By performing experiments on models of large markets we hope to verify our intuitions and gain further insight to what problems can arise in large open markets with small means of control, and which remedies are effective to avoid and remove those problems.
To do this we are developing a workbench for simulating systems of agents which produce services and sells them on to other agents (see fig. 3).
Figure 3: Screenshot of a commerce simulation workbench
Outline of the current model: The market world is a 2-dimensional surface populated with actors who can be buyers or sellers. The actors range of sight can be constrained either by a maximum length, or by walls subdividing the surface into rooms. The actors move about randomly and for each step they choose who to do business with among the visible actors. If the actors move into another room they can't see anyone in their previous room.
A seller is modeled as an entity offering a service for a fixed price. The service provided by the seller has a particular value. This value is not known to the buyer when he/she is deciding on doing business or not. An actor who is offering a service with a high price and a low value is either malicious or inefficient. (The two are the same in this model.) Currently all producer produce the same service to avoid side effects caused by the trade of other services.
To produce a service, the seller must pay a certain amount of money, for example half that of the value. Remember, an agent buying a service has no way of knowing the value of the service, it can only look at the price. After the transaction is done, the value of the provided service is added to the capital of the buyer.
Every actor gets an initial capital. For every turn of the simulation the actor has to pay a certain price to participate. When the actor is out of money it is removed from the simulation.
The number of actors is fixed, therefore new actors can only be introduced when another actor is removed. This is done by making a copy of an actor who has a large capital and inserting it at random into the world matrix.
By running simulations on market systems where the actors have different methods for selecting cooperation partners we can see which of these methods that succeed in choosing good sellers, that is, sellers with larger produced value than price plus buyer's participation cost.
If the system stabilises when there is a population of non-malicious sellers in many different price categories and where the price is correlated to the value of the service then the actors have managed to auto-organise them self into a market where a buyer can choose the quality of what he/she is asking for simply by looking at the price. Malicious agents can't make money on their services, since they only provide a very low value.
So far we have studied what happens when an actor pick anyone, the closest seller, the cheapest seller, sticks to a favourite seller or to a locally recommended seller. We find that by disallowing buyers to choose from all sellers a larger number of seller agents can co-exist. Furthermore, by creating regions of commerce (rooms), the system is able to more quickly get rid of malicious agents. This happens as the actors in regions infested by malicious actors go bankrupt, making the malicious actors go bankrupt too. All of this is very dependent on the migration level between the different regions. High migration leading to an almost global environment is much more sensitive for bad agents.
___________________________________________________
Lars Rasmusson