I use different returns depending on what I'm trying to model. The nature of the question determines what to use.
For example, I have savings goals for this year.
Now, some random stranger **might** think I have the handsomest beard ever and give me a bazillion dollars for that reason, but I don't include that possibility in my plan. I view the stock market in any given year pretty much the same way except that it's the results of millions of people's decisions for reasons that are equally inexplicable. So I ignore stock returns for my savings goals.
For example, one savings goal for this year is to save the max allowed in our two 401Ks. That's $46,000. I don't include employer matches in my goal because I don't control what they will do. I don't include market losses or returns in that goal because that's not in my control. My job is to enlist 46,000 green workers in my 401K and put them to work for me.
If, however, I'm doing long term trend analysis, including market returns (or losses) is appropriate. Notice I'm not setting a goal, I'm investigating what the result would tend to be in different situations. I tend to study 2 of the 3 main ones. #1, what's the average result? #2, if things were on the bad side, what would the results tend to be? For scenario #1, the average result, I use the average return after subtracting inflation. That's 7% for the US market. For scenario #2, a below average result, I use a lower number like 3%. I don't bother with scenario #3, an above average result, because if #1 and #2 aren't in the shitter :) then the other scenario is bound to make me happy.
If I want to stress test my situation I head over to CFireSim.com and put in the numbers. That will test my plan against historical data. But again, that's exploration and analysis, not goal setting.