It doesnt have to be very accurate to be useful. in fact i'll be you could probably produce a useful and reasonably accurate approximation function using only:
- the company
- the initial price
- the initial review scores
It would actually be a fun statistics project for anyone in school looking for an independent study assignment.
You'd just have to scrape the data for a bunch of games, to get the company name, initial price, initial review scores, release date, and find a source that has historical game price data. And then see what kinds of functions you could fit to the data.
The hard part would probably be collecting the historical/training data.