In theory I agree, but I very rarely run into people who actually do so on their own. Even fewer who do something recognizable as contributions to a research field, in the sense of writing up something of peer-reviewed paper quality, which indicates the author is familiar with existing research, appropriately explains how the new contribution relates to concepts and techniques that already exist, and presents the results in a convincing format (whether that's statistical analysis or a good case study or whatever). It doesn't even have to actually be published; I'm fine browsing papers on arXiv too, or things circulated as whitepapers or tech reports.

They exist, but they're uncommon, and many of the people doing that kind of work are in sort of para-academic jobs, like long-time librarian at a big research library, or senior staff member at a (government or corporate) research lab. Even in an environment like Google, the majority of the papers and paper-like writeups seem to come from either people with PhDs, or people with long track records in a quasi-academic environment, like Google's sizable stable of ex-Bell-Labs researchers.

I don't really say that out of any particular love for institutionally tied research, but more out of the opposite, a frustration that it's so rare to find DIY research that really contributes to a research field. Some of that, I assume, is just incentives: if what you earn money on involves getting something to work, all that really matters is that you come up with a technique that works. Understanding how your technique relates to existing techniques, explaining whether it's completely novel or a variant of an established one, and doing the detailed analysis to figure out why precisely it outperforms existing techniques (e.g. is there one particular tweak that's critical, and if so, why is it critical?), is not always something incentivized in that context, but is important to advancing the state of knowledge in a field.