In previous parts I mentioned about the importance of forecasting in RM and fact that it’s used in such processes as overbooking or evaluating RM implementation. Therefore appropriate choice of forecasting method is one the most important decision. Moreover as Lee (1990) shows that 10 per cent improvement in forecasting can contribute to 3 per cent increase in Revenue. Following methods are the most popular and widely used in practice.
But before we can move to actual methods and their application. Let’s spend some time on data I used to model the demand. Data was generating randomly on non-existing hotel. The reason for such approach is fact that the goal of it is not a comparison of methods but explaining how to use them in practice. Therefore, I made following assumptions.
- The hotel has 100 rooms, and average occupancy of 90%
- The data comes from 3 years period
- Reservations are tracked 30days before arrival date
- The date is 31/12/2013
Moreover, the dataset will be divided into two sets. Training one and the second one, which will be useful to evaluate forecast. But please keep it mind, that I am only showing the method to create and evaluate forecast and the result is meaningless as the data was randomized.
Therefore, I might have sometimes referred to my Master Thesis were I conducted full empirical research on actual sales data of an existing hotel. Mostly to present the efficiency of particular method.
The data is available here: Dataset
Some of the methods will not be calculated in Excel (as it would be too difficult) but in special econometrics software. Among many different I chose Gretl software as it is easy to use and it is open source software. However, don’t worry, when certain model will be calculated, I will include detailed description so you can follow my steps. I assume that installation will not be a problem, but If you had any issues let me know.
Gretl might be download from here: Gretl
There are several ways to compare different forecasting methods. Most popular rely on calculating error between actual and forecasted values and are summarized below:
I chose to use Mean Absolute Percentage Error as it clearly shows the difference between forecast and actual values in percentages.