Entrepreneur: Start & Grow Your Business

A system comparison approach to distinguish two nonseparable and nonnested agricultural household models.


by Sonoda, Tadashi

Numerous empirical studies have emphasized the importance of nonseparable agricultural household models (AHM), in which a farm household simultaneously determines its production organization with its consumption choice (Bardhan and Udry 1999). Those studies devoted their efforts exclusively to showing the superiority of a nonseparable model to the corresponding separable one. Furthermore, they were able to relate the separable model to a special case of the nonseparable one. For example, Jacoby (1993) tested a relation by which the shadow wage is equal to the market wage in the separable model. Benjamin (1992) tested a relation by which the demand for farm labor is unresponsive to changes in household composition in the separable model. Using that simple relation, they used a nested test to make a partial comparison of the two models. (1)

However, no studies have compared two nonseparable models to show that one is superior to the other. This consequence arises probably because we cannot expect a simple relation in comparing two nonseparable models. Instead, we will observe a less familiar, nonnested relation, which can be inferred by comparing the following two common hypotheses, with special attention paid to labor supply functions and shadow or "internal wages" (Sonoda and Maruyama 1999).

One hypothesis, corresponding to heterogeneous labor supply, is denoted as "HET." Under this hypothesis, members of a farm household confront different disutilities from working on and off the farm: they supply heterogeneous farm and nonfarm labor. In addition they freely supply their time to the market for nonfarm labor, but the household does not demand nonfamily farm workers and operates its farm by self-employment. Lopez (1984) tested this hypothesis for Canadian farm households and found the associated nonseparable AHM superior to the corresponding separable one.

The other hypothesis corresponds to a restricted labor market and is denoted "RES." Under this hypothesis members of a farm household are indifferent to working on and off the farm; they supply a single type of total labor (the sum of farm and nonfarm labor). In addition the household operates its farm by self-employment just as under the HET hypothesis. But its members face restricted hours of nonfarm work because nonfarm employers offer a higher than equilibrium wage to use the resulting excess supply of labor as a worker discipline device. Sonoda and Maruyama (1999) tested it for farm households in Japan and found a similar result to Lopez (1984).

Under the HET hypothesis, the household has separate supply functions of farm and nonfarm labor with their respective own wages--an internal wage for farm labor and the market wage for nonfarm labor. In this case we find the internal wage to be relevant to the household simply because it operates its farm by self-employment. Under the RES hypothesis, the household has a single supply function of total labor with a single own wage--another internal wage for total labor. Now we find the internal wage to be relevant to the household, partly because it operates its farm by self-employment and partly because it faces restricted hours of nonfarm work. In this way we face serious inconvenience in comparing AHM under the two hypotheses: they include distinct internal wages and have different numbers of labor supply functions.

For appropriate comparison of those models, it is necessary to apply a nonnested test to determine the distinct internal wages in the two models. Furthermore, comparison of AHM under the HET and RES hypotheses is practical for evaluating the behavior of Japanese rice-farming households, as explained below. For these reasons this study demonstrates a method of distinguishing nonseparable AHM under the two hypotheses. In addition this paper provides an economic reason why we need to distinguish between them in terms of their comparative statics analysis.

We address this issue using data from Japanese rice-farming households during 1982-1991. Sonoda and Maruyama (1999) used the same data to find AHM under the RES hypothesis to be a better model than the corresponding separable AHM. However, the following observations suggest further support for AHM under the HET hypothesis.

During Japan's period of high economic growth (from mid 1950s to early 1970s), opportunities for nonfarm employment increased in and near rural areas. In addition mechanized systems in rice farming had been widely adopted by the early 1980s. According to Hayami (1986), rice-farming households have adopted a division of labor within the household in adapting to these situations: adult males primarily work at nonfarm firms, whereas housewives and elderly household members play a major role in farming. (2) In addition their small paddy field (about one hectare on average) allows them to operate their farm through self-employment. (3) This situation seems to be well described by specialization within the household: one member works off the farm for a market wage, and others work on the farm through self-employment. In this case there is no reason for the market wage (of one person) to be equal to the internal wage (of other family members).

The next section compares optimality conditions for AHM under the HET and RES hypotheses and explains the importance of their distinctions. The third section introduces a system comparison approach to distinguish nonseparable AHM under them. The fourth section applies the framework to data of Japanese rice-farming households and reveals the HET hypothesis to be better. That section also presents a comparison of the respective elasticities of the internal wages and quantity variables under the two hypotheses. The final section presents salient conclusions of this study.

Comparison of AHM under the HET and RES Hypotheses

Japanese rice-farming households operate their farms on small paddy fields; many of them depend solely on workers from their own families. For this reason we assume that a farm household does not hire nonfamily workers.

We first introduce AHM under the HET hypothesis. A farm household devotes Lf hours to its own farm and supplies [L.sub.m] hours to market work at nonfarm firms. A distinctive assumption under HET is heterogeneity of the two types of time: (4)

(1) u = u(C, [t.sub.f], [t.sub.m], G), [t.sub.f] [equivalent to] [T.sub.f] - [L.sub.f] > 0, [t.sub.m] = [T.sub.m] - [L.sub.m] >0

where C and G, respectively, denote the amount of purchased consumption commodities and the vector of household characteristics; [T.sub.f] and [t.sub.f] ([T.sub.m] and [t.sub.m]), respectively, represent the endowed time and leisure hours of farm (nonfarm) workers.

Another distinctive assumption under HET is that the farm household can freely supply its time to nonfarm firms at the market wage w. When it also uses competitive markets for consumption and farm commodities and variable production factors other than family labor (seed and seedlings, fertilizers, feed, agricultural chemicals, fuel, light, heat, and processing materials), the household faces the following budget constraint:

(2) rC = pX - qF + w[L.sub.m] + V

where X and F, respectively, denote the amounts of farm commodity and variable inputs other than labor. Also, p, q, and r, respectively, denote the market prices of farm commodity, other inputs and consumption commodities; V denotes nonlabor income.

Furthermore, farm production technology of this household is expressed as:

(3) X = f([L.sub.f], F, K)

where the vector K includes fixed inputs (farm machinery and land) and shift factors (including policy variables) of the function f(x).

When the farm household maximizes its utility function (1) subject to constraints (2) and (3), the optimality conditions are written as:

(4) p([partial derivative]f/[partial derivate][L.sub.f]) = [w.sup.*]

(5) p([partial derivative]f/[partial derivative]F) = q

(6) X = f([L.sub.f], F, K)

(7) [partial derivative]u/[partial derivative]C = [lambda]r

(8) [partial derivative]u/[partial derivative][t.sub.f] = [lambda][w.sup.*]

(9) [partial derivative]u/[partial derivative][t.sub.m] = [lambda]w

and

(10) rC + [w.sup.*][t.sub.f] + [wt.sub.m] = [M.sup.HET] [equivalent to] pX - [w.sup.*][L.sub.f] - qF + [w.sup.*][T.sub.f] + w[T.sub.m] + V

where [lambda] denotes the Lagrange multiplier that is associated with budget constraint (2). (5) The "internal wage" [w.sup.*] satisfies the following relation:

(11) p([partial derivative]f/[partial derivative][L.sub.f]) = [w.sup.*] = r([partial derivative]u/[partial derivative][t.sub.f])/([partial derivative]u/[partial derivative]C).

Next, we introduce AHM under the RES hypothesis. A distinctive assumption under RES is the homogeneity of farm and nonfarm labor, which might be expressed as:

(12) u = u(C, [t.sub.t], G), [t.sub.t] [equivalent to] [T.sub.t] - [L.sub.f] - [L.sub.m] > 0

where [T.sub.t] and [t.sub.t], respectively, denote the endowed time and leisure hours of all workers.

Another distinctive assumption under RES is that the farm household faces a restricted market for nonfarm labor: it supplies fixed hours [[bar.L].sub.m] of nonfarm labor at the going wage w. When it uses competitive markets for consumption and farm commodities and other variable inputs, its budget constraint is expressed as:

(13) rC = pX - qF + w[[bar.L].sub.m] + V.

When the farm household maximizes its utility function (12) subject to constraints (3) and (13), the optimality conditions are written as:

(14) p([partial derivative]f/[partial derivative][L.sub.f]) = [w.sup.**]

(15) p([partial derivative]f/[partial derivative]F) = q

(16) X = f([L.sub.f], F, K)

(17) [partial derivative]u/[partial derivative]C = [[lambda]'r

(18) [partial derivative]u/[partial derivative][t.sub.t] = [lambda]'[w.sup.**]

and

(19) rC + [w.sup.**][t.sub.t] = [M.sup.RES] [equivalent to] pX - [w.sup.**][L.sub.f]- qF + [w.sup.**][T.sub.t] + (w - [w.sup.**])[[bar.L].sub.m] + V

where [lambda]' denotes the Lagrange multiplier that is associated with budget constraint (13). (6) The internal wage [w.sup.**] satisfies the following relation:

(20) p([partial derivative]f/[partial derivative][L.sub.f]) = [w.sup.**] = r([partial derivative]u/[partial derivative][t.sub.t])/([partial derivative]u/[partial derivative]C).

We next compare the optimality conditions under the HET and RES hypotheses. First, the endogenous internal wage appears on both production and consumption sides of the conditions. Therefore, the AHM under the two hypotheses are both nonseparable.

Second, the two internal wages are determined in a different manner. Equation (11) shows that [w.sup.*] is determined to equate the value of the marginal product of farm labor (MPL) with the marginal rate of substitution (MRS) that is associated with farm labor. On the other hand (20) shows that [w.sup.**] is determined to equate the same MPL to the MRS that is associated with total labor.

Third, when we specifically examine the production side, (5) and (6) coincide with (15) and (16). Equation (4) includes a different internal wage [w.sup.*] from another internal wage [w.sup.**] in (14), but both are unobservable. Therefore, system (4)-(6) seems to be indistinguishable from system (14)-(16) in empirical analyses.

Finally, when we specifically examine the consumption side, (7)-(10) under the HET hypothesis yield a demand system of three commodities. Equations (17)-(19) under the RES hypothesis yield a demand system of two commodities. In particular the former yield supply functions of farm and nonfarm labor with their respective own wages [w.sup.*] and w, whereas the latter yield a supply function of total labor with its own wage [w.sup.**].

Consequently, we can distinguish AHM under the HET and RES hypotheses only on their consumption side. The remainder of this section describes the importance of their distinction by examining responses of the internal wages and quantity variables to a fall in the price r of the consumption commodity.

Figure 1(a) shows how internal wage [w.sup.*] is determined under the HET hypothesis. Therein, [L.sup.D([tau]).sub.f] and [L.sup.S([tau]).sub.f] , respectively, represent the demand and supply curves of farm labor at time [tau] (= 0, 1); they determine the work hours [L.sup.(tau)].sub.f] and internal wage [w.sup.*([tau])]. In addition, [L.sup.D([tau]).sub.m] and [L.sup.S([tau]).sub.m] represent similar curves of nonfarm labor, and they determine work hours [L.sup.([tau]).sub.m]. The curve [L.sup.D([tau])].sub.m] is a horizontal line at the competitive market wage [w.sup.([tau])], while the curve [L.sup.S([tau]).sub.m] is depicted as a vertical line for ease of visual interpretation. (7)

A fall in the price r only affects the supply curves and causes them to shift leftward if the income effect is greater than the substitution effect. Consequently, the internal wage rises from [w.sup.*(0)] to [w.sup.*(1)]. We follow Sonoda and Maruyama (2000) to express the response of the internal wage [w.sup.*] to a change in an exogenous variable s as:

(21) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

Therein, [L.sup.S.sub.f] [equivalent to] [T.sub.f]-[t.sub.f] defines the supply of farm labor, and [([partial derivative]Q/[partial derivative]s).sub.const] represents the response of quantity Q with the internal wage fixed. On the right-hand side of (21), terms in the numerator represent shifts in the demand and supply functions of farm labor, whereas terms in the denominator represent their slopes.

Figure 1(b) shows how internal wage [w.sup.**] is determined under the RES hypothesis. The demand curve for total labor comprises the demand curve [L.sup.D([tau]).sub.f] for farm labor and that for nonfarm labor (the horizontal segment of length [[bar.L].sub.m] at w = [w.sup.([tau])]). This composite curve and the supply curve [L.sup.S(tau)].sub.t] determine total work hours [L.sup.([tau]).sub.t] = [L.sup.([tau]).sub.f] + [[bar.L].sub.m] and internal wage [w.sup.**([tau])]. Compared with figure l(a), curve [L.sup.S([tau])].sub.t] has a gentler slope than curve [L.sup.S([tau]).sub.m] but has a steeper slope than curve [L.sup.S([tau]).sub.f]. In addition work hours and internal wages at time 0 have identical levels in figures 1(a) and (b).

A fall in the price r only affects the supply curve. We set its shift equal to the sum of the shifts in farm and nonfarm labor in figure l(a). Then, the internal wage rises sharply from [w.sup.**(0)] to [w.sup.**(1)]. Consequently, [w.sup.**] is expected to respond more than [w.sup.*] under the assumptions described above, which will be verified in the empirical analysis. We express the response of internal wage [w.sup.**] to a change in an exogenous variable s as:

(22) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [L.sup.S.sub.t] [equivalent to] [T.sub.t] - [t.sub.t] denotes the total supply of labor.

[FIGURE 1 OMITTED]

Different responses of the internal wages in (21) and (22) engender different responses of quantity variables, which might be expressed as:

(23) [partial derivative]Q/[partial derivative]s = [([partial derivative]Q/[partial derivative]s).sub.const] + ([partial derivative]Q/[partial derivative][w.sup.*])([partial derivative][w.sup.*]/[partial derivative]s)

and

(24) [partial derivative]Q/[partial derivative]s = [([partial derivative]Q/[partial derivative]s).sub.const] + ([partial derivative]Q/[partial derivative][w.sup.*])([partial derivative][w.sup.**]/[partial derivative]s)

On the respective right-hand sides of (23) and (24), [([partial derivative]Q/[partial derivative]s).sub.const] represents a direct effect of a change in an exogenous variable s, whereas ([partial derivative]Q/[partial derivative][w.sup.*(*)])([partial derivative][w.sup.*(*)]/[partial derivative]s) reflects an internal wage effect. The former is identical, but the latter differs under HET and RES. In particular if [w.sup.**] is more responsive than [w.sup.*], as shown in figure 1, the internal wage effect tends to be greater under the RES hypothesis.

A System Comparison Approach to Distinguish AHM under HET and RES

We employ a method of two-step estimation for our nonseparable models, as did Sonoda and Maruyama (1999). When estimating AHM under HET, we first simultaneously estimate the optimality condition (5) for other variable inputs and the production function (6). This conjecture yields estimated parameters of the production function, which in turn yield the estimated internal wage [[??].sup.*] as [w.sup.*] = p([partial derivative]f/[partial derivative][L.sub.f]) and the estimated full income [[??].sup.HET] from its definition. Subsequently, we estimate a system of commodity demand functions that is derived from optimality conditions (7)-(10), given [[??].sup.*] and [[??].sup.HET]. We employ a similar method for estimating AHM under RES.

In this case it is natural to compare production and consumption sides of AHM separately. Based on the previous discussion, we compare only the consumption side of AHM under HET and RES. Before starting the comparison, we specify the production technology and consumption preference and explain their estimation method.

Specification and Estimation Method of the Production Side

Let ALL = {[L.sub.f], F, K} be composed of all production and shift factors. The vector K consists of the real capital stock RK, total area planted A, the intensity rate SAP of the set-aside program (see the next section) and ETT = exp(TT) (TT: time trend). Also, let ALL/{F} be constructed by excluding F from ALL. We specify the production function as:

(25) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

In this case the optimality condition for other variable inputs is expressible as:

(26) qF/pX = [b.sub.F] + [summation over (z[member of]ALL])] [b.sub.F,z] ln z.

We estimate (25) and (26) simultaneously using the generalized method of moments (GMM) under both the HET and RES hypotheses. Consequently, we obtain the same estimates of the production function parameters and the same estimated internal wage, [[??].sup.*] = [[??].sup.**] = p([partial derivative]f/[partial derivative][L.sub.f]), under the two hypotheses.

Specification, Estimation Method, and Comparison of the Consumption Side

Under the HET hypothesis we obtain three demand functions for consumption commodities and leisure hours of farm and nonfarm workers. Similarly to Shively and Fisher (2004), we specify an almost ideal demand system (AIDS) and estimate the following two share equations for leisure hours of farm and nonfarm workers:

(27) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

Under the RES hypothesis we obtain two demand functions for consumption commodities and total leisure hours. We specify an AIDS model and estimate the following single share equation for total leisure hours:

(28) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

The two demand systems are estimated using GMM after imposing adding-up, homogeneity and symmetry restrictions.

Next, we compare systems (27) and (28). They differ not only in their dependent and independent variables but also in the number of equations. Consequently, they are complicatedly nonnested, as stated in the first section. Nonnested models have often been tested using the J test of Davidson and MacKinnon (1981) or the Cox test of Pesaran (1974).

The J test is usually used to compare systems that have the same number of equations and the same dependent variables. Using that test, Lopez (1984) attempted to compare separable and nonseparable models which have different numbers of equations. His method requires a complicated transformation of the models to apply the J test (see [22] of Lopez). Moreover, this transformation might not maintain properties of the original models, which might engender inappropriate comparisons of the models. (8)

In contrast the Cox test does not require this type of strict correspondence between two systems; instead, it compares their likelihoods. As Mizon and Richard (1986) indicate, it can be interpreted as an encompassing test that evaluates the extent to which the null model explains an important characteristic (likelihood, parameters, etc.) of the alternative. In our case we can use a Cox-type test of Smith (1992), which compares their GMM objective functions instead of their respective likelihoods.

We evaluate the GMM objective function for model (28) in two ways when we test the null model (27) against the alternative (28). (9) One is to evaluate it on the assumption that model (28) is correct. The other is to evaluate its probability limit, assuming that model (27) is correct. We interpret that model (27) can explain model (28) if the difference between the two values is sufficiently small. Smith (1992) shows that the difference is asymptotically normal. The definition and computation of the Cox-type statistic are complicated. For that reason they are explained in the Appendix.

Empirical Analysis

Although the use of micro data is preferable in estimating AHM, this study uses aggregate data for the following reasons. Lopez (1984) and Sonoda and Maruyama (1999) use aggregate data to estimate AHM under the HET and RES hypotheses. Furthermore, most Japanese studies have used aggregate data to estimate AHM (e.g., Kuroda and Yotopoulos (1980) and Arayama (1986)). For that reason, estimating AHM under the two hypotheses using Japanese aggregate data at least makes practical sense.

Use of aggregate data renders the discrete choices of farm households unobservable, which might cause a major problem. In our case the problem is related to whether Japanese farm households employ nonfamily workers in rice farming or not, and whether their members work at nonfarm firms or not. As inferred from the hours of hired farm labor and the number of nonfarm workers shown in table 1, only a very limited number of farm households use significant hours of nonfamily labor and supply none of their members' labor to nonfarm firms. Consequently, we are not likely to observe a serious difference between the two AHMs estimated from our aggregate (but not so highly aggregated) data and those estimated from the original micro-data.

Data

We use aggregate data for rice-farming households in Japan, which are mainly adapted from the Survey of Farm Household Economy by Types of Farm Households (FHET) (Japan, Ministry of Agriculture, Forestry and Fisheries 1982-1991). We also use data from the Statistics of Prices and Wages in Rural Areas (PWRA) (Japan, Ministry of Agriculture, Forestry and Fisheries 1982-1991) and the Annual Report on the Consumer Price Index (RCPI) (Japan, Management and Coordination Agency 1982-1991).

Annual aggregate data are available for seven paddy-scale classes in eight regions for 1982-1991. (10) The paddy-scale classes include 0.5-1.0, 1.0-1.5, 1.5-2.0, 2.0-2.5, 2.5-3.0, 3.0-5.0, and 5.0 ha and larger. The regions include Tohoku, Hokuriku, Kanto-Tosan, Tokai, Kinki, Chugoku, Shikoku, and Kyushu, which cover almost the entire area of Japan. We would have 10 (years) x 7 (scale classes) x 8 (regions) = 560 observations if data were to constitute a complete time-series cross-section. However, our incomplete data include only 283 observations, reflecting the actual situation: a small number of farm households operate large paddy areas. (11)

Table 1 shows the means and standard deviations of variables used in empirical analyses. The price of consumption commodities is obtained from the general index by region in RCPI. The Divisia price indexes of other variable inputs and capital goods are constructed using expenditures for relevant items in FHET and the price indexes for the same items in PWRA. Using FHET data, the price of rice is obtained by dividing the gross revenue derived from rice sales by the amount of rice produced.

The nonfarm wage rate is obtained by dividing the salaries and wages in nonagricultural gross income by the total hours of nonfarm work. Empirical studies employing micro-data commonly replace the wage rate, when thus constructed, with the predicted one in estimating their labor supply function to allow for measurement errors in work hours (e.g., Sahn and Alderman 1988). Instead, we choose a standard solution to the measurement error problem in estimating AIDS models: we use a set of instrumental variables that includes exogenous factors affecting the nonfarm wage. One reason for our choice is that variations in a predicted wage are smaller than those in the original wage, particularly when we use aggregate data, which often makes estimated parameters less precise. Also, the FHET surveys work hours of farm households by asking them to record their daily work hours throughout the year, from which we infer that those work hours do not include overly large errors.

The amount of rice produced, hours of family farm labor and total area planted are obtained from FHET. Quantities of other variable inputs, capital stock and purchased commodities are obtained by dividing their expenditures or value in FHET by their respective price indexes. The endowed time of farm, nonfarm and total workers is estimated as [T.sub.i] = 16 x 365 x [N.sub.i] (i = f, m, t), where [N.sub.f] and [N.sub.m], respectively, represent the numbers of farm and nonfarm workers; also, [N.sub.t] [equivalent to] [N.sub.f] + [N.sub.m]. Other production costs OC aside from the cost qF of other variable inputs and the number NF of household members are available from FHET for subsequent analysis.

Finally, a set-aside program in Japan was introduced to assuage the mounting surplus of rice in the late 1960s. Since that time, the government has set annual targets for fallow paddy areas, and rice-farming households have complied with that program to receive compensatory payments and avoid a sharp decline in the price of rice. The intensity rate SAP of the set-aside program is defined by the farm household's fallow paddy area as a fraction of its total paddy area. Its data are obtained from FHET.

Estimation of Production Side and Variations in the Internal and Non farm Wages

To deal with our incomplete time-series cross-section data, we simply assume fixed effects and express the coefficients [b.sub.x] and [b.sub.F] in (25) and (26) as:

(29) [b.sub.z] = [b.sub.z,0] + [summation over (r)] [[pi].sub.z,r][RD.sub.r] + [summation over (s)][[rho].sub.z,s][SD.sub.s] (z = X, F;r = 1, ..., 7;s = 1, ..., 6)

where [RD.sub.r] and [SD.sub.s], respectively, denote regional and paddy-scale dummies. After substituting (29) into (25) and (26), we estimate them using GMM.

In (25), (26) and (29), exogenous variables suitable for instrumental variables are a constant term, lnRK, lnA, lnSAP, TT, p, q, [RD.sub.r], and [SD.sub.s]. Substitution of (29) into (25) produces cross terms between dummies and endogenous regressor lnF. Therefore, we construct a basic set of instruments by adding [RD.sub.r] x TT and [SD.sub.s] x TT to the exogenous variables described above. (12) Furthermore, to allow for cross and quadratic terms between endogenous regressors lnF and ln[L.sub.f] in (25), we construct cross and quadratic terms among lnRK, lnA, lnSAP, TT, p, and q and combine them with the basic set in various ways. We eventually choose plnRK, plnA, plnSAP, [(In A).sup.2], [p.sup.2], qlnRK, and qlnA to combine them with the basic set. (13) The combined set yields the estimation results shown in table 2, (14) which: satisfies monotonicity (in all inputs) and concavity (in variable inputs) of the production function; passes the overidentifying restrictions (OIR) test of Hansen (1982) at the 5% level; and yields the most stable t-values of the parameters.

We can estimate the internal wage from this result, table 1 shows the mean of the internal wage (687 yen/hour) at about one-half that of the nonfarm wage (1302 yen/hour). The remainder of this subsection describes these wages as exhibiting different and sufficient variations for our purpose of identifying the two demand systems. Detailed variations in the internal wage are examined systematically in the final subsection.

The sources of variation in the nonfarm wage w are attributed to education, age, sex, region, and macroeconomic factors, whereas those in internal wage [[??].sup.*] are attributed to all factors affecting the farm household's behavior. Consequently, the latter is expected to vary in a more complicated manner. Actually, the coefficients of variation of w and [[??].sup.*] are 0.22 and 0.26, respectively, and their correlation coefficient is 0.21.

Next, we examine variations in nonfarm wages in greater detail. The FHET does not provide data for education, sex, and age. Therefore, we attempt to explain those variations using our available information. First, the FHET provides data showing the shares of white-collar permanent workers among all nonfarm workers. According to Fukui (1993), this type of variable is useful as a proxy for the educational level because farm household members with a higher education level tend to engage in white-collar jobs. Second, the Survey of the Farm Household Economy (Japan, Ministry of Agriculture, Forestry and Fisheries 1982-1991) (15) shows that the share of males among nonfarm workers is quite stable during 1982-1991, at about 60% for the eight regions examined here. Therefore, a gender-related variable does not vary so much and might not strongly affect variations in the nonfarm wage. Finally, it might be difficult to identify the effects of age for our aggregate data because we often expect two nonfarm workers in each farm household, as shown in table 1.

With this information in mind, the logarithm of the nonfarm wage is regressed on a constant term, regional dummies, time trend and the shares of white-collar and blue-collar workers. The result shows that all coefficients are significant at the 5% level, and the adjusted coefficient of determination is 0.79. All regional dummies have significant coefficients because the eight regions cover almost the entire area of Japan, and regional differences in the nonfarm wage have expanded along with economic development. The time trend has a positive coefficient, implying an upward trend in nonfarm wages. Finally, both shares of white-collar and blue-collar workers have positive coefficients; the coefficient for the former is 2.7 times larger than that for the latter, verifying our previous inference related to educational level.

Estimation and Comparison of Consumption Systems

When we allow for fixed effects, the farm household's characteristics and its taste changes, we specify coefficient [[alpha].sub.i] (i = f, m, t) in (27) and (28) as:

(30) [[alpha].sub.i] = [[alpha].sub.i,0] + [[alpha].sub.i,2]FSHARE + [[alpha].sub.i,3]TT + [summation over (r)] [[pi].sup.i.sub.r][RD.sub.r] + [summation over (s)][[rho].sup.i.sub.s][SD.sub.s]

where NF and FSHARE, respectively, represent the number of household members and the share of farm workers among all workers. After substituting (30) into (27) and (28) and imposing the theoretical restrictions, we estimate each system using nonlinear GMM with the [[alpha].sub.0] parameter in ln [P.sup.k] (k = HET, RES) set equal to zero.

In (27), (28) and (30), exogenous variables suitable for instruments are: a constant term, r, w, NF, [N.sub.f], [N.sub.m], TT, [RD.sub.r], and [SD.sub.s], where [N.sub.f] and [N.sub.m] are used in place of FSHARE = [N.sub.f]([N.sub.f] + [N.sub.m]) to obtain precise estimates of parameters. These variables are combined with exogenous variables included in the full income (ln RK, lnA, ln SAP, p, q, and OC) to construct a set IV-1 of instruments. (16) Hours [L.sub.m] of nonfarm work are exogenous only under the RES hypothesis. Therefore, this variable is not included in view of the criterion stated in the Appendix. In addition we use another set IV-2 to take care of measurement errors in the nonfarm wage. This set is constructed by replacing the wage w in set IV-1 with the shares of white-collar and blue-collar workers in nonfarm workers. (17)

The estimation results satisfy concavity in all the cases when we estimate the demand systems (27) and (28) using sets IV-1 and IV-2. Three of the four results pass the OIR test at the 5% level, and the remaining one (system [28] for IV-l) passes it at the 1% level. Furthermore, most coefficients are significant at the 5% level.

Using these results, we can compute the Cox-type statistic in (A.1) of the Appendix. In testing the HET hypothesis against RES, we noted that the statistic [[lambda].sub.HET,RES] is equal to 1.17 [0.24] for set IV-1 and 0.87 [0.39] for set IV-2, where the p-value is shown in brackets. Conversely, in testing the RES hypothesis against HET, we found that the statistic [[lambda].sub.RES,HET] is equal to 4.47 [0.00] for set IV-1 and 2.98 [0.00] for set IV-2. Therefore, the RES hypothesis is rejected, but the HET hypothesis is not rejected for both sets of instruments. We conclude that the HET hypothesis is more appropriate in our statistical tests.

Finally, our Cox-type test used "generated regressors" (the estimated internal wage and full incomes), which might engender biases in nonnested tests, as McAleer and McKenzie (1991) suggest. One way to examine this possibility might be to use a similar method to Newey's (1984) and compare uncorrected and corrected standard errors of the parameters. table 3 compares two sets of t-values of the AIDS parameters under the HET hypothesis, where the set IV-2 of instruments is used for estimation. (18) The two sets of t-values differ greatly for about one-third of the parameters, but they do not give different results in tests of significance at the 5 % level, suggesting that biases in our nonnested test are not serious. Similar results are verified under the RES hypothesis.

Comparative Statics Analysis

We now compare price elasticities of internal wages and quantity variables under the HET and RES hypotheses. We can use (21)-(24) for this purpose. The responses of factor demand and output supply on the right-hand sides of these equations are obtained by differentiating the optimality conditions (4)-(6) (or [14]-[16]) and using estimated parameters of the production function. These responses are identical under the HET and RES hypotheses because we use the same estimation method on the production side.

Similar responses of labor supply are obtained in the following way. For our AIDS models we can express the full income elasticity of leisure demand: [t.sub.i] (i = f, m, t), under hypothesis k (= HET, RES), as [[eta].sup.k.sub.i] [equivalent to] [partial derivative] ln [t.sub.i]/[partial derivative]ln[M.sup.k] = 1 + ([[beta].sub.i]/[S.sup.k.sub.i]). Under the HET hypothesis price elasticities of leisure demand: [t.sub.i] (i = f, m) are then expressible as: (19)

(31) [([partial derivative] ln [t.sub.i]/[partial derivative] ln y).sub.const] = [([partial derivative] ln [t.sub.i]/[partial derivative] ln y).sub.M] + y [H.sub.y] [[eta].sup.HET.sub.i] / [M.sup.HET] (y = p, q, r, w, [w.sup.*]).

The first term on the right-hand side represents the price elasticity of leisure [t.sub.i] with full income fixed, whereas the second term represents the effect of changes in full income, with [H.sub.p] = X, [H.sub.q] = -F, [H.sub.r] = O, [H.sub.w] = [T.sub.m], and [H.sub.w*]. = [T.sub.f] - [L.sub.f]. For example, [([partial derivative] ln [t.sub.f]/[partial derivative] ln [w.sup.*]).sub.M] = -1 + {[[gamma].sub.f,f] - [[beta].sub.f]([[alpha].sub.f] + [[gamma].sub.f,f] ln [w.sup.*] + [[gamma].sub.m,f] ln w + [[gamma].sub.c,f] ln r)}/[S.sub.f].

Under the RES hypothesis, price elasticities of leisure demand are expressed as:

(32) [([partial derivative] ln [t.sub.t]/[partial derivative] ln y).sub.const] = [([partial derivative] ln [t.sub.t]/[partial derivative] ln y).sub.M] + y[R.sub.y][[eta].sup.RES.sub.t]/[M.sup.RES] (y = p, q, r, w, [w.sup.**])

where [R.sub.p] = X, [R.sub.q] = - F, [R.sub.r] = O, [R.sub.w] = [L.sub.m], and [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

Price elasticities of labor supplies are obtained by multiplying those of leisure demands by - [t.sub.f]/[L.sup.S.sub.f] or -[t.sub.m]/[L.sub.m] under HET and by -[t.sub.t]/[L.sup.S.sub.t] under RES.

The price elasticities of quantity variables thus evaluated, along with their nonlabor income elasticities, are shown in panel (i) of table 4. Elasticities in this panel are substituted into (21) or (22) to yield elasticities of the internal wages, which are in turn substituted into (23) or (24) to yield elasticities of quantity variables, including internal wage effects. Their computed values are shown in panel (ii) for the HET hypothesis and in panel (iii) for the RES hypothesis.

First, panel (i) of table 4 shows that the internal wage elasticity of farm labor supply, 0.38, is much greater than that of the total labor supply, 0.14. Therefore, the supply function of total labor is actually steeper than that of farm labor, as depicted in figure 1. This result reflects the fact that household members work much longer hours at nonfarm firms than on their farm. It also reflects the fact that the supply of nonfarm labor responds less flexibly than that of farm labor.

One rationale for the latter is that it seems more difficult for self-employed farmers to adjust their time at nonfarm firms than on their own farm. Another explanation suggests that principal (nonfarm) workers are likely to supply their labor less flexibly than nonprincipal (farm) workers, reflecting a commonly observed pattern of labor division mentioned in the first section.

Next, the market price elasticities of the total labor supply are greater in absolute value than that of the farm labor supply when the internal wages are fixed, as shown in panel (i). These relations imply that a shift in the supply curve [L.sup.S.sub.t] of total labor is greater than the corresponding shift in the supply curve [L.sup.S.sub.f] of farm labor in figure 1. We note two points to interpret that difference. First, panel (i) of table 4 shows that the difference in shifts among the supply functions of farm, nonfarm and total labor clearly reflect the size of their respective income elasticities. (20) Second, the estimated income elasticity of the total labor supply -0.35 is inferred to be biased downward because it is greater in absolute value than both the income elasticities of farm and nonfarm labor supply -0.26 and -0.21. These points suggest that the greater shifts of the supply function of total labor result from the biased income elasticity of the total labor supply.

In view of (21) and (22) as well as figure 1, the steeper slope and the greater shifts of the supply function of total labor imply that the internal wage [w.sup.**] responds more sensitively to changes in market prices than the other internal wage [w.sup.*] does. Panels (ii) and (iii) of table 4 verify this relation.

In view of (23) and (24), such greater elasticities of internal wage [w.sup.**] cause greater internal wage effects on quantity variables. Actually, most elasticities of demand for farm labor and other variable inputs include much greater internal wage effects in panel (iii) than in panel (ii). The elasticity of output supply includes moderately greater internal wage effects in panel (iii) because internal wage effects on the demands for farm labor and other variable inputs are mutually offsetting to some degree.

Conclusion

This paper describes a system comparison approach to distinguish nonseparable AHM under the HET and RES hypotheses. Using a two-step estimation, we found their consumption side to be distinguished because they yield demand systems that not only have different dependent variables but also different numbers of equations. We proposed an empirical procedure to apply the Cox-type test of Smith (1992) to make an appropriate comparison of the nonnested systems.

Our specific comp