feasibility up to one to two years. Figure 3 plots the estimated differ-
ence in the discounted cost of the two systems for each investment year
beginning in 1980.
 A very slow decline in photovoltaic system costs characterized the
pessimistic scenario. Furthermore, no real increase in utility electric
rates was allowed. Under this scenario, the first year of economic
feasibility for buy-back ratios of 0.25 and 0.50 does not occur until
after year 2000. With a buy-back xatio of 0.75 the initial year of
feasibility is 1998. The effect of increasing the buy-back ratio on the
initial year of economic feasibility is similar to the preceding scenarios.
Each 0.25 increase in the ratio moves the initial year of feasibility up
by one to two years. For each investment year, the difference is dis-
counted costs of the two systems under comparison is plotted in Figure 4.
 The importance of the cost behavior of photovoltaic systems over
time can also be inferred from the results of the simulation. For given
buy-back ratios, the base model cost projections result in initial
feasibility occurring an average of five years sooner than if system
costs follow the pessimistic cost projections. Similarly, optimistic
cost projections result in the initial economic feasibility of photovoltaic
systems occurring an average of six years sooner than if base cost pro-
jections held. These results imply that the effects of commercializing
the manufacture of photovoltaic arrays, and the ensuing cost declines,
have a significant impact on when photovoltaic powered irrigation systems
are economically viable.
 Overall, the results of the simulation model appear internally
 consistent. Changes in the buy-back ratio generally have the same effect
 on the initial year of economic feasibility for a wide range of changes
 in system costs over time. Furthermore, the effects of changing the
 cost projections for system cost generate similar responses in the initial
 year of feasibility for differing buy-back ratios. Because the analysis
 of these types of problems require considerable abstraction and assumptions
 in the absence of any statistical measures, such consistency is extremely
 significant in giving the above results considerable credence.