It's definitely situational. On the street, in broad daylight, in most contexts, I am somewhere in the yellow range. If I have to go into a parkade to get my car, it goes into red automatically, and I have weapons immediately available (I do not like parking garages!!) This is also true when I'm in other enclosed spaces (e.g., on a bus) where there's little chance of assistance if things start getting weird. At night, I'm in red under most circumstances.
I think the crucial skill that needs to be developed is, for most people, the ability to identify strange behavior in other people, things even subtlely different from normal patterns of action and reaction. The sucker punch is always a possibility, so you need to be ready to react when someone imposes on your space in a way that could be threatening (whether or not they show obvious intent), just as seeing a driver on the road getting a bit too close to your car should immediately pop your defensive driving skills to the top of your stack.
In principle, yes, you want to minimize the amount of time you spend in the red zone. The problem is that it's only in hindsight that you know which were the genuinely threatening situations and which were harmless. The incidence of the former is only a minute fraction of the frequency of the latter—but you don't know which is which while you're in the situation. It's like working with corrupted data: if you know that only a tiny amount of the data you're working with is corrupt, but you don't know which data are the baddies, then you can't trust your results. So I find it better to anticipate trouble that, statistically anyway, will very likely never occur—just in case it does. Doing it that way is costly in terms of stress, but, I'm betting, not as costly as what I'll have to pay if, as Andy M.'s signature puts it, I'm caught nappin'.