Distributing Demand Accross a Hub and Spoke Network in a Multiplayer Airline Simulation

The complex nature of the airline industry presents several unique and difficult situations regarding the management, distribution, and prediction of revenues and costs, and the relationship between the two.  Much has been done in recent years to gain an understanding of how costs and revenues behave in such a special environment through the workings of yield management and other computer-driven algorithms.  While the industry struggles with the trial and error method of dealing with the problem, academia is beginning to acquire the technology to simulate the same situation.  These simulations, which are geared towards airline managers and students, are not new in theory, but have yet to benefit from the new techniques and technology available from the industry’s research are available to improve the fidelity of it’s own simulations.  In this writing, we will take a close look at Airline Empires, an online airline management simulation, its current simulation of an airlines revenues, the relationship of cost to the revenues, and the inaccuracies with the current system.  Much time will be spent exploring suggestions to improve the fidelity of this simulation by applying some of the knowledge gained by the industry in the area of revenue managements.

  Several new generation airline management simulations have been introduced to the market as massive multi-player online games (MMPOG).  Several examples on airline management MMPOG’s are AirlineSim.de, by a German company of the same name, Efzed’s Airline Online, and the new Airline Empires, created by the author of this paper.  The advantage of such a situation is that a very large number of players can be used to develop a virtual economy, which is what Airline Empires is based on.  While being available online attracts a large number of players into one simulation, computer capacity and performance limits the number of features and fidelity available to the MMPOG’s.  This loss in fidelity can be more than compensated by the virtual economy and macro competition that is possible.

Airline Empires is the first airline simulation which allows decisions made by one player to affect the opportunities to another.  Aircraft can be bought and sold between players, limited gate space requires deal-making between the players of the simulation, and most importantly, revenues are highly susceptible to the competition created by the other players.  This method of managing the revenues awarded to the players is a first of its kind.  Until Airline Empires, the ticket price, frequency, aircraft, or facilities of another player’s route had no bearing on the revenues generated by another.  The current revenue algorithm within Airline Empires is highly sensitive to both the integral market conditions of a city pair, and the competition present in this market. 

A unique demand equation was developed expressly for Airline Empires, which is used to calculate load factors.  Several factors are taken into account, including city demand, frequencies offered, amenities offered, possible connecting passengers, competing flights, and competing fare’s.  In a no competition situation between two cities in Airline Empires, a weighted-average of the two cities actual yearly passenger enplanements as reported by the Bureau of Transportation Statistics in table T100 is used to generate the initial demand between the cities.  Distance is factored into the equation, although at a much diluted effect.  The frequency offered by a player is then used to simulate the “S-Curve” theory present in airline bookings, which offers that as daily frequencies are offered between two cities pairs increases, the number of bookings will increase in an “S” shaped pattern until roughly 10 flights have been added, where the additional bookings drops off for every flight added (Peters). 

Figure 1 - Example load factor distribution for a typical route

This is used to develop a demand equation with the y axis as demand (number of bookings) and the x-axis as the fare.  The fare is then multiplied by the load factor and capacity of the aircraft to generate the total revenues from the flight.  As the fare increases, the number of bookings drops off until there is nobody left.  When the player changes the fare, a different x value is simply calculated, however when competition is introduced, the entire equation must be altered.

            With the introduction of competition into the Airline Empires demand equation, the demand curve shifts noticeably.   As shown in Figure 2, the entire graph shifts to the left since there is more capacity available at the same demand.  This results in a much lower fare required to achieve profitable load factors;  The same drop in yield the major airlines are complaining of today.  The specifics of how much the demand equation is moved, and all factors included is proprietary, of course, but it should be easy to see how competition is simulated in the revenue environment.

Figure 2 - Example load factor distribution for a route with competition

            While this system was revolutionary to online airline management simulations, it does have some major drawbacks.  Firstly, by taking an average of the cities yearly enplanements to calculate demand, accurate demand between airports is impossible.  Most specifically, a player inaugurating a flight between two extremely close, yet large cities, will develop a respectable profit.  For example, New York’s LaGuardia and JFK airports will have a very high load factor since both of these airports generate a large number of enplanements, but very few, if any, passengers are actually flying between these two airports.  The same situation results when operating a flight between to distant cities.  Hong Kong generates a very large number of enplanements every year, and South Bend, IN develops enough to be included in the game, however there are actually very few passengers who fly that route.  The SBN-HKG route in the current Airline Empires system would have a comparable number of passengers as the SBN-ATL route, which in reality, has many more enplanements.

            Another problem with the current demand equation is the handling of connecting passengers.  Currently, when a hub is designated by a player, all routes through that city are affected by a percentage of passengers entering the city.  For each passenger that flies into or out of the city, they are added to the total city value for that airline at half of the fare.  For instance, if you have two full 100-seat routes between your hub in Atlanta and Birmingham and Charlotte, the passenger base for each city would increase by 100 passengers (100 for each flight divided by two).  By adding the total number of passengers who fly into the city to the cities yearly enplanements, an increase in load factor will result, as is found in the industry.  Since each connecting passenger takes two flights for every one ticket they purchased, their ticket revenue is divided by two.  While this does add incentive for players to develop a hub-and-spoke system, it falls far short of simulating the true revenues generated in an accurate connecting environment.  It is from this inaccuracy that the next generation of revenue management simulations will be derived.

            The premise behind most airline revenue management systems is to maximize revenue by catering to the high yield business passenger, yet minimizing unused capacity by offering otherwise unsold seats at a discount.  Revenue management systems are among the most complex computer systems in the world, and attempting to simulate both the market and the reaction of one of these systems is a comparably complex task.  It is no wonder then that an accurate simulation of actual airline revenue generation does not currently exist. 

            In most airline simulations, and in a majority of the airlines themselves, it is attempted to compare profitability of a flight by comparing the revenue from the tickets sold to the operating and fixed costs of the operating flight.  When operating in a solely origin and destination (O&D) market, this is very simple since all revenues are associated with that one flight.  Where the equation becomes complex is when you factor in connecting passengers.  The problem is how do you prorate the ticket to find what percentage of the revenues are designated to which flights (Baldanza, 65) .  Simply dividing the revenue by the number of flights is not enough since a fare of $1,000 to fly from Louisville to London, England can hardly be divided into two $500 segments.  Likewise a distance proportion is not entirely accurate either since several ticket prices are set according to O&D markets, and while distance is a factor, it is not the most driving factor in determining the fare.  One solution is to separate costs and revenues by reporting costs per flight, and revenues per city pair.  While this is a more accurate way to determine total profitability of an airline, it makes it difficult to view the profitability of a single flight.  Fortunately, this is not as central to a simulation since we do not necessarily need to separate revenues per flight (Baseler, 80).

            The first step in achieving this new, more accurate revenue generating system, is to  find actual (or at least representative) O&D data.  Since the only reliable O&D data is available from the BTS, only U.S. cities will be reported.  With the O&D data, we no longer need to find a weighted average of passenger enplanements at both cities, rather we now have a representation of the total number of passengers traveling on that distinct market. 

            Since the O&D data does not discriminate regarding time of day, it is necessary to divide the passengers according to the time the flight operates.  An accurate representation of this was achieved by designating a percentage of the daily O&D passengers grouped by timeframe.

Table 1 – O&D distribution throughout the day

            Another factor to the accurate assignment of revenues is to adjust for the frequencies offered.  For this a deviation percentage from the original demand is used.

Table 2 – Demand deviation due to frequency

This takes into account the passengers preference for a flight convenient to the time they want to depart.  The higher the frequencies, the more likely you will offer a convenient flight to the passenger, and more passengers will choose to fly your route.  This is the driving force behind the “S-curve,” and its requirements are satisfied through the above deviations. 

            Another factor to include when determining how many passengers will fly on a particular flight is whether or not the flight is a non-stop flight.  Since passengers tend to prefer non-stop over connecting flights, a deviation cumulative to the frequency deviation can be developed.

Table 3 – Demand deviation due to number of connections

This accommodates the notion that a passenger will choose an available non-stop with all other factors equal at least 40% of the time over a two connection trip.  Other factors that can be taken into consideration are the passengers preference for jets over turboprops, amenities, airport lounges, etc.  For the sake of simplicity in the following example, we will only use the two deviations.

            Airline Empires is written in the PHP programming language and currently uses the open-source database program MySQL to store the large amounts of information.  Both of these development tools are known for their speed and performance, which is critical when choosing a programming language and database program for a complex program, such as this.  In order to store the information into the MySQL program, a simple database schema needs to be set up.  While this example will use MySQL as the database, the schema is easily transferred to any database program available.

            In order to record the revenues for each passenger group, we will need four tables:  one to store the aircraft information, such as speed, seats, and cost, which we will call the “aircraft_static” table, since it will not need to be changed once it is created, one to store the individual aircraft information such as owner and age called “aircraft”, one to store the individual flight information such as the city pairs, time, and cost called “flights”, and one to store the passenger revenue data which we will call “passengers.”  The important thing to notice is that all costs are calculated and stored separately from the revenues.  This satisfies the earlier criteria of accurately reporting revenues.

            To better illustrate how the system will work, an example is necessary.  First we will determine the passenger loads for a non-stop and connecting flight with no competition.  Airline ‘X’ operates out of four cities, Chicago Midway (MDW), St. Louis (STL), Kansas City (MCI), and Oklahoma City (OKC) with a hub in STL.  There are three flights we will use for this example, and the information entered into the “flights” table of the database will resemble the following:

Table 4 – “Flights” table

Aircraft #1 is a 100-seat aircraft, while aircraft #2 has 50 seats.  The O&D demand between the individual markets will be:

Table 5 – Hypothetical O&D demand and average fares derived from BTS database

The first task of the revenue program is to find the O&D passengers traveling non-stop on each flight.  For flight #1 (MDW-STL) we will need to take the daily O&D passengers between the two cities, and adjust them for time of day, frequency, and connections which we designated in tables 1, 2, and 3.  By interpolating between the departure (10%) and arrival (14%) percentages, we can find a percentage of the daily O&D passengers willing to fly on this flight at 12% of the original 500 daily passengers, or 60 passengers.  Since there is only one frequency available during the day, we then reduce the 60 passengers by 15% which leaves us with 51 passengers.  Next, by adjusting for the nonstop reward of + 20%, we have a final count of 61 passengers who will board flight #1 in MDW bound for STL.  Since there are 100 seats available, we can carry all 61 passengers.  We repeat the same formula through all non-stop flights, careful to ensure that no more than the capacity of the aircraft are allowed.  This results in the following:

Table 6 – “Passengers” table

             Once all O&D passengers have been placed, we need to calculate remaining capacity for connecting passengers.  Once again, we will use flight #1 to take a close look at the details of how to establish how many passengers will choose to connect through STL on the MDW-MCI and MDW-OKC markets.

            After we have established that there are seats remaining on the flight after all O&D passengers have been seated (39 open seats in this case), we then ask the database to retrieve all flights from which a passenger on flight #1 can connect.  For this example, the criteria for a connecting flight is one that departs between 30 minutes and 2.5 hours from the arrival time of the first flight.  The two flights that are returned are flight #2 (STL-MCI) and flight #3 (STL-OKC).  Each connecting market is then ranked by its average fare, since the airline will always prefer to seat the higher paying customer.  In this instance, there are only two connecting flights available, flights #2 and #3, and the MDW-MCI route is the highest fare at $140 so those passengers will be accomodated first.  By taking the daily demand derived from Table 5, and adjusting for time of day, frequency, and the connection, the total number of passengers who will fly between MDW and MCI is 31.29 passengers.  Unfortunately, there are only 29.79 seats available flight #2 from STL-MCI, so we are limited to that number of passengers.  Therefore:

Table 7 – “Passengers” table

By using this same methodology for the MDW-OKC market, we see that we can only fit another 1.2 passengers on Flight #3 due to the 50-seat capacity and 48.8 O&D passengers.  This results in the following flight loads:

Table 8 – Flight Loads

Revenue can then be easily calculated by multiplying the average fare by the average number of passengers for each market.

            While this process of distributing demand is quite revolutionary to airline simulations, it does have some serious drawbacks.  First, this system assumes that you know exactly how many passengers from each market will be willing to fly on your flight.  This allows you to reserve only as many seats as you can fill, and does not take into account no-show passengers, demand spill, or demand variations due to seasonal fluctuations.

Secondly, the storage capacity required to handle such large files is immense.  The O&D database available from the BTS itself is over 1.2 gigabytes, and the database required to store all market combinations could easily become as large.  As well as capacity, server speed will be a large hurdle since these calculations, while small, will place a large strain on the server when repeated thousands of times.

            With the generation of the hub and spoke system in the airline industry, it has become infinitely more difficult to distribute revenues over connecting flights.  While airlines struggle with this problem, it is entirely possible for airline management simulations to use current technology to provide an accurate representation of what takes place.  By using the preceding formulas to distribute demand and revenues to aircraft within the simulation, more accurate airline management simulations can be created to better train future managers, and to provide insight to the behaviors of such a market.  Airline Empires, while already revolutionary with its competition-based demand equations, will have to develop this style of demand distribution to remain competitive in the coming years. 

REFERENCES

Baldanza, Ben (2002). Measuring Airline Profitability. New York: McGraw-Hill: Retrieved from Handbook of Airline Economics pp 61-73.

Baseler, Randy (2002). Airline Fleet Revenue Management – Design and Implementation. New York: McGraw-Hill: Retrieved from Handbook of Airline Economics pp 77-106.

Belobaba, Dr. Peter P. (2002) Airline Network Revenue Management: Recent Developments and State of the Practice. New York: McGraw-Hill: Retrieved from Handbook of Airline Economics pp 141-156

Bureau of Transportation Statistics. DB1BMarket database.  Retrieved December 2, 2004 from http://www.bts.gov.

Caves, R. E. (1995). European airports and airline network strategies - their mutual relationship. United Kingdom: Loughborough University of Technology: Loughborough. Retrieved December 2, 2004, from Aerospace & High Technology Database database.

PETERS, H. J. (1975). Contribution to routing aircraft and to the economy of air transportation (11 Aircraft (MT); 3 Air Transportation & Safety (AH) No. ESA-TT-222; DLR-FB-74-25) December 2, 2004, from Aerospace & High Technology Database database.

PETERS, H. J. (1974). Routing of-aircraft and economics of flight operations [ph.D. thesis - tech. univ. stuttgart].DLR-FB-74-25Retrieved December 2, 2004, from Aerospace & High Technology Database database.