Many people make the mistake of thinking the first estimate needs to be accurate, which is rather stupid, becasue it’s rather difficult, especially with anything RnD type projects. So start coards. Can you do it in one year? How about a month? a week? a day? At some point uncertainty sets in. Then you have your first estimate. Then do whatever it takes to remove uncertainty so you can refine your estimate. If it’s prototypes and exploration, then go that’s what it takes.

Regarding past behavior: Generally a good gauge, except when your team changes or the nature of the project does and you end up comparing apples with pears. Othewise it’s a good approach, if you bothered to collect some project metrics and post mortems.