210 likes | 439 Views
Better Unconstraining of Airline Demand Data for Improved Forecast Accuracy and Greater Revenues. AGIFORS--RM Study Group Berlin, April 2002 Larry R. Weatherford, PhD University of Wyoming Stefan P ölt, PhD Lufthansa German Airlines. Outline of Presentation. I. Introduction
E N D
Better Unconstraining of Airline Demand Data for Improved Forecast Accuracy and Greater Revenues AGIFORS--RM Study Group Berlin, April 2002 Larry R. Weatherford, PhDUniversity of Wyoming Stefan Pölt, PhD Lufthansa German Airlines
Outline of Presentation I. Introduction II. Review of Common Unconstraining Methods III. Comparison of Unconstraining Performance IV. Comparison of Revenue Performance V. Conclusion
I. Introduction • One of the major factors that affects forecast accuracy is the inability to observe the true (unconstrained) demand • Had several presentations lately (Pölt, 2000; Weatherford, 2000; Zeni, 2001) that discussed different unconstraining methods • What’snew here? Want to quantify the revenue impact of using the more sophisticated unconstraining methods
II. Review of Common Unconstraining Methods Going to look at 6 commonly used methods: A. Naïve 1—use all data (open and closed) B. Naïve 2—use only “open” data C. Naïve 3—replace “closed” with larger of actual, avg. of “open” D. Booking Profile E. Projection Detruncation F. Expectation Maximization
III. Comparison of Unconstraining Performance Intuitively, it makes sense that more statistically sound procedures like the “Projection Detruncation” and “Expectation Maximization” methods should do a better job than the “Naïve” methods at estimating the true unconstrained demand, but the question is how much better and is it worth the effort? Of course, one of the real problems in performing this analysis is that if one uses real airline data, we never know what the true unconstrained demand is and therefore are not able to accurately compare all 6 methods
Leads us to use simulated data--randomly generated “true unconstrained demand” and also randomly generated “booking limits” that determine whether or not we observe the true unconstrained demand or some constrained value. Then, we can make an honest evaluation of how much better one method does than another and how close it came to the true unconstrained demand (because we secretly know what that is).
A. Simulated Data Sets We’ll look at 2 different data sets (1000 observations each): 1. Simulated #1, unconstrained mean = 20, % unconstrained varies from 0% to 98% 2. Simulated #2, unconstrained mean = 4, % unconstrained varies from 0 to 98%
IV. Comparison of Revenue Performance So, we know EM & PD are the most robust unconstraining methods …but how much revenue impact does it make when we integrate such an improved unconstrainer in our overall RM system?? Let’s test it using real airline (major US) data—2 representative markets (business, leisure) Process: historical bookings data available, unconstrain data, generate forecasts of demand, establish optimal booking limits, interface bkg limits with random arrivals, calculate actual revenue, [ Repeat for 1100 weeks of departures ]
Details: Leg Optimization generated by EMSRb Forecasts simply used average of past unconstrained obs. # of reading days/dcps = 10 # of fare classes = 5 (see next slide) Tested at multiple demand/capacity ratios (0.9 to 1.5)
C. Feedback loop on estimating unconstrained mean (true = 3.6)
V. Conclusion The type of unconstrainer you’re using can make a BIG revenue difference. Business: 1.3 – 2.9% on high demand legs Leisure: 2 – 12% on high demand legs Questions?