Airbnb Revenue Showdown
Estimating and predicting revenue for Airbnb rentals in Seattle and Boston.
Introduction
Did you ever wondered how much could you make renting your property trough Airbnb? Perhaps even buy a new apartment and pay it with the rentals?
In this article we will look into two hot markets for Airbnb rentals: Seattle and Boston and see wich neighbourhoods and property types generate more revenue.
Finally, we will try to create a model to help us predict revenue.
About the Data
The dataset we used didn’t include directly how many days each listing were booked and for what price, so we had to estimated it.
Thanks to Inside Airbnb this task became a little easier.
In short, we utilized reviews and some m̶a̶g̶i̶c̶a̶l̶ ̶n̶u̶m̶b̶e̶r̶s̶ approximations to calculate days booked for each listing. For example, we considered that 50% of the people that rent through Airbnb leave a review.
Part I: What Neighbourhoods Generate More Revenue?
More importantly than to know wich city generates more revenue, we need to identify wich neighbourhoods of each city that contributes more for this metric.
In addition to this, we want to know what kind of property is best suitable for these areas, otherwise we might end up listing for rental a Camper/RV in Downtown Boston.
Since our data have different timeframes, we selected the last full year that we had available (2015) to do this analysis.
Here we can see that Broadway is the leader with it’s estimated revenue above 2.5 million USD followed by South End.
We can identify that Seattle’s revenue is concentrated in two main neighbourhoods, Boston on the other hand is a little more uniformly distributed.
Part II: What Kind of Properties is Best Suited for Each Neighbourhood?
We had a little spoiler in Part I about this question, but it’s worth to look at the numbers.
Figure 2 shows a table containing the property type that most generated revenue in each neighbourhood of each city, alongside with the listing quantity and an averaged revenue per listing.
From this we can observe that in Boston most listings are apartments and in most cases, the average revenue is higher than Seattle’s.
This indicates that if you want an apartment to put it for rental, you will be better of in Downtown or Back Bay in Boston, if you instead want a house, go with West Queen Anne.
Of course a higher average revenue probably also means that the purchase price of the property will also be higher, so you will have to considerar that in your decision.
Part III: Can We Predict Future Revenue?
Short answer: yes, we can. The key here is to know how good we can be with our predictions.
For this task, we utilized a simple machine learning algorithm called Lasso. It’s basically a linear regression with a L1 regularization. We also did a GridSearch on it to try to get better predictions.
We used the R2 score along with Cross Validation to see how well our model handled our data. Oversimplifying it, the R2 score can go up to 1, meaning that it’s a “perfect” predictor.
Here are the results:
Lasso Evaluation:
R2 train score: 0.85156950972085
R2 test score: 0.8470433348956904
Cross Validation R2 mean: 0.8350424187389971
Cross Validation R2 standard deviation: 0.020411665666851028This means that our R2 is on average 0.83, which is pretty good for a simple model.
It’s worth mentioning that the best way to validate these result is to put it to the test with data that we didn’t use neither in the train or test datasets so we can be more confident that we didn’t had any Data Leakages.
Conclusion
Recapitulating, we saw the 10 neighbourhoods of each city that had more revenue, from which Broadway is our champion.
We also observed that Boston market is dominated by apartments and have a overall higher average revenue.
And finally we obtained a fairly good score creating a model to predict future revenue.
These conclusions are very informal and need a more formal study. So we can still ask:
How much could you earn with Airbnb rentals?
To see the step by step of this analysis, see the link in my Github.
