the global market model is an abstraction, not a simulation, designed back in 2002 for a game to be played by computers that had as little as 128 MB RAM.
Expect behavior to be abstracted, simplified, and in some cases not following proper economic behavior if such behavior would hinder gameplay performance on low-end systems of the time of development or add yet further layers of management to an already overly micro-managed gameplay system.
And remember, the emphasis in the term AI for Victoria should be on the artificial, it has no intelligence whatsoever and thus whatever cheats, abstractions or elidations are needed to make the computer remotely competitive towards game end are vital to the gameplay experience.
Paradox's AIs have improved vastly in the six years that have followed since the development of Victoria, so if a new game based on the Victoria time-period was developed using the EU3 engine, I expect the AI would be much more competent. As for Victoria, we simply have to accept that it is a relic of an earlier period in Paradox history, with all the limitations, problems and inconsistencies that implies.
Still doesn't take away from the fact that Victoria is a lot of fun with lots of replay value, but in the end it is preferable not to look too closely at the AI behavior, given the age of the game and the much stricter limitations to minimum prerequisites the game operated under at time of development (128 MB Ram, 450 MHz processor).