vector of exogenous variables for producer s.
Due to the information asymmetry, θ is inherently unobservable to the government
and the econometrician. Following the stochastic frontier approach, I treat ln θ as a random
variable. Assume that v and θ satisfy the following moment conditions:
M1. E [vs|zs] = 0;
M2. E [v03] = 0;
M3. E [ln θ∣zs] = -σp2∕π;
M4. E £(lnθ - E [ln θ])3] = σ3 (1 - 4∕π) p2∕π.
Under M1, the vector vs has a mean-zero disturbance vector uncorrelated with
the instruments. In addition, M2 states that the disturbance for Eq. (5) is symmetrically
distributed. Assumptions M3 and M4 require that ln θ be uncorrelated with the instruments,
and that its mean and skewness correspond to those of - ∣y∣, where y is a random variable
distributed N (0, σ2).
As noted by Aigner et al. (1977), these distributional assumptions have two practical
implications. First, consider a least squares estimator that ignores the type-dependent com-
ponent of the error structure, mistakenly using the moment conditions E [v0 - ln θ∣zs] = 0.
By treating the expected compound error as mean zero, rather than mean σ∖∕i2∕π, this
regression effectively replaces Eq. (5) with
N i----∖ N-1 w
ln C (we,q*,α,θ) = (β0 + σ∙∖∕2∕π) + P βi ln — + βq ln q* + βa ln a
n=1 wN
+βaa (ln a)2 + (υ0 - ln θ - σ p 2 n´ . (8)
Such an estimator generates consistent estimates of all parameters in the cost function except
βo, which is upwardly biased by σ∙ξ∕2∕π. The second practical implication is that the