This paper investigates the industrial effects of true state dependence, the sales-to-capital ratio and unobserved heterogeneity on the rate of investment in plant, machinery and equipment (PME) in Swaziland. A range of fixed and random effects estimators are compared. In all the methods used, the first-order autoregressive – AR(1) – model with unobserved firm-specific effects and the sales-tocapital ratio have insignificant coefficients. Similarly, the impact of unobserved firm-specific characteristics underlying investment decisions is also insignificant. Most interestingly; however, our novel result is that we show how missing investment values reduce the probability of investing under both exogeneity and endogeneity assumptions. Missing investments at time t-1 reduce the likelihood of investing at time t by [-5.56%, -4.91%] depending on regressor exogeneity or endogeneity assumptions, respectively. By interacting missing investments with labour, a probability of up to 0.55% of capital substitution for labour is estimated. Furthermore, notice that the Generalized Method of Moments (GMM) and multilevel methods naturally assume a single investment regime by default. When an endogenous switching regime model of investment with unobserved separation is estimated, it produces negative but significant results in both regimes. Notably, all the switching regime results are scale-independent, where firm size is defined as the inverse of the previous period capital stock. However, a Wald-test of equation independence across regimes confirms a single investment regime produced by the GMM and multilevel models.
