Predicting Hong Kong Airbnb prices
Airbnb went public on Dec 10th 2010 and its share price has doubled on the very first day. No, I was not lucky enough to be part of that but its news got me interested in looking at its website once again after having been stuck in Hong Kong due to COVID-19 and away from travelling for so long.
Hong Kong is the world most expensive place to buy a home. It is such a small place but you may be surprised to know that Airbnb has over 7200 listings for it in Nov 2020 when compare to metropolitan cities such as Seattle (8700 listings) and Boston (3500 listings) in the same period.
Despite this, I come to know that most of the Airbnb listings in Hong Kong could be illegal under Hong Kong law.
It is quite intriguing. I have decided to take a deeper look at Hong Kong’s Airbnb data (available at insideairbnb.com) to find out more:
- Is there any COVID-19 impact on Airbnb market in Hong Kong?
- What is the most common type of Airbnb rental property? Where is the most popular area?
- Any price seasonality trend?
- Can we predict the Airbnb prices and figure out a way to provide pricing insight?
COVID-19
Before looking at 2019 numbers I would have thought over 7,000 listings in 2020 is amazing. But if I compare Nov 2020 data with Nov 2019 there is actually a 42% drop from 12,485 listings to 7,226! Before COVID, there was 12,485 listings on Airbnb for Hong Kong!
If we look at the availability information, we can see that availability percentage have increased from 47.3% in 2019 to 69.8% in 2020.
What does this tell us? The number of listings have dropped 42% with remaining vacancy increased 22%!
Actually availability is only a proxy of vacancy as ‘Unavailable’ does not necessarily implies confirmed and paid booking.
Nevertheless, this is a very bloody market in 2020!
What’s popular?
In 2019, there were 5,912 ‘Entire home/apt’ and 5,605 ‘Private room’ listed on Airbnb for Hong Kong. But in 2020, the numbers have reduced to 2,602 and 3,975 respectively.
‘Private room’ used to be on par with ‘Entire home/apt’ but I guess in this pandemic situation many property owners with ‘Entire home/apt’ have converted their short term rentals to longer term rentals thus removing listings from Airbnb or those rental arbitrators have ceased their operations.
Where?
From the charts below, we can clearly see that most of the listings are concentrated in Yau Tsim Mong, Central & Western HK Island and Wanchai area, for both 2019 and 2020. Although the numbers have changed due to COVID but the pattern remain the same.
I have also put the listings on the map for a better perspective: 2019 vs 2020
Let’s zoom in..
Price Trend
I grouped the calendar data by months and took an average on them to create monthly averaged prices and plotted the charts below, it seems like price remain mostly flat throughout the year except perhaps slightly cheaper in November and December with the peak in October.
The Prediction Model
Since there are more listings data available in 2019 I have decided to use that instead of the 2020 dataset. I have successfully built a prediction model based on Random Forest Regression algorithm. The model’s performance is measured by something called R² score. The R² score for the model was above 0.85 (the closer it gets to 1 the better in general), which seem to be decent.
However when I tried to fit the 2020 dataset into the prediction model and the corresponding R² score was -0.107 which is close to zero, which essentially suggested a failure.
The Conclusion
While a model was built with promising R² score performance using 2019 dataset, fitting the 2020 dataset into the model produces a very different outcome.
Something is clearly wrong, either the modelling algorithm selected is not the right choice, or the data covering longer time horizon should be used. Having said that, I also strongly believe that the 2020 dataset is likely very much distorted due to COVID and therefore the model was not able to correctly predict the prices in 2020.
Future Improvement
Dataset for a longer time horizon should probably be used to build a better model or maybe there is a better modelling algorithm. Given this is my first attempt I hope I could revisit this topic in the future when I am better equipped with different skills. Comparing other regions’ dataset could also give us better insight and help us build a better model for Hong Kong.
I also hope in the future if I can combine with other models such as an Automatic Valuation Model (AVM) I can provide better insights such as investment rental yield for different properties for Airbnb in Hong Kong.
