Projects / Whether more listings help or hurt hosts in a vacation rental market
Whether more listings help or hurt hosts in a vacation rental market
Course project, Michigan Ross MBA. Instrumental-variable regression (two-stage least squares) in R on 281 local markets. Not client or employer work.
| Situation | A vacation rental marketplace operates in 281 local markets. Its growth teams add listings to make each market larger. |
|---|---|
| Complication | More listings can help hosts (more travelers come) or hurt them (more competition). A plain comparison of markets cannot tell, because hosts enter the markets where demand is already good. |
| Question | What is the effect of more listings on the bookings of a host? |
| Answer | They hurt. 10% more listings cut the booking confirmation rate by 1.7 points. A plain regression shows no effect. |
A host's bookings depend on demand and on the chance that an inquiry is confirmed; listings act on the second part
- Bookings of a host
- Travelers who inquire
- Inquiries per day in the marketFixed: held constant in the model
- Chance that an inquiry is confirmed
- Listings in the marketThe lever of the growth team
Do more listings cut the confirmation rate?
Yes. 10% more listings: 1.7 points less, from an average of 21%.
- Mix of listingsFixed: rooms, apartments, executive listings, location
- Nights occupied
- Occupancy rate
Does occupancy fall too?
Not clear. 0.25 points less, but p = 0.13.
Lever what the decision changes Limit what the answer must respect Fixed outside the decision
Booking confirmation rate, for 10% more listings
-1.7 pts
from an average of 21%; clear at the 5% level
The same effect in a plain regression
+0.1 pts
the competition effect is hidden, because good markets attract listings
Strength of the instrument
F = 12.8
above the usual minimum of 10, but not by much
A plain regression shows no effect; the instrument shows that more listings cut confirmations
| Outcome for hosts | Method | Effect | p-value | Reading |
|---|---|---|---|---|
| Booking confirmation rateaverage 21.0% | Plain regression | +0.10 points | 0.55 | not statistically clear |
| Booking confirmation rateaverage 21.0% | With the instrument | -1.71 points | 0.04 | clear at the 5% level |
| Occupancy rateaverage 4.1% | Plain regression | +0.04 points | 0.36 | not statistically clear |
| Occupancy rateaverage 4.1% | With the instrument | -0.25 points | 0.13 | not statistically clear |
Standard errors are robust to heteroskedasticity. Controls: inquiries per day, average number of rooms, share of apartments, share of executive listings, share near adventure sites and national parks.
The model
Question
Does a larger market bring each host more bookings (more travelers come) or fewer (more hosts compete)?
Problem
Listings are not random. Hosts enter the markets where demand is already good, so a plain regression mixes the two directions.
Method
Two-stage least squares. New listings from a migration of hosts move the market size for a reason that is not local demand.
- The occupancy effect has the same sign (-0.25 points) but is not statistically clear (p = 0.13).
- The first-stage F statistic is 12.8. The instrument is usable, but the estimates are not precise.
- For the marketplace, the result points to demand: more travelers, not more hosts, are what the weak markets need.
- The instrument is valid only if the new listings affect confirmations through market size alone. The data cannot test this.