Better unconstraining of airline demand data in revenue management systems for improved forecast accuracy and greater revenues

Revenue impacts of demand unconstraining and accounting for dependency

“…Polt (2000), Weatherford (2000) and later Zeni (2001) conclude that EM is the most robust method for unconstraining. Weatherford and Polt (2002) carry out an extensive simulation study to examine the impact of the demand estimates obtained by using six methods of unconstraining viz., naïve methods, booking profile method, projection de-truncation method and a method based on EM algorithm on the revenue. Constraining (per cent of total observations that are censored) the demand levels between 0-98 per cent, they find that for lower levels of truncation, naïve methods are superior, but, as the level of censoring increases, methods based on projection de-truncation and EM algorithm yield more robust results.…”

“…At the next step, we generate 100 observations (B i , Y i ) where, i ¼ 1, 2, y, 100 (Typically, airlines maintain records of about two years' worth of historical departures to make their next forecast. As the data are seasonal by day of week, they only have one observation for each day of the week (Weatherford and Polt, 2002). So, they would have about 104 (2 years  52 weeks/year) Mondays worth of observations available to make the next Monday's forecast).…”

“…The issue of demand un-constraining is handled by using Expectation-Maximization (EM) algorithm given by Dempster et al (1977). Weatherford and Polt (2002) document that for a 12.5-25 per cent demand constraining, the demand estimation using EM algorithm leads to a revenue increase of 2-5 per cent. As the percentage of data being constrained goes past 60 per cent, the estimates of means increase by about 20 per cent, yielding 2-12 per cent improvement in revenue performance.…”

“…As the percentage of data being constrained goes past 60 per cent, the estimates of means increase by about 20 per cent, yielding 2-12 per cent improvement in revenue performance. Weatherford and Polt (2002) used the EM algorithm for independent demands across (two) fare classes in an airline. The assumption of independence of demands across fare classes is quite common in the revenue management literature.…”

Dynamic VaR Measurement of Gold Market with SV-T-MN Model

“…Subsequently, Huiming et al [10,11] further studied the MCMC method based on Gibbs sampling to obtain the best SV model parameters estimation results. Since the SV-T model is difficult to estimate and the out-of-sample volatility prediction, the SV-T model is transformed into a linear state space without loss of any information [12][13][14][15][16][17]. Standard Kalman filter can obtain the better state estimation only in the linear Gaussian state space.…”