b). Joint Model of Residential Location Choice and Real Estate Price (RLC-REP, connecting to Economy and Land Use Systems): simulates the interactions and negotiations between several typical agent types in a dynamic housing market, which are grouped into seeker (i.e., investor, buyer and renter) and offeror (e.g., seller, and landlord) agents: see (Zhuge and Shao, 2018b) for model specification. The joint model is used to find new locations for household agents and also to update real estate prices.
The potential influence of cost-related factors on the adoption of electric vehicle: An integrated micro-simulation approach
2020, Journal of Cleaner ProductionCitation Excerpt :In brief, SelfSim-EV is an integrated urban microsimulation platform incorporating the purchase and use behaviours of EV, as well as the EV-related behaviours, actions and decision-makings of agents. More detailed introduction to SelfSim-EV and its key components (e.g., transport facility development model) can be found in the work of (Zhuge and Shao, 2018a, 2018b; Zhuge et al., 2018a, 2018b, Zhuge et al., 2019, 2019d). The introduction to the datasets for the model initialisation and calibration can be found in (Zhuge and Shao, 2019; Zhuge et al., 2018a, 2018b, 2019a, 2019b).
Exploring the future electric vehicle market and its impacts with an agent-based spatial integrated framework: A case study of Beijing, China
2019, Journal of Cleaner ProductionCitation Excerpt :The outputs of SelfSim-EV simulation in 2015 were used as the inputs of the Beijing scenario here, and can be viewed as an agent- and GIS- based virtual Beijing. Note that all of the facilities (including both transport and activity facilities) in SelfSim-EV are assumed to be located at road nodes (or interactions), meaning that agents need to travel from one node to another in order to perform activities in different node-based facilities (Zhuge and Shao, 2018a, 2018b; Zhuge et al., 2018a, 2018b). Fig. 4 shows a specific element in the virtual Beijing: each dot in the map, which is based on road node or interaction, represents a residential building with the capacity information attached.
Investigating the factors influencing the uptake of electric vehicles in Beijing, China: Statistical and spatial perspectives
2019, Journal of Cleaner Production
Chengxiang Zhuge is currently studying for his Ph.D. degree in Geography at the University of Cambridge, United Kingdom. He is an urban geographer with particular interest in modelling dynamic urban and economic systems. His research is focused on modelling, implementing, testing and applying an agent-based Integrated Land Use-Transport (ILUT) model – SelfSim – that is Java-based. SelfSim was firstly applied to a Chinese medium-sized city, Baoding to simulate how its land use-transport system evolved from 2007 to 2014, and is currently being used to investigate the Electric Vehicle market in Beijing, China.
Prof. Chunfu Shao is a professor from the Beijing Jiaotong University. He has devoted 30 years to the area of traffic and is of experience especially in Transportation Planning, Traffic Management, ITS and Traffic Security. He fulfilled more than 20 Japanese projects like urban transportation planning for the Ministry of Land Use and Transport and the urban transportation planning for the local government in Japan. Since he came back to China in 1999, Prof. Shao has undertaken more than 100 projects, published more than 200 papers and 7 books, and obtained 4 patents.