An inter prediction method performed by a decoding apparatus includes: receiving information on a motion vector difference (MVD) in units of integer samples; deriving a first motion vector predictor (MVP) in units of fraction samples on the basis of neighboring blocks of a current block; deriving a second MVP in units of integer samples on the basis of the first MVP; determining a first motion vector (MV) in units of integer samples for the current block on the basis of the second MVP and the MVD; determining a second MV in units of fraction samples for the current block on the basis of the first MVP and the first MV; and generating a prediction sample for the current block on the basis of the second MV. The present invention is capable of increasing the accuracy of the MV while reducing the amount of data allocated to the MVD.