The present invention relates to a device and a method for coding a multi-view video, a residual prediction method, according to the present invention, comprising the steps of: deriving a disparity vector of a current block; performing residual prediction on the current block on the basis of the disparity vector; and generating a residual prediction sample, wherein the disparity vector is not derived on the basis of a neighboring block of the current block, but is derived from a preset default vector. According to the present invention, the effect of prediction may be increased by performing residual prediction on the current block according to a certain condition, not only when the disparity vector is induced but also when the disparity vector is not induced from the neighboring block of the current block.