Pricing a Short-Term Rental: Finishing the Model
In 2023 I wrote a short note on optimal pricing of short-term rentals. It found the price that maximizes revenue per night: the price at which occupancy falls by one percent for each percent the price rises. Then it stopped, because the rule needs the demand curve, and I did not know how to estimate one. I now have the rest of the model, written up as an 11-page tutorial. You can download the PDF.
How it was made
I had an AI coding assistant (Claude) work through the open question and write it up, starting from my note. Everything in it is checked against a simulated market. That market is a made-up cabin whose true demand is known, so every estimate can be compared with the truth. No real listing data went into it.
What it says
The demand curve in my note is a special case. The tutorial models the chance that an unsold night books on a given day. That chance depends on the price, on how far ahead the stay is, and on season, weekday and local events.
Your own booking history gives the wrong answer if you read it naively. You raise prices when demand is high, so high prices and full calendars show up together. In the simulated cabin, the monthly scatter of occupancy against price slopes upward, by 0.7 occupancy points for every $10. Even a regression that controls for season puts the price sensitivity at $226, against a true $110.
The fix is a small random price nudge. Moving each day’s price up or down by up to 15% at random costs about 1% of profit. A two-stage fit on the nudged data then recovers $97 ± 18.
A dynamic program turns the estimate into a price for each night, on each day it is on sale. The price starts high and falls as the stay gets closer.
Here is how each way of pricing the cabin did, as a share of the profit you would make if you knew the true demand:
| Pricing | Share of best profit |
|---|---|
| Dynamic program, estimate from the nudges | 98.7% |
| My 2023 rule, one price per night, true curve | 96.0% |
| Best single fixed price | 94.1% |
| The simulated host’s rule of thumb | 84.3% |
| Dynamic program, naive estimate | 74.8% |
The last row is the one that surprised me. A clever pricer fed a bad estimate does worse than a simple rule of thumb.
The catch
One listing learns slowly from its own nudges. Pinning the price sensitivity down to within 10% takes about 21 years of data at a 20% nudge. The good news is that profit is flat near the best price. An estimate 10% off loses under 2%, and 25% off loses about 5%. Starting from an estimate taken from comparable listings, even one that is 33% off, the simulated cabin earned 97% to 99% of the best profit in its first year.
The tutorial ends with a weekly routine a single host can run: log the calendar every day, nudge the prices at random, refit, and let the dynamic program set the prices.
Limits
It is a model, checked on a simulation. It does not cover competitors reacting to your prices, the platforms’ search ranking, cancellations or reviews. I have not run it on our own listings yet.
As always: this is what worked (and didn’t) for us, not legal, tax, or investment advice.