Differences in coefficients of xtabond vs. xtlsdvc

Hi everyone,
I am working on a balanced panel of 79 universities (N) and 17 periods (T). I am exploring the impact of several variables on the number of annual enrollments at German universities. The dependent variable reflects the delta (percentages) of number of enrollments. As I am assuming a trend effect of enrollment rates of the past year on the current year, I am trying to specify a dynamic panel model. As my data is medium T and medium N a LSDVC or an Arellano Bond estimation should lead to consistent results. However, and very surprising to me, the coefficients of the lagged dependent variable strongly differ between the models. Could this be correct? If yes, which model should I prefer ou should I report both models?

Here are the commands and results of the lagged dependent variable:

xtabond enrollment_no [ind. variables], lags(1) vce(robust) twostep
L1.enrollment Coeff: 0.0250851 | p-value: 0.597

xtlsdvc enrollment_no [ind. Variables], initial(bb) lsdv vcov(100)
L1.enrollment Coeff: -0.1032591 | p-value:0.005