Why don't high-ranking journals go solo?
Many journals are publishers' own products, "going solo" makes no sense. It's like asking why Gmail does not segregate from Google. In some cases, professional societies hire publishers to take care of their official publishing organ, I imagine your question relates to these cases. Otherwise, there have been cases of the entire editorial board leaving for another journal, I've heard it happened to a Frontiers journal over concerns on quality, but in this case the journal is still there, just edited by other people.
Running a professional-level journal is not trivial and there are reasons to favor established organizations, commercial or not, to handle that part. These include administration, secure web hosting, a long-term back-up strategy, typesetting, distribution, submitting accurate and complete article metadata to third parties like Pubmed or Web of Science, printing when applicable, etc. and of course finding the money to do all of these.
For many editors, it's a pretty straightforward decision to outsource these hurdles to a specialized organization, be it for profit or not, especially if there is an ongoing issue-free relationship going on for years.
Actually, some journals do successfully go solo, just as you suggest. A nice high-profile example is the Journal of Machine Learning Research, a top-ranked journal that formed when the entire editorial board of Machine Learning resigned to create this free alternative.
This points to the main reason why traditional journals have much inertia. The reason that JMLR could work is because:
- A large fraction of the key players in the community coordinated to make the change (thereby immediately granting the new journal a high academic reputation) and,
- They were able to arrange sufficient support from their home institutions to bear the start-up costs of organizing the journal.
These are both difficult to arrange, requiring quite a bit of coordination and personal investment, and so it is not surprising that it is rare to happen.
I am going out on a limb here and disagree with Darrin. I think there are plenty of academics who would be both, capable and perfectly willing to run a university- or self-published journal. I think it is an illusion that academics want to do only research, all the time. A lot of (tenured) academics do plenty of things that require lots of time and don't directly contribute to their research, be it writing entry-level text books, maintaining scientific software, communicating their work to the broader masses through events or magazine articles, running for offices in their university or various societies, etc. etc. I fail to see how running a journal would be so different to these activities that no-one would take up the task.
The main reason, in my opinion, why this rarely (although not never) happens is because of legal issues. Most journals (and, in computer science, conferences) are mostly identified through their name, and this name is owned by whoever currently publishes it. For instance, the editorial board of the IEEE Transactions on Software Engineering (TSE) is not free to just decide that the journal now goes self-published. Sure, the editorial board can decide to jointly quit and start a new journal, but it is not guaranteed that the community would see this new journal as a continuation of TSE. Much more likely, the new journal would need to start building a reputation from scratch, which is not easy at all. TSE, in the meantime, would continue even with a completely new editorial board, because I can guarantee you that there would be many qualified new people waiting in the wings for a chance to get into the board of the most important journal in their discipline.