$\sum y_i=\sum\hat y_i$ holds precisely when there's an intercept

In case of regression of $Y_i$ on $X_i$ with fitted values $\hat y_i, i=1,\dots,n$ we get $\sum_{i=1}^ny_i=\sum_{i=1}^n\hat y_i$: