I am running a first difference regression. I have seen the nocons option included on several posts about first difference regressions on this forum. My R-Squared skyrockets from 0.12 to 0.94 when I use the nocons option and all the controls suddenly become statistically significant at the 1% level. However, I have also read that R-Squared becomes misleading with the nocons option. Can somebody explain whether I should be including a constant and what the reasons for the sudden increase in precision and goodness of fit are?