1. Introduction

Since its inception in 2008, Airbnb has made great strides in the short-term lodging industry [1]. The selection of lodging by consumers is influenced by housing prices, which may exhibit discernible spatial autocorrelation [2]. To investigate the influence of spatial and locational factors on Airbnb prices, this article: (1) visualizes the overall distribution of prices; (2) conducts Global Moran's I analysis on prices; (3) conducts a GWR model using a semi-logarithmic model with four categories and a total of 10 explanatory variables.

2. Literature Review

Given that hosts are in charge of pricing their own listed properties, analyzing the elements that influence Airbnb prices is significant for researchers and industry professionals [3]. The hedonic price model, a popular approach in this discipline, hypothesizes that the price is reflective of the consumer's assessment of the property's disparate features [4]. Two primary elements that may explain the differential pricing include physical attributes (e.g. the number of bathrooms, etc.) and location-specific variables (e.g. proximity to amenities, the desirability of the neighborhood, etc.) [5].

Besides, factors related to the host or customers such as the characteristics of the host [6], host responsiveness [7], customer ratings, and their review comments [8] are also essential. Real estate prices are highly spatially dependent, exhibiting a high degree of homogeneity in attributes across homes within the same geographic area [9]. The traditional Ordinary Least Squares (OLS) model fails to account for the spatial dependence of Airbnb prices, thus rendering the results of the price model statistically biased [10,11]. An academically accepted approach to tackling spatial autocorrelation (SAC) is the application of geographically weighted regression (GWR), which can provide a spatial regression model and perform spatial prediction, while indicators such as Moran's I enable the test of SAC [12].

Learning from the literature, this article takes London as the study area, based on the Airbnb price data in London, selects 10 explanatory variables from the four dimensions of property characteristics, location characteristics, neighborhood characteristics, and transaction characteristics, uses Global Moran's I to test SAC and constructs the GWR model to the influence of spatial and locational factors on Airbnb prices.

3. Study Area and Data

3.1. Study Area

The city of London is situated in the southeast region of England, the coordinates of its central area being 51°30' north latitude and 0°5' east longitude [13]. Fig.1 shows London's 32 boroughs, underground stations, and green space areas.

Figure 1: London Profile.

3.2. Data Source and Variables Selection

The final data set selected in this article encompasses information about Airbnb listings in London over the 12 months preceding December 10, 2022, with a total of 51,218 valid values collected. The data was sourced from Airbnb's official website http://insideairbnb.com/, and is pre-processed in QGIS, RStudio, and ArcGIS according to the following principles:

(1) Removes missing values and hard-to-quantify text information.

(2) Adopts the semi-logarithmic model, which takes the natural logarithm of Price and generates a new variable (LnPrice).

(3) Deals with categorical variables in terms of their types.

3.3. Selection and Measurement of Variables

3.3.1. Explained Variable: Listing Price (Unit: £)

There are 71,938 transaction prices in the original data set. Firstly, 13 outliers whose value equals 0 are discarded, and the natural logarithm (LnPrice) is taken for the remaining 71,925 samples. The sample distribution of LnPrice is shown in Fig.2. Besides, the distribution of LnPrice mean across London boroughs is visualized in Fig.3. The presence of spatial heterogeneity is apparent, as evidenced by the higher prices observed in areas closer to London's center, particularly in the south-western region.

3.3.2. Explanatory Variables

Considering the problem of possible multicollinearity, this article finally selected 10 explanatory variables and a categorical variable as shown in Table 1 below.

Table 1: Variable Clarification.

Apart from the full-sample regression, this article also conducts four regressions separately based on the room type to examine its influence on price. The corresponding distribution of LnPrice based on room type is shown in Figures 4-7 below. The descriptive statistics of the variables selected in this article are shown in Table 2.

Table 2: Descriptive Statisctics.

4. Methodology

4.1. Global SAC Test of Airbnb Price

SAC pertains to the level of correlation that exists between a particular attribute value within a regional unit and the corresponding value in a neighboring unit. This article uses the indicator Global Moran's I (Equ.1) to calculate the global SAC [12].

\( I=\frac{n}{\sum _{i=1}^{n}\sum _{j=1}^{n}{w_{ij}}}\cdot \frac{\sum _{i=1}^{n}\sum _{j=1}^{n}{w_{ij}}({x_{i}}-\bar{x})({x_{j}}-\bar{x})}{\sum _{i=1}^{n}{({x_{i}}-\bar{x})^{2}}} \) (1)

where w_ij represents the spatial weight value; n is the total number of spatial units. The value is between -1 and 1. At a significant level, Moran's I>0 indicates that there is a positive SAC, and the prices reflect the cluster effect. This paper uses ArcGIS to conduct a global SAC test on LnPrice and obtains Global Moran's I=0.20, P-Value<0.0001, which proves that the spatial distribution of London Airbnb prices is clustered. The result is shown in Fig.8 below.

Figure 8: Results.

4.2. GWR Model

The GWR model applies the spatial weight matrix to the linear regression model, enabling captures of spatial heterogeneity [12]. Its general form is shown in Equ.2.

\( {Y_{i}}={β_{0}}({u_{i}},{v_{i}})+\sum _{k=1}^{K}{β_{k}}({u_{i}},{v_{i}}){X_{ik}}+{ε_{i}} \) (2)

where u_i and v_i are spatial coordinates; \( {β_{0}}({u_{i}},{v_{i}}) \) is the intercept term; β_k is the estimated coefficient of the k^th influencing factor of the i^th point on the explained variable \( {Y_{i}} \) ; \( {ε_{i}} \) represents the error term.

The regression coefficient is obtained according to Equ.3.

\( {β_{j}}({u_{i}},{v_{i}})={({X^{T}}W({u_{i}},{v_{i}})X)^{-1}}{X^{T}}W({u_{i}},{v_{i}})Y \) (3)

where \( W({u_{i}},{v_{i}}) \) is the weighting matrix when the i^th spatial point characterizes the model.

Its formulation is shown in Equ.4.

\( {W_{ij}}={e^{-\frac{1}{2}{(\frac{{d_{ij}}}{b})^{2}}}} \) (4)

where \( {d_{ij}} \) represents the distance between \( i \) and \( j \) , and \( b \) is the bandwidth.

Thus, based on the aforementioned fundamental framework, this article constructs the following benchmark semi-logarithmic model (Equ.5) and uses RStudio to realize it.

\( LnPrice={β_{0}}({u_{i}},{v_{i}})+\sum _{k=1}^{K}{β_{i}}({u_{j}},{v_{j}}){X_{ij}}+{ε_{j}} \) (5)

5. Empirical Results

5.1. Global and GWR Model Results

5.1.1. Model Evaluation

The tables below show the global model and GWR of the full sample and classification discussions. It is noteworthy that the "bedroom" variable of shared rooms is not included in the regression, as all samples have a value of 1. Compared to the global model, the R² value of the GWR model of the full sample is 0.5720318, the Adjusted-R² value is 0.5320542. Combined with the results of the classification discussion, the results comprehensively show that the goodness-of-fit for the GWR model is notably higher than the global model for the study of the spatial impact of Airbnb prices, particularly in relation to the Private Room type.

Table 3: Results Full Sample.

Significance stars

* p<0.1,** p<0.05,*** p<0.01

Table 4: Results Classification.

Table 5: GWR Full Sample.

Table 6: GWR Classification.

6. Conclusion and Discussion

After implementing the global Moran's I analysis and GWR model on the sample data, this article draws a conclusion that LnPrice presents clustered characteristics. In addition, the impact of some explanatory factors presents obvious spatial instability, especially for variables related to location and neighborhood presents, that is, their influence on LnPrice is either positive or negative. In addition, this article also conducts a classification discussion to integrate the impact of Room_Type on LnPrice.

However, constrained by the availability of data, this article acknowledges the presence of the following limitations: (1)The non-uniform distribution of room types within the sample population has the potential to impact the efficacy of the model; (2) The absence of significant factors, such as the age of the dwelling, within property-related explanatory variables is notable; (3) Factors that reflect the amenity, such as supermarkets and shops, are neglected in neighborhood-related explanatory variables; (4) Area_Green Space and Distance_Station are seriously affected by dimensions. Based on the above considerations, this article suggests that future research can expand the sample size to improve model interpretation, incorporate additional POI data, and standardize independent variables to avoid dimension effects.