Are Airbnb management companies really worth it?

Photo: iStockphoto

Having risen from humble beginnings, Airbnb has grown to become a household name in the span of a little over a decade. Despite having been rocked by the global coronavirus pandemic, it is still valued at $18 billion as of April 2020. A key part of Airbnb’s sales pitch concerns attracting tourists who want “unique, authentic places to stay and things to do … powered by local hosts.” But does this actually happen in real life?

Property management companies have popped up in recent years to capitalize on Airbnb’s dominance in the short-term rental market. For a rate per night, plus a fixed fee, homeowners can let a professional rental company manage their properties to increase their rental revenue — typically, through increasing occupancy rates. But does it work?

In this article, I take to the data to find out, using a Seattle Airbnb property listings dataset from the halcyon days of 2016. In particular, I ask the following questions:

1. Who are the hosts with the most property listings in Seattle?
2. What are the occupancy rates of these listings?
3. Are occupancy rates of investor-owned listings higher than that of individually-owned listings?

For simplicity, we will assume that all investor hosts are, in fact, actually property management companies. Looking at the host URLs of the biggest hosts usually reveals this. The questions above are asked in service of our underlying business problem: are property management companies worth it for individual hosts to go through to rent out their properties?

Note: below I use ‘property management company’ and ‘investor’ (and ‘individual’ and ‘non-investor’) interchangeably.

Data

To answer the questions above, we use two datasets covering (i) property listings and (ii) an availability calendar. The listings dataset provides a full description of each listing (unit characteristics, host information, average review score, etc.), and the availability dataset provides 365 days of availability data (2016–2017) for a set of host listings. Using the listings dataset, we can find the hosts with more than 10 properties:

Hosts with the most

In the listings dataset, some of the biggest hosts were listed under first names — for example, Global Luxury Suite properties (502) were listed under the name Kara, and Turnkey properties (354) were listed under the name Bo. These indicate which hosts are explicitly property management companies. Some of the others may also be property management companies but I was unable to verify this from the information available.

Because we will be using the availability calendar to make statistical inferences about the characteristics we should make sure that there is a significant overlap between the properties in the listings dataset and those in the availability calendar. A straightforward way to do this is by taking a ratio of the properties represented in the availability calendar to the host’s total number of listings, otherwise known as a representativeness (R-) index.

Given the big drop in R-index values past a certain point, we can feel comfortable using quite a high threshold (90%) to isolate which hosts we want to include.

Taking these two steps allow us to first isolate which investors have the most listings and then trim that list down to those who are best represented in the availability dataset. This gives us a sample of investor hosts about whom we can make more confident and informed inferences:

Occupancy

The listings dataset contains information on the number of days a unit is available over the next 30 days from when the data was scraped, which is presumably meant to be a snapshot into the occupancy rates of each listing. However, this might not give us as accurate of a look into this metric as we would like, given the time of year it was scraped, not accounting for seasonal changes in availability, etc. More accurate results would likely be yielded from using the availability calendar. Plotting the average occupancy rate over each investor host, calculated as the average number of days unavailable in a year over all a host’s listings:

On average, investor host listings have an occupancy rate of 23%. Property listings hosted by the rest of the population (i.e. individual hosts) have occupancy rates of 42%. The data tell us that individually-hosted listings are occupied, on average, 19 percentage points more of the time than investor listings.

Test

Finally, we want to ask the underlying question of this exercise: are property management companies worth it? Or in other words, are occupancy rates meaningfully different between the two groups? Even though we can probably make a pretty good guess of the answer with the information available to us at this point, we should test this hypothesis methodically.

We use a two-sample independent t-test which tells us whether there is a statistically significant difference between the mean of two groups which are independent from one other. This fits our situation fairly well as we have two independent groups — company and individually-hosted listings — for which we have two sets of occupancy rates. To give us a sense of whether this is the right direction we should be taking, we can use a box-plot to visualize the spread of data points within both groups:

Immediately we can tell that the variance in occupancy rates, and the median occupancy rates, of individual-hosted listings are greater than in the case of company-hosted listings. This provides us with additional evidence in favour of using the two-sample t-test as described above.

The results of this test are obtained by using the Scipy stats library and the ttest_ind module. We test the following:

Null hypothesis

H0: Investor mean occupancy rate = Non-investor mean occupancy rate

Alternative hypothesis

H1: Investor mean occupancy rate < Non-investor mean occupancy rate

The null hypothesis is that the means of both groups are the same, while the alternative hypothesis is that non-investor-hosted listings have, on average, higher occupancy rates than investor-hosted listings do.

The test produces a statistic of -3.127 with a p-value of 0.09 (or a one-sided p-value of 0.005), allowing us to reject the null hypothesis in favour of the alternative at conventional confidence levels. We can conclude that on average, the listings of non-investors have higher occupancy rates than investor-hosted listings.

Evaluation

Investor-hosted listings had mean occupancy rates of 23% in 2016, while non-investor listings had mean occupancy rates of 42%. What does this tell us, and how can we explain this? It is perhaps the case that non-investors simply have more popular listings. The ‘local host’ effect could result in higher review scores for individuals, boosting their occupancy rates over time. While outside of the scope of this article, it would be interesting to see whether there is as dramatic of a difference in average review scores or revenue of investor vs. non-investor listings.

The results of this exercise should give pause to any current Airbnb host considering a property management service for their listings. Perhaps it is the case that being a local host adds value to an Airbnb listing, through providing a personalized touch that guests seem to appreciate. Maybe the investor-host model just doesn’t work for Airbnb. We are limited of course by the scope of the data used, but this can point us in an interesting direction if we follow trends over time.

Lastly, I’d like to acknowledge the good work that folks at Inside Airbnb do to make open source data more accessible (http://insideairbnb.com/get-the-data.html).

Statistical and programmatic details can be found in my GitHub repo. ❤