In this paper we estimated primal panel data models for risky production technologies on
an unbalanced panel of Norwegian salmon farms. We examined the risk properties of
inputs and the patterns of technical change for both the deterministic and risk part of the
technology. Furthermore, we analyzed how different specifications of farm-specific effects,
different functional forms and different estimators influenced the empirical results.
The linear quadratic provides mean production function elasticity estimates that are very
similar to those derived from the translog, despite different specifications of the farm-specificeffect. When we compare the estimated Just-Pope models we find that the empirical results
in several respects are not dramatically affected by the choice of functional form or estimator.
However, the inclusion of farm-specific effects in the model had a profound effect on the
estimated rates of technical change. This is particularly the case for the risk part of the
production function. The choice of estimator (FGLS vs ML) also has an effect on estimated
marginal risk effects of inputs.
For salmon aquaculture we find that feed and fish input have risk-increasing effects, while
labor input has a risk-decreasing effect. According to Ramaswami these results imply that a
risk averse salmon farmer will use less feed and fish input and employ more labor than a risk
neutral firm. The estimated models also predict that an expansion of the scale of operations
at a given farm site lead to a substantial increase in output risk. This result implies that site
diversification, i.e., use of several sea sites, may be a sensible strategy for risk averse firms.
Such a response to production risk has also been observed in Norwegian salmon farming.
The empirical results clearly support the use of flexible models of technical change.
Large year-to-year fluctuations in the rate of technical change are found both for the mean
part and risk part of the production technology. Shifts in biophysical conditions along the
Norwegian coast and economic conditions are believed to be important factors behind theobserved time-patterns of technical change.In section 1 it was argued that in an analysis of technical efficiency under production risk,
it is not sufficient to merely measure the change in mean productivity over time. When
producers are risk averse, one also has to measure the change in output risk. We find that
from 1985 to 1993 technical change has lead to higher mean output, but also higher output
risk for the average salmon farm. However, when technical change is analyzed in termsof the first-order stochastic dominance criterion, it is found that the improvement in mean
output dominates the increase in output risk. This means that risk neutral and risk averse
producers agree that there has been an improvement in technical efficiency during the data
period.
It can be argued that the conclusions presented here regarding the structure of production
risk and the nature of technical change are relatively robust, since they generally hold across
a variety of models. The results support the hypothesis that differences in observed output
in salmon farming can be explained both by farm heterogeneity and by stochastic shocks.
It is believed that the empirical evidence provided here should motivate further research
into the nature of technical change in risky production sectors.