Additional Demand
Additional Demand Calculation:
Additional Demand employs a statistical algorithm, This algorithm, operates as following terms.
Bookings, enclose Cancellations and Changes, across various data points.
A “window” defines the data points used in the process, and the system employs two different windows.
The property-specific window is determined based on the property’s booking pattern through data analysis.
The window is a rolling window, initiating data processing once there are bookings in the window, and the rate program is closed.
A rolling window / Rolling-window analysis (Time-series model)
Means that Additional Demand runs daily, with the start of the “days left” changing each day.
Most properties use a 3 X 5 X 15 window, while those with late bookings may use a 7 X 3 X 21 window.
The numbers represent the number of arrival dates, days left before the arrival date, and the total points for calculation, respectively.
For a property with a 3 X 5 X 15 window:
The model uses bookings for 3 arrival dates (same day of the week), considering the booking pattern for 5 days before each arrival date, totaling 15 points.
For instance, if producing additional demand numbers on May 1st for the arrival date of Tuesday, May 24th, using a 3 X 5 X 15 window:
With 23 days left before arrival, booking information for 5 days left for the 3 closest historical arrival dates (May 10, May 17, and May 24) is used.
This rolling window concept ensures that each day the start of the “days left” changes. For example, on May 19th, producing demand for the arrival date of May 24th, the bookings from May 10, May 17, and May 24 would be considered.
Additionally, in a 7 X 3 X 21 window, with 7 arrival dates, 3 days left, and 21 points, data points would be taken from specific dates.
The model also considers holidays differently.
For example, in a 3 X 5 X 15 window, the model uses only the last number (total points) for holiday calculations and excludes the first two numbers. This accounts for the dissimilarity of holidays compared to other data points.
The example provides insight into the dynamic nature of the rolling window and how the algorithm adapts to different scenarios, incorporating the most recent information for accurate unconstraining
results.