Model Selection

This article explains the models in RoomPriceGenie to forecast demand, along with how they are working.

What is "model selection"?

RoomPriceGenie has 3 different methods for predicting demand for your property.

Why 3?! Because we have found three very good and totally independent ways to predict demand. By combining them, we get a very robust prediction of demand.

1. Local market/competitor hotel pricing
2. Local AirBnB pricing and occupancy
3. A Bayesian PickUp based model, using booking data only (similar to the model used by other modern RMSs)

On top of these methods, you are then able to add your own adjustments to the hotel and AirBnB prices in the market to get the profile of pricing that works for your hotel.

But different hotels have different requirements. The model selection enables us to set weights for each component.

The end result is the most robust pricing in the market— prices you can rely on.

What are the models that can be used in the algorithm?

• Hotel Competitors
  

  This is usually the biggest contributor to the demand calculation. By using the information on prices from 10 hotels, many with experienced revenue managers, we get the best view of demand for your hotel. It also helps us position your property well relative to the other hotels in your market. We will be cheaper than them if you have a lot of rooms left to sell, but more expensive when you are selling out fast.

• AirBnB Competitors
  

  Your competition is not just hotels. We get a different picture of demand when we include AirBnB listings. And the nice thing about our AirBnB data is that we get to know the occupancy as well as the prices they are charging. By combining current occupancy and prices charged, we get a really robust view of demand for AirBnB properties.

• Bayesian Pickup Model

  This is a similar type of algorithm to that used by RMSs for large hotels. It uses the price you are currently charging and then looks at how many bookings you are receiving at that price. If you are receiving fewer than expected, it adjusts the expected fair price downwards. If more, it adjusts it upwards. By doing this, the fair price gets more accurate every day. This fair price is used as a weighting in the algorithm.

What happens after?

The base price for the algorithm is determined when the 3 models are combined in the weighting that we choose. 

On days where there are not enough competitor hotels giving prices, then the other two models take over more. In this way, we can adjust for less information from competitors.

But this is just the basis. We then look at how you are performing over the whole month and see if the assumptions made were correct. We readjust pricing with the demand we see, meaning that we correct if we got anything wrong with our original demand prediction.

Finally, we optimize your profit based on the number of rooms you have left. This is where we match demand with supply to get the perfect price for you to optimize profit after variable costs are taken into consideration.

Under which circumstances would you use different weightings?

Bad competitor hotels: if your competitors are not good at pricing, not offering prices through the year, or not similar to you in clientele, we may want to lean less on this model.

If you offer vacation rentals: if your main competitors are listed on AirBnB, you may want this to form the strongest basis of your pricing strategy.

If you want to react more to pick-up than to your competitors changing prices: we can weight the Bayesian pickup model higher than the competitor hotels.

How do you change the weightings?

Your revenue management support at RoomPriceGenie has a lot of experience setting the weights. Together, you can discuss what is best for your hotel and set the weights for each model within the algorithm.