Esoteric, proprietary formats have a high risk factor. Long term preservation goes beyond administrative archives and includes niche software used to make e.g. interactive objects which may need to be preserved for legal or other reasons.
Long term preservation also extends beyond the 2-5 year horizon, when one could make a reasonably accurate risk assessment. The legal framework may demand that some (parts of) archives need to be preserved for the next 30-50-100 years or even eternity (no end date specified). Just consider documents related to nuclear installations, machinery, tooling,... which go well beyond ubiquitous office file formats.
> But files should be kept in their original format.
Substitution is choice which needs to be considered carefully. Hence why emulation is a big topic in digital archiving.
> Changing format is the best way to lose information, for zero preservation gain.
It's not that black and white. Ultimately, bitstreams need to be translated to an output device. A big question is "Am I seeing the message in a digital context such as what one would have seen 30 years ago?" There's a can of worms discussion regarding the "essence" of digital information, the representation on screen, and the bits stored on disk.
Changing formats usually is a question of judicious and pragmatic decision making based on valorisation of the material as well as the context from which it originated.
What's more important? Being able to consume the information? Or being able to consume it in its original form? What do we lose and what do we gain? There are no absolute answers to be had in that regard: it's always a per-case assessment.