Jun 19, 2005 - Scientific innovations will never be a panacea in this regard ... quality would reduce the performance of the creamery'scostlycreamseparator,.
Information Technologies and New Agricultural Products to Increase and Sustain Rural Incomes
By David A. Hennessy*
Presented at New Markets Workshop Sonoma Valley Inn
June 19, 2005
*Center for Agricultural and Rural Development and Department of Economics at Iowa State University, Ames, IA 50011-1070. Comments of GianCarlo Moschini are appreciated.
1
Introduction “ Me a s ur ewha ti sme a s ur a bl e ,a ndma keme a s ur a bl ewha ti snots o. ”Galileo Galilei “ Tome a s ur ei st oknow. ”Lord Kelvin “ I fy ouc a nnotme a s ur ei t ,y ouc a nno ti mpr ovei t . ”Lord Kelvin “ I nphy s i c a ls c i e nc et hef i r s te s s e nt i a ls t e pi nt hedi r e c t i onofl e a r ni nga nys ubj e c ti st of i nd principles of numerical reckoning and practicable methods for measuring some quality connected with it. I often say that when you can measure what your are speaking about, and express it in numbers, you know something about it; but when you cannot measure it, when you cannot express it in numbers, your knowledge is of a meager and unsatisfactory kind; it may be the beginning of knowledge, but you have scarcely in your thoughts advanced to the state of Sc i e nc e ,wha t e ve rt hema t t e rma ybe . ”Lord Kelvin Tongue in cheek, let’ s modify this last quote a bit to given what I think is half-true. “ I nbusiness the first essential step in the direction of economic profit from a market is to find principles of numerical reckoning and practicable methods for measuring some quality connected with it. … … ; but when you cannot measure it, when you cannot express it in numbers, your knowledge is of a meager and unsatisfactory kind; it may be the beginning of profit, but you have scarcely in your thoughts advanced to the state of Prosperity, whatever the ma t t e rma ybe . ”Apologies to Kelvin One further comment ont ha tquot a t i onr e ga r dst hewor d‘ uns a t i s f a c t or y . ’TheLatin root is in c ommonwi t ht ha tof‘ f a c t or y ’whe r et h el a t t e ri st odopl ur a lpr oc e s s e sa ndwhe r e‘ s a t i s f a c t or y ’ is to do enough. So the idea that performance and a systematic approach are linked is old. Good information, and so measurement, is essential in securing acceptable performance. This is especially true for new products, where the potential for revenue enhancement and cost reduction are largely unknown. It is perhaps ironic that some of the major steps toward industrialization, such as development of water power for wheat milling and the assembly line in meat packing, had agricultural context. Yet these innovations were taken up more readily elsewhere, and I contend that the primary reason was the failure to understand the nature of raw material at hand.1 The 1
Carriquiry (2004) provides an interesting discussion on the need for beef tenderness measurement technologies in meeting consumer demands.
period 1850 through to the present saw major advances in the chemistry, physics, and biology of raw materials, but the economic yield from these advances were different in kind across sectors. The innovations created new sectors throughout the economy, and then helped to advance these sectors through cost reductions and product development. In agriculture, innovations were largely about cost reduction at the production and processing phases. That is changing toward an emphasis onpr oduc tde ve l opme nt .Ke l vi n’ sthird quote concerns Science and not Engineering. In a laboratory, control may be easy. In practice, control over large volumes of material may be impossible. In the case of biological organisms, this is particularly true because organisms are distinct. The science of Genetics has provided the form of innovation that allows for advances in both scientific knowledge and engineering control, and so has advanced the rate of product development in food markets. The intent of this paper is to discuss some economic roles for information technologies, largely measurement or monitoring technologies, in how food is produced and processed. The paper first presents standard ideas in the theory of incentives under asymmetric information and discusses their relevance to food product development. We then turn to two models intended to convey insights on determinants of efficiency under symmetric information, and how more information can affect product development. We also discuss a variety of specific issues regarding how information technologies determine the nature of supply and demand for food.
Information Technologies and Stylized Information Asymmetry Problems A consequence of our partial understanding of food is the importance of information related market failures in food markets. Scientific innovations will never be a panacea in this regard because innovations create new demands for information that may be difficult to fulfill, just as it facilitates the fulfillment of existing demands. Following the standard delineation, as in MachoStadler and Pérez-Castrillo (2001), we will describe four broad classes of issues in the economic
2
information failure and comment on their relevance in food markets.
Moral Hazard: The problem here is non-verifiable and non-imputable effort on the part of an agent in fulfilling the terms of a contract with a principal. If the production process were completely understood and all other factors wereme a s ur a bl e ,t he na na ge nt ’ sa c t i onsc oul dbe imputed even if they could not be directly measured. So moral hazard involves two sorts of measurement issues. One illustration of some moral hazard issues that can arise in food markets is that of late 19th Century competition between Danish and Irish creameries for the growing English dairy market. The Danish share of the British butter market increased from 10.3% to 36.6% over 1881-1900 whe r e a st heI r i s hs ha r ede c l i ne df r om24. 5% t o16. 8% ( He nr i ks e na ndO’ Rourke (1999)). Henriksen and Hviid (2004) claim that primary reasons for this change in market share were a) how the cooperative system enforced commitments to deliver, and b) how it monitored and punished milk adulteration. Scale-biased technical change in butter production shifted processing off the farm to the local creamery. Identity was lost during transportation and processing so that the farmer could not be compensated for quality produce. By 1895, the Gerber test was available to test for fat content, and it was conducted through analysis of random on-farm samples. While Henriksen and Hviid have nothing to say about testing in Ireland, they do assert that a) the cooperative format was better able to administer the test than private creameries because supplying members were often used to oversee and lend credibility to testing, and b) the cooperative movement was stronger in Denmark than in Ireland, in part because longterm supply contracts were deemed a restraint on trade under British law. A second moral hazard problem regarding farm organization and performance has been raised by Allen and Lueck (1998). The importance of nature as an input in farming creates considerable noise in production. Technical aspects of modern farming may suggest potential for substantial gains from specialization, even in the face of seasonal tasks. But specialization
3
would involve principal-agent relations in the presence of noise-induced monitoring difficulties. They contend that, at least as far as crop agriculture is concerned, the family farm is organizationally efficient in eliminating the need for hierarchical monitoring of agent activities. In neither case are new product markets directly at issue. Step back, though, and consider the downstream consequences of a monitoring technology. In the case of milk, uncertain and unreliable milk quality would reduce the performance of the c r e a me r y ’ sc os t l yc r e a ms e pa r a t or , the extent of gains from other centralized processing activities, distribution efficiencies, and so the overall quality of produce. In the case of an on-farm labor monitoring or environmental control technology, such an innovation may allow for labor specialization efficiencies to be exploited so that returns on human capital investments and overall produce quality are likely to increase.
Adverse Selection: The problem here is that an agent possesses relevant private information before a relationship is initiated, and uses that information to advantage. The problem cannot be overcome by measuring agent behavior because it is the nature of the agent that is at issue. When quality is costly to produce and consumers cannot observe quality, then producers have little incentive to produce quality. Producer types well-positioned to produce quality, but at a cost, will not be rewarded for doing so and may either not incur the cost (moral hazard) or leave the market to other types. The idea of adverse selection in market relations explains the following mechanism. Danish veterinary health officials categorize cattle herds into three Salmonella Dublin status classes (Danish Zoonosis Centre, 2004). Commencing about three years ago, this information on herd types became public knowledge, and animal purchasers appear to use the information when trading. Herds, and especially herds with the best rating (Level I), have used this information to trade only with Level I herds. This example illustrates a way in which a public monitoring program, one that might be infeasible and in any case inefficient for a private trader to operate
4
(Barzel, 1982), can overcome the adverse selection problem to provide higher levels of safety to the consumer. Other public efforts to overcome moral hazard and adverse selection problems in food markets have posed difficulties for international trade, especially with regard to trade between low income and high income countries. Jaffe and Henson (2004) provide anecdotal evidence that some food export sectors in low income countries can, depending on circumstances, respond to meet higher standards in wealthier importing countries whereas others cannot. Public monitoring and enforcement can sort types out and/or change actions.
Signalling: This is an approach that a producer can take to overcome an adverse selection problem. While unable to credibly signal its type directly to a food consumer, a grower may do so indirectly by undertaking actions that have little to do directly with product quality but are not as costly for more desirable types. As has been pointed out by Bureau, Gozlan, and Marette (2001), voluntary testing to signal quality will be less costly to producers of better quality food. In the context of trade liberalization to admit imports from countries where it is more difficult to detect low quality (unsafe) product, liberalization benefits domestic consumers but may reduce overall domestic welfare by increasing the need to engage in costly testing. An interesting related idea is provided in Lapan and Moschini (2004), who consider the introduction of a substituting good (call it GM) that some consumers consider to be inferior to an existing good. In order to garner a premium relative to the GM good, the traditional product must signal by engaging in costly identity preservation. That the introduced good is weakly inferior in the eyes of consumers matters because the cost of signaling a distinction between the growers falls on producers of the traditional good and not on producers of the newer good. Because incentives are poorly aligned, advent of the newer technology may reduce welfare.
Collective Reputation: Regional, varietal, branding, and other designations have proven to be
5
enduring marketing devices for producers seeking to benefit from strong demand for the connoted attributes. As has been pointed out by Tirole (1996), there is a collective reputation for the group as a whole. However, there are often measurement problems in that the complete hi s t or yofapa r t i c ul a rg r owe r ’ sbe ha v i orma ynotbea va i l a bl et ot hema r ke t . To the extent that this is true, there will be an incentive to free-ride on quality enhancing actions. In turn, the g r oup’ sc o l l e c t i vepa s tbe ha vi ora f f e c t se a c hme mbe r ’ spr e s e ntbe ha vi orand is likely to affect the behavior of new members. Thus, the performance of a product in the marketplace will be determined by the collective history of producer behaviors. Winfree and McCluskey (2005) adapted the idea of dynamic reputation to a dynamic game over common property (i.e., reputation). Their direct concern was with collective reputation for Washington Apples, where Red Delicious Apple eating quality is held to have declined in recent years. They suggest that the decline was due to problems in grading, where the emphasis is on measuring color, shape, and size. Measurable attributes more strongly correlated with eating quality, such as Brix (basically, sugar content) and acidity, are not measured. This, they argue, provides Washington apple growers that do not brand further with insufficient incentive to maintain collective reputation by engaging in costs to support eating quality attributes.
Symmetric Information, New Products, and Efficiency The intent of models I and II to follow is to demonstrate that information about raw materials is closely linked to performance in processing, and so to the incentive to invest in additional processing activities.
A. Cost reduction and de-skilling: Model I There are two raw material types, A and B. One worker on a processing line must make a single decision concerning the raw materials. The fraction of units of raw materials that are type A is
, while the residual are type B. Workers are indexed by skill parameter , the cardinal
6
meaning of which is as follows. If a worker is of skill type , then fraction of type A animals signal as type A, i.e., the probability of signaling as type A is . Restrict 0.5 and [0,1 ] so that sA makes sense as a signal for A. Likewise, fraction of type B animals signal as type B, i.e., the probability of signaling as type B is . And we restrict
0.5 with [0,1 ] so that [0,1 max[, ]] . We set the upper bound on as
1 max[, ] . A larger value represents more competence as a decision-maker. Table 1 summarizes the data. The loss to the processor from an incorrect diagnosis on the part of the worker is . Now a ppl yBa y e ’ st he or e m;
(1)
( ) P ( A | sA, , ) ; 1 2 ( )(1 ) P ( B | sB , , ) . 2
We assume throughout that workers have incentives compatible with the minimization of processor expected loss. Receiving sA: If the signal is sA and the worker chooses B, then the expected loss to the processor is ( ) /[1 2 ] . If the signal is sA and the worker chooses A, then the expected loss is (1 )(1 ) /[1 2 ] . With signal sA, then the worker should choose2 (2)
A
if
1 sA , 1
B otherwise.
The expected loss conditional on signal sA is
2
When sA , then the worker is assumed to choose B. 7
(3)
min 0, sA L( sA, , ) (1 ) . 1 2 1 2
Receiving sB: If instead the signal is sB and the worker chooses A, then the expected loss to the processor is ( )(1 ) /[ 2 ] . If the signal is sB and the worker chooses B, then the expected loss is (1 ) /[ 2 ] . With signal sB, then the worker should choose3 (4)
B if
sB , 1
A otherwise.
The expected loss conditional on signal sB is (5)
min 0, sB L( sB, , ) (1 ) . 2 2
Note that 0.5 , 0.5 , and 0 ensure that sA 0.5 sB . Clearly, Remark 1. If sB ( sA ), then the worker should assume the raw material is of type A (B) regardless of signal observed. If (sA ,sB ) ,t he nt hewor ke r ’ sopt i ma lde c i s i onwi l lb e consistent with the signal observed.
Ex-Ante Expected Loss Function: Next consider unconditional expected loss. Taking expectations, the unconditional expected loss is L ( , )
(6)
(1 ) min 0, sA ) min 0, sB . (1
The expected loss function (ELF) is piece-wise linear in , and may alternatively be written as
3
When sB , then the worker is assumed to choose A. 8
(7)
L ( , ) (1 )
if
sA ;
if
sA sB ;
if
sB .
Remark 2. Expected loss increases with on [0, sA ] and decreases with on [sB ,1] . It increases (decreases) with on (sA ,sB ) whenever ( ).
Figure 1 depicts expected loss in the dimension for 1 and . We provide two values of , 0.1 and 0.15 . Regarding remark 2, notice that if then a processor would prefer to be at sA than at sB . This is because raw material type B is more heavily weighted in the mix at sA , and condition indicates better worker performance in detecting type B than detecting type A. All else equal, ( ) suggests a bias on the part of the processor in favor of procuring type B (type A) raw materials. The effect of an increase in is immediate and obvious. Observe that dsA / d0 dsB / d , while d L ( , ) / d 0 on sA sB and d L ( , ) / d0 otherwise.
Remark 3. Expected loss decreases with .
Thus, the region on which loss is sensitive to expands when increases and the magnitude of the loss is reduced (weakly) everywhere when increases. Uniformity in raw materials: The relation between and aside (see remark 2 above), whether A or B dominates in the mix of raw materials, should be of little relevance to worker performance in our model. This is because we have not distinguished between the materials in any way except to assert that they are different in a manner that affects treatment and that a worker may differ in his ability to detect types. Remark 2 suggests that uniformity in raw materials is good for 9
processor profits. To be clearer on what uniformity in raw materials means, we offer a definition of‘ mor euni f or m. ’ Definition 1. Raw material is said to become more uniform if (0, 0.5] and then decreases or if [0.5,1) and then increases.
Since d L ( , ) / d0 on sA 0.5 and d L ( , ) / d0 on sA , a reduction in the value of on (0, 0.5] weakly increases the value of d L ( , ) / d. Since d L ( , ) / d0 on 0.5 sB and d L ( , ) / d0 on sB , an increase in the value of on [0.5,1) increases the value of d L ( , ) / d. This logic allows us to assert: Proposition 1. Uniformity in raw materials and worker skill are substitutes.
This point has been made in the context of pesticide use on crops by Vandeman (1995), among others. Now assume that worker skill can be purchased at price w() : [0,1 ] , a continuously differentiable, strictly increasing, and convex function. Then the processor seeking to minimize expected cost will establish (8)
C () min [0,]
L ( , ) w(),
with minimizing argument * () . When solving this program it is useful to write the ELF in another way;
(9)
if 0.5 and sA 1 (1 ); if 0.5 and sA 1 (1 ), L ( , ) or if 0.5 and sB (1 ) ; (1 ) if 0.5 and sB (1 ) .
Demand for skills should vary inversely with the value of , but our model carries with it more
10
structure than that. To pursue the role of in demand for skills, let us assume that 0.5 . The other case can be treated symmetrically. A diagram of L ( , ) along [0, ] has L ( , ) on [0, sA ] and L ( , ) on (sA , ] . Figure 2 depicts the function. If w(sA ) , then the convexity and strict monotonicity properties of w() ensure that the
processor will choose * () 0 .4 This is because function L ( , ) w() is increasing along
[0, ] whenever w(sA ) . If w() , the same reasoning establishes that * () or * () 0 . This is because L ( , ) w() is increasing in on [0, sA ) and decreasing in on (sA , ) , so the set of possible global minimizers can be restricted to the pair of end-points {0, } . If w(sA ) w() , then L ( , ) w() has an interior local minimum on (sA , ) . A similar analysis can be engaged in when 0.5 . Proposition 2. For a given skill-wage schedule w() there exists a least l 0 such that * () 0 l and a greatest u l such that * () {0, } u . If (l ,u ) , then * () {0 (sA , )} .
Note, it is never optimal for * () sA because sA would have to be the local minimizer on [sA , ] and * () 0 would then be the global minimizer. For the case drawn in figure 2, it is fairly clear that * () . The reason is that is high relative to the cost of procuring more skill in the market. Proposition 1 suggests that worker skill should decrease with an exogenous increase in
11
uniformity of raw materials. But how the change occurs is important when seeking to understand incentives to remove labor from the processing activity. To better understand some consequences of more uniformity in raw materials on the demand for worker skill, note that dsA / d1 0 . A consideration of marginal effects, see figure 3, establishes the following observations. Letting 2 1 , notice that difference
(10)
( 2 1 ) L ( , 2 ) L ( , 1 ) 2 ( ) 1 ( )(1 2 )
on [0, sA ( 1 )]; sA on (sA ( 1 ), ( 2 ));
on [sA (2 ), ];
is (weakly) increasing in . Proposition 3. Let (0, 0.5] and 2 1 . If * (1 ) , then either * (2 ) or
* (2 ) 0 . If * (1 ) (0, ) , then either * (2 ) * (1 ) or * (2 ) 0 . If * (1 ) 0 , then * (2 ) 0 .
Automation: An automation technology becomes available. The technology replaces the worker with capital where the replacement cost is K , with K w(0) so that automation can save on labor costs. The automation technology has no capacity to distinguish between types, and so the expected loss under automation is D () K . In figure 4, both C () and D () are graphed on [0, 0.5] . From (7), i.e., 1 , as well as the relation between restricted and unrestricted cost functions, when (sA , 0.5) then C () increases more slowly than does D () . If the expected loss functions do cross as increases, then they cross just once and C ()
4
w(sA ) means dw() / d evaluated at sA .
12
crosses from above D () to below C () .5 Therefore, Proposition 4. Let (0, 0.5] and K w(0) . More uniformity in raw materials can make it optimal to change from the labor-intense technology to the automation technology, but it cannot make it optimal to change from the automation technology to the labor-intense technology.
The proposition is informative in the following sense. Propositions 1 and 3 suggest that the skill premium will shrink as raw materials uniformity increases. At face value, this substitution may look like good news for the less skilled, even if the increasing substitutability between wor ke r ss t r e ng t he nst hee mpl oy e r ’ sc a pa c i t yt oe xt r a c ts ur pl us .Butr a wma t e r i a l suni f or mi t y also facilitates automation, and so biotechnology and chemical innovations that promote input uniformity may eventually be bad for the less skilled worker.
B. Conditioning with information on raw materials: Model II A firm handling a single input can engage in n processing steps at cost C ( n) , an increasing, twice continuously differentiable and strictly convex function. We have in mind the situation where an additional step might involve turning wooden planks into furniture, cutting blown glass for fine crystal, aging wine, or fitting finished product for a more demanding export market. We allow the number of processing steps to assume any non-negative real value. Benefits per unit of successfully processed raw material amount to B ( n) , an increasing, twice continuously differentiable and strictly concave function. The raw material can come in M types, with shares sm 0 ,
mM
sm 1 , and m M {0,1, ... , M 1} . The types could be
distinguished by physical, biological, or chemical attributes, e.g., heat content, genetic origin, or
5
A similar analysis applies when 0.5 . 13
acidity. Collectively, the share simplex coordinates are described as S ( s0 , s1 , ... , sM 1 ) . Due to comparative familiarity with types, the number of things that can go wrong during a processing step when processing a given type decreases with the share of that type in the raw material. We index the number of things that can go wrong with type m by g ( sm ) , a positive and decreasing function that may be considered to be an index of scale efficiency in learning-bydoing. Thus, processing will be better geared toward accommodating the mth type if sm is comparatively large. The probability that one unit of type m in the lot is affected by one of these potential failure sources is (0,1) . Each failure source is independent so that, in share form, the probability the unit does not fail at a step is (1 ) g ( sm ) and the probability it does fail is 1 (1 ) g ( sm ) . With 1 and upon aggregating over types, the unconditional (i.e., not type conditioned) probability that a failure occurs on a unit at a given step is (11)
mM
g ( sm ) sm 1 g ( sm ) 1 m sm , M
and the unconditional probability a unit does not fail the step is (12)
I ( S ; ) m smg ( sm ) ;
1 .
M
where I ( ) [0,1] and I ( S ; ) |0 1 . Product is tested at the end of n steps only so that one cannot terminate the process early in the event of a failure. Failures are independent events across steps so that the unconditional probability a unit survives the entire process is (13)
n . I ( S ; )
Failed product has value 0 so that expected profit per unit is (14)
V ( n; S , ) B ( n) I ( S ; ) C ( n) B ( n)e nLn[ I ()] C ( n). n
14
The first- and second-order conditions with respect to the extent of processing are (15)
(15.1)
Bn (n)e nLn[ I ()] B ( n)Ln[ I ( )]e nLn[ I ( )] Cn ( n) 0;
(15.2)
Bnn ( n)e nLn[ I ()] 2 Bn ( n)Ln[ I ( )]e nLn[ I ( )] B ( n) Ln[ I ( )] e nLn[ I ( )] Cnn ( n) 0. 2
Insert (15.1) into (15.2) to obtain (16)
Bnn ( n)e nLn[ I ()] Bn ( n)e nLn[ I ( )] Cn ( n) Ln[ I ( )] Cnn ( n).
This is certainly negative, so that the second-order condition is satisfied at any interior optimum. Due to continuity, any interior optimum must therefore be unique. Henceforth we assume an interior solution and label it as n n* , and as n n* ( S ) when the emphasis is needed. The cross-derivative of (15.1) with respect to Ln[ I ( )] is n* Bn ( n n* )e n Ln[ I ()] B( n n* )e n Ln[ I ()] n* B ( n n* )Ln[ I ( )]e n Ln[ I ( )] *
(17)
*
*
n* Bn ( n n* ) B ( n n* ) n* B( n n* )Ln[ I ( )] e n Ln[ I ( )] *
n*Cn ( n n* ) B ( n n* )e n Ln[ I ()] 0, *
where (15.1) has been employed. Consequently, d 2V ( n; S , ) / dndI ( ) 0 when n n* . We will use this observation shortly. We seek to understand how the share simplex allocation vector S affects the incentive to process. To this end the concept of majorization is relevant. Definition 2. (Marskall and Olkin, 1979, pp. 10 and 59) Vector Q n is majorized by Q n (written as Q Q ) if
qi 0 q( k n and i) i 0 ( i ) k
k
N 1
i 0
where the q( q i ) i 0 ( i ) N 1
q( i ) are defined as order statistics, q(0) q(1) ... q( N 1) . A Schur-convex function U (Q ) : n satisfies the statement: U (Q ) U (Q ) whenever Q Q .
To illustrate, let S (0.1, 0.3, 0.6) and S (0.65, 0.05, 0.3) . Then the cumulants under S
15
are 0.05 , 0.35 , and 1 . Since 0.05 0.1 , are 0.1 , 0.4 , and 1 , while the cumulants under S 0.35 0.4 , and 1 1 , the definition asserts that S S . The definition captures the idea of is more concentrated. As it happens, S concentrates more homogeneity/uniformity because S
on s0 but the definition is symmetric in placing no preferences for any coordinate, i.e., quality type. The concept has been used extensively in the literature on income inequality because it is an alternative presentation of the Lorenz curve in discrete form (Dasgupta, Sen, and Starrett, 1973; Shorrocks, 1983). and g ( Proposition 5. If S S ) is concave, then firm profits and extent of processing are than under S larger under S . than under S (i.e., is Schur – Proof of Proposition 5. Statistic I ( S ; ) is larger under S n
than under S convex) if I ( S ; ) m smg ( sm ) m sme g ( sm )Ln () is larger under S . M
M
From Marshall and Olkin (1979, p. 11), this is true for all majorizing vectors if and only h( sm ) sme g ( sm ) Ln () is convex in sm . The derivatives are
hsm ( sm ) e g ( sm )Ln () sm g sm ( sm )Ln()e g ( sm )Ln ( ) ;
(18)
2 hsm , sm ( sm ) 2 g sm ( sm ) sm g sm ( sm ) Ln() sm g sm , sm ( sm ) Ln()e g ( sm )Ln () sign
2
2 g sm ( sm ) sm g sm ( sm ) Ln() sm g sm , sm ( sm ) 0.
Since h( sm ) is convex, we have V ( n* ( S ); S , ) V ( n* ( S ); S , ) V ( n* ( S ); S , ) after reoptimizing. Finally, from (17), an increase in the value of Ln[ I ( S ; )] , i.e., of I ( S ; ) , due to increases the marginal value of processing and so increases the extent of processing S S
when processing and input homogeneity are complements in production.
16
G
Notice that sm g ( sm ) concave suffices to ensure that hsm , sm ( sm ) 0 . The condition that g ( ) be concave is reasonable in that it may be interpreted as diminishing marginal gains from learning by doing. is partial, not being able to Entropy and information content of raw materials: Relation S S
rank all pairs of weighting coordinates on the unit simplex. A specific function form on g ( sm ) , rather than say the set of decreasing and concave functions as above, would be necessary in order to completely rank the weighting coordinates in terms of their implications for processor profits and decisions. Summary statistics, such as (higher) variance or (lower) entropy rather than S S , provide a complete ordering on data. But that complete ordering will provide an
inappropriate level of exactness when there is only limited knowledge about the context being studied, as was the case with g ( sm ) above. The capacity to control inputs is predicated upon information and technologies using that information. Most overtly, one might receive signals about raw material and then use the information to physically sort the materials into homogeneous lots. Alternatively, as in Chalfant et al. (1999), the information may be embedded in the technology where a sieve or grader sorts existing raw materials. In other cases, technology can be used during the production of the raw materials to endow the raw materials with information regarding the extent of order or homogeneity on the raw materials. This is the case in the manufacture of steel, and when genetics are used to control the nature of the beast to be born. Sorting mechanisms can employ economic incentives, as when providing agents with a menu of contracts in order to ensure that agents with private information deliver similar (say higher quality) inputs. The most widely used statistic intended to depict order in a system, be it regarding energy flows or information flows, is the Boltzmann-Shannon entropy statistic. Theoretical motivation
17
is provided by Weitzman (2000) concerning its use as a measure of ecological diversity. For our purposes we write it as (19)
E ( S ) m smLn( sm ); M
sm 0 m M ;
mM
sm 1.
It is concave in the sm , measuring the extent of disorder rather than the extent of order in the implies E ( S system. Note that S S ) E ( S ) . We ask whether a reasonable technology
and circumstances on the nature of uncertainty exist such that our index I ( S ; ) and the (inverse) entropy index E ( S ) are essentially the same? The answer is in the affirmative when
0 , i.e., when failure due to a given cause and at a given step is rare and when g ( ) takes form g ( ) k Ln( sm ) .6 Proposition 6. If 0 and g ( ) k Ln( sm ) , then firm profits are increasing in the extent of raw materials uniformity as measured by the negative of Boltzmann-Shannon entropy. Proof of Proposition 6. Bearing in mind that e x e0 1 x 1 x in the neighborhood of x 0 , take a first-order Taylor series expansion of e g ( sm ) Ln (1) near 0 to obtain the
approximate change in value as 1 g ( sm )Ln(1 ) 1 g ( sm )Ln(1 ) plus terms of order two and higher so that
(20)
mM
sme g ( sm )Ln (1) 1 Ln(1 )m sm g ( sm ) M
Ln(1 )m sm k Ln( sm ) kLn(1 ) Ln(1 )m sm Ln( sm ). M
M
Now take a first-order Taylor series expansion of B ( n) I ( S ; ) near 0 ; n
n 1
n
(21)
6
B ( n) I ( S ; ) B ( n) nB( n) I ( S ; ) |0 Ln(1 )m sm g ( sm )
M
B ( n) nB ( n) I ( S ; ) |0 Ln(1 ) k m sm Ln( sm ) . n 1
M
Weitzman posed a problem that was broadly similar, and our proof is similar to that for his
18
But Lim0 nB ( n) I ( S ; ) nB ( n) so that n 1
(22)
B ( n) I ( S ; ) B ( n) nB( n)Ln(1 ) k E ( S ) . n
This is decreasing in the value of E ( S ) .
G
Notice that g ( ) k Ln( sm ) is convex and so the conclusions in Proposition 1 do not follow. However, apply the last line in (8) to g ( ) k Ln( sm ) and obtain (23)
2 1 2 g sm ( sm ) sm g sm ( sm ) Ln() sm g sm ,sm ( sm ) 1 Ln() 0. sm
Upon reconsidering the proof of Proposition 5, it can be seen that the assertion applies for g ( ) with E ( S k Ln( sm ) also. However, we cannot replace S S ) E ( S ) in the proposition
when not in the neighborhood of 0 .
Some Issues A. Traceability Demand for bundling an agricultural product with its production history has grown in recent years. This is partly because product information is likely a luxury attribute, growing with consumer income. It is also partly because of emerging systemic problems that include commingling and infectious disease in the presence of expanding international trade in food and feed. Hobbs (2004) identifies three functions for a traceability system. In her terminology, the reactive traceback function allows for a more efficient and effective recall of product in ex-post reaction to a problem. The liability function is also ex-post in nature, allowing for identification of the origin of a problem in order to assign liability. It should reduce the costs of monitoring and enforcement. The quality verification function is ex-ante in providing consumers with Theorem (p. 255). 19
documented information on the product before consumption. This reduces consumer transactions costs. Whi l eHobbs ’c onc e r nwa swi t hi nf or ma t i ona s y mme t r y ,i nt hel a r g e rpi c t ur eof strategic management one should pose another function for tracing. Limited knowledge in life sciences means that we know little about health consequences, deterioration, and other problems to with food use. As suggested in Clemens and Babcock (2002), tracing will allow producers and processors to better understand and adapt to consumer needs. Hennessy, Miroanwski, and Babcock (2004) point to the use of on-farm information systems and environmental control in production to learn and innovate. Animal-level and lot records that are linked across the supply chain should be nigh essential in this management strategy.
B. Franchising Food franchises want products that are consistent with the franchise brand, be that quick and tasty or the idealized traditional Sunday meal. Ritzer (2004) has decomposed the McDonaldization phenomenon into four dimensions. The efficiency dimension involves meeting consumer needs for a meal and respite at low time and effort costs on the part of the customer while making efficient use of franchise resources. Calculability requires that the consumer have a firm sense of the deal being offered, in terms of price, time commitment, meal satisfaction, comfort, and nutrition. Workers too should have a strong sense of employment requirements. Predictability involves reliability in delivering on these implicit contracts with the consumer and employee. Control through non-human technology is the tenet that the workplace physical infrastructure and corporate protocols, in addition to social culture, ensure that the realistic expectations of employers, customers, and workers are met. The success of food franchising suggests that these perspectives form a coherent business strategy. Placement of food in this retail channel requires that the product be highly standardized so that it is readily prepared according to a given set of simple instructions. It requires that taste
20
and other discernable quality attributes are consistent with brand perception and consistent over consumption experiences. In addition, it requires that the raw materials facilitate automation in preparation and service. The information demands on food attributes from this retail channel are strong, and these demands are likely to increase as more becomes known about how to identify and design foods that meet franchise needs.
C. Antibiotics Antibiotics are used in animal production to remedy a disease outbreak, to prevent disease, and as a growth promotant in the healthy animal. Concerns exist that widespread use in animals has encouraged resistance to antibiotics used in human medicine. In response, in recent years antibiotic use in animal production has been curtailed through legislation, and by voluntary commitments on the part of food production chains, see Hayes and Jensen (2003). Quite apart from effects on mean performance of resources at production and processing phases, antibiotics promote consistency across animals. Thea nnounc e me nti nJ une2003t ha tMc Dona l d’ s Corporation would voluntarily phase out use of some antibiotics by its animal suppliers was seen as a landmark decision concerning the direction of animal production. In order to continue receiving the consistency they expect, retail food channels that prohibit such production control technologies will have to impose other forms of control.
D. Industrialization We bs t e r ’ sThi r dNe wI nt e r na t i ona lDi c t i ona r ya s s e r t st ha t‘ i ndus t r i a l ’me a ns‘ t opr oduc eby s y s t e ma t i cl a bor . ’I fas y s t e mi st owor ke f f i c i e nt l y ,t he nt he r emus tbeme a s ur e me nt .Ast h e breadth of demands from increasingly affluent consumers grows, increasingly systematic approaches will be necessary to ensure that the consumer gets what she expects. This will be true for highly processed foods, for GM-f r e ef oods ,f or‘ or ga ni c ’f oods ,f or‘ f a i rt r a de ’f oo d s , and even for a food designation requiring that produce came from a traditional family farm.
21
Ironically, the existence of industrial farming places the onus on non-industrial farms seeking surplus for that attribute to demonstrate the case by carefully documenting their activities. They must adopt some parts of the industrial approach. Capitalism and industrialization have been associated ideas in modern history due partly to the fact that factories are capital intensive. By contrast with labor, capital tends to be inflexible so that standardization is at a premium (Boehlje, 1996). To the extent that a new product requires additional capital inputs, control over and so information on inputs will complement these capital inputs. The food raw material base will have to better integrate with processing in order to provide higher-end processed products.
Summary Who will gain from the advent of information technologies that assist in bringing new products to market? Those with most to gain are consumers, whose demands are being addressed. Whether all consumers will be better off is another issue. In the fast food and mainstream products segments, the resonance that activists have achieved with a significant segment of consumers suggests that some negative externalities are being created as far as the preferences of these consumers are concerned. Externalities frequently mentioned include animal welfare issues, information failures regarding what foods are labeled as, strong perceptions about what others should eat, concerns about the social costs of poor diets, or reduced liquidity in markets f orf oodsd e e me d‘ whol e s ome ’byt hi sc ons ume rs e g me nt . Consumers with alternative preferences can also benefit from information technologies. Apart from strengthening incentives in the production and processing of alternative foods, these technologies assist in verifying the authenticity of the food being offered for sale. Whether the price and information benefits outweigh disutility arising from consuming a food that is known to originate from a highly controlled production environment is for these consumers to decide. The decisions they make in this regard will determine organizational forms on alternative farms,
22
e.g., whether Community Supported Agriculture with significant consumer participation through worker share discounts assumes a significant fraction of production. As usual, the processors and retailers that gain will be those that obtain by foresight or good fortune comparative advantage on the supply side of the market. This is true also of producers. Growers that deliver information processors and consumers want, do so without ceding the rights to the information, and without becoming involved in costly bargaining games will benefit most. They would be particularly well positioned to gain from the provision of such information if competition in supply is restricted, as when situated in a small geographical location that consumers associate with high quality. In some cases, though, information technologies and technologies that substitute for informed animal-level husbandry may contribute to marginalizing a whole class of growers.
References Al l e n,D. W. ,a ndD.Lue c k.“ TheNa t ur eoft heFa r m. ”Journal of Law and Economics 41(October 1998):343–386. Ba r z e l ,Y.“ Me a s ur e me ntCos ta ndt heOr g a ni z a t i onofMa r ke t s . ”Journal of Law and Economics 25(April 1982):27–48. Boe hl j e ,M.“ I ndus t r i a l i z a t i onofAg r i c ul t ur e :Wha ta r et heI mpl i c a t i ons ? ”Choices 11(First Quarter 1996):30–34. Bureau, J.-C. ,E.Goz l a n,a ndS.Ma r e t t e .“ Qua l i t ySi g na l l i nga ndI nt e r na t i ona lTr a dei nFood Pr oduc t s . ”Wor ki ngPa pe r01-WP 283, Center for Agricultural and Rural Development, Iowa State University, August 2001. Carriquiry, M. “ Guaranteed Tender Beef: Opportunities and Challenges for a Differentiated Agr i c ul t ur a lPr oduc t . ”Working Paper 04-WP 371, Center for Agricultural and Rural Development, Iowa State University, September 2004. Cha l f a nt ,J . A. , J a me s ,J . S. ,La voi e ,N. ,a ndR. J .Se xt on.“ As y mme t r i cGr a di ngEr r or and Adverse Se l e c t i on:Le monsi nt heCa l i f or ni aPr uneI ndus t r y . ”Journal of Agricultural and Resource Economics 24(July 1999):57–79. Cl e me ns ,R. ,a ndB. A.Ba bc oc k.“ Me a tTr a c e a bi l i t y :I t sEf f e c tonTr a de . ”Iowa Ag. Review 8(Winter 2002):8–9. Danish Zoonosis Centre. Annual Report of Zoonoses in Denmark, 2003. Ministry of Food, Agriculture and Fisheries, Søborg Denmark, 2004. Dasgupta, P., Sen, A.K., and D. Starrett. “ Notes on the Measurement of Inequality.”Journal of Economic Theory 6(2, 1973), 180–187. Ha y e s ,D. J . ,a ndH. H.J e ns e n.“ Le s s onsCa nBeLe a r ne df r om Da ni s hAnt i bi ot i cBa n. ”
23
Feedstuffs 75(September 15, 2003): p. 1 and p. 16. He nne s s y , D. A. ,Mi r a nows ki ,J . A. ,a ndB. A.Ba bc oc k.“ Ge ne t i cI nf or ma t i oni nAg r i c ul t ur a l Productivity and Product Develop me nt . ”American Journal of Agricultural Economics 86(February 2004):73–87. Henriksen, I., and M. Hviid. “ Cont r a c t si nt heGove r na nc eoft heEa r l yDa ni s hDa i r ySe c t or . ” Unpublished working paper, School of Law, University of East Anglia, Norwich, U.K., June 2004. He nr i ks e n,I . ,a ndK. H.O’ Rour ke .“ Pr i c e s ,Ma r ke tI nt e g r a t i ona ndQua l i t y :TheLa t e19th Ce nt ur yI nt e r na t i ona lBut t e rMa r ke t . ”Mimeo, Institute of Economics, University of Copenhagen, Denmark, 1999. Hobbs ,J . E.“ I nf or ma t i onAs y mme t r ya ndt heRol eofTr a c e a bi l i t ySy s t e ms . ”Agribusiness 20(Autumn 2004):397–415. Jaffe, S., and S. Henson. “ St a nda r dsa n dAgr o-Food Exports from Developing Countries: Re ba l a nc i ngt heDe ba t e . ”Wor ki ngPa pe rWPS33 48,Wor l dBa nkPol i c yRe s e a r c hSe r i e s , Washington D.C.: 2004. La pa n,H. E. ,a ndG.Mos c hi ni .“ I nnov a t i ona ndTr a dewi t hEndog e nousMa r ke tFa i l ur e :The Ca s eo fGe ne t i c a l l yModi f i e dPr o duc t s . ”American Journal of Agricultural Economics 86(August 2004):634–648. Macho-Stadler, I., and J.D. Pérez-Castrillo. An Introduction to the Economics of Information: Incentives and Contracts, 3rd ed. Oxford: Oxford University Press, 2001. Marshall, A.W., and I. Olkin. Inequalities: Theory of Majorization and its Applications, San Diego, CA: Academic Press, 1979. Ritzer, George. The McDonaldization of Society, Revised New Century Edition. Thousand Oaks, CA: Pine Forge Press, 2004. Shor r oc ks ,A. F.“ Ra nki ngI nc omeDi s t r i but i ons . ”Economica 50(February 1983):3–17. Ti r ol e ,J .“ AThe or yofCol l e c t i veRe put a t i on( wi t ha ppl i c a t i onst ot he persistence of corruption a ndt of i r mqua l i t y ) . ”Review of Economic Studies 63(January 1996):1–22. Va nde ma n,AnnM. ,1995,‘ Ma na g e me nti naBot t l e :Pe s t i c i de sa ndt heDe s ki l l i ngof Agriculture.’Review of Radical Political Economics, 27(3, September), pp. 49B59. We i t z ma n,M. L.“ Ec onomi cPr of i t a bi l i t yve r s usEc ol og i c a lEnt r opy . ”Quarterly Journal of Economics 115(February 2000):237–263. Winfree, J.A., and J.J. McCluskey.“ Collective Reputation and Quality. ”American Journal of Agricultural Economics 87(February 2005):206–213.
Table 1. Probabilities in Identifying Animal Type from Signal If type is A B
then signal is sA with probability 1
24
then signal is sB with probability 1
