Modeling firms' R&D expenditures often become complicated due to the zero values reported by a significant number of firms.
The maximum likelihood (ML) estimation of the Tobit model, which is usually adopted in this case, however, is not robust to
heteroscedastic and/or non-normal error structure. Thus, this paper attempts to apply symmetrically trimmed least squares
estimation as a semi-parametric estimation of the Tobit model in order to model firms' R&D expenditures with zero values.
The result of specification test indicates the semi-parametric estimation outperforms the parametric ML estimation significantly.