Academic journal article Current Politics and Economics of South, Southeastern, and Central Asia

A Temporal Case-Based Procedure for Cancellation Forecasting: A Case Study

Academic journal article Current Politics and Economics of South, Southeastern, and Central Asia

A Temporal Case-Based Procedure for Cancellation Forecasting: A Case Study

Article excerpt

(ProQuest: ... denotes formulae omitted.)

1. INTRODUCTION

In many industries, it is a common phenomenon for customers to book their desired products or services in advance, especially when the capacity is limited. For example, customers may reserve limited debuted records or railway seats many days before they actually get the record or use the service. In most situations, customers, not uncommon, may also cancel their reservations before they use the product or service with or without paying penalties. In the airline industry, cancellation fees are usually applied when passengers choose to cancel their reservations. Industries such as hotels and rental car companies allow their clients to cancel bookings without charging a penny if cancellations are done within a time period such as 24 hours before using the service. An extreme example is in the restaurant industry which restaurateurs even charge nothing for no-shows in most situations. One severe problem of cancellation is that customers who book late are denied if the number of bookings has reached the capacity. These late and denied booking clients, however, attempt to pay more for their desired services than early booking customers. Most of denials will switch to other alternative services as a consequence. A powerful solution to tackle the abovementioned problem is overbooking which operators sell more seats (higher than the capacity) to compensate the anticipated cancellations. Smith et al. (1992) indicated that without overbooking controls, American Airlines may have 15 percent spoiled seats on sold-out flights.

Revenue management (RM) is widely utilized to help operators arrange perishable capacity so that maximized revenues can be achieved. Kimes (2005) has indicated that using RM may bring 3-5% increase of revenues in the hotel, rental car, and airline industries. In RM, cancellation forecasting offers essential information to determine the volume of overbooking. The capacity and the volume of overbooking form the so-called pseudo capacity which is an important constraint while deciding the allotment of resources. As a result, the accuracy of cancellation forecasting is critically related to the performance of revenue management systems. Lee (1990) has shown the value of accurate forecasting in RM which 10% increase of predictive accuracy can bring 0.5-3% improvement of revenues for high demand flights. Ridel and Gabrys (2007) also mentioned that 10% reduction of forecast errors can bring 2-4% of additional expected revenues for airlines.

Starting from historical booking models, which bases on conventional time series perspective and utilizes solely cancellation numbers on the service day such as exponential smoothing and moving average (Weatherford and Kimes, 2003), more sophisticated models are also available in the literature. For example, ARIMA, neural networks, and machine learning algorithms are all common and potential alternatives. In M3-competition, Makridakis and Hibon (2000) have compared the performance of 24 time series techniques on 3003 time series data. These 24 tested methods can all be potential tools for cancellation forecasting.

Another important stream to predict cancellations (or no-show/standby/refund/exchange) is based on the use of passenger name records (PNR), which is the most detailed information collected in airline reservation systems. Four major types of features are usually extracted from PNR data for model construction: flight attributes, airport attributes, seasonal influences, and passenger attributes (Neuling et al., 2004). Flight attributes include information related to flight itself such as whether a trip is long- or short-haul and also the frequency. For example, long-haul passengers are less likely to cancel their reservations because they have planned their trips more thoroughly in advance. Another fact is that passengers are prone to make cancellation if the frequency is high since the cost of switch is going to be insignificant. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.