Agriculture Reference
In-Depth Information
9. For example, see Laffont and Matoussi (1995), or Hayami and Otsuka (1993) for agricultural cases, and Garen
(1994) or Kawasaki and McMillan (1987) for business cases.
10. The linear mean-variance utility function is routinely used, especially in agriculture (Chavas and Holt 1990,
Pope and Just l991).
11. Kawasaki and McMillan (1987) derive a similar optimal sharing parameter in their model of sales compensa-
tion.
12. This prediction is not unique to the risk-sharing model.
13. We do this because in North America land contract shares tend not to be continuous, as shown in chapter 5.
14. Cheung (1969) first suggested this distinction in costs between cash and share contracts.
15. Predictions 6.5 and 6.6 assume that neither
V
c
>V
s
nor
V
s
>V
c
hold for all parameter values.
16. Even when data on prices is relevant, it is not clear that historical price variability data are an appropriate
measure of a farmer's forward-looking price variability because of continual changes in market conditions.
Historical measures of yield variability are more reliable because they are mostly determined by long-term natural
forces such as weather and pest populations.
17. The possibility of a negative correlation between yield and price might suggest that sharing output can actually
mitigate risk. However, because all of the farmers in our sample operate in competitive world markets, this is
highly unlikely. Indeed, calculations of price-yield correlations at the state or province level indicate that negative
relationships are not common (see table A.11). More important, because our tests exploit regional and county (or
parish) yield variability, even negative statewide price-yield correlations are not relevant.
18. Alternatively, one could, in principle, test the model with data on individual risk preferences. Ordinarily such
data would seem impossible to obtain, although some (Gaynor and Gertler 1995) use self-reported risk preference
measures.
19. For instance, a recent study of executive compensation contracts by Garen (1994) notes the “endogeneity
problem,” but then uses industry-wide R&D expenditures as a measure of the exogenous variability, claiming that
“settings in which R&D is important should display a greater variance in returns in investment opportunities.”
Kawasaki and McMillan (1987) and Lafontaine (1992) use proxy variables that are endogenous to firm behavior
as well. Sotomayer, Ellinger, and Barry (2000) use farm yield that also depends on farmer behavior. Interestingly,
in their recent survey Chiappori and Salanie (2000) do not mention the empirical difficulty of obtaining exogenous
risk measures.
20. Rosenzweig and Binswanger (1993), confirming many studies, find that the timeliness of monsoons is an
important variable in explaining crop yields in India but that rainfall amounts are not good explanatory variables.
In chapters 8 and 9, timeliness costs are an an important issue in farm organization and ownership.
21. This requires that the total output of the region be bounded. This model assumes the covariance between effort
e
and natural parameters
θ
is zero, which is consistent with the standard production technology assumed in the
principal-agent literature.
22. For other types of agriculture, heterogeneous regions may generate enough idiosyncratic risk that performing
this test may not be possible. Lafontaine and Bhattacharyya (1995) argue that such heterogeneities plague
franchising studies of risk sharing. On the other hand, if regions were perfectly homogeneous, one might expect
relative performance contracts for farm production. Such contracts are never observed in our data set but are found
in chicken production where (homogeneous) technology and inputs are provided by a single supplier to many
growers (Knoeber 1989).
23. Some of these data are not available for British Columbia because its regions tend to encompass the entire
provincial production for the crops we examine (see table A.7). Also, our measures of crop yield variability are not
consistent between British Columbia and the other three states because of data limitations. In particular, British
Columbia yield data are not available for smaller, county-like areas.
24. Higgs (1973) conducts a similar exercise using aggregate data from eleven Southern states for 1910. Consid-
ering only two crops (corn and cotton) he finds a positive effect on CV but notes the U.S. Census data does not
distinguish between “sharecroppers” or unskilled laborers without capital, with “share farmers” who provide their
own capital.
