agricultural & applied economics association - … · agricultural & applied economics...

14
Agricultural & Applied Economics Association A Dynamic Analysis of Land Prices Author(s): Jean-Paul Chavas and Alban Thomas Source: American Journal of Agricultural Economics, Vol. 81, No. 4 (Nov., 1999), pp. 772-784 Published by: Oxford University Press on behalf of the Agricultural & Applied Economics Association Stable URL: http://www.jstor.org/stable/1244323 . Accessed: 11/04/2013 14:18 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. . Agricultural & Applied Economics Association and Oxford University Press are collaborating with JSTOR to digitize, preserve and extend access to American Journal of Agricultural Economics. http://www.jstor.org This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PM All use subject to JSTOR Terms and Conditions

Upload: phamkhanh

Post on 20-May-2018

214 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

Agricultural & Applied Economics Association

A Dynamic Analysis of Land PricesAuthor(s): Jean-Paul Chavas and Alban ThomasSource: American Journal of Agricultural Economics, Vol. 81, No. 4 (Nov., 1999), pp. 772-784Published by: Oxford University Press on behalf of the Agricultural & Applied Economics AssociationStable URL: http://www.jstor.org/stable/1244323 .

Accessed: 11/04/2013 14:18

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

Agricultural & Applied Economics Association and Oxford University Press are collaborating with JSTOR todigitize, preserve and extend access to American Journal of Agricultural Economics.

http://www.jstor.org

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 2: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

A DYNAMIC ANALYSIS OF LAND PRICES

JEAN-PAUL CHAVAS AND ALBAN THOMAS

A dynamic model of land prices is developed. It derives arbitrage asset prices under nonadditive

dynamic preferences, risk aversion, and transaction costs. The model nests as special cases risk

neutrality, time-additive preferences, the static capital asset pricing model (CAPM), as well as the dynamic consumption-based CAPM. The model is applied to the analysis of U.S. land prices for the period 1950-96. The econometric results provide evidence showing that U.S. land price patterns are inconsistent with risk neutrality or with the static CAPM model. No strong evidence was found against time-additive preferences. The econometric findings indicate that both risk aversion and transaction costs have significant effects on land prices.

Key words: asset pricing, land, risk, transaction cost.

Agricultural land prices in the United States have fluctuated widely over the last decades. They have been subject to a sharp apprecia- tion in the 1970s, followed by a significant depreciation in the early 1980s. This has stim- ulated much research on the factors influenc- ing land prices. A common approach has been to rely on capitalization formulas, where land prices are equal to the discounted expected value of the stream of future farm income or rent (e.g., Burt, Alston, Melichar). While it is clear that farm income can help explain a sig- nificant portion of land price movements, such a simple approach has left many issues un- settled. For example, Featherstone and Baker; Clark, Fulton, and Scott; and Falk have pre- sented evidence that expected farm income (or rent) does not explain some farmland price variations. This indicates that simple capital- ization formulas fail to provide an accurate representation of land prices. Similarly, Just and Miranowski found that real land values and recent changes in returns to land had op- posite trends. This suggests a need to examine other factors (besides expected farm income) that affect land pricing.

There is empirical evidence that land prices are also affected by risk and risk aversion (e.g., Just and Miranowski, Chavas and Jo- nes). The capital asset pricing model (CAPM)

allows for the effects of risk and risk aversion on asset prices (e.g., Barry, Robison, and Neartea). Yet, Bjornson and Innes have found that standard asset pricing models applied to land yield some anomalies. For example, they show that mean returns to farm assets are low- er than those on comparable risky nonagri- cultural assets. Also, using a time-varying risk premium, Hanson and Myers found that CAPM models fail to explain adequately land price variations. This suggests that there are factors not captured by a CAPM representa- tion of asset prices that play significant roles in the determination of land prices. A possible explanation for failures of arbitrage-based models to adequately predict land prices is that the capital market is segmented, hence leading to a costly arbitrage process between farm and nonfarm equities. In particular, transaction costs may act as barriers to non- farm capital flow in agriculture.

Using a CAPM model, Shiha and Chavas uncovered statistical evidence that transaction costs have significant effects on land prices. Just and Miranowki argued that transaction costs are important and incorporated them in their land price model under risk aversion. However, they just assumed that transaction costs influenced land prices without testing for their statistical significance. This suggests a need for further research evaluating such ef- fects.

There have also been concerns raised about the nature of intertemporal preferences and their effects on asset valuation (e.g., Koop- mans, Epstein, and Zin 1991; Barry, Robison,

The authors are, respectively, professor of agricultural and applied economics, University of Wisconsin, Madison, and assistant profes- sor of economics, INRA, University of Toulouse, France.

This research was supported in part by a Hatch grant from the College of Agricultural and Life Sciences, University of Wisconsin, Madison. The authors would like to thank anonymous reviewers for helpful comments on an earlier draft of the article.

Amer. J. Agr. Econ. 81 (November 1999): 772-784

Copyright 1999 American Agricultural Economics Association

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 3: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

Chavas and Thomas A Dynamic Analysis of Land Prices 773

and Neartea; Knapp and Olson). For example, this has been illustrated in part by the so- called equity premium puzzle: Why is the his- torical return on stocks so much higher than the return on Treasury bills? Unless one is willing to assume very high levels of risk aversion, risk aversion alone does not seem able to explain this difference (see Kocher- lakota for an overview of this literature). It has become known that this puzzle can pos- sibly be resolved if one is willing to relax some key assumptions found in standard ex- pected utility models that are time additive. In particular, Epstein and Zin (1989, 1991) have developed an asset pricing model that relies on nonadditive nonexpected utility models. This generalized specification is very flexible in the sense that it allows for a sep- arate characterization of intertemporal substi- tution and risk aversion (a feature not shared by the time-additive expected utility model). Epstein and Zin have investigated the impli- cations of their model for asset pricing and presented empirical evidence showing that their dynamic representation of preferences can help solve the equity premium puzzle (Ep- stein and Zin 1991, Epstein and Melino, Ko- cherlakota). This raises the question, Could a proper understanding of land prices require dynamic models that are more general and less restrictive than the traditional time-additive expected utility models?

The objective of this article is to develop and apply a model of asset pricing that in- corporates risk aversion, transaction costs, and dynamic preferences that are not restrict- ed to the traditional time-additive expected utility models. The model generates arbitrage- pricing rules that are then used in an analysis of U.S. land values for the period 1950-96. This involves an econometric estimation of the derived asset pricing rules, generating use- ful information about the factors influencing land values in the United States.

The article is organized as follows. A model of asset pricing is first developed. It extends the Epstein-Zin approach by incorporating transaction costs into their dynamic prefer- ence specification. It shows how risk aversion, non-time-additive preferences, and positive transaction costs influence asset pricing. The model nests as (restrictive) special cases the following situations: risk neutrality, time-ad- ditive expected utility models, and zero trans- action costs. This provides a basis for an in- vestigation of the relevance and effects of such factors on land prices. Our analysis then

presents alternative asset pricing formulations that are particularly convenient for empirical analysis. The formulation also shows that our model can generate as special cases some pricing rules that are similar to those obtained from the static CAPM model as well as from the intertemporal consumption-based CAPM. This can provide useful insights on the em- pirical relevance of these two models. We then discuss the data and the econometric approach of our analysis applied to U.S. land values for the period 1950-96. The model is estimated in its structural form, using the generalized method of moments (GMM) method. The econometric results are presented next. First, we evaluate the validity of our econometric representation of land prices, looking at good- ness-of-fit measures and the orthogonality properties of the instruments used in the GMM method. Second, the econometric re- sults provide statistical evidence about the ef- fects of risk aversion, nonadditive preferenc- es, and transaction costs on U.S. land prices. Implications of our findings for land pricing rules and the functioning of land markets are then discussed. Finally, concluding remarks are presented.

The Model

Consider an agent making consumption and investment decisions over time. Let y, > 0 be consumption at time t, and at =

(aot, alt, a2t,

. aKt) > 0 be a (K + 1) vector of assets, where akt is the kth asset held by the agent at time t, k = 0, 1, ..., K. Let ao, be a riskless asset, while (al,, a2t,, .... aKt) are risky assets whose future returns are not fully predictable. Denote by m, = (mo,, mt,, m2t, ..., mKt) a (K + 1) vector of investments, where mkt is the investment (or disinvestment if negative) in the kth asset made by the agent at time t. The state equation representing asset dynamics at time t is

(1) akt = ak,t-1 + mkt,

k = 0, 1, 2, ..., K.

Denote by q, the market price of the con- sumption good y,, and by (Pi,, p2t, ... PKt) the vector of market prices for the risky assets at time t. Let vkt be the transaction cost of buying or selling one unit of the kth risky asset akt at time t, k = 1, 2, ..., K. The riskless asset ao, is assumed to have price unity, and

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 4: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

774 November 1999 Amer. J. Agr. Econ.

to yield a next-period return of r,,, ao,, where rt+, is the risk-free rate of return. The risky assets (alt, a2t -...* aKt) generate an uncertain next-period return denoted by the differentia- ble function rr,+l(al,, a2t,, . , aK). The agent's budget constraint at time t is

(2) rao,~1 + Trt,(a1,1, ..... alt-, ) K

Sqy, + mor + [Pk, + Vkt(mk,)]mk,. k=l

Equation (2) states that, at time t, the return from riskless investment (rtao,t-_) as well as risky investment (Tr,) must equal consumption expenditures (q,y,), plus investment cost in the riskless asset (mo,) as well as risky assets

(kK=-1

[Pkt + vkt(mkt)]mkt). The function vkt(mkt) in equation (2) rep-

resents the unit transaction cost of buying or selling the kth risky asset mkt. The absence of transaction costs would correspond to a sit- uation where vkt = 0, that is, where market transactions are done costlessly. Alternative- ly, we can expect vkt # 0 if market transactions in mkt are costly due to information cost, bro- kerage fees, time cost, etc. In this case, we will consider the following specification for vkt(mkt):

(3) vkt(mkt) = vk > 0 if mkt,> 0,

= 0 if mk = 0,

= vk < 0 if mkt < 0

where vk+ and (-vk-) are, respectively, unit transaction costs for investing (mkt > 0) and disinvesting (mkt < 0), k = 1, 2, ... , K. Such a specification has the following intuitive in- terpretation. When mkt > 0, transaction costs mean that the unit net price received by the agent for buying mkt is higher since (Pkt + Vk ) > Pkt. Alternatively, when mkt < 0, trans- action costs mean that the unit net price paid by the agent for selling mkt is lower since (Pkt + Vk) < Pkt. This reflects a situation where transaction costs tend to reduce the income of market participants under any circumstance. This suggests that transaction costs provide a disincentive for agents to participate in mar- kets.

Note that the budget constraint (2) is con- tinuous everywhere. In the absence of trans- action costs (vkt = 0), it is also differentiable everywhere. However, in the presence of transaction costs (vk, $ 0) given in equation

(3), it is differentiable everywhere except at the point mkt = 0. Indeed, at mkt = 0, equation (3) implies a different slope of the budget con- straint with respect to mk, comparing an in- crease versus a decrease in mk, from zero. The implications of this "kinked" budget con- straint under transaction costs will be explored below.

Assume that the agent receives utility from consuming y,, yt1, Yt+2, .... In period t, cur- rent consumption y, is deterministic, but future consumption y,?, Y,+2,

...., is uncertain. We

assume that the lifetime utility function rep- resenting the agent's preferences at time t has the following recursive form (Koopmans):

(4) U, = W(y,, yt+, yt,+29 ...) =

U(y,, M(U,?

III))

where M(U,,+II,)

is an aggregator function representing the certainty equivalent of future consumption given the information I, avail- able to the agent in the planning period. Fol- lowing Epstein and Zin (1989, 1991), we will assume the following specification for equa- tion (4):

(5a) U, = [(1 - P)y," + 3MtP]"P,

for 0 # p < 1,

= (1 - P)log(y,) + p log(M,),

for p = 0

where

(5b) M, = M(U,,,+ I) = (EUt+1,)l()

if 0 xot < 1, = exp[E,log(U,.,)] if xo = 0

yt > 0, 3 = 1/(1 + 8), and E, is the expectation operator based on the information available at time t. When future consumption is deter- ministic, this specification is a utility function exhibiting a constant intertemporal elasticity of substitution u = 1/(1 - p) and a rate of time preference 8. Also, the parameter o can be interpreted as a relative risk aversion co- efficient with the degree of risk aversion in- creasing as o falls (Epstein and Zin 1989, 1991). The specification (5) is flexible since it permits the separation of intertemporal sub- stitutability issues from risk aversion (Epstein and Zin 1989, 1991). Also, when o = p # 0, equation (5) reduces to U, = [(1 - P3)E, •j>o

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 5: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

Chavas and Thomas A Dynamic Analysis of Land Prices 775

dollars

marginal value of investment (MVI) MV2

Pkt + Vk zone of "no investment"

S+ Vk

0 mk mkt(3) mt(1) mk(2)

Figure 1. Investment under transaction costs

3jyo,+J]"1, which corresponds to the familiar expected time-additive utility specification.

We assume that the function U(y,, M(Ut+,,,)) is differentiable and bounded for all feasible (Yt, at, mt). The individual's decisions can then be represented by the following recursive op- timization problem:

(6) V,(at-1) =

max{U(yt, M,(Ut+l)):

equations (1) and (2)},

= max U

rtao,t_

+ 7r,

- (ao,

- ao,t1)

K

- (Pkt + Vk,)(ak, t- akt-1) k=l

- qt, Mt(Vt+,(at))

where V,(at,_) is the indirect objective func- tion (or value function) at time t, and ar, =

Wr,(al,,_ , ..., a,,_1). Under appropriate dif- ferentiability assumptions, the first-order necessary conditions for a, in equation (6) are

(7a) a0,: (JU/lay)/q, = (aU/8M,)(aM,/tao,)

(7b) ak: (U/ayt)(pkt + vkt)Iq, =

(aU/Mt)(aM,/lakt), if mkt 5 0

(7c) (aU/ay,)(pkt vk )/qt

- (aU/aMt)(aMt/akt)

- (OU/ay,)(Pkt + v)/qt, if mki = 0

for k = 1, 2, ... , K. Equations (7b) and (7c) reflect the kinked budget constraint at mkt = 0 under transaction costs. As illustrated in fig- ure 1, the presence of transaction costs (vk, t 0) in equation (3) generates three possible sce- narios. Scenario 1 is obtained under equation (7c), when the marginal return to investment is moderate and no investment takes place (mkt = 0). Scenarios 2 and 3 correspond to equa- tion (7b). Scenario 2 is obtained when the marginal return to investment is high, stim- ulating investment (mk, > 0). And scenario 3 is obtained when the marginal return to in- vestment is low, inducing disinvestment (mk, < 0). Figure 1 illustrates that the zone of "no investment" tends to increase with higher transaction costs.

The value function V,(a, 1) in equation (6) is continuous everywhere. In the absence of transaction costs (vk, = 0), it is also differ- entiable everywhere under appropriate regu- larity conditions. And in the presence of trans- action costs given in equation (3), it is dif- ferentiable everywhere except at the boundary points where mink, switches away from zero. Indeed, at the boundaries between market par- ticipation and nonparticipation (i.e., at the switching points between

mkt, > 0 and

mkt, =

0, and between mk, < 0 and mk, = 0), the

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 6: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

776 November 1999 Amer. J. Agr. Econ.

budget constraint is kinked and so is the value function. However, in the language of mea- sure theory, these points of nondifferentiabil- ity of

V,(a,_l) are "few" and occur "with

measure zero." Thus, the value function V,(a,_-) is differentiable "almost every- where," that is, everywhere except at the few points where a kink exists.

At points of differentiability (i.e., almost everywhere), the envelope theorem applied to equation (6) yields

(8a) aV,/aoa,_1 = (U/lay,)(1 + r,)/q,

and

(8b) V,laaka,_ = (aU/ay,)[(aOr,/aak,_-l) + (Pk, +

Vk,)]/q,, if mkI # 0

(8c) = (aU/&y,)[(Tr,/&ak,_t-) + pktlq,,

if mt, = 0.

Equations (8) link the marginal utility of the assets a,_, to the marginal utility of con- sumption y, at time t. Also, we need to eval- uate the derivatives aM,l/aa, in equation (7). From equation (5b), the derivative of M, with respect to some argument y of

U,+, is M,/lay

= M-"Et[U;-,'(aU,t+,ay)]. Assuming that mk,,

0 0, use this expression after substituting equations (8) into equations (7) to obtain

(9a) (U/lay,)/q, = (aU/aM,)

x (M -oE,[U+-,'(aU/ay,+,)(1 + r,+,)/q,+,])

(9b) (U/ly,)(pk, + vkt,)q, = (aU/aM,)

x {M,-oE,[U,O+,'1(aU/ay,+,)

x [(aI,+,l/aak,) + (Pk.t+ + Vk,t+)]/qt+ll]},

if mkt,

0

(9c) (3U/3y,)(pk, + v )/q,

- (aU/aM,)

x {M: -,E,[U?+.,'(aU/ay,,

)

X [(aTI,+ ,l/ak,)

+ (Pk.t+ + Vkt+l)]/qt+I]}

- (aU/Iy,)(Pkt + vj)/q,, if mk, = 0

k = 1, ..., K. Equations (9) are Euler equa- tions characterizing the optimal price dynam- ics corresponding to optimal investment under rational economic behavior. Equation (9a) is the arbitrage equation related to investments in the riskless asset and equations (9b)-(9c) are the price arbitrage equations for the kth risky asset. Note that the effects of monetary values in equations (9) are in real terms: they are all deflated by the consumer price q, or

q,+,, thus expressing all monetary effects in

terms of their purchasing power. Equations (9) provide the basic specification for the eco- nomic analysis presented below.

Note that substituting (9a) into (9b) yields (Pkt +

Vkt).E,[Ul(aU/ay,+l)(1 + r, l)/q, l],

(10a) = E, { Ut'(aU/lay,+,)

x [(a7rt,?/ak,) + pk,,t + vk,,t+]lq,+1

k = 1, 2, ..., K. An interesting special case of equation (10a) is worth noting. Assuming that a = p = 1, we obtain the standard time- additive model under risk neutrality. Indeed, in this case, Ut+0-1 (aU/ay,j+) = 1 - 13, and equation (10a) becomes

(10b) (pkt + Vk,)

= E,{[(a, t+/aakt) + Pk,t+ + vk,t+ ]q,?1} + E,[(1 + rt,1)/q,1+,]

k = 1, ..., K. This is a standard arbitrage pricing formula. For example, in the case where the rate of return

r,+, and prices q,+,

are known at time t, equation (10b) states that current asset prices must equal the discounted value of next-period return (awr,+/laak,) plus next-period asset prices (Pk,t+I + vk,t+1), the discount factor being 1/(1 + r,+ ). This is the intuitive result that, at the optimum, marginal discounted return plus marginal capital ap- preciation equals zero. However, it is worth stressing that equation (10b) holds only under restrictive conditions. Indeed, comparing equations (10a) and (10b), it is clear that as- suming time-additive model under risk neu- trality does affect arbitrage pricing.

In the more general case allowing for risk aversion and nonadditivity over time, equa- tion (10a) provides the appropriate arbitrage pricing formulation. It implicitly involves the covariance between the utility

U+4,' and the

marginal utility of income at time (t + 1). While this covariance is zero under risk neu- trality (or

= 1), it becomes nonzero under risk

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 7: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

Chavas and Thomas A Dynamic Analysis of Land Prices 777

aversion (a < 1). Then, risk aversion can pro- vide incentive for consumption and/or income "smoothing." Also, implicitly through the term U, 1, equation (10a) allows the parameter p (reflecting the elasticity of intertemporal substitution) to affect asset price dynamics. The economic assessment of these effects is discussed next.

Specification

Equations (9) provide a basis for the anal- ysis of dynamic arbitrage pricing under gen- eral conditions. However, they are not in a form that can be used directly. The main issue is the empirical tractability of the term U t+I(aU/ay,t+) in equations (9). In this sec- tion, we derive relationships that express this term in a form that is econometrically tractable.

From equation (5a), note that aU/ay,?, =

UI+((1 - P3)yr 1. It follows that U;+-1(aU/ay,+1)

= U,-f(l - P)y,11.

Note that the term U,•-p = 1 whenever ao = p, which corresponds to

the expected utility specification. Thus, in this (restrictive) special case, there is no need to evaluate U,+, in the econometric specification of equations (9a) or (9b). However, any de- parture from expected utility (i.e., . # p) re- quires an evaluation of U,+ . One way to do this is to solve numerically the dynamic pro- gramming problem (6) for the value function

V,+,(a,). However, this can be complex. An

alternative is to try to obtain analytical results on the value function without obtaining an explicit solution to equation (6). This second alternative is explored here.

Denote the agent's wealth at time t by A,, where

K

(1la) A, = ao, + 1 (Pk, + kt) ak,. k=l

Define the gross return at time t as G, = (1 + r,)a0,t-l + EI (Pk, + Vk,)ak,t + -t(al,,t-,

.. aK,,). Then, the gross rate of return on wealth at time t is R, = G,/A,_I. It follows that the budget constraint (2) can be alternatively written as

(lib) A, = R,A,_I - q,y,.

Note that equation (1 la) involves only vari- ables at time t, implying that all the dynamics are captured by equation (1 ilb). In the lan- guage of dynamic programming, treating

(1 lb) as the state equation, this means that the variables (y,, ao,

..... aK,) can be treated

as "control variables," while A, is the "state variable." This implies that the dynamic pro- gramming problem (6) can be alternatively expressed as

V,(A,_i) = max U

R,A,_tI - ao,

At,at

K - (Pk, + vkt)akt

k=l

+ q,,

M,[V,t+(A,)]},

where the value function is now written

V,(A,_,). Using the envelope theorem at points of differentiability (i.e., almost everywhere), we obtain aV,/aA,_, = (aU/ay,)(R,/q,). Using equation (5a), it follows that

(12) aV,I/IaA,

= (aUt+llayt+,)(R,+l/qt+) = Utl+f(1 - 3)yt,+P1(Rt+,/qt+,).

We now make the following assumption.

ASSUMPTION Al. The return function 7r,(al,, a2,t ..* aKt) is linear homogenous in (al,, a2,,

?. . ,

akt).

Note that the utility function specification U, in equation (5) is linear homogeneous in the consumption stream. Under assumption (Al), the budget constraint in equation (2) is also homogenous in (y,, a0,, al,,

a2,,...., aK,). Giv-

en the homogeneity of equations (Ila) and (1 b), this implies that the value function V,(A,_1) is linear homogeneous in A,_,, that is, that V,(A,_-) is proportional to

A,_l.' Under

assumption (Al), this means that the value function can be expressed as V,1, =

y,,lA,. Combining this result with equation (12) yields

V,+, =

U+p(1l -

p)yp(R,+l/q,+l)At,,or, for p # 0,

(13) U,+,

= [(1 - f3)y,y+1A,R,,+/q,+ll/P.

In the case where p # 0, it follows that

' To see that, let V(A) = maxb (b, A), where f(b, A) is linear homogenous in (b, A) and satisfies Xf(b, A) = f(Xb, XA), for all X > 0 and all feasible (b, A). It follows that V(XA) = {maxb, f(b, XA)} = {X maxbf(b, A)} = XV(A), thus proving that V(A) is linear homogenous in A.

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 8: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

778 November 1999 Amer. J. Agr. Econ.

(14) Upll (a Ulayt l)

= U,"+P(1 - P)y+I1

= [(1 - P)yP,-IARt+l Iq,+

I](O-P)/

x (1 - 13)ytpI.

Using equations (5a), (13), and (14), equa- tions (9) can be alternatively written as (see the derivation in the appendix)

(15a) 1 = E,(({qtq/qt+I)Y(yt+ I/yt)Y(P 1)

X (Rt+1) -Y'(1 + rt+,)}

(15b) pkt + vkt = E,{(p3q,/q,+I)-(y, l/y,)(P-'I)

X (R, l)-

X (arr,+ laakt + Pk,t+ 1

+ Vk,+1)},

if mkt, 0,

pkt + V k E,{(pq,/q,+,)-(yt+ ly,)Y-1)

X (R,+,))-I

X (awt+,/aakt + pk,t+ +

vk,t+1)

(15c) Pk + Vk, ifmkt = 0

k = 1, ..., K, where y = a/p. Again, equa- tions (15b) and (15c) are illustrated in figure 1, where the marginal value of the kth asset is MV =

E,{(Pq,/q,+,)/(y,+m/y,)Y(P-)(R,m)Y-I (ar~+,,laak, + Pk,t+l + vkt+1)}. As discussed above, this identifies three possible regimes: regime 1 corresponding to a moderate MV where no investment takes place (mk = 0); regime 2 corresponding to a high MV and positive investment (mk > 0); and regime 3 corresponding to a low MV and disinvestment (mk < 0).

Expressions (15) give general asset pricing relationships where the term

[(pq,/q,+,)Y(y,,1/ y,)Y(P•-)(R,+,)-)Y] is used to discount future payoffs in the determination of current asset prices. This includes as special cases some well-known formulas for asset pricing. For example, having y = 0 generates the static capital asset pricing model (CAPM) where the discount factor becomes

1/R,+1, which de-

pends only on the gross return on wealth and not on the growth rate of consumption (y,+,/ y,). In this case, the riskiness of an asset is measured in equations (15) by the covariance

of its return with the overall return (see Jen- sen). Alternatively, having y = 1 gives the intertemporal consumption-based capital as- set pricing model (CCAPM) where the dis- count factor is [(pq,/q,, )(y,,I/y,)'P•-], which depends only on the growth rate of consump- tion (Y,+i/y,) (and not on the return R,,+). In this case, the riskiness of an asset is measured in equations (15) by the covariance of its re- turn with the growth rate of consumption (see Merton, Breeden). This is a situation where risk management under risk aversion clearly involves "consumption smoothing" activi- ties.

In the general case where y # 0 and y # 1, equations (15) give asset pricing formulas that generalize the CAPM model and the CCAPM model. Indeed, in equations (15), the factor

[(pq,/q,,m )-(y,, /y,)-Y'P)(R,,,)V-']

used to discount future payoffs in the determination of current asset prices depends both on overall return R,,, and on consumption growth (y,,I / y,). In this case, the riskiness of an asset is measured by the covariance of its return with both the overall return (as measured by R7; ') and consumption growth (as measured by

(Yt+l/Yt)yp-"1)). Finally, note that equations (15) are con-

sistent with the results obtained by Epstein and Zin [e.g., equation (16) in Epstein and Zin 1991, p. 268)]. Our article extends their analysis in two main directions. First, while Epstein and Zin assume implicitly that q, =

1 for all t, we incorporate explicitly the role of inflation in the determination of asset pric- ing. In equations (15), this is measured by the term (q,/q,+,)- which reflects the relative change in consumer prices. Second, we intro- duce transaction costs in Epstein and Zin's analysis and derive their implications for dy- namic asset pricing. This is reflected by the terms vkt, and vk,t+I in equations (15). It gives a basis for investigating empirically the role of transaction costs in the determination of asset prices.

Equations (15) provide a parametric model of dynamic asset pricing that can be estimated econometrically. The empirical investigation of this model in the context of land prices is presented next.

Data and Estimation

The above model was developed at the mi- croeconomic level. Unfortunately, time-series data on investment, consumption, and asset

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 9: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

Chavas and Thomas A Dynamic Analysis of Land Prices 779

prices are quite rare at the micro level. The best time-series data available for analyzing the determinants of land prices are at the ag- gregate level (e.g., USDA). Such data con- straints are the main reason why we focus our empirical analysis here at the aggregate level. This raises the issue, Are the microeconomic implications derived above also valid at the aggregate level?

First, consider the situation in the absence of transaction costs (vk, = 0). As one moves from a micro to an aggregate analysis, we expect to find significant differences in wealth across agents. However, if we adopt the as- sumption that agents are similar otherwise, then homothetic preferences given in equation (5a) imply that aggregate behavior is consis- tent with the behavior of a representative agent (e.g., Epstein and Zin 1989, p. 956). In this case, all decisions are proportional to wealth and the pricing rule in equations (15) depends only on relative quantities or wealth. This means that equations (9) or (15) would apply as well at the aggregate level, as if they were associated with a representative agent. Thus, in the absence of transaction costs (where Vk, = 0), equations (15) can be ex- pected to provide a useful basis for aggregate analysis.

In the presence of transaction costs (vk, t 0), aggregation becomes a more complex is- sue. The problem arises if we aggregate across both buyers and sellers that face dif- ferent objective economic conditions. As re- flected in equation (3) under transaction costs, buyers pay a higher net purchase price while sellers receive a lower net selling price. Because they face different net prices, buyers and sellers may have different decision rules, suggesting that aggregate behavior would de- pend on each group's decision rules. Also, the implications of these decision rules for aggregate asset pricing can be complex. Such complexities create significant challenges for empirical modeling. Here, we propose to handle such complexities as follows. As ar- gued above, in the context of a representative agent, we expect equations (15) to provide a good basis for an aggregate analysis of asset pricing in the absence of transaction costs. We also expect equations (15) to give some basis for investigating the role of transaction costs at the aggregate. As long as decision rules are not too different among market par- ticipants, one can expect the parameters x, I3, p, and y in equations (15) to represent some "average behavior" of participants in

asset markets. However, one expects the im- pact of transaction costs on aggregate be- havior to differ from equations (15). The rea- son is that, given equation (3), equation (15b) differs between buyers (where vk, = vk > 0 when mk, > 0) and sellers (where vk, = v~ < 0 when mk, < 0). Since both buyers and sell- ers are present in the market, the aggregate effects of transaction costs on asset pricing should be some (weighted) average of their micro effects across market participants. For example, one might expect transaction costs to affect asset prices in a way similar to v +

in equations (15) in a "buyers' market" (when buyers' behavior dominates), but sim- ilar to vk in a "sellers' market" (when sell- ers' behavior dominates). We will rely on such arguments in the specification, estima- tion, and testing of transaction cost effects in the aggregate econometric analysis pre- sented below.

Our analysis focuses on the aggregate analysis of annual land prices in the U.S. between 1950 and 1996. Data on land values (p,, measured in 1,000 dollars per acre), ag- gregate land quantity (Q,, measured in mil- lion acres), gross rate of return on farm eq- uity (R,), and net farm income per acre (rr/ a, measured in 1,000 dollars per acre) were obtained from USDA publications. Under assumption (Al), constant return to scale means that we can measure the marginal re- turn arlaa by the average return rr/a. Also, the price of consumption goods (q,) is mea- sured by the Consumer Price Index (CPI) reported by the Bureau of Labor Statistics. The risk-free interest rate r, is measured by the interest rate on U.S. Treasury bills. Fi- nally, we calculated consumption y, by in- terpreting (q,y,) as disposable personal in- come of the farm population (measured in billions of dollars), as reported by the USDA.2 These data, summarized in table 1, provide a basis for estimating equations (15).

We estimate model parameters using the system of equations (15a)-(15b) [rather than equation (15b) alone], where equation (15b) represents arbitrage pricing for land. Such an approach allows us to obtain more efficient parameter estimates by exploiting all the mod- el information. This implies that parameter restrictions across equations are imposed and

2 Disposable personal income is calculated from net farm income, plus nonfarm income, minus taxes.

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 10: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

780 November 1999 Amer. J. Agr. Econ.

Table 1. Descriptive Statistics

Standard Variable Mean Deviation Minimum Maximum

q, (= 1 in 1982-84) 0.66 0.42 0.26 1.52 y, (billion dollars) 60.43 8.92 40.19 93.89 Q, (million acres) 1,085.11 79.02 974.00 1,206.00 p, (1,000 $/acre) 0.37 0.26 0.07 0.76 a, (billion dollars) 416.09 263.13 115.70 816.40 R, 1.04 0.07 0.86 1.24 7r,/a, (1,000 $/acre) 0.023 0.014 0.009 0.055 r, 0.054 0.028 0.009 0.141

Note: Number of observations is 45.

are easily tested for by investigating possible model misspecification.

The Generalized Method of Moments (GMM) is the most popular estimation pro- cedure when dealing with Euler equations. GMM estimates are fairly easy to obtain and are consistent provided an adequate set of in- strument variables is used. It is known (e.g., Epstein and Zin 1991, Hansen and Singleton) that the choice of instruments is an important issue with GMM, as parameter estimates can be sensitive to competing sets of instruments. This suggests a need to assess the relevance of instruments used in GMM estimation.

Based on the orthogonality condition E u(O) 'W = 0, the GMM estimation method in- volves minimizing the distance:

(16) u(O)'W(W'QW)- W'u(0)

where 0 is the (p x 1) vector of parameters, W is a (2T x k)-instrument matrix, 11 is the (2T x 2T) variance-covariance matrix asso- ciated with the above orthogonality condition, and u(0) is the (2T x 1) vector of residuals defined as

(17) u(O) = (u,,(0), u,12(), . . . U,(),

u21(0), u22(0), .. 2T. )) where3

ut, = 1/q, - (Pq,1/qI)-,(y,,Ily,)t -

X (R,+,)Y'-I(1 + r,+,)/q,

u2t = P, + vt (tqtq,,+I) (yt+, /y,)vct-1

X (R,+ )Y-

X (a7r,+t/ aa+ +

p,+ +

v,+ ).

Several remarks are in order. First, the gen- eral form of the GMM estimator defined above allows for different instruments in each equa- tion. We choose to use the same set of instru- ments for equations (15a) and (15b); hence, the instrument matrix W is constructed by du- plicating the original (T x k) set of instruments. Thus, there are a total of (2 x k) identification restrictions. In general, one wants to choose instruments among variables that are known to both the econometrician and the economic agents at the time of their decisions. Such in- struments would satisfy the orthogonality con- dition Eu(O)'W = 0 and reflect the nature of price and economic expectations used in de- cision making. The relevance of the instru- ments and the associated orthogonality con- ditions can be assessed by the Hansen test, evaluating the validity of the overidentification restrictions. Second, consistent and robust pa- rameter estimates obtain using a two-stage pro- cedure. In the first stage, the variance-covari- ance matrix 11 is consistently estimated by a robust procedure allowing for heteroskedastic- ity and serial correlation of residuals (Newey and West). The matrix fi is then replaced by its estimate in the GMM criterion and final parameters are estimated.

Before estimation, we need to specify a parametric structure for transaction costs in equation (15b). At the micro level, equation (3) implies that transaction costs v, are positive (negative) when assets are sold (bought). As argued above, one expects this pattern to hold on the average at the aggregate. As a result,

3 For simplicity, we suppress the subscript k on asset equity and price.

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 11: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

Chavas and Thomas A Dynamic Analysis of Land Prices 781

Table 2. GMM Estimation Results, 1950-96

Param- Standard eter Estimate Error

p 0.7561** 0.0352 y 1.0622** 0.0633 p 0.9779** 0.0053 cm, 0.0172* 0.0064

C, 0.1366 0.1086 Oa 0.8031** 0.0744 R2 equation (15a): 0.9980

equation (15b): 0.9303

Notes: * and ** denote a parameter significantly different from zero at the 5% and 1% levels, respectively. Instruments used: (1, p,_t, y,/y,_,, q,/ q,1_, Rt,).

in aggregate analysis, we hypothesize that transaction costs are proportional to the vari- ation in the volume traded. This generates the following specification for transaction costs:

Model MI: v, = cAQt if AQ, > 0,

ScmMAQt if AQ, < 0,

where Q, is land quantity at time t, AQ, = Q, - Qt,_l, and c, and cm are parameters. Hence, marginal transaction cost is propor- tional to a positive change in land quantity, with parameter c, > 0, and is proportional to a negative change in land quantity, with parameter cm > 0. Asymmetry in the mar- ginal transaction cost is allowed for when the parameters c, and cm are different. Al- ternative models obtain by appropriate re- strictions on model M 1:

Model M2:

v, = cAQ, (symmetric transaction costs, proportional to changes in quantity);

Model M3:

v, = 0 (no transaction cost).

Table 3. Hypothesis Testing

Model M2 is obtained by imposing the re- striction

cp = cm = c on model Ml. The hy-

pothesis of no transaction costs of model M3 obtains through the restriction c, = Cm = 0 imposed on M1.

Parameters to be estimated are therefore 0 = (., p, "y,

Cp, Cm) for the unconstrained model Ml. Note that we follow here the notation of Epstein and Zin (1991) for structural param- eters, where the risk aversion coefficient ae is embedded in the parameter y = o/p.

Estimation Results

The model is estimated using the two-stage GMM procedure described above. Several sets of instruments were considered to obtain consistent parameter estimates. The selected set of instruments is (1, pt-1, y,/yt-1, qt/q,_1,

R,_I) for equations (15a) and (15b). It was

found that increasing the size of these in- strument sets with any other combination of instruments improved efficiency of parame- ter estimates but often failed the Hansen overidentification-restrictions test. Hence, we exploit ten orthogonality restrictions for the model, leaving five overidentification re- strictions. Model performance was also en- hanced when serial correlation of residuals was taken into account in the computation of the variance-covariance matrix ni. We set the number of lags to four in the Newey and West robust estimation procedure.

Parameter estimation results are in table 2, and tests of the various hypotheses are presented in table 3. Using the Hansen test of the overidentification restrictions, we do not find statistical evidence against the or- thogonality conditions between the instru- ments and the error terms. This suggests that the model specification and the choice of instruments appear appropriate. By using a rather large number of overidentification re- strictions, parameter efficiency is increased

Test Statistic p-Value

Overidentifying restrictions (Hansen test) X2(5) = 3.0769 0.6881 No transaction costs (c, = Cm = 0) X2(2) = 7.5591 0.0228 Symmetric transaction costs (c, = Cm) X2(1) = 1.2407 0.2653 Expected utility (y = 1) X2(1) = 0.9643 0.3261 p = 1 X2(1) = 47.7823 0.0000 P= 1 X2(1) = 17.0069 0.0000 Risk neutrality (a = 1) X2(1) = 7.0020 0.0081

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 12: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

782 November 1999 Amer. J. Agr. Econ.

while orthogonality restrictions are not re- jected. Statistical fit is good for both equa- tions, with an R2 equal to 0.9980 for equa- tion (15a) and 0.9303 for equation (15b). Thus, our model has a high explanatory power for land prices.

The structural parameter a (= y-p) is es- timated together with its asymptotic stan- dard error, using consistent estimates for the parameters y and p. Risk neutrality is tested by considering the null hypothesis Ho:

a- = 1. A Wald test statistic rejects this hypoth- esis at the 5% significance level, hence pro- viding strong evidence that market partici- pants are risk averse. Recall that risk aver- sion increases as parameter ax decreases away from 1. Estimates of xo and p are very close, indicating that the hypothesis of ex- pected utility may not be rejected. This is confirmed by a Wald test statistic [distrib- uted as a X2(1)] which fails to reject the null hypothesis (Ho: y = 1) at the 5% significance level. Our results therefore indicate that there is no strong evidence of departure from the expected utility hypothesis. This con- trasts with Epstein and Zin (1991) who re- ported strong rejection of this hypothesis for most of their instrument sets. This differ- ence may be due to the fact that Epstein and Zin used more disaggregate consumption commodities in their analysis. Finally, our estimate of the parameter p is close to 1. It is larger than the one estimated by Epstein and Zin. The intertemporal elasticity of sub- stitution is accordingly higher: a = 1/(1 - p) = 4.10. Again, this difference may be related to the fact that it may be easier to uncover substitution possibilities while working with more aggregate commodities. The hypothesis that parameter p is equal to 1 is strongly rejected at the 5% level.

Our estimate of the parameter y also in- dicates that, while y is close to 1, it is sta- tistically different from zero at the 5% level. As discussed above, this suggests a strong rejection of the static CAPM model in favor of the dynamic consumption-based CAPM model. This suggests that the effects of risk and risk aversion can be more appropriately characterized through "consumption smoothing" rather than "income smooth- ing." Such a result can help explain previ- ous findings by Bjornson and Innes or Han- son and Myers concerning the inadequacy of the static CAPM model in explaining land price variations.

The rate of time preference is computed at

8 = 1/P - 1 = 0.0226 for model Ml, about 2.2%. This seems reasonable, recalling that the effects of all monetary values in our mod- el are in "real terms." Although parameter p is very close to 1, the hypothesis that this parameter is equal to 1 is strongly rejected, because it is estimated with sufficient pre- cision.

We now turn to the results concerning pa- rameter estimates related to transaction cost. Using a Wald test, the absence of transaction costs (c, =

cm = 0 in model M3) is rejected

at the 5% level against the alternative that transaction costs are proportional to changes in quantity (model Ml). In particular, the pa- rameter Cm (related to negative changes in land assets) is significantly different from zero at the 5% level. This provides statistical evi- dence that transaction costs have significant effects on land prices. To evaluate these ef- fects quantitatively, consider the 1980s and the 1990s periods. During these periods, ag- gregate farm land decreased over time. Esti- mated at mean values for each period, our model shows that transaction costs were on average 12.98% of land value in the 1980s, and 7.54% in the 1990s.

Finally, the hypothesis of symmetry in mar- ginal transaction cost is also tested (c, = c, in model M2). The Wald test statistic indicates that symmetry cannot be rejected at the 5% level. Thus, we do not find strong evidence that transaction costs differ between periods of land increase versus land decrease. Such a result may be due to the fact that the estimated value of cm in model Ml is rather small in magnitude, whereas parameter c, is not sig- nificantly different from 0 at the 5% level. This latter finding may be due to the fact that only a few sample observations were relevant for the estimation of

cp: increases in aggregate

land quantity took place only during the 1950s. Further empirical analysis may be needed to confirm whether our evidence of symmetry in transaction costs holds true in different situations and with different data.

Conclusions

In this article, we analyze land price variation using a general pricing model, including as special cases most popular specifications found in the literature. This approach is mo- tivated by the need to consider more flexible pricing models that can better explain land price movements, as capitalization-based or

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 13: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

Chavas and Thomas A Dynamic Analysis of Land Prices 783

CAPM models have often been rejected when applied to land prices. Regarding farmers' be- havior, risk neutrality and time-additive pref- erences are easily tested. Transaction costs are explicitly introduced in the pricing formula and are specified as asymmetric functions of quantity changes.

Our empirical findings uncover statistical evidence that both risk aversion and trans- action costs affect land prices. The fact that transaction costs are a significant component of land prices suggests the potential for capital market segmentation. Transaction costs would affect the functioning of the capital market in agriculture as well as the return to farm assets (e.g., Shiha and Chavas, Bjornson and Innes). Also, we find statistical evidence against the static CAPM model. This can help explain the inadequacies of the static CAPM model in explaining land price variations (e.g., Hanson and Myers). This stresses the need to capture adequately the effects of risk, risk aversion, and transaction costs on land prices in order to understand better the functioning of land markets.

Finally, we do not uncover strong evidence against either the time-additive preferences or the dynamic consumption-based CAPM. Con- trary to Epstein and Zin, we do not find the additive expected utility model overly restric- tive. This may be due to the fact that our analysis relies on a consumer good that is more aggregate than that in Epstein and Zin (1991). Epstein and Zin worked with more disaggregated commodities and obtained a negative relationship between durability of goods and intertemporal substitution. Because we use aggregate data on consumption, pos- sibilities of substitution may be more impor- tant, explaining why the intertemporal rate of substitution is higher in our case.

Our results supporting the dynamic CAPM model indicate that agricultural households manage risk through consumption smoothing activities. This stresses the importance of con- sumption data in analyzing the effects of risk on land values. While such information exists at the aggregate level, it is often not available at the micro level. Our findings stress the need for collecting better micro data in the farm sector. This would allow for more refined em- pirical analyses. For example, building on our conceptual model, it would be interesting to model transaction costs at the micro level by considering a regime-switching framework. At every time period, individual farmers would be in any of three situations: sell, buy,

or do nothing. The presence of transaction costs could then explain who participates in the land market, with implications for land transactions and variations in land prices. In- vestigating such issues appears to be a good topic for future research.

[Received July 1998; accepted January 1999.]

References

Alston, J.M. "An Analysis of Growth of U.S. Farmland Prices, 1963-82." Amer. J. Agr. Econ. 68(February 1986):1-9.

Barry, P.J., L.J. Robison, and G.V. Neartea. "Changing Time Attitudes in Intertemporal Analysis." Amer. J. Agr. Econ. 78(Novem- ber 1996):972-81.

Bjornson, B., and R. Innes. "Another Look at returns to Agricultural and Nonagricultural Assets." Amer. J. Agr. Econ. 74(February 1992):109-19.

Breeden, D.T. "An Intertemporal Asset pricing Model with Stochastic Consumption and In- vestment Opportunities." J. Financ. Econ. 7(September 1979):265-96.

Burt, O.R. "Econometric Modeling of the Cap- italization Formula for Farmland Prices." Amer. J. Agr. Econ. 68(February 1986):10- 26.

Chavas, J.P., and B. Jones. "An Analysis of Land Prices under Risk." Rev. Agr. Econ. 15(May 1993):351-66.

Clark, J.S., M. Fulton, and J.T. Scott, Jr. "The Inconsistency of Land Values, Land Rents, and Capitalization Formulas." Amer. J. Agr. Econ. 75(February 1993):147-55.

Epstein, L.G., and A. Melino. "A Revealed Pref- erence Analysis of Asset Pricing under Re- cursive Utility." Rev. Econ. Stud. 62(1995): 597-618.

Epstein, L.G., and S.E. Zin. "Substitution, Risk Aversion, and the Temporal Behavior of Consumption and Asset Returns: A Theo- retical Framework." Econometrica 57(July 1989):937-69.

Epstein, L.G., and S.E. Zin. "Substitution, Risk Aversion, and the Temporal Behavior of Consumption and Asset Returns: An Empir- ical Analysis." J. Polit. Econ. 99(April 1991):263-86.

Falk, B. "Formally Testing the Present Value Model of Farmland Prices." Amer. J. Agr. Econ. 73(February 1991):1-10.

Featherstone, A.M., and T.G. Baker. "An Ex-

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions

Page 14: Agricultural & Applied Economics Association - … · Agricultural & Applied Economics Association and Oxford University Press ... the static capital asset pricing model ... discuss

784 November 1999 Amer. J. Agr. Econ.

amination of Farm Sector Real Asset Dy- namics: 1910:85." Amer. J. Agr. Econ. 69(August 1987):532-46.

Hansen, L.P. "Large Sample Properties of the Generalized Method of Moments Estima- tors." Econometrica 50(July 1982):1029-54.

Hansen, L.P., and K.J. Singleton. "Generalized Instrumental Variables Estimation of Nonlin- ear Rational Expectations Models." Econo- metrica 50(September 1982):1269-86.

Hanson, S.D., and R.J. Myers. "Testing for a Time-Varying Risk Premium in the Returns to U.S. Farmland." J. Emp. Financ. 2(Sep- tember 1995):265-76.

Jensen, M.C. "Capital Markets: Theory and Ev- idence." Bell J. Econ. and Manage. Sci. 3(Autumn 1972):357-98.

Just, R.E., and J.A. Miranowski. "Understanding Farmland price Changes." Amer. J. Agr. Econ. 75(February 1993):156-68.

Knapp, K.C., and L.J. Olson. "Dynamic Resource Management: Intertemporal Substitution and Risk Aversion." Amer. J. Agr. Econ. 78(No- vember 1996):1004-14.

Kocherlakota, N.R. "The Equity Premium: It's Still a Puzzle." J. Econ. Lit. 34(March 1996): 42-71.

Koopmans, T.C. "Stationary Ordinal Utility and

Impatience." Econometrica 28(April 1960): 287-309.

Melichar, E. "Capital Gains versus Current In- come in the Farming Sector." Amer. J. Agr. Econ. 61(December 1979):1085-92.

Merton, R.C. "An Intertemporal Capital Asset Pricing Model." Econometrica 41(Septem- ber 1973):867-87.

Newey, W.K., and K.D. West. "A Simple, Posi- tive Semi-definite, Heteroskedasticity and Autocorrelation Consistent Covariance Ma- trix." Econometrica 55(May 1987):703-8.

Shiha, A.N., and J.P. Chavas. "Capital Segmen- tation and U.S. Farm Real Estate Pricing." Amer. J. Agr. Econ. 77(May 1995):397-407.

U.S. Department of Agriculture, Economic Re- search Service (USDA). Economic Indica- tors of the Farm Sector, National Financial Summary, 1993. Washington DC, ECIFS-13- 1, 1993.

Appendix

Derivation of Equations (15)

When p # 0, note that equation (5a) gives aU,/aM, = U-P3MP-', and aU,/ay, = Uh-P(1 - P)y-'l, im- plying that

(Al) (aU,IaM,)I(aU,Iay,)

= P3M-'I[(1 - P)y-'].

Also, equation (5a) gives MP = [Uf - (1 - p)yp]/ p. Using equation (13), this yields

(A2) M, = {[(1 - P)yp-'A,_,R,/q,

- (1 - 13)y,]/3}l'Ip.

Combining equations (Al) and (A2) with equation (14) gives

[( U/laM,)/( U/lay,)]M, -

Utl ( U/layt+, 1 )q,/q, +1

= {1/[(1 - P3)y,-']}P3M-'

x [(1 - P)yp-_lAR,, /qt+l]-" ')(1 - P)

using equations (Al) and (14)

= {1/[(1 - P)y-'] x {[(1 - 3)y•'A,_R,/q,

- (1 - f3)y]/Pf3}(P'-Y)'P

x [(1 - )yp-_'ARt l/ql]'" •-,'P

x (1 - P)y-' q,/q,+

,

using equation (A2)

= (q/q ,t+l)c"/"(ytlyt+

)"x/P)-x

x [(R,A,_, - q,y,)/(R,+lA,)1

"-(/''

= (pq, /q,+ l)c"P(ytl/y,+ )I-op) *(Rt+, )'('P-) ,

using equation (11b).

Substituting this expression into equation (9) gives equation (15).

This content downloaded from 144.92.122.218 on Thu, 11 Apr 2013 14:18:10 PMAll use subject to JSTOR Terms and Conditions