In this paper, we formulate the forest representation learning approach named casForest as an additive model, and show that the generalization error can be bounded by $\mathcal{O}(\ln m/m)$, when the margin ratio related to the margin standard deviation against the margin mean is sufficiently small. This inspires us to optimize the ratio. To this end, we design a margin distribution reweighting approach for the deep forest model to attain a small margin ratio. Experiments confirm the relation between the margin distribution and generalization performance. We remark that this study offers a novel understanding of casForest from the perspective of the margin theory and further guides the layer-by-layer forest representation learning.