It's somewhat interesting that periodic/cyclic patterns always seem to emerge in my basic simulated economies when AI agents don't have access to historical data
In the real world there is publicly available information: market news and analysis, talks at conventions etc, insider knowledge about shipment quantities, state funding to start businesses or increased taxes/regulations to lower economic activity in that sector etc.
For example: if one station has a very high demand of a product, and all traders would quickly buy that product elsewhere and travel (at the same time) to the station with the high demand, then there would be a sudden oversupply and resulting pricedrop. The market would always react very cyclical.
To reduce that strong fluctuation, there could be a newsboard that indicates how many supply-ships with that commodity are already on the way to that station, and thus for other traders it will indicate that the demand for that product will soon fall - making a trip with that product likely to be unprofitable.
-> So traders will early on notify the demanding station, that they will promise to deliver a certain amount of goods.
Not every trader needs to plan according to that newsboard, especially if he is close by with the required goods. But it will keep the majority of traders from all running towards that same temporary arbitrage possibility.
Also traders could remember previous profitable trade-runs, and then when considering several alternatives, value the previously successful runs higher. So every trader will have a personal evolving set of preferred trade-routes.
Thats why it would be good to have the game run a bit the trade-simulation before the player starts the game: so that the economy can stabilize over a certain period (removing unrealistic price-differences that would not occur in an actual grown economy)