News:

Use the "Forum Search"
It may help you to find anything in the forum ;).

Reconsidering the speed bonus

Started by jamespetts, December 28, 2012, 01:35:31 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

jamespetts

Some time ago, I sought to check whether the speed bonus system in Simutrans was realistic or not, and undertook a quick check based on figures from a book on transport economics. That check appeared to show that, with the default speed bonus of 18%, the formula in Simutrans was indeed about right.

However, Asaph, in his work to balance Pak64.Experimental, has looked into this issue in some more detail, and shown far greater potential fluctuations in the income that can be generated by various different speed bonus settings than appear realistic (see here for an explanation of how the current speed bonus code works).

The current basic speed bonus code has not been changed from Standard, save that the speed bonus rating (the percentage figure shown next to the item in the list of goods) now fluctuates depending on the distance travelled. However, in Standard, the speed bonus is based on the theoretical maximum speed of the convoy, whereas, in Experimental, it is based on the actual average point to point speed.

I should be interested in any views on how (and, indeed, whether) to replace the existing Standard-based speed bonus formula with something more realistic. My aim is to balance the various income related matters of the game before starting to work on cost balancing (the latter of which will require some quite extensive changes to the code as well as heavy work in the calibration itself). This is why physics (which greatly affect how much revenue that any given combination of vehicles can earn), timing calibrations, and realistic generation of passengers/mail have been and must be (as the case may be) done before the work on cost balancing can start.

Some things to consider are listed below.


  • Should the speed bonus use the existing formula but be capped? This has the disadvantage that it might climb quickly and then level off, giving inadequate incentive to differentiate speed above the cap, although it might be that a better taper can be used by reducing the speed bonus rating: this will need to be investigated
  • Alternatively should we consider a linear formula similar to what we already use for, for example, the comfort bonus, whereby we define a minimum and maximum amount of bonus/penalty (perhaps the percentage of the rating itself can entail + or - the given percentage at most, so that goods/passengers with an 18% speed bonus rating can only receive, at maximum, + or - 18% of the total fare on account of a speed bonus or penalty)
  • We need to consider how, if we change the formula substantially, paksets designed for Standard will cope with the change
  • Do we need to show the different speed bonus rating at different distances in the list of goods window, for example by having a new "effective speed bonus rating" column?
  • Do we need to change how we calculate the "base" speed bonus? Currently it is determined by the date and values provided in speedbonus.tab - but other possibilities include comparison to an overall average speed, comparison to an overall average speed for the particular type of goods and/or waytype (air, rail, narrowgauge, water, etc.), comparison to an overall average speed from goods/passengers transported to/from a particular town/factory, comparison to an overall average speed from goods/passengers transported within a particular distance range, or something else entirely? It ought be borne in mind that the more complicated the mechanism, the harder that it is to code, and the more memory that has to be consumed; and the more criteria that are added, the more likely it is that any given speed will be compared only with itself, rather vitiating the point of having a bonus in the first place. An average over a number of years might help with this, but this is then prone to changes in technology giving everyone an excess of profit for a while as the speed bonus waits to catch up

All potentially useful thoughts welcome.
Download Simutrans-Extended.

Want to help with development? See here for things to do for coding, and here for information on how to make graphics/objects.

Follow Simutrans-Extended on Facebook.

asaphxiix

The way I look at it, the first question is whether, and how much speed bonus is needed in theory. The second question, how to implement it, and on what basis.

From some research I've done, I've found no clear indication that the sheer speed of line makes a substantial consideration in its price. Trains are more expensive than buses, yes, but also allow for a much more comfortable and reliable journey. Planes, on the other hand, are generally not more expensive than trains in Europe (probably less, from what I've gathered). Time is only one of many factors influencing travel price, and cannot easily be isolated to determine a factor, even when combined with travel distance.

So a passenger going from London to Edinburgh will pay pretty much the same price for a train and a plane ticket, while the plane goes 5 times faster than the train. The airports are not in the cities and you have to get there early for the check in and security checks, so in the end the whole journey between city centers will take about 4 hours on the plane and 5 hours on the train. You also need transportation to the airports, add another few pounds each end, make the plane more expensive for the whole trip in about the same proportion as it is faster, if no account is made for the reduced comfort of the flight and intra-city travel.

So there are two problems here: one, is that IRL a line's speed is not reflected in pricing, and even simple time value for whole journeys plays only a very limited factor in pax's willingness to pay a higher price.

From a very comprehensive study about time value in transportation, which estimates total travel costs (including even the hours worked to pay for it, as well as the effects of comfort, reliability, distance etc.).

"The true value of changes in travel speed can be difficult to determine because people
tend to have fixed travel time budgets, typically devoting about 70 daily minutes or 8
weekly hours to personal travel. As a result, increased travel speeds tend to increased
travel rather than save time.22 For example, if a highway or transit improvement increases
travel speeds, commuters often accept longer distance commutes. As a result, the true
benefits are increased mobility and improved location options, not travel time savings.23"


I think this states just why journey time tolerance is a better way of rewarding faster routes in the game. it does not affect revenue per passenger, but rather the number of passengers. Also, it's based on time/distance between source and destination, and disregards routing altogether.

Even if we ignore the portion of the journey problem (revenue per trip rather than whole journey), if we look at the game, giving a better price for a faster car is trying to isolate the effect of time alone on public transportation price, which is not a realistic cause, I believe. Comfort rating is a good step in this direction, but there are many other factors in pricing.

If we do want to isolate speed as a pricing factor and determine its rating, we may have better luck finding the answer in toll roads rather than in public transportation:

In another city in my corrupt country, they forfeited a lane on the airport highway (busiest place in the country in peak hours) to a private investor. The 13km drive is charged according to the traffic state on the regular lanes, and ranges between 7-75 NIS, with the average fee for a drive during peak hours 25 NIS.
The route from the airport to Tel Aviv by the regular lane (40 minutes) is 4 times longer in a given peak hour (or: fast lane, 10 minutes is 75% shorter in duration), and adds 128% (1.92 NIS/km) to the total car cost (maintenance, fuel, equity, insurance etc - 1.5 NIS/km), of those 13km alone in an average economic car. So from the base travel cost, we would be adding 1.70% to the travel cost for every percent of the long journey time that was saved, and for a 50% time save (twice as fast), the speed bonus would be 85.33% of base cost, or 140% of the median hourly wage rate in Israel (35 NIS/2 for half an hour saved). So far the road seems to be meeting usage expectations.

The Carmel Tunnels in Haifa are quite a bit cheaper - for saving 22 minutes (30 -> 8 minutes) in rush hours and 6km of the drive distance, they charge 14 NIS, isolating the speed here makes for 5 NIS for the 22 minutes, a much better deal the fast lane. However this example is not as "pure" as the first.

In summary
, I an not sure that Simutrans Ex. needs speed bonus at all, but rather that time tolerance and competition are sufficient for this purpose, given that other pricing factors are not determined, and more importantly, that price is not a factor in people's decision whether to go. If used, I think speed bonus (and possibly, or if necessary, the whole pricing) should be calculated for the total time saving on a whole journey rather then a trip leg, and include waiting times as well, (like I've said before, it's important that waiting times are more accurately and dynamically calculated than they seem to be currently, if possible. This was discussed in another thread); it should use a rather linear, date based formula for transparency, and reach a maximum of less than 100% of the base rev for a 100% speed bonus (twice as fast as the base).


Quote from: jamespetts on December 28, 2012, 01:35:31 AM
goods/passengers with an 18% speed bonus rating can only receive, at maximum, + or - 18% of the total fare on account of a speed bonus or penalty)
this is what I had in mind when I heard the term 'speed bonus'.

I realize that this may require changing the pricing module to journey based, and that this is not in agreement with the current game design, but it may be worth sacrificing the pricing method a bit here for the realism of the journey time bonus. In any case, the only reason pax have to stray from the most direct route is time, so they would never choose to from London to Oxford via Birmingham, unless they have to take a horse carriage on the direct route... I think for the purpose of the game, this would be rather realistic.


jamespetts

I am not sure that you are correct that the journey time does not influence prices very much: what of the example of Concorde, whose only advantage was speed, and which was able to command vastly higher ticket prices on account of it? Short haul air against rail transport in Europe is probably not a good comparison, since short haul air transport faces much stiffer competition for the same routes than the train does, especially where state regulation distorts the rail market more than the air market.

We can test the price of higher and lower speed trains against each other in the UK, as there are some routes where the same journey can be made on different trains at different speeds at different costs. Take, for example, London to Ashford in Kent. If you put in a journey on the National Rail Enquiries website from "London" to "Ashford International", you will see different sets of journeys: some starting in London Victoria and Charing Cross, and some starting in St. Pancras International. The former cost £26.20 off peak, wheras the others cost £31.30 off peak. The former journey is about one and a half hours and the latter about 40 minutes using the new high speed rail link built for the Channel Tunnel. It really would not be realistic not to have price determined by journey speed at all, I don't think.

As to this:

Quote
I realize that this may require changing the pricing module to journey based, and that this is not in agreement with the current game design, but it may be worth sacrificing the pricing method a bit here for the realism of the journey time bonus. In any case, the only reason pax have to stray from the most direct route is time, so they would never choose to from London to Oxford via Birmingham, unless they have to take a horse carriage on the direct route... I think for the purpose of the game, this would be rather realistic.

I am not sure exactly what you mean. What do you imagine by a "journey time bonus"? In real life, people do not generally have through ticketing for their entire trips - the local 'bus price will not depend even slightly on how fast that one's air or train journey to the airport/main railway station in a town was. Through ticketing to the London Underground from national railway stations is the exception rather than the rule, and has only existed since the 1980s, over 30 years since the railways were first nationalised. It is far more realistic to have a price per journey leg than per end to end journey. If you meant something else, I am afraid that I am not sure what it was.

The real question to answer, however, is not, I think, whether to have a speed bonus, but what it should be. Concorde suggests a potentially very high difference in price, whereas the London to Ashford link suggests a lower difference; that can be accounted for by the difference in the overall journey time. It might be that simply modifying the speed bonus ratings to reach higher ratings at only very long journey times and having lower ratings overall, more realistic figures can be achieved. The question then becomes: do we need a cap on top of that, and, if so, what should it be? Does a maximum multiplier/divider of four seem sensible?
Download Simutrans-Extended.

Want to help with development? See here for things to do for coding, and here for information on how to make graphics/objects.

Follow Simutrans-Extended on Facebook.

asaphxiix

This is quite a complicated discussion.

Quote from: jamespetts on January 03, 2013, 12:34:26 PM
I am not sure that you are correct that the journey time does not influence prices very much: what of the example of Concorde, whose only advantage was speed, and which was able to command vastly higher ticket prices on account of it?
The concord is one example, a good one because it features the difference between the same mode and locations, with only time difference, but very extreme as well, I think. We should remember however that the Concorde was a. a very luxurious and unique product; b. tickets prices were far from constant and as airliners usually do, prices were often dropped to fill the plane. Fares alone are not the best determiner of speed factor, I think. There are so many other substantial considerations.

Quote from: jamespetts on January 03, 2013, 12:34:26 PM
Short haul air against rail transport in Europe is probably not a good comparison, since short haul air transport faces much stiffer competition for the same routes than the train does, especially where state regulation distorts the rail market more than the air market.
True, but it does match the game's current scale and function, pretty much. And state regulation is a part of reality, even if not simulated in the game.

Quote from: jamespetts on January 03, 2013, 12:34:26 PM
We can test the price of higher and lower speed trains against each other in the UK, as there are some routes where the same journey can be made on different trains at different speeds at different costs. Take, for example, London to Ashford in Kent. If you put in a journey on the National Rail Enquiries website from "London" to "Ashford International", you will see different sets of journeys: some starting in London Victoria and Charing Cross, and some starting in St. Pancras International. The former cost £26.20 off peak, wheras the others cost £31.30 off peak. The former journey is about one and a half hours and the latter about 40 minutes using the new high speed rail link built for the Channel Tunnel. It really would not be realistic not to have price determined by journey speed at all, I don't think.
In contrast, from London to Peterborough, http://ojp.nationalrail.co.uk/service/timesandfares/London/PBO/240113/1215/dep, a 45 minute journey costs less than a 75 minute journey, not sure if the route is the same, but the slow route through Sandy seems pretty direct. From scouting for different tickets through the Kingdom, it really doesn't seem like the time duration (or number of changes) has any effect on the price, while timing (peak/off peak) and booking option (advance/refundable), two elements that are absent from the game, are much more prominent factors.

Quote from: jamespetts on January 03, 2013, 12:34:26 PM
I am not sure exactly what you mean. What do you imagine by a "journey time bonus"? In real life, people do not generally have through ticketing for their entire trips - the local 'bus price will not depend even slightly on how fast that one's air or train journey to the airport/main railway station in a town was.

Through ticketing to the London Underground from national railway stations is the exception rather than the rule, and has only existed since the 1980s, over 30 years since the railways were first nationalised. It is far more realistic to have a price per journey leg than per end to end journey. If you meant something else, I am afraid that I am not sure what it was.

When I buy a ticket in this website above, the ticket may include changing. This, in game terms, means a journey with two (or more) trips, doesn't it? Same goes for regional bus/rail passes (for short and medium distance), and airline tickets, where I pay for a p2p journey regardless of stops in the middle, which are much more meaningful in flying, and take hours, as well as the increase in flight duration. For this (London-TLV through Frankfurt vs. direct route), the fare will be no less and no more, although of course my inclination to purchase a slow ticket for a large price may be decreased, and I'll most likely look for other, faster flights (time tolerance). This again returns to the question of whether the game is simulating fares as they are determined, or rather the economic process of revenue generation (what people actually pay). Either way, I don't think through tickets are unrealistic for the game's simulation, even with multiple modes.


Quote from: jamespetts on January 03, 2013, 12:34:26 PM
The real question to answer, however, is not, I think, whether to have a speed bonus, but what it should be. Concorde suggests a potentially very high difference in price, whereas the London to Ashford link suggests a lower difference; that can be accounted for by the difference in the overall journey time. It might be that simply modifying the speed bonus ratings to reach higher ratings at only very long journey times and having lower ratings overall, more realistic figures can be achieved. The question then becomes: do we need a cap on top of that, and, if so, what should it be? Does a maximum multiplier/divider of four seem sensible?

I still think speed bonus is fun; but I think its effect should not be great.

jamespetts

Quote from: asaphxiix on January 03, 2013, 01:24:53 PM
This is quite a complicated discussion.
The concord is one example, a good one because it features the difference between the same mode and locations, with only time difference, but very extreme as well, I think. We should remember however that the Concorde was a. a very luxurious and unique product; b. tickets prices were far from constant and as airliners usually do, prices were often dropped to fill the plane. Fares alone are not the best determiner of speed factor, I think. There are so many other substantial considerations.

I don't think that, aside from the speed itself, Concorde was any more luxurious than first class cabins on subsonic airliners, nor does there seem to be any reason to think that prices on Concorde were dropped more regularly or to a greater extent than on other airliners. It might be an extreme example: but we actually have Concorde in the game thanks to Milko/Giuseppe, so we need to be able to have a system that accounts for that sort of extremity.

QuoteTrue, but it does match the game's current scale and function, pretty much. And state regulation is a part of reality, even if not simulated in the game.

The difficulty with this is that we introduce arbitrary distortions if we use regulation affected data without taking into account or simulating the effect of regulation. We need in principle to compare like with like.

QuoteIn contrast, from London to Peterborough, http://ojp.nationalrail.co.uk/service/timesandfares/London/PBO/240113/1215/dep, a 45 minute journey costs less than a 75 minute journey, not sure if the route is the same, but the slow route through Sandy seems pretty direct. From scouting for different tickets through the Kingdom, it really doesn't seem like the time duration (or number of changes) has any effect on the price, while timing (peak/off peak) and booking option (advance/refundable), two elements that are absent from the game, are much more prominent factors.

The difference between those tickets is that the cheaper ticket can be booked in advance (which is less expensive), whereas the longer route is not available for advance booking. I chose the high speed rail as a specific example of where there is direct competition between different speeds of transport. Another good example would be to compare fares on the London Midland slow trains from Euston to Birmingham with the Virgin West Coast fares on that same route. Where the comparison is between local stopping trains and long distance trains run by the same operator on the same route, the comparison of price is not a useful one, as the two trains have a different purpose, and are not really in competition: if the prices on the stopping train were reduced, some passengers for the long distance run would take the stopping train, possibly making it too crowded for the passengers who actually need to use the stopping train, so there is no reason to send a price signal in that case. In Simutrans, this does not much matter, since passengers will take the fast train in any event.

QuoteWhen I buy a ticket in this website above, the ticket may include changing. This, in game terms, means a journey with two (or more) trips, doesn't it? Same goes for regional bus/rail passes (for short and medium distance), and airline tickets, where I pay for a p2p journey regardless of stops in the middle, which are much more meaningful in flying, and take hours, as well as the increase in flight duration. For this (London-TLV through Frankfurt vs. direct route), the fare will be no less and no more, although of course my inclination to purchase a slow ticket for a large price may be decreased, and I'll most likely look for other, faster flights (time tolerance). This again returns to the question of whether the game is simulating fares as they are determined, or rather the economic process of revenue generation (what people actually pay). Either way, I don't think through tickets are unrealistic for the game's simulation, even with multiple modes.

There is a great difference between, on the one hand, changing trains on the same railway company's route and having a through ticket, and, on the other, arriving by air and being able to use one's airline ticket to catch a local 'bus. The difficulty is that it is not easy in Simutrans to draw the line between the two, and it is more realistic in any event to compute fares by journey leg than otherwise (after all, why should an airline passenger pay more or less to use the local 'bus than a local commuter?).

QuoteI still think speed bonus is fun; but I think its effect should not be great.

The real question is: exactly how much is "not great"? How do we calibrate it?
Download Simutrans-Extended.

Want to help with development? See here for things to do for coding, and here for information on how to make graphics/objects.

Follow Simutrans-Extended on Facebook.

asaphxiix

#5
Quote from: jamespetts on January 03, 2013, 01:42:04 PM
I don't think that, aside from the speed itself, Concorde was any more luxurious than first class cabins on subsonic airliners, nor does there seem to be any reason to think that prices on Concorde were dropped more regularly or to a greater extent than on other airliners. It might be an extreme example: but we actually have Concorde in the game thanks to Milko/Giuseppe, so we need to be able to have a system that accounts for that sort of extremity.

**
Quote
On Concorde, a return flight across the Atlantic could easily cost £8,000, compared with £3,000 or so in vastly more comfortable first class on a Jumbo Jet.

When Concorde returned to service in 2001 after its Paris crash, it simply failed to convince enough people to fly.
(http://news.bbc.co.uk/2/hi/business/2935337.stm)

I think all in all, Concorde was a very high-end luxurious brand name, like a Bentz or Apple products, you're not only paying for the actual product. I'm no economist, but I think there are a number of factors that disqualify this example, also the fact that it had no competition within its supersonic category, which of course set its prices up.

Quote from: jamespetts on January 03, 2013, 01:42:04 PM
The difficulty with this is that we introduce arbitrary distortions if we use regulation affected data without taking into account or simulating the effect of regulation. We need in principle to compare like with like.
This is true; which is why I say that since there are so many price-determining factors that cannot be simulated in the game, treating speed should be done with much caution.

Quote from: jamespetts on January 03, 2013, 01:42:04 PM
The difference between those tickets is that the cheaper ticket can be booked in advance (which is less expensive), whereas the longer route is not available for advance booking. I chose the high speed rail as a specific example of where there is direct competition between different speeds of transport. Another good example would be to compare fares on the London Midland slow trains from Euston to Birmingham with the Virgin West Coast fares on that same route. Where the comparison is between local stopping trains and long distance trains run by the same operator on the same route, the comparison of price is not a useful one, as the two trains have a different purpose, and are not really in competition: if the prices on the stopping train were reduced, some passengers for the long distance run would take the stopping train, possibly making it too crowded for the passengers who actually need to use the stopping train, so there is no reason to send a price signal in that case. In Simutrans, this does not much matter, since passengers will take the fast train in any event.
this is actually what I was trying to say. How did we end up agreeing here? The situation in life where one has a large variation in alternatives that differentiate in duration and price alone is quite rare, I think. The Oxford example is a good one, I suppose, but even there you can see that for a substantial difference in duration, the price difference is quite meager, much less than the fast lane example from earlier.

Quote from: jamespetts on January 03, 2013, 01:42:04 PM
There is a great difference between, on the one hand, changing trains on the same railway company's route and having a through ticket, and, on the other, arriving by air and being able to use one's airline ticket to catch a local 'bus. The difficulty is that it is not easy in Simutrans to draw the line between the two, and it is more realistic in any event to compute fares by journey leg than otherwise (after all, why should an airline passenger pay more or less to use the local 'bus than a local commuter?).

Theoretically, such a journey price calculation could be limited to mode (not sure what the benefit of that would be though). But still, the game is not a copy of reality, it has limitations, and I think journey price calculation could be a good way to mitigate those somewhat. As for realism, when two persons who bought through tickets or area passes ride the same vehicle, it may well be that they paid very different rates (if divided by their total journey rate) .

jamespetts

We seem to agree in principle, I think, that speed can and does affect price, but perhaps rather less than the current code and pakset configurations produce in practice. What we need to know is, in reality, how much that speed affects price so that we can adjust the paksets and code accordingly. The only way of doing this, I think, is to look at and aggregate real life examples.

One thing mentioned above in connexion with Concorde is the question of competition. The current speed bonus works by assuming a given level of speed by a sort of generic competition at any given time. One issue to consider is whether there is any practical way of modelling actual competition in the game for these purposes. This will be hard to do unless we can think of an effective simplification that does not introduce distortions - itself not an easy task.

Aside from the question of competition, we have a few real world examples: the Oxford train and 'bus comparison, the Ashford International high speed versus ordinary rail comparison and the Concorde comparison. Can we find more? Do they all fit into a normalised range or band from which we can make generalisations so as to produce a workable model?
Download Simutrans-Extended.

Want to help with development? See here for things to do for coding, and here for information on how to make graphics/objects.

Follow Simutrans-Extended on Facebook.

asaphxiix

I think the influence of many other factors in pricing will be a great obstruction to such generalisations. The fast lane example is better I think, even though (or because) it's not taken from public transportation, since it isolates speed alone (if omitting the differences in cost of driving 90 km/h to driving in congestion which shouldn't be too great).

jamespetts

But other things also isolate the speed alone - Concorde isolates speed alone, as does the differential in price between the High Speed tickets from London to Ashford International and the ordinary services from Charing Cross, as well as the difference in price between the London Midland services from Euston to Birmingham New Street and the Virgin West Coast services on the same route. There are no other relevant differences in each case. The London to Oxford example also has speed as a primary factor, since, as discussed, comfort is about equivalent on both routes, although subjective perceptions of comfort and convenience may vary.
Download Simutrans-Extended.

Want to help with development? See here for things to do for coding, and here for information on how to make graphics/objects.

Follow Simutrans-Extended on Facebook.