Messages

1

General Chat / Load and profit: does the algorithm make any sense?

« on: February 27, 2008, 10:29:17 am »

That being said, pax could be willing to pay higher when their is less capacity on the route.

Nice theory, but it does not work, IMHO. I think it's a mistake to consider that passengers know the size of the planes, the number of competitors, the price of all seats etc. on a given route. They know how much a trip costs, in average, on this route, and they check if the proposed price is reasonable or not. Most of them don't care about the size of the plane, and at least in Europe, it is completely impossible to know if the plane is fully booked or not; this information is not available when you order a ticket, and even partially hidden afterwards (with some companies, you can chose your seat among a selection of seats, not all of them). For instance, it happened to me once to pay an expensive ticket, assuming that the plane was full, and we were 10 in a 150 seats aircraft. Too bad :-)

In any case, if your suggestion was true, real airlines would tend to decrease the frequency of flights, to fill them as much as possible with high ticket price. It's not the case, at least in Europe: you often end up in 50% loaded planes running 4 times a day. I'm really convinced that high frequencies brings you extra passengers, because all experienced travelers have already missed a plane, and know how painful it is to wait for 24 hours because your company runs only a daily trip. High frequency is a service, as much as free meals or good coffee.

Actually, the game correlates reasonably well in regards to which routes are profitable and how frequency plays in.

I can't agree, definitely. OK, I am not experienced in this game, and I wouldn't dare being too harsh with the developers, but I really feel that one should avoid the reaction: "I don't want to change anything, so I will find any reasons explaining why this bug is actually a feature of the game". I won't explain again the pax number bug, but my feeling is that the algorithm does not need fine tuning, it needs to be reconsidered from scratch.  Here is a list of unexpected behaviors:

* When you don't have any competitors on a route, the optimal strategy is to run only one flight with 100% load and high ticket price. This shouldn't be, in real world, you should have an optimal frequency, with an intermediate ticket price, and an intermediate load.
* When you are in a saturated market, the optimal strategy is to run low-frequency flights to hundred of destinations, and you earn more money in going to ridiculously small airports. This shouldn't be. In the real world, you should never make profit, even with small planes, between two very small airports: such flights simply do not exist (with a few exceptions, for instance if you are lucky enough to travel to a hub for a low-cost company), you always need a transfer through a bigger airport in this case.
* When you look at the schedule of large airports, you can see only small frequency flights run by the same company. This shouldn't be. Just go to any big airport, and you will see that most routes are 3, 4, 5, up to 10 frequency routes.
* On highly competitive routes, competition is so harsh that the price drops and the route is basically "screwed up". This shouldn't be. In the real world, companies are competing, but the average ticket price remains reasonable (keep in mind that the 1€ tickets are not average prices, most passengers on this routes pay much more than that, and in the few cases where a low-cost company tries to break the market, it is almost never on highly competitive routes: it is rather done to attract people to a small countryside airport).

I acknowledge that it is possible to play the game as it is today. It is probably even possible to have fun doing that. However, it is not a good simulation, and experienced players might tend to be conservative because they know how to play. For instance, there is a post asking "why are you stupid enough to run 2-frequency routes?" in this forum, and I think the right answer is "because most players still think this game is a simulation, and they behave as if they were managing a real airline in the real world".

2

General Chat / Load and profit: does the algorithm make any sense?

« on: February 26, 2008, 08:11:51 pm »

Hi all,

First, sorry if the topic has been already debated, I'm quite new to the game.

I played in two extreme situations: a more or less saturated market at the end of the last round, and a completely free market now. And I've been really surprised by the behavior of the load/profit algorithm, and its relationship with the ticket price. With the beginning of the new round, I could play with these parameters on a completely empty marker, which is better to understand the behavior of the simulation. And I found strange things, some of them can be discussed, but others look as real bugs.

Experiment done on the route LYS-CDG before any competition, my plane is the default BAE-146-100, 94 seats.

A) The function load = f(ticket price) is a negative exponential, truncated at 100% load. That means a sharp angle at the optimal price. The load obviously varies with the frequency of the flight. I guess it's really easy to reverse-engineer the exponential by a non-linear regression, I was too lazy to do that.

B) I then considered the function passengers=F(ticket price)... and big big surprise: it also depends on the frequency. For instance, when the ticket price is at 600€, the load is 82% (77 passengers) when the frequency is 1, and... 26% (49 passengers!!!!!) when the frequency is 2. In other words, if your flights are more frequent, less passengers will travel every day! In my opinion, this does not make sense at all, one would expect exactly the opposite (when flights are more frequent, it is easier to travel because you have less schedule constraints).

From there, everything converges to show why we have such an unrealistic bias towards low frequencies: in any case, whatever the route, whatever the competition, you'll achieve the optimal profit/flight by 1) putting the frequency as low as possible (direct consequence of point B) and 2) setting the price ticket at the maximum price leading to a load of 100% (consequence of the sharp angle described in A). All other fees (leasing price for planes, misc fees...) mainly scale with the number of flights, so they won't impact the profit/flight. The only exception is the gate rental price, which keeps constant whatever the frequency of your flights; however, this effect  is not strong enough to compensate the huge "frequency" effect on the number of passengers. There are some more complicated features (for instant, the load slightly influences the costs), but they are far too small to change the big picture.

I really think these weaknesses in the simulation algorithm directly lead to a game where the only optimal strategy is to run 0.5 or 1-frequency flights loaded at 100%, and to decrease the price in order to get these 100% in case of competition. The consequences are crystal clear: on competitive routes, the prices will drop down to a few euros; and all companies will run a lot of flights to a lot of destinations, instead of having high frequencies in important routes. Of course, some players will claim that they play in a different way, but actually they just play poorly, and they could earn much more money with the optimal strategy.

Of course, there can be long discussions claiming that it's not a bug, it's a feature, it regulates competition, blah blah blah. The fact is that it is completely anti-intuitive, and probably all beginners spend a lot of time before understanding that the game is strongly biased towards this strategy.

In my opinion, the game would be much more interesting if several strategies were viable, depending on the competition, on the route, on the size of the company, etc. I think the "passenger number bug" should be fixed, to get at least as many passengers a day with a fixed ticket price, whatever the frequency of the flights; and the load function should be smoother (a negative sigmoid for instance, with an asymptote at 100% --100% load is actually non-reachable. This would lead to very interesting strategies, in which the best profit might be achieved with intermediate loads and intermediate prices, etc.

Hope it helps.