Basing capex on rent is a non-starter. It would be less fuzzy to have it based on percent of median rent for the market, since older lower end units cost much more proportionally. Still, what I do (which let's be frank, is still somewhat trash) mirrors other comments: square footage, age and life of components, and vacancy rate based.
Square Footage: No comment required, although multi stories and multi units cost somewhat less than single famile ranch house style. They simply have less exterior attack surface to defend.
Age and life of components: I sourced average useful life for a laundry list of components from InterNACHI and a few other sources to fill in, as well as my own professional experience. I suppose you can use median age of housing in a market for an extremely wide overview as well.
Vacancy rate: Turnover budget is your primary capex expense on virtually every property, and if units are turning over faster, you need to include a bump for that. Stats are available for average length of residency for individual markets, and this is a better stat to hang your hat on here. I use 2.5 years if I have not looked it up for the individual market. I note as an aside that too many people calc actual vacancy without taking into account their PM fees on each vacancy, which adds a month of economic vacancy in general. For these reasons I calculate general capex, opex, and turnovers as 3 separate buckets in my spreadsheets.
I agree that the NAA surveys are a goldmine for this. Remember smaller properties will be at least 20% more expensive since there is no economy of scale - either on the structure itself or the vendor relationships. This is offset by stricter MF standards, but if your stats don't fall above 25% listed rent for opex and capex (+ turnovers if using 3 buckets) based on their 20%, you can be certain you are too optimistic for a decent business plan. Many properties will cost more than this.
