First of all, how do we make predictions? We usually look at the past and project it into the future.
If you look at past returns for bonds over ten year holding periods, they stack up pretty nicely in a shape close to a “bell curve”. This would suggest you might with some confidence predict around 5% with a reasonable low side of 4% and upside of 7%.
Unfortunately many predictions of bond returns are using this methodology and the conclusion is wrong. (I believe most defined benefit pension plans use this methodology when determining a discount rate which is used to predict funding needs).
Reality has played an incredible trick on us.
The last time interest rates were as low as they are now was in the late 1940s. Interest rates rose to a very high level peaking in about 1981. The length of the period of rise and then decline is pretty much equal. So when you stack returns and put them on a graph you get something that looks like a normal distribution. Statisticians use this data to determine a mean (or average) return and how likely we are to have that return or something similar in the future. The last time I looked there was a historical mean of about 5.5% with a standard deviation of about 2 points on each side which is a range of 3.5% to 7.5%. Unfortunately, it isn’t going to happen. If I buy a 10 year treasury note today, I know I will get a return of about 2.4%. That is the appropriate predictor of returns.
The lesson from this is before you accept a statistical prediction, be sure to consider other factors to see if it makes sense. To predict the future using the past, generally speaking, you need to have a past where things migrate toward something. If they vary too far one way they trend back and then if they vary too far the other way they trend back. That is called a “central tendency”. Consider that interest rates which dictate the returns on bonds are affected by inflation. Anyone want to try and predict inflation? If the answer is “no” then the past will not be a reliable guide for future bond returns. Current yields will work better.