Escolar Documentos
Profissional Documentos
Cultura Documentos
Athanasius Zakhary, Heba Abdel Aziz, Mohamed Saleh Faculty of Com uters and !nformation, Cairo "ni#ersity Giza, E$y t n%el$ayar&fci'cu%edu%e$ , athanasius%it&$mail%com , heba%ds&$mail%com , saleh&salehsite%info Amir Atiya Faculty of En$ineerin$, Cairo "ni#ersity Giza, E$y t amir&alumni%caltech%edu Hisham El Shishiny !(M Center for Ad#anced Studies in Cairo !(M Cairo )echnolo$y *e#elo ment Center Giza, E$y t shishiny&e$%ibm%com ABSTRACT !n this a er a ne+ hotel re#enue mana$ement model is ro osed to ro#ide hotel mana$ers +ith a fle,ible and efficient decision su ort tool for room re#enue ma,imization% )he ro osed system is com osed of an ad#anced room demand forecast model and an o timization model that takes $rou reser#ations into account% )he system also allo+s for e, ert inter#ention to u date system arameters and out uts in order to inte$rate domain kno+led$e +ith com uter models% A case study based on three different sets of reser#ation records of simulated hotel data demonstrated the #iability of the system for $i#in$ useful recommendations related to acce tin$ room re-uests in order to ma,imize hotel room re#enue% .ey /ords0 1e#enue Mana$ement System, Hotel 1oom *emand Forecastin$, 2 timal Hotel 1oom Allocation
1. INTRODUCTION 1e#enue Mana$ement 31M4 is commonly racticed in the hotel industry to hel hotels decide on room rate and allocation% Hotel re#enue mana$ement is ercei#ed as a mana$erial tool for room re#enue ma,imization, i%e% for attem tin$ to sell each room to the customer +ho is +illin$ to ay the hi$hest rice so as to achie#e the hi$hest re#enue 3 El Gayar et al% 56674% A re#enue mana$ement system a lies basic economic rinci les to ricin$ and controllin$ rooms8 in#entory and e,ecutes t+o main functions0 Forecastin$ and o timization% )he forecastin$ system attem ts to deri#e future demand usin$ historical data and current reser#ations acti#ity ro#ided by the reser#ations system% (ased on the demand forecast and the remainin$ ca acity, the o timization function sets the a ro riate controls on the rooms8 in#entory% /ith the inte$ration of so histicated information technolo$y a roaches and their effecti#e combination +ith Machine 9earnin$, *ata Minin$, Statistics, 2r$anizational )heory and business e, erience and kno+led$e: ne+ models of Hotel 1e#enue Mana$ement Systems can be de#elo ed and used to ro#ide hotel mana$ers +ith effecti#e means to achie#e an o timal le#el of re#enue by sellin$ hotel rooms at #arious rice le#els to different cate$ories of customers 3Choi and Cho, 56664% )he use of hotel re#enue mana$ement systems is re orted to substantially increase re#enue of hotels 3Shehata, 566;4% 2n the other hand, e,istin$ 1M soft+are roducts are not based on inte$rated ad#anced com uter methods and so histicated rediction models, +hich make their results far from bein$ com rehensi#e or o timal% Further de#elo ment and enhancements in hotel 1M models is needed and is e, ected to ha#e si$nificant im act on the tourism industry%
!n this a er a ne+ 1M model is ro osed to ro#ide hotel mana$ers +ith an effecti#e tool for room re#enue ma,imization% )he ro osed 1M model is com osed of an ad#anced room demand forecast model and an o timization model that addresses $rou reser#ations% )he system also allo+s for e, ert inter#ention to u date system arameters and out uts to inte$rate domain kno+led$e +ith com uter models in order to ro#ide a fle,ible and effecti#e decision su ort tool for hotel re#enue mana$ers% )he a er is or$anized as follo+s0 Section 5 re#ie+s research efforts conducted in the field of hotel re#enue mana$ement and outlines roblems addressed by this study% Section < ro#ides an o#er#ie+ on the function of the ro osed 1M system% Sections = and ; are de#oted to outlinin$ the details of the forecastin$ system and the o timization model ro osed in this +ork, res ecti#ely% Section > resents the results of a case study% Finally conclusion and future +ork are resented in Section ?% 2. RELATED WORK AND PROBLEMS ADRESSED *etailed and accurate forecasts are crucial to re#enue mana$ement% !naccurate redictions lead to subo timal decisions about the rate and a#ailability recommendations roduced by the re#enue mana$ement system that in turn ha#e a ne$ati#e effect on hotel re#enue% !n addition, accurate forecastin$ can also hel hotels in better staffin$, urchasin$ and bud$etin$ decisions 3/eatherford and .imes, 566<4% 1e#enue mana$ement forecastin$ methods fall into one of three ty es% Historical bookin$ models, ad#anced bookin$ models and combined models% Historical bookin$ models only consider the final number of rooms or arri#als on a articular stay ni$ht% Ad#anced bookin$ models only include the build u of reser#ations o#er time for a articular stay ni$ht% Combined models use either re$ression or a +ei$hted a#era$e of historical data and ad#anced bookin$ models to de#elo forecasts% A re#ie+ of forecastin$ methods for all three ty es is found in /eatherford and .imes, 566< and Frechtlin$, 566@% !n this study articular interest is de#oted to reser#ation data as it is #ery rich and contains #ery useful information indicatin$ the actual demand to come% A recent study that com ares se#eral #ariations of ad#anced bookin$ models is found in Zakhary et al% 5667% Generally, #ery little +ork +as ublished on room demand forecastin$% 1esearch on re#enue mana$ement forecastin$ is mainly based on the airline industry% Most ublished +ork on hotel re#enue mana$ement systems deal +ith room ricin$ and allocation% 1efer to Chian$ et al%, 566? for a recent sur#ey% !n this a er an alternati#e a roach to room demand forecast is su$$ested% )his a roach is based on modelin$ hotel arri#als from its first rinci les and relatin$ it +ith other #ariablesA arameters like reser#ations, cancellations, seasonality and simulatin$ that for+ard to obtain the forecast% 2ur ro osed method is based on analyzin$ reser#ation data and e,tractin$ information' basically distributions of arameters' from ast reser#ation to feed them in a simulator to $enerate future reser#ation% )he ro osed forecastin$ module uses both current bookin$s and historical reser#ations and arri#als to deri#e accurate forecast% )o the best of our kno+led$e, there is no ublished +ork that simulates future reser#ations and uses it for forecastin$ as the ro osed system does% !n addition, modelin$ hotel arri#als as an interrelated rocess of many other factors and roducin$ density forecast rather than oint forecast, is a no#el a roach to hotel room demand forecastin$% 2n the other hand, hotels offer the same rooms to different ty es of $uests% /hile hotel mana$ers +ould like to fill their hotels +ith hi$hly rofitable $uests as much as ossible, it is $enerally necessary to allo+ for less rofitable $uests in order to re#ent rooms from remainin$ #acant% An im ortant decision to be made is +hether to acce t a bookin$ re-uest and $enerate re#enue no+ or to reBect it in antici ation of a more rofitable bookin$ re-uest in the future 3Goldman et al%, 56654% Findin$ the ri$ht combination of $uests in the hotel such that re#enues are ma,imized is the to ic of re#enue mana$ement% 1e#enue mana$ement is therefore defined in the hotel industry as the rocess of selecti#ely acce tin$ or reBectin$ customers by rate, len$th of stay and arri#al date to ma,imize re#enue 3Cinod, 566=4: by o timally matchin$ demand to a#ailable su ly 3rooms4 to accommodate the most rofitable mi, of customers% Methods for o timal ca acity utilization ran$e from sim le rule'based heuristics to so histicated mathematical ro$rams +ith hundreds of decision #ariables% !n$old et al% briefly re#ie+ed a subset of related mathematical ro$rammin$ a roaches and mar$inal re#enue a roaches% )alluri and Can 1yzin, ro#ided a com rehensi#e treatment for the classic, e,act and heuristic a roaches to sin$le resource ca acity control% Sin$le resource ca acity control deals +ith o timally allocatin$ ca acity of a hotel room for a $i#en date at different rate classes% 2n the other hand, the roblem of mana$in$
room ca acity on consecuti#e days +hen customers stay multi le ni$hts is referred to as the net+ork control ca acity roblem% )his deals +ith a mi, of customers ha#in$ different len$ths of stay and share the ca acity on any $i#en day% )he net+ork ca acity control roblem is si$nificantly more com le, than the sin$le'resource roblem and therefore its solutions rely mainly on a ro,imations% )alluri and Can 1yzin discussed some net+ork ca acity control methods in details% !n this +ork $rou reser#ations are of articular interest% Grou reser#ations D unlike indi#iduals' are reser#ations filed by a lar$er $rou of eo le and can either be totally acce ted or totally denied% Most o timization models ro osed in the literature i$nored $rou reser#ations and assumed a re'defined robability distributions to re resent $uest arri#als 3see (rumelle and /alczak, 566< and Slyke and Eoun$, 56664% Not takin$ $rou reser#ations into account is a critical limitation to the model: as $rou reser#ations can sometimes constitute the main stream for $uests accommodated in the hotel, articularly in hotels tar$etin$ tourist attractions% 2n the other hand assumin$ a robability distribution for hotel arri#als can be a rather restricti#e constraint in the real a lication of the hotel business: as hotels arri#al characteristics can ha#e a dynamically chan$in$ and e#ol#in$ nature% )he 1M model resented in this +ork treats the shortcomin$s discussed abo#e% !n articular the method resented in this a er e,tends e,istin$ o timization techni-ues for hotel re#enue mana$ement to address $rou reser#ations and uses Fforecasted demandF arri#als $enerated from the data% More details on the forecastin$ and the o timization models +ill be ro#ided in sections = and ; res ecti#ely% 3. OVERVIEW OF THE PROPOSED REVENUE MANAGEMENT SYSTEM )he ro osed 1M model is com osed of an ad#anced room demand forecast model and an o timization model that addresses $rou reser#ations% )he system also allo+s for e, ert inter#ention to u date system arameters and out uts in order to inte$rate domain kno+led$e +ith the com uter model to ro#ide a fle,ible and effecti#e decision su ort system for re#enue mana$ement% )he system erforms the follo+in$ functions0 @% Scans and analyses historical bookin$ and occu ancy atterns and current reser#ations and fits a -uantitati#e forecastin$ model 3forecastin$ module4% 5% Gredictions are used as in uts to an o timization module to make rate and allocation decisions% <% System users D +hich are mainly the hotel re#enue mana$ers' are allo+ed to chan$e system arameters and out uts and to assess ho+ these chan$es affect the final hotel re#enue% )his can hel them in makin$ better decisions re$ardin$ ricin$ and allocation% )he ro osed forecastin$ module takes as in ut recorded reser#ation records from the hotel database% Each reser#ation record contains all the arameters that ortray a certain reser#ation like Arri#al *ate, 1eser#ation *ate, 9en$th of Stay, 1oom )y eH% )his data is then analyzed and used to e,tract many arameters and com onents like )rend, Seasonality, (ookin$ Cur#e, CancellationsH% )hese arameters and com onents are then used to $enerate for+ard reser#ation records that +ould take lace in the future% )he future $enerated reser#ation records are assed +ith all their attributes to the o timization module as an in ut% (y analyzin$ these $enerated reser#ation cases one can obtain realistic erce tions for 2ccu ancy, Arri#als, and 1e#enue in the future% )he o timization module hence determines rates and allocations accordin$ to the estimated demand% At any sta$e the re#enue mana$er can o#erride system out uts, like for e,am le the forecasted demand% )his mi$ht be necessary in order to include ne+ e#ents in the surroundin$ en#ironment% After o#erridin$ system out ut by the re#enue mana$er, the system is re'run based on the ne+ chan$es and the reser#ation system is fed +ith the modified system out ut% Fi$ure @ illustrates the relationshi bet+een the ro osed system and the hotel en#ironment includin$ the e,istin$ 1eser#ation System, 1e#enue Mana$er and the Hotel *atabase% 4. ROOM DEMAND FORECAST USING SIMULATION Forecastin$ is often considered the most critical art of re#enue mana$ement% )he -uality of re#enue mana$ement decisions, such as ricin$, ca acity control, or o#erbookin$, de ends on an accurate forecast% Chian$ et al% re orted that a 56I reduction of forecast error can translate into a @I incremental increase in re#enue $enerated from the re#enue mana$ement system% )his section resents a ne+ a roach for room
demand forecast% )his a roach is based on analyzin$ reser#ation data and e,tractin$ information' basically distributions of arameters' from ast reser#ation to feed them in a simulator to $enerate future reser#ations%
saves Reservations on DB
$istening to Updates
R!%!(*! M&(&)!$
DB
F')*$! 1 Gro osed 1e#enue Mana$ement System and its relationshi +ith the Hotel En#ironment =%@ Hotel *ata Hotels store many forms of data like arri#als, reser#ation data, folio history, occu ancy rates and more% )he definitions and a#ailability of these data may differ from one hotel to another% )here are < forms of data that are of articular interest to this study0 1. Historical arri#als0 +hich are the final numbers of $uests that arri#e in the hotel at a certain date in the ast%
2. 1eser#ation 1ecords 31140 a detailed #ersion of reser#ation data% A 11 contains all the arameters that
ortray a certain reser#ation like 3Arrival Date, Reservation Date, Length of Stay). )his data should e,ist for all the $uests that used the hotel in the ast% 1ecords must contain at least the abo#e stated three arameters ' to be useful% 2ther arameters like Cancellation Date, Customer Type, Room Type can be also useful% <% (ookin$ Matri,0 a com act #ersion of reser#ation data% !t $i#es the buildu of reser#ations for e#ery day in the ast alon$ its bookin$ horizon% )he days in the future +ill ha#e incom lete buildu s% Considerin$ all the arri#al dates in the ast in addition to those that +ill take lace in the future and uttin$ them in ascendin$ se-uence +ill $i#e the bookin$ matri,%
)he 1eser#ation 1ecords3114 form can be considered the most im ortant data format as the other t+o forms of data, stated abo#e, can be $enerated from this Fra+F data% Many other forms could be $enerated by analyzin$ the 1eser#ation 1ecords of the ast% =%5 Forecastin$ Simulator )he forecastin$ simulator consists of t+o main modules: the analysis module and the simulation module% )he analysis module takes as an in ut the historical reser#ation records% !t then analyzes this data and uses it to e,tract many arameters and com onents like 3trend, seasonality, bookin$ cur#es, cancellations dynamics and len$th of stayH4% Certain distributions can be assumed a riori or can be set by the user: 3refer to Zakhary et al, 566?4% An alternati#e a roach is to deduce the distributions and the arameter #alues from the data directly% For e#ery distribution or arameter, a suitable a roach is de#ised to estimate it% For e,am le, the seasonal inde, is estimated usin$ the multi licati#e seasonal decom osition 3Frechtlin$, 566@4, the len$th of stay and room ty e distributions are estimated usin$ a sim le fre-uency based distribution estimator%
)he effects of the a#era$e daily reser#ations 3 i days rior to arri#al4 and the trend and seasonality is estimated by assumin$ that the net reser#ation rate b(i) e-uals a normalized reser#ation rate bn(i) times a trend com onent tr(t) times a seasonal com onent s(t), +here t is the arri#al date, as follo+s0 b(i) bn(i)!tr(t)!s(t) )he reser#ation rates bn(i) are estimated by $rou in$ the reser#ation cur#es to hi$h, medium and lo+ accordin$ to their corres ondin$ seasonality inde,es, and estimatin$ a tem late for each seasonal inde,% !n summary, the com onents e,tracted by the Analysis "o#ule can be cate$orized into < classes0 Easily E,tracted Com onents0 such as (ookin$ Horizon, Arri#al )ime series and Cumulati#e (ookin$ Matri,% - "ser *efined Com onents0 such as Forecastin$ !nter#al, and A#ailable 1ooms% - !ntelli$ently e,tracted Com onents0 such as Seasonal !nde,es, )rend, *uration of Stay *istribution, Cancellation 1ate, and (ookin$ Cur#e )em lates% A detailed list of all com onents and ho+ to com ute them can be found in Zakhary et al% 566?% Fi$ure 5 de icts the o eration se-uence of the Forecastin$ Simulator%
Raw Data
Reservation Date 1 1 : Current Date Arrival Date 5 20 : Current Date +7 Length of Stay 4 5 : 2 Room Type 2 1 : 4 Price Class 1 3 : 2 Cancellation Date : ... ... ... : ...
Processed Data
Seasonality Indexes Trend oo!ing C"rve Templates Length of Stay Pro#a#ilities Price Class Pro#a#ilities Room Type Pro#a#ilities $
A n a ly s is Fed as Input
%orecasting Sim"lator
ts u tp u
Raw Data
Reservation Date 1 1 : Current Date Current Date +1 Arrival Date 5 20 : Current Date +7 Current Date +7 Length of Stay 4 5 : 2 3 : 4 Room Type 2 1 : 4 2 : 2 Price Class 1 3 : 2 3 : 1 Cancellation Date : Current Date +5 : ... ... ... : ... ... : ...
F')*$! 5 2 eration Se-uence of the Forecastin$ Simulator )he Simulation Module takes the arameters and com onents estimated by the Analysis "o#ule as an in ut to $enerate for+ard reser#ation records that +ould take lace in the future% )he future $enerated reser#ations are assed +ith all their attributes to the $ptimi%ation "o#ule as an in ut% Analyzin$ these $enerated reser#ation cases: one can obtain realistic redictions for occu ancy, arri#als, and re#enue in the future% 2ne ad#anta$e of this simulation a roach is that one obtains distributions of future key arameters 3reser#ations, arri#als, etc4% )his is erformed by runnin$ the simulator many times from the current 3deterministic4 startin$ oint 3determined by the current sna shot of e,istin$ reser#ations4, thereby roducin$
many aths, each obtained by the different random com onents of the future reser#ations rocess% )his is a fa#orable and a re-uested feature for the subse-uent o timization module in the ro osed 1e#enue Mana$ement System% Fi$ure < de icts the a#era$e daily forecast for three months $enerated from the forecastin$ simulator% )he in ut data is obtained from actual historical reser#ations of Plaza Hotel1 in Ale,andria, E$y t% )he blue continuous cur#e re resents the a#era$e of @666 arri#al cur#es $enerated from the simulator out ut% )he #ariance of @666 forecasted arri#als are also lotted #ersus the actual arri#als% )able @ com ares the results of the forecastin$ simulator to classic forecastin$ techni-ues% !n articular, it com ares the o ular icku method, the e, onential smoothin$ method and the common racticed Fsame day last yearF method to the forecastin$ simulator: based on different statistical error measures 3MAE, 1MSE and SMAGE4 to forecast both arri#als and occu ancy% !t is +orth notin$ that the icku method cannot roduce forecast for the occu ancy because it does not take ad#anta$e of the len$th of stay% !t is ob#ious that the simulator forecasts out erform the other methods% )his is besides its su eriority in modelin$ the hotel arri#al intrinsically from the first rinci les, hence allo+in$ for Scenario Analysis based on control arameters and also roducin$ density forecasts rather than sim le oint forecast%
70 Simulator Forecast for t e Arriv al timeseries of !la"a #otel Actual $ccu%a&c' (imeseries )ea& Forecaste* $ccu%a&c' (imeseries + ,- 1 S(D Dev iatio& .o&fi*e&ce /a&*s
60
50 Number of Arrivals
40
30
20
10
0 0
10
20
30
40 Dates
50
60
70
80
90
F')*$! 3 A sam le out ut from the forecastin$ simulator T&+"! 1 Com arison of Arri#al )ime Series
Arrival M! ,o-.E$$o$ M!&#*$! Fo$!/&# '() S'0*"& o$ Gicku E, onential Smoothin$ Same &Day ' last (ear MAE 11.2112 @5%@;;> @7%7555 @<%5>?7 RMSE 13.1331 @=%J@J@ 5<%;775 @?%<J57 SMAPE 4.4113 6%==;< 6%?>5J 6%;6JJ Occupancy MAE 15.3665 NA <;%=;;> 56%<J@@ RMSE 21.1324 NA =5%@J<< 5>%J=67 SMAPE 4.3322 NA 6%7@;5 6%===6
Glaza Hotel, Ale,andria, E$y t is a ) stars, "i#'si%e, business as *ell as sea'si#e hotel. http+,,***.pla%aegypt.com,home.htm
3. ROOM ALLOCATION MODEL )he room allocation model uses the deterministic model described in literature 3(aker and Collier, 566<, (ertsimas et al., 566;, Goldman et al., 56654, +hich re laces demand by its estimation and obtains the o timal allocation of the rooms o#er the e, ected demand, i%e% it treats demand as if it +as deterministic and e-uals it to its e, ectation% )he deterministic model has three dimensions to define a stay in the hotel: 3a,9,k4, +here a is the first ni$ht of the stay, 9 is the len$th of the stay and k the rice class% Further, it denotes the set of stays that make use of ni$ht l by Nl, +here Nl K L3a,9,k40 a K l 0 aM9'@N% )he contribution in the +ork described in this section is mainly addin$ a forth dimension to the ori$inal model to re resent the size of a $rou 3$4 in a hotel +hile takin$ into consideration the inte$rality constraints, +hich are t+o roblems often i$nored in the literature 3(aker and Collier, 566<, (ertsimas et al%, 566;, Goldman et al%, 56654% )he follo+in$ are the arameters used in the model alon$ +ith the final form of the modified deterministic model% . a , L , - , g K )he number of rooms allocated to a stay of ty e (a,L,-,g)
Ra , L , - , g K )he re#enue obtained from a stay of ty e (a,L,-,g) # a , L , - , g K )he e, ected demand of a stay of ty e (a,L,-,g)
f=
SubBect to
a , L ,- , g
a , L ,- , g
. a , L ,- , g
l
a , L , - , g /l
gP.
a , L ,- , g
Cl
. a , L ,- , g # a , L ,- , g a , L, - , g . a , L ,- , g Q a , L, - , g
)he maBor difference in this model '' beside the addition of the fourth dimension '' is the first constraint L
a , L , - , g /l
gP.
a , L ,- , g
Cl N% Here the number of rooms allocated to each ty e of stay is multi lied by its
associated $rou size% )he obBecti#e of this model is to ma,imize the total re#enue under the restriction that the total number of reser#ations for a ni$ht does not e,ceed the ca acity of the ni$ht% Moreo#er, in order to re#ent #acant rooms, the number of rooms allocated to each ty e of stay is restricted by the le#el of the demand, +hich in this model is re laced by its e, ectation% )o incor orate stochastic demand, the randomized model found in 9ai and N$, 566; is used%, i%e% +e obtain the simulator out ut a number of times: and for each the modified deterministic model for $rou s D described abo#e' is run, before and after acce tin$ any s ecific re-uest% )he out ut for the randomized model +ill be the a#era$e o ortunity cost% Hence a re-uest +ill be only acce ted if its re#enue is lar$er than 3or at least e-ual to4 the a#era$e o ortunity cost roduced by the randomized model% )o formally describe the idea of an o ortunity'cost based control, consider a re-uest for a stay (a,L,-,g), +hich arri#es at a oint in time 0t1, and its re#enue is 0r1 (for the case of $rou reser#ations4%% )his re-uest is acce ted +hen the follo+in$ condition holds0 r R fbefore ' fafter
/here fbefore is the o timal total re#enue $enerated, at time t, by the model before acce tin$ the re-uest% And fafter is the o timal total re#enue $enerated, at time t, by the model after acce tin$ the re-uest% )he only difference bet+een the o timization model before and after acce tin$ the re-uest is the ca acity #ector Cl: i%e% in the model arisin$ after a re-uest is acce ted, the ca acity #ector is subtracted by the amount of reser#ed ca acity that must be allocated% 6. A CASE STUDY
)his section resents the results of the ro osed o timization model on a simulated hotel case study% Here a hotel mana$er +ants to identify the o timal allocation for rooms to each ty e of stay a year ahead% )he forecast simulator Ddescribed re#iously' rocesses reser#ation records that took lace in the last < years in the hotel and e,tracts arameters that are used to $enerate for+ard reser#ation records that +ill take lace in the comin$ year% From this $enerated data, the o timization model uses the follo+in$ attributes0 Arri#al *ate, 9en$th of Stay, Grice Class, and Grou Size% )he future time horizon of the study is hence <>; ni$hts, the ma,imum len$th a $uest can stay is ? days, and the ma,imum number of ersons in a $rou is @6 ersons% Moreo#er, the hotel mana$er has se$mented the $uests into < rice'classes 3+here each class is char$ed differently: the first class is char$ed for S=66, the second for S566 and the third is char$ed for S@664% )he hotel roomsT ca acity is @;66 rooms% !t is assumed that before the onset of the model that the initial room occu ancy is >6I for the first month: and this occu ancy rate decreases @6I 3of total ca acity4 each month% (ased on the abo#e information, and based on the demand scenarios $enerated by the simulator, the randomized model is sol#ed as described re#iously% )his model is cate$orized as a lar$e'scale inte$er ro$rammin$ model% )o sol#e this model a so histicated mi,ed inte$er lar$e'scale ro$rammin$ sol#er by )2M9A( is used 3Ed#all and GUoran, 566>4 +orkin$ under the Matlab en#ironment ?%; 3)he Math+orks, 566>4% )he reader may refer to 3Abdel Aziz et al% 56674 for more details on the im lementation% )he ro osed model is com ared to the a sim le Vfirst'come'first'ser#eW 3FCFS4 model for all three demand scenarios: +here in the FCFS the first re-uest to come is to be acce ted 3if thereTs a#ailable ca acity4 not considerin$ +hether or not it +ill ma,imize the total re#enue of the hotel% )able @ summarizes the re#enue $ained from usin$ either the FCFS model or the ro osed o timization model for the three in#esti$ated demand scenarios% )he last column in the )able calculates the increase in re#enue $ained +hen usin$ the su$$ested o timization model o#er sim ly acce tin$ all re-uest $i#en a#ailability of rooms% )able 5 Com arison bet+een results of FCFS X the ro osed model FCFS
Scenario @ S@5<,5<;,<66
P$o7o#!0o-!"
S@<7,7==,666
Scenario 5
S@5<,?<J,;66
S@<>,J56,666
14.638
Scenario <
S@5;,JJ<,J66
S@<J,<7=,=66
14.628
2b#iously, the modified randomized model has contributed in ma,imizin$ the total re#enue of the hotel under different demand scenarios +hen com ared to the sim le FCFS model% )he im ro#ement in re#enue is at least @6I as seen from )able @% !n fi$ure =3a4 the total number of the rooms allocated by the model #s% their e, ected demand alon$ the year is de icted% Another im ortant illustration, fi$ure = 3b4, is the total number of rooms occu ied in the hotel: +hich is e-ual to the allocated rooms by the model lus the ones already reser#ed before runnin$ the model% !t
can be seen from both fi$ures that the total room allocation ne#er e,ceeds the ca acity constraint and neither e,ceeds their e, ected demand% )here are other cate$ories of information that can be of interest to the re#enue mana$er of the hotel: for e,am le kno+in$ ho+ lon$ the maBority of the $uests like to s end in his hotel, or +hich rice class is more beneficial for him, or he may +ant to kno+ the distribution of the $uests amon$st the $rou sizes associated +ith the model% Fi$ure ; sho+s these distributions to a total number of 5@@7; of o timal reser#ations that are $enerated after sol#in$ the modified model% Fi$ure ;3a4 sho+s that most of the o timal reser#ations are of a len$th of stay of < days% Also most reser#ations are from the first rice class, +hich is the hi$hest class 3refer to Fi$ure ;3b44% Finally as obser#ed from fi$ure ;3c4, the maBority of the hotelTs $uest ha#e a $rou size of <%
3a4 3b4 Fi$ure =3a40 )otal number of rooms #s% total number of forecasted demand Fi$ure =3b40 )otal rooms 3allocated X reser#ed4
6 @ 5 < = ; > ?
6 @ 5 <
;66
!$o9#%&'( + 0 * N
5666
L!() , o9 # &:
P$'/! C"&##
G$o*7 S';!
3a4
3b4
3c4
Fi$ure ;0 Number of reser#ations accordin$ to 3a4 different len$ths of stay, 3b4 different rice classes, 3c4 different $rou sizes
Finally, as e, lained before, to com ute the o use the follo+in$ formula r R fbefore ' fafter
)o outline the o eration of the model, an e,am le is $i#en as follo+s% !t is re-uired to determine +hether to acce t or reBect a s ecific reser#ation on ni$ht @<J +ith len$th of stay e-uals ; and a $rou size of =% Hence the o ortunity cost is calculated for each of the three demand scenarios and the follo+in$ o ortunity cost is obtained Y7=6, @@76, and @666Z for the first, second and third demand scenario, res ecti#ely% )he a#era$e o ortunity cost of this re-uest +ould then be @66>%>> dollars% !n this case the 1M system +ould recommend to the mana$er to acce t this re-uest, if its re#enue is at least e-ual to @66>%>>% 5. SUMMARY AND FUTURE WORK )his a er resents a re#enue mana$ement model that is based on ad#anced forecastin$ and o timization% )he ro osed forecastin$ module attem ts to model the hotel reser#ation rocess from first rinci les% !n articular it models hotel arri#als as an interrelated rocess of stochastic arameters like reser#ations, cancellations, duration of stay, no sho+s, seasonality, trend, etc% !n our a roach +e attem ted to model all these rocesses as faithful as ossible and simulate for+ard in time the actual rocess of reser#ations to obtain the forecast% )he distinct ad#anta$es of our ro osed forecastin$ can be summarized in the follo+in$ oints0 First, it roduces the density of the forecasts, and hence also confidence inter#als% !t also allo+s for scenario analysis like for e,am le e,aminin$ the effect of o#erbookin$ on future arri#als% Finally, the sensiti#ity of the arri#al forecast due to chan$es of some control #ariables can be estimated% )his, of course is a #ery useful feature for decision su ort in re#enue mana$ement% Also it is +orth notin$ that the forecastin$ model deri#es the necessary demand that the o timization model needs directly from data +ithout the need of rior assum tions% !n the future the forecastin$ module +ill also incor orate other statistical and machine learnin$ forecastin$ methods 3Ahmed et al% 566J and Zakhary et al% 56674 usin$ both reser#ation data and historical arri#als% 2n the other hand, the ro osed o timization module e,tends e,istin$ o timization techni-ues for hotel re#enue mana$ement to address $rou reser#ations +hile includin$ inte$rality constraints and usin$ Fforecasted demandF arri#als $enerated from the data% )he o timization model is based on lar$e'scale inte$er ro$rammin$ model to o timize decision rules for acce tin$ reser#ations% 1esults sho+ that the model is able to $enerate effecti#e recommendations to ma,imize re#enue% For future research se#eral more issues need to be addressed% For e,am le, considerin$ cancellations or $uests no sho+s% 2#erbookin$ olicies com ensate for $uests that +ould cancel or Bust not sho+ u % !n $eneral, an o#erbookin$ limit should be determined in order to balance bet+een minimizin$ the loss of re#enue caused by em ty rooms, and the loss of customers8 satisfaction caused by not findin$ a#ailable rooms des ite reser#in$ beforehand 3 !$naccolo and !nturri, 56664% Also, this a er mainly focuses on the allocation of rooms to different se$ments and assumes fi,ed rices for each se$ment% Ne#ertheless, Fen$ and [iao ro osed a method +here the ricin$ and ca acity allocation decisions are carried out inde endently% So another ossible e,tension of this +ork is the inte$ration of ricin$ and ca acity allocation decisions% AKNOWLEDGEMENT= )his +ork is art of the V*ata Minin$ for !m ro#in$ )ourism 1e#enue in E$y tW research roBect +ithin the E$y tian *ata Minin$ and Com uter Modelin$ Center of E,cellence% /e +ould like to e, ress our thanks to Mr% Hossam Shehata for su ortin$ us +ith the necessary domain kno+led$e in the hotel industry% /e +ould also like to e, ress our dee $ratitude to *r Hanan .attara and FGlaza HotelF family, Ale,andria, E$y t for ro#idin$ us +ith the reser#ation data to test and de#elo our models%
REFERENCES
Abdel Aziz, H% X Saleh, M% XEl Gayar, N% X El 'Shishiny, H% 356674A 1andomized Model for Grou 1eser#ations in Hotel8s 1e#enue Mana$ement System% Groceedin$s of !NF2S85667, March 5?'5J, % 5='<6% Ahmed, N% X Atiya, A% X El Gayar, N% X El'Shishiny, H% 3566J4% An Em irical Com arison of Machine 9earnin$ Models for )ime Series Forecastin$F, acce ted for ublication, to a ear in Econometric 1e#ie+s% (aker, )%.% X Collier, *%A% 3566<4% )he benefits of o timizin$ rices to mana$e demand in hotel re#enue mana$ement systems% Groduction and 2 erations Mana$ement, Col% @5, % ;65D;@7%
(ertsimas, *% X de (oer, S% 3566;4 % Simulation'based bookin$ limits for airline re#enue Mana$ement% 2 erations 1esearch, Col% ;<, %J6D@6>% (rumelle, S% X /alczak, *% 3566<4% *ynamic airline re#enue mana$ement +ith multi le semi'Marko# demand% 2 erations 1esearch% Col% ;@, %@<?D@=7% Chian$, /%C X Chen, \%C%H% X [u, [% 3566?4% An o#er#ie+ of research on re#enue mana$ement0 current issues and future research% !nternational \ournal of 1e#enue Mana$ement% J?'@57 Choi, )%E% X Cho, C%, 356664% )o+ards a kno+led$e disco#ery frame+ork for yield mana$ement in Hon$ .on$ hotel industry% Hos itality Mana$ement, #ol% @J, % @?'<@% Ed#all, M%M% and GUoran, A%3566>4% )2M9A( ]"!C. S)A1) G"!*E% El Gayar , N% X Henda+i, A% M% X Zakhary, A% X El'Shishiny, H%356674% A Gro osed *ecision Su ort Model for Hotel 1oom 1e#enue Mana$ement% !CGS) !nternational \ournal on Artificial !ntelli$ence and Machine 9earnin$, !CGS)'A!M9% S ecial !ssue on Com utational Methods for the )ourism !ndustry% 5<'57% Fen$, E% X [iao, (% 3566>4% !nte$ration of ricin$ and ca acity allocation for erishable roducts, Euro ean \ournal of 2 erational 1esearch, Col% @>7, %@?D<=% Frechtlin$, *% 3566@4 Forecastin$ )ourism *emand0 Methods and Strate$ies% (utter+orth Heinemann, 2,ford% Goldman, G% X Frelin$, 1% X Gak, .% X Giersma, N% 356654 Models and techni-ues for hotel re#enue mana$ement usin$ a rollin$ horizon% \ournal of 1e#enue X Gricin$ Mana$ement, Col% @% 56?D5@J% !n$old, A and McMahon'(eattie, "% and Eeoman, !% 3566<4% Eield Mana$ement% 5 edn, Continuum, 9ondon 9ai, .'.% and N$, /'9% 3566;4% A stochastic a roach to hotel re#enue o timization% Com uters X 2 erations 1esearch, Col% <5, %@6;JD@6?5% Slyke, 1%C% X Eoun$, E% 356664% Finite Horizon Stochastic .na sacks +ith A lications to Eield Mana$ement% \ournal of 2 erations 1esearch% @;;'@?5 )alluri, .%)% and #an 1yzin, G%\% 3566;4% )he theory and ractice of re#enue mana$ement% S rin$er Science M (usiness Media, !nc% )he Math/orks 3566>4 D MA)9A( and Simulink for technical com utin$: +++%math+orks%com Cinod, (% 3566=4% "nlockin$ the #alue of re#enue mana$ement in the hotel industryT, \ournal of 1e#enue X Gricin$ Mana$ement, Col% <, %@?7D@J6% /eatherford, 9%1% and .imes, S%E% 3566<4% A com arison of forecastin$ methods for hotel re#enue mana$ement% !nt% \ournal of Forecastin$, Col% @J, %=6@D=@;% Zakhary, A% X Atyia A% X El Gayar, N% X El'Shishiny H% 3566?4% Forecastin$ Hotel *emand "sin$ Simulation% )echnical 1e ort% *ata Minin$ for !m ro#in$ )ourism 1e#enue in E$y t research roBect +ithin the E$y tian *ata Minin$ and Com uter Modelin$ Center of E,cellence% Zakhary, A% X El'Gayar, N X Atyia, A% 356674% A Com arati#e Study of )he Gicku Method and its Cariations "sin$ A Simulated 1eser#ation Hotel *ata% !CGS) !nternational \ournal on Artificial !ntelli$ence and Machine 9earnin$, !CGS)'A!M9, S ecial !ssue on Com utational Methods for the )ourism !ndustry% @;'5@% Zhan$, *% X Co er, /%9% 3566;4% 1e#enue mana$ement in arallel fli$hts +ith customer choice (eha#ior% 2 erations 1esearch, Col% ;<, %=@;D=<=%