Building a Short Term Rental (STR) Forecasting Algorithm

The STR industry has witnessed remarkable growth, with many companies racing to provide insights and their own forecasting algorithms. These algorithms empower property owners, hosts, property management companies and investors to optimize their rental strategies or identify new properties through data-driven decision-making. In this blog post, we will delve into the key components of a robust short-term rental forecasting algorithm, and some approaches focusing on comparable unit selection, building occupancy, and pricing.

Common Methods: 

• Trailing Twelve Month Comparable Units 

• Competitive Set Daily Occupancy and Rates

Comparable Unit (CompSet) Selection

One crucial aspect of building an effective short-term rental forecasting algorithm is the selection of comparable units for analysis. Comparable units are properties that share similar characteristics and are located in the same or similar neighborhoods. By considering factors such as property type, size, amenities, location, and guest capacity, algorithms can identify suitable comparables. This allows for accurate benchmarking and analysis of market trends, enabling hosts to make informed decisions based on a comprehensive understanding of their property's competitive landscape.

Trailing Twelve Month Comparable Units 

With the CompSet some approaches involve simply averaging major KPIs over a specific period. For example, someone might average RevPan, Occupancy, ADR or Rents over a Trailing-Twelve Month (TTM) period and blend or adjust those for the target unit. There are a lot of nuances buried in the simplification of these metrics, but it does help offset some seasonality issues. This method can be "good enough" for very dense markets with a lot of identical inventory, for example a stack of condos that are identical. Some builders may take the period down to the week or month level to allow more flexibility. 

CompSet Daily Occupancy and Rates

Another approach is to build the CompSet at the daily level allowing one to select the basket of most comparable units on a given day. This method allows you to at the daily level have the largest sampling size of identical inventory and avoid throwing out a very comparable unit due to some seasonal blocks. 

Once you have your daily concept the methodology behind the daily occupancy and rates can be as simple or complex as you desire. You can simplify by having a straight forward expected value equation of probability of being booked (occupancy) multiplied by payoff (rate) giving you the expected rent for a given day. 

While occupancy is easier to build and adjust based on feature differences, rates can be where you go down a rabbit hole. You can either simply take the average of the basket of rates per day or you can leverage a pricing algorithm to price how you would in the market based on booking window, occupancy, length of stay, demand vs supply and so on. The later option would give more precision but could come with a false sense of accuracy or a feedback loop. 

Target v CompSet Differences 

One of the most important adjustments that will need to be factored into your roll ups, overlays or algorithms will be the differences in the target property vs the CompSet and the amount the market may be wrong. Without accounting for these differences you are just building a fancy averaging machine. 

Important factors to model into your adjustments would be unit level amenities versus CompSet and how that impacts bookings. Some of those features may be: 

• Size of Unit (Bedroom, Bathroom, Square footage) Versus CompSet

• Occupancy of Unit and Occupancy versus Bathrooms 

• Unit Specific Amenities (Pets, Hot Tub, Pool, Games, etc)

• Unit Lacking Expected Utilities (Wifi, AC, Parking, etc)

• Common Amenities Differences (Shared pool, shared hot tub, etc) 

• Unit Location and distance to key market amenities (ie Ocean Side of Inland Side of Highway 1)

• Unit Location Features (Water View, Water Front, Mountain View, Ski Ski Out, etc) 

• Unit Design or Luxury against CompSet (Good design makes a world of difference) 

• Review Scores and Counts are something very important guests review and consider 

Some of these features can be easier isolated to determine the marginal impact of an extra bedroom or occupant while others will be more cross correlated and difficult to isolate. It is important to contract of unit differences! 

Target v CompSet Sophistication 

Another caveat to performance outside of unit differences will be marketing and yield management strategies. If you assume the market is always right simply averaging the market rates might be a fine strategy. If you assume that you have pricing power over the market or the market is priced incorrectly then you may be able to charge higher rates during certain periods to earn more during high demand periods or charge less in lower demand periods to get higher occupancy. These assumptions should be built into your model to account for performance. Or in a highly dense market you may filter out units that you do not believe perform on your level. Whichever method you choose, it is very important to track assumptions against reality. 

Discounts & Refunds

One hidden item that is not captured in scrape data are discounts and refunds. These will include last minute discount, special rates requested by the guests or if an incident happens during the stay any refunds adjusted. While more one off on the aggregate these begin to add up, and are important to factor in. It is naive to operate as if you will not have discounts or refunds, so it is important to factor these earlier. There are a few ways to handle, either discounting the final projection after adjustments or by building a range around your projection. 

Continuous Improvement 

Building an effective forecasting algorithm requires continuous improvement and user feedback loops. Regular updates ensure the algorithm incorporates new data, adapts to evolving market dynamics, and integrates user feedback or business logic. By comparing forecasted outcomes with actual results and collecting feedback from your team, the algorithm's performance can be evaluated and refined. This iterative process ensures the algorithm remains relevant and reliable in a dynamic short-term rental market.

Closing Out

Change is the only constant, especially in this business. Due to this, it is important to always reevaluate your approach and to refine over time. It is also very important to keep in mind that you will never be 100% accurate and past performance doesn't always guarantee future results. Being with in an acceptable median error should be tolerable. 

You should of-course be honest with yourself and your prospects that a forecast is not a guarantee and prospects should pay more attention to "How" a manager will deliver performance rather than "What" the forecast says as anyone can put numbers on a paper but how are you going to put numbers on the score board is more important. 

A famous forecaster once said: "It's tough to make predictions, especially about the future." - Yogi Berra 

Curious to talk more about STR forecasting and approaches or want to get your hands on some data? Schedule time with an expert!