Notes - The Nature of the Farm - page 233

25. The two states are not clones though. Nebraska has a greater fraction of land similar to the Corn Belt and South
Dakota has more land similar to the High Plains. Also, climate and soil tend to improve as one travels from north
to south. Because the statewide crop CV rankings differed between the two states, the coefficient estimates from
the pooled sample could not be used to directly test the effects of crop riskiness mentioned earlier.
26. Unfortunately, data limitations in calculating yield variability prevent us from estimating these effects in British
Columbia using the specification that follows.
27. At the same time, because these farmers all face essentially the same market prices, there is no variation in
prices across our sample of contracts.
28. We also use MEAN as a control variable when STD is used to approximate σ

2
ij

.
29. The size of some Louisiana crop samples (mostly sorghum and wheat) were further restricted because we were
not able to identify the parish for some farmers. Regional yield statistics are not available in British Columbia, so
this test could not be performed. The samples ( n j ) were generally smaller for COUNTY CV and STD, compared to
REGIONAL CV and REGIONAL STD, because the states do not calculate yields for counties when total output is
below a threshold; therefore COUNTY CV and COUNTY STD were not always available even when REGIONAL
CV and REGIONAL STD were available.
30. We are unable to estimate a wheat equation for Louisiana because of the small number of observations.
31. We use other methods to test the robustness of these estimates. First, we estimated OLS regressions using
ln(s/

as the dependent variable. We also estimated the same 46 equations by expanding the sample to include
cash rent contracts and counting them as 100 percent share contracts. For each cash contract, we set

1 − s)

s = 0.999 in

, is defined for all contract observations. Next we estimated
the 46 equations without control variables, using only the CV or STD variables, and with a smaller set of control
variables than used in table 6.6. Finally, we used Heckman's two-step estimation method in order to control for the
contract choice selection problem. None of these alternative specifications change the findings reported in table
6.6 in any significant way.
32. Stiglitz (1974) is one of the earliest papers to make the point about risk-sharing and futures markets.
33. It is possible that futures markets have arisen for those crops that have highly variable yields, although this
seems unlikely. Simple inspection of our data shows that futures markets are found for nearly all widely traded
and storable crops. Furthermore, Pirrong (1995) and Williams (1987), argue that futures markets exist to reduce
the transaction costs of measurement and trading commodities.
34. These coefficients, along with the ones mentioned in the next paragraph in the text, are not reported.
35. This is hardly surprising, however, since we also found that rice land is nearly always cropshared (table 6.3).
36. Prendergast (2002) develops a similar model in which greater output variability lowers the returns to input-
based contracts, thus increasing the use of output-based contracts.
37. Additional evidence across crops is found in appendix A (table A.12).
38. Risk preference assumptions are also crucial in franchising models (Lafontaine 1992). In the standard model—
the franchisee is risk averse but the franchisor is risk neutral—risk-sharing implies that greater exogenous
variability will result in more fixed payment contracts (hired managers). If, as some have argued, risk preferences
are reversed (Martin 1988), the contract choice prediction is reversed.
39. Demsetz and Lehn (1985) develop a similar argument in their study of outside shareholders in a corporation.
Greater uncertainty makes the risk cost of concentrated shareholding (less diversification) greater, and this is the
effect that most focus on. Demsetz and Lehn argue that more uncertainty increases the value of shareholder mon-
itoring of the manager, and so increases the benefit of concentrated shareholding. They also present evidence that
this monitoring effect dominates. Greater uncertainty is linked to greater concentration of outsider shareholding.
40. It should be noted that it was not possible for us to directly test risk-sharing predictions against those from
our transaction cost model because of the different data requirement for each test. The clearest tests of the
risk-sharing model used contract data for specific crops. The transaction cost predictions, however, required that we
use contracts for land with differing crops to account for differences in soil exploitation and output measurement
costs.

order to insure that the dependent variable, ln(s/ 1 − s)