Customized regression model for Airbnb dynamic pricing | Hacker News Reader

Customized regression model for Airbnb dynamic pricing

blog.acolyer.org

59 pointsfeross9 comments

This is really interesting. I've found that Airbnb under-prices my listing pretty significantly, so I go through and set the weekend prices to roughly 2x of what smart pricing recommends, and they all book up anyway.

This makes me think that smart pricing errs on the side of giving guest a better value and not on the side of maximizing profits for the host.

This may be good for hosts in the long run, however, but I wish it could tell me specifically how it was weighing the factors involved in my listing. For instance:

- seasonal demand: high, weight: 3

- special case demand (annual conventions, etc): low, weight 2

- local lodging availability: scarce, weight: 5,

- percentage of the time guests choose your listing vs others when both are available: 75, weight: 3

- competitiveness based on incentives for length of stay: 8, weight: 3

- your typical guest price sensitivity: low, weight: 5

It would also be very cool for Airbnb to offer beta testing of different smart pricing algorithms. Supply and demand is volatile, so just because a unit averages $x does not mean it can't sometimes fetch 5x (see Uber's surge pricing). Smart pricing never does this, so I suspect there is no notion of demand surges built into the algo, even though they obviously occur.

> This makes me think that smart pricing errs on the side of giving guest a better value and not on the side of maximizing profits for the host.

It's very similar to the economic relationship between a realtor and a home seller. The home seller wants top dollar, the realtor doesn't care; they want deal flow, and a few thousands or tens of thousands of dollars in price difference impacts their commission little.

Airbnb would rather have a property available to be booked, and booked, then not booked with a host attempting to get top dollar. "For The Good Of The Platform"

> Supply and demand is volatile, so just because a unit averages $x does not mean it can't sometimes fetch 5x (see Uber's surge pricing). Smart pricing never does this, so I suspect there is no notion of demand surges built into the algo, even though they obviously occur.

There are laws against such price gouging in many US states, e.g. to prevent hotels from charging very high prices during natural disasters, or from discriminatory rates. Undoubtedly airbnb will not implement such protections unless forced to.

I think OP was talking more along the lines of, say, Atlanta during super bowl weekend, or Random, Wyoming during a total eclipse.

Would be AWESOME if they can someone hook into the "amount of tourist" in the country: be in flight-information, flight-occupancy-rates (check for full bookings ?), local tourism services. , Trending Google Searchs(dates+location services,visa requirments), TripAdvisor (searches amount) for location. twitter hashtags (many ppl i see post 3 more week before going to xyz) #mysuper-holiday-location. Feature engineering :) Got to love it

What should one study if you wanted to "get up to speed" with these types of models? Like which branches of statistics deals with this? It looks like you need to combine time-series with some kind of demand forecasting?

Say I wanted to do something like this to Widgets and adjust their prices over time, assuming I can collect data on how other people are pricing similar Widgets in the "market" (using this to estimate demand/supply), etc.

I'm just not sure what this type of field is called so I get get up to speed with state of the art.

One of the keywords you're looking for is "statistical forecasting."

It's a fairly specialized data science skill since it requires experience with some fairly specific techniques (e.g. dealing with seasonality, autocorrelation, etc. within data has all kinds of interesting solutions).

This looks a lot like getting short term profit increases for losing long term control. I wouldn't do it, at least not with just Airbnb suggestions.

What's the point of generating a price plot without a y axis?

Reply on news.ycombinator.com