In the first observation, DeltaEAs is set to zero, and the SUM function in the last statement of the data step will also set EAs_avant to zero (sum of a missing value and zero).
Then, in all observations where MT_EVT is negative (which it is in your example dataset), this formula is calculated:
DeltaEAs=EAs_avant*(MT_EVT/MT_EA_AVMVT)
Since EAs_avant is zero, the result of this is also zero.
So the SUM of EAs_avant (0) and DeltaEAs (0) will still be zero.
And so the zero perpetuates throughout your dataset.