Beruflich Dokumente
Kultur Dokumente
An Econometric Approach
Edgardo Moscardi and Alain de Janvry
Attitudes toward risk among peasants in Puebla, Mexico, are derived from survey data
in a model of safety-first beha vior. The measurements of behavior toward risk obtained
Downloaded from http://ajae.oxfordjournals.org/ at Penn State University (Paterno Lib) on May 9, 2016
are then explained by a set of socioeconomic and structural variables that characterize
peasant households. Knowledge of the determinants of attitudes toward risk is, in turn,
useful for the purpose of tailoring technological recommendations to particular
categories of peasants.
A number of recent studies have evidenced tend to be followed whenever the satisfaction
the importance of risk on decision making by of basic needs may be at risk (Scandizzo and
peasants (Dillon and Anderson, Wolgin, Dillon).
Wiens) and farmers (Lin, Dean, and Moore; Assuming that the safety-first model holds,
Boussard and Petit). Attitudes toward risk are the degree of risk a version manifested by indi-
major determinants of the rate of diffusion of vidual peasants can be derived from observed
new technologies among peasants and of the behavior. Given a production technology, the
outcome of rural development programs. If risk associated with production, and market
they are going to be effective, new tech- conditions, the observed level of factor use
nologies and rural development programs reveals the underlying degree of risk aversion.
need to be tailored to the attitudes toward risk In this fashion, attitudes toward risk can be
of particular categories of peasants. For this measured for relatively large samples of peas-
purpose, it is important to identify the specific ants. Explanations for differential degrees of
determinants of behavior toward risk and to risk aversion among peasants can then be
quantify their impact on decision making. sought by relating the measures of risk aver-
This paper starts by examining the extent to sion obtained to a number of variables that
which risk may be responsible for discrepan- characterize the peasant household, its in-
cies between peasants' demand for fertilizer come-generating opportunities, and its posi-
without risk and actual demand under risky tion in the political economy.
conditions. For this purpose, risk is intro- This indirect approach to the measurement
duced in a model of economic decision making and explanation of risk aversion is introduced
as a safety-first rule. According to this rule, an as an alternative to the direct approach devel-
important motivating force of the decision oped by von Neumann and Morgenstern. In
maker in managing the productive resources this direct approach the subject is asked to
that he controls and, in particular, in choosing make decisions in reaction to a large number
among technological options is the security of of randomly arranged hypothetical bets and
generating returns large enough to cover sub- insurance schemes. This approach has serious
sistence needs. As recent empirical studies of difficulties resulting from the fact that the sub-
peasant behavior indicate, safety-first criteria jects have different degrees of utility or disutil-
ity for gambling (the very method used to re-
veal their preferences) and that the concepts
Edgardo Moscardi is an economist with the International Center of probability are by no means intuitively ob-
for Wheat and Com, Mexico, and Alain de Janvry is an associate
professor of agricultural and resource economics. University of
vious (Edwards). In addition, the method is
California. Berkeley. very time consuming to administer (Lin,
Giannini Foundation Paper No. 455. The authors are indebted Dean, and Moore). As a result, the measure-
to Donald Winkelmann and George M. Kuznets for helpful sug-
gestions and to the International Center for Corn and Wheat for
ments of risk aversion thus obtained have
financial assistance. been few, and attempts at relating these mea-
Moscardi and de Janvry Peasants' Attitudes Toward Risk 711
Downloaded from http://ajae.oxfordjournals.org/ at Penn State University (Paterno Lib) on May 9, 2016
dicted behavior that are not due to risk in
deduced from the observed levels of product
production are accounted for by the model.
and inputs by solving equation (2):
and 1971. The data consist of 1,142 observa- area and is also the largest component of vari-
tions from twenty-five trials conducted in one able costs, its marginal productivity derived
of the production systems of the project from the farm experiments is used to calculate
known as Deep Soils of the Popocatepetl. the risk-aversion parameter K for each farm
Each basic observation consists of informa- household. This marginal productivity is mea-
tion on yield (Y) in kilograms per hectare, ni- sured for the particular soil conditions of each
trogen doses (N) in kilograms per hectare, peasant and for a level of phosphorus corre-
phosphorus doses (PO) in kilograms per hec- sponding, on the expansion path, to the ob-
tare, soil organic matter (M) in percentage, served doses of nitrogen used.
soil phosphorus content (P) in kilograms of The derived marginal productivity together
P20S per hectare, and soil p H (PH) in units. with the coefficient of variation of yield and
Due to the existence of high correlations product and factor prices for 1971 provide a
among the exogenous variables, the param- value of K for each farm household using
eters of the generalized power production equation (3). A sample of the results obtained
function are estimated using ridge regression is shown in table 1. The estimated frequency
(Hoerl and Kennard). The results are log Y = distribution of K obtained from its cumulative
0.98 + 0.34 log N - O.OIM log N + 0.04 log distribution function using Schlaifer's fractile
PO - 0 Plog PO + 0.01 log M + 0.32 log P + rule is shown in figure 1. The results show a
1.71 log PH; R2 = 0.6, ridge parameter » 0.2. distribution of risk aversion highly skewed
The information about fertilizer levels actually toward the risk averters and centered around
used by farmers, as well as location and socio- K = 1.12. This value is close to that estimated
economic characteristics of peasant house- by Scandizzo and Dillon for a similar popula-
holds, was obtained from a random survey tion of subsistence peasants in northeast
conducted in 1971. From this survey, a sample Brazil but following the direct approach of von
of farm households was selected in order to Neumann and Morgenstern.
meet two requirements that would insure that Since Kataoka's safety-first rule implies
the decision model postulated applies. These constrained optima for values of risk a version
requirements are that the farmers knew about greater than zero, the marginal rates of return
fertilizer, had used it in the past, and culti- associated with diminishing degrees of risk
vated maize as their major agricultural activity aversion can be calculated. The results are
and had no effective family labor constraint in shown in table 2. Clearly, the net benefit curve
the sense that the difference between available becomes practically fiat for expenditures
family working time and time allocated to off- above $600 per hectare and the marginal rate
farm, income-generating activities was greater of return drops drastically. If one thinks of net
than the labor requirements to grow com using benefits as returns for risk taking and defines a
the economically optimum level of fertilizers. risk premium as the percentage amount that a
Forty-five of the fifty-five farm households in farmer requires before exposing himself to a
the survey were retained on the basis of these variable income, such a premium is very high
criteria. for an average farmer (K being about 1), thus
Since nitrogen is agronomically the most discouraging nitrogen fertilization rates be-
important input for increasing yields in the yond some 60 kilograms per hectare.
Table 1. Economic Optimum, Farmers' Use of Fertilizer, and Risk-Aversion Measure, Puebla,
1971
Economic Optimum Farmers' Use
Farmer Risk
Identification N N Measure
0.10
Higher levels of education have generally
been associated positively with risk taking. In
0.05 the area under study, the average number of
years of schooling is, however, very low-2.4
0~-=-0.725::-:0:-:.5::::0--:0::-:.7::5:-:-LO~0;:-:'1~.-21-~=.-55:-:0~1.7'::5:-:-2.~OO-:--J- K years-indicating attendance at elementary
Downloaded from http://ajae.oxfordjournals.org/ at Penn State University (Paterno Lib) on May 9, 2016
school without graduating. Due to this low
Figure 1. Frequency distribution of the risk- level of education, the impact of schooling on
aversion measure K risk taking can be expected to be positive but
not of great importance.
Two opposing interpretations can be given
Determinants of Attitudes Toward Risk to the relationship between risk taking and
family size. One is that, the larger the size of
Three classes of variables are used to define the family, the higher the subsistence con-
the socioeconomic characteristics of the peas- sumption needs and, given a fixed amount of
ant households: the nature of the household, land, the lower the willingness of the farmer to
its income-generating opportunities, and its assume risks. In this case, family size reflects
access to public institutions. An explanation the consumption needs of children. Another
for the differential degree of risk aversion interpretation is that family size reflects the
among peasant households can be sought from labor capacity of the peasant household. A
these characteristics. larger family implies greater availability of
Age and years of schooling ofthe household labor on the farm, which is particularly impor-
head and family size are included in the first tant at harvest time when there is a labor
class of variables. It is generally assumed that, shortage in the region and a greater capacity
other factors being the same, older farmers to generate off-farm income. As a result, the
tend to be less prone to take risks than capacity of the farmer to assume risks in-
younger ones. This should be particularly true creases with family size. The data support this
in subsistence agriculture where age can second interpretation: most family members
hardly imply greater on-the-job experience are actually engaged in productive activities.
(which may be thought to be positively asso- A verage family size is 5.4 members with only
ciated with risk bearing), since, on the one 1.5 members below ten years of age.
Table 2. Marginal Rates of Return for Diminishing Degrees of Risk Aversion in Fertilizer Use,
Puebla, 1971
Table 3. Mean and Standard Deviation of Socioeconomic Variables for Farmers with Different
Attitudes Toward Risk, Puebla, 1971
Off- Farm Land
Family Income under Solidarity
Group Age Schooling Size (U.S. $) Control (Ha.) Group (%)
Downloaded from http://ajae.oxfordjournals.org/ at Penn State University (Paterno Lib) on May 9, 2016
Standard deviation 10.0 2.1 2.8 183 1.9
Intermediate risk (N2 = 21)
Mean 48.0 2.8 6.4 135 2.9 14
Standard deviation 14.0 1.7 2.5 378 3.9
High risk (N3 = 18)
Mean 58.0 1.4 4.9 70 1.8 11
Standard deviation 13.0 1.4 3.2 182 1.6
explore quantitatively the relation between Puebla Project with results essentially similar
risk a version and the household characteris- to those presented here (Moscardi).
tics defined above. This multivariate tech- For predictive purposes, a regression equa-
nique is more adequate here than regression tion was run between risk a version and the
analysis for two reasons. First, the interval of same socioeconomic variables characterizing
variation of K is truncated between 0 and 2 for peasant households using a combined sample
the behavioral condition (1) to be satisfied with of seventy-six cases from two regions. As can
a coefficient of variation of yield of 0.5. Sec- be seen from table 5, the results also support
ond, it is useful for the design of rural devel- the hypothesis that the risk-bearing capacity
opment programs to classify peasants into of peasants can be explained by their socio-
groups with similar attitudes toward risk and economic and structural characteristics. Par-
identify the defining socioeconomic charac- ticularly significant for that purpose are the
teristics of those groups. extent of land under control, off-farm income,
Downloaded from http://ajae.oxfordjournals.org/ at Penn State University (Paterno Lib) on May 9, 2016
The risk parameter K has been used to and membership in a solidarity group.
classify peasants into three groups: low-risk (0
< K < 0.4), intermediate-risk (0.4:$ K:$ 1.2),
and high-risk aversion (1.2 < K < 2). The Conclusions
general hypothesis that a systematic relation-
ship exists between attitude toward risk and Estimation of risk aversion for a relatively
the socioeconomic characteristics of peasant large sample of farmers, following the indirect
households, as well as the particular hypothe- method outlined above, shows that risk aver-
sis on each variable discussed above, is re- sion is indeed responsible for substantial dif-
jected or accepted to the extent that the dis- ferences between the demand for fertilizer
criminant function estimated assigns the peas- without risk and actual demand. Risk pre-
ant households to the same groups as does the miums are high, discouraging the use of high
classification variable K. Table 3 shows the rates of fertilizer under safety-first behavior.
mean and standard deviation of the socioeco- Knowledge of the value and distribution of
nomic variables for the three groups defined in K is important for the construction of models
terms of attitude toward risk. Particularly re- of peasant behavior and of aggregate supply
-markable is the association between risk tak- response. Particularly interesting is the fact
ing and off-farm income, land under control, that, following the much less data-demanding
and membership in a solidarity group. indirect method, results similar to those ob-
Table 4 makes it possible to compare the tained by direct use of the von Neumann-
classification obtained by discriminant analy- Morgenstern technique have been obtained.
sis with that established on the basis of risk Beyond random discrepancies, it is possible
aversion. Of the total number of observations to establish a systematic relationship between
assigned to .each group on the basis of risk risk aversion and a number of socioeconomic
aversion, 83% remained in the risk-neutral variables that characterize peasant house-
group, 86% in the low-risk group, and 82% in holds, their access to income-generating op-
the high-risk group. No reclassification oc- portunities, and their relation to public institu-
curred between the extreme groups. This tions. Knowledge of attitude toward risk for
analysis was repeated on a sample of thirty- particular categories of peasants defined by
one peasant households in another area of the these socioeconomic variables makes it possi-
Table 4. Households Classified According to Risk Aversion and Discriminant Analysis, Puebla,
1971
Discriminant Analysis Groups
1 5 I o 6
2 o 15 6 21
3 o 2 16 18
Totalb 5 18 22 45
Table 5. Regression Estimates: Risk Aversion and Managerial Behavior." J. Farm Econ. 42
(1960): 118-32.
versus Socioeconomic and Structural Charac-
Hoerl, A. E., and R. W. Kennard. "Ridge Regression:
teristics Biased Estimation for N onorthogonal Problems."
Technometrics 12 (1970):55-67.
Regression
Variables in Logs Coefficient t-value International Center for Wheat and Corn. The Puebla
Project: Seven Years of Experience, 1967-1973. El
Intercept 2.079950 1.9248 Batan, Mexico, 1974.
Z2
Z, = schooling
= age -0.121252
0.632502 -0.4496
1.0728 Econometrica
Kataoka, S. .. A 31Stochastic
(1963):181-96.
Programming Model."
Z3 = family size -0.011686 -0.0424 Lin, W., G. Dean, and C. Moore ... An Empirical Test of
Z5 = off-farm income -0.091616 -2.0901 Utility vs. Profit Maximization in Agricultural Pro-
Downloaded from http://ajae.oxfordjournals.org/ at Penn State University (Paterno Lib) on May 9, 2016
Z6 = land under control -0.477450 -3.0141 duction." Amer. J. Agr. Econ. 56 (1974):497-508.
Z7 = solidarity group -0.365100 -2.3103
R2 = 0.37 Magnusson, G. Production Under Risk: A Theoretical
F = 6.78
Study. Uppsala, Sweden: Almqvist & Wiksells, 1969.
Moscardi, E. R. .. A Behavioral Model for Decision
ble, in turn, to determine packages of techno- Under Risk Among Small-Holding Farmers." Ph.D.
logical and institutional practices optimally tai- thesis, University of California, Berkeley, 1976.
lored to peasants' economic behavior. Such Officer, R., and A. Halter. "Utility Analysis in a Practi-
packages should greatly enhance the chances cal Setting." Amer. J. Agr. Econ. 50 (1968):257-77.
of success of rural development programs. Pratt, J. W. "Risk A version in the Small and in the
Large." Econometrica 32 (1964): 122-36.
[Received January 1977; revision accepted Pyle, D., and S. Turnovsky. "Safety-First and Expected
May 1977.] Utility Maximization in Mean-Standard Deviation
Portfolio Analysis." Rev. Econ. and Statist. 52
(1970):75-81.
Roy, A. "Safety First and the Holding of Assets."
References Econometrica 20 (1952):431-49.
Scandizzo, P. L., and J. L. Dillon. "Peasant Agriculture
Arrow, K. Essays in the Theory of Risk-Bearing. and Risk Preferences in Northeast Brazil: A Statisti-
Chicago: Markham Publishing Co., 1971. cal Sampling Approach." Paper presented at CIM-
Barraclough, S., ed. Agrarian Structures in Latin Amer- MYT Risk Conference, EI Batan, Mexico, 9-15
ica. Lexington, Mass.: D.C. Heath and Co., 1973. March 1976.
Boussard, 1., and M. Petit. "Representation of Farmers' Schlaifer, R. Probability and Statistics for Business Deci-
Behavior Under Uncertainty with a Focus-Loss sions. New York: McGraw-Hill Book Co., 1959.
Constraint." J. Farm Econ. 49 (1967):869-80. Telser, L. "Safety First and Hedging." Rev. Econ. Stud.
de Janvry, A. "The Generalized Power Production Func- 23 (1955-56): 1-16.
tion." Amer. J. Agr. Econ. 54 (l972a):234-37. von Neumann, J., and O. Morgenstern. Theory of Games
--. "Optimal Levels of Fertilization Under Risk." and Economic Behavior. Princeton, N.J.: Princeton
Amer. J. Agr. Econ. 54 (l972b): 1-10. University Press, 1947.
Dillon, J., and J. Anderson. "Allocative Efficiency in Wiens, T. "Peasant Risk Aversion and Allocative Behav-
Traditional Agriculture and Risk." Amer. J. Agr. ior: A Quadratic Programming Experiment." Amer.
Econ. 53 (1971):26-32. J. Agr. Econ. 58 (1976):629-35.
Edwards, W. "Behavioral Decision Theory." Annu. Rev. Wolgin, J. "Resource Allocation and Risk: A Case Study
Psychol. 12 (1961):473-98. of Smallholder Agriculture in Kenya." Amer. J. Agr.
Halter, A., and C. Beringer. "Cardinal Utility Functions Econ. 57 (1975):622-30.