Agribusiness Management · Agri-sector specialisation · also AEC-502, AEC-604 Advanced Production Economics
The economics of turning inputs into farm output: production functions, the three basic relationships, optimum input use, efficiency, scale and risk.
In the exams: agricultural master's entrance examinations, factor-product relationships (heavy); agricultural economics PhD entrance and fellowship examinations, production economics (heavy); agricultural scientist eligibility tests (heavy)
Agricultural production economics is the economics of turning inputs into farm output. It takes the theory of the firm and applies it to crops and livestock: how yield responds to fertiliser, water, feed or labour; how one input can replace another; how land and labour should be divided between enterprises; what scale of farm is cheapest; how efficient real farms are compared with the best; how new technology changes all of these; and how risk and the family nature of most farms change the answers. Its tools are the production function, marginal analysis, cost and profit functions, and the econometrics used to estimate them.
For a practitioner the subject is not abstract. Every fertiliser recommendation that says 'apply the dose where the last kilogram just pays for itself' is the rule MVP = MFC. Every ration formulated at least cost, every enterprise mix chosen to spread risk, every resource-use-efficiency study that tells a ministry fertiliser is under-used in one district and over-used in another, and every efficiency score produced by frontier analysis rests on the relationships set out here. Price changes, subsidies and new varieties are all analysed as shifts in these functions.
Worldwide, the subject is taught as a core Master's course in agricultural economics (in the Indian ICAR syllabus as AEC-502, with an advanced PhD course AEC-604) and as the production half of intermediate micro-economics elsewhere. Entrance and fellowship papers test the three relationships, the stages of production, MVP = MFC and MC = MR, isoquant properties and Cobb-Douglas algebra with one-step numericals; PhD and scientist-recruitment papers go further into efficiency frontiers, duality, technical change and risk programming. This manual is written at the level those exams need and in the order a university course teaches it. Worked numbers are illustrative, not data from any country.
Agricultural production economics is the application of the theory of the firm to farming. It studies how inputs (land, labour, water, seed, fertiliser, feed, machinery services, management) are transformed into outputs (grain, milk, meat, fibre), and how a farmer who wants to reach some goal, usually maximum profit subject to limited resources and risk, should choose what to produce, how to produce it and how much to produce. The central tool is the production function, a technical relationship written Y = f(X1, X2, ..., Xn), which gives the maximum output obtainable from each combination of inputs with a given technology. Every later idea in the subject, from the optimum fertiliser dose to the measurement of efficiency and the dual cost function, is derived from this one relationship together with prices. Earl Heady's Economics of Agricultural Production and Resource Use (1952) set the subject out in the form still taught; Heady and Dillon's Agricultural Production Functions (1961) made the estimation of response functions a standard research method. Production economics organises farm decisions around three questions. What to produce is answered by the product-product relationship, which compares outputs that compete for the same resources. How to produce is answered by the factor-factor relationship, which finds the least-cost combination of inputs for a given output. How much to produce is answered by the factor-product relationship, which relates one variable input to one output and finds the profit-maximising level of that input. A fourth question, when to produce, brings in time, discounting and risk. The three relationships are the backbone of every syllabus in the subject, and exam papers at every level ask which relationship answers which question.
The two subjects share their theory but differ in purpose and level of abstraction. Production economics is concerned with general principles of resource use and with the behaviour of a representative farm or a whole sector: how output responds to input and prices, what determines supply and input demand, how technology and scale change productivity. It works with continuous functions, calculus and econometric estimation, and its results feed policy, for example fertiliser subsidy analysis or the supply response of a crop. Farm management is concerned with the individual farm as a business: records and accounts, budgeting, farm planning, linear programming of a particular farm, and the day-to-day organisation of labour and machinery. It applies production-economics principles to one set of resources with discrete choices. A useful short form is that production economics is the 'why' and farm management the 'what to do on this farm'. Both are micro-economic; neither deals with national income or market structure, which belong to macro-economics and marketing.
The static theory assumes a known production function, known input and output prices, a price-taking farm in competitive markets, a single-period decision and a profit-maximising objective. Each assumption is relaxed in later chapters: estimation deals with an unknown function, risk with unknown prices and yields, the household model with objectives other than profit, and technological change with a shifting function. Time is handled by the usual three runs. In the short run at least one input, typically land or a building, is fixed, so the farm works on one short-run production function and faces fixed costs. In the long run all inputs are variable and the farm chooses its scale. In the very long run technology itself changes. The distinction matters for decisions: in the short run the farm should keep producing as long as revenue covers variable cost, even if it does not cover fixed cost, because the fixed cost is paid either way; in the long run every cost must be covered or the resources should leave the enterprise.
The form chosen for a production function is not neutral: it fixes in advance what shapes of response the data are allowed to show. A form that cannot reach a maximum will never tell a farmer that too much fertiliser lowers yield; a form with a constant elasticity of substitution will never reveal that two inputs are close complements. The choice is made on three grounds: agronomic or biological reasoning about the response, statistical fit with the data, and the economic properties the analyst needs, such as a finite optimum or flexible substitution. Heady and Dillon's rule of thumb still holds: try several forms and choose on theory and fit together, not on R-squared alone. Studies that fitted several forms to the same fertiliser trial have found similar fit statistics but economic optimum doses that differ substantially, so the recommended dose can depend as much on the form chosen as on the data. Reporting the optimum from more than one plausible form is good practice.
The linear form Y = a + bX has a constant marginal product b, so it implies no diminishing returns and no finite optimum; it is useful only over a narrow range. The quadratic Y = a + bX - cX^2 has a marginal product b - 2cX that falls linearly and becomes zero at X = b/2c, the point of maximum yield. It shows stages II and III but, with a positive intercept, its average product falls from the start, so it cannot show stage I. It is the most widely used form in fertiliser response trials, though it is known to overstate the optimum dose when the true response flattens into a plateau. The square-root form Y = a + bX + cX^0.5 has a response that rises steeply at low doses and falls slowly. The cubic Y = a + bX + cX^2 - dX^3 is the smallest polynomial that shows all three stages of production. The Spillman or Mitscherlich form Y = M - AR^X (0 < R < 1) approaches an upper asymptote M; each equal increment of input adds a constant proportion R of the previous increment, which is Spillman's statement of diminishing returns. Linear-response-and-plateau models, in which yield rises linearly to a plateau and stays there, are used in soil fertility work because they match Liebig's law of the minimum.
The Cobb-Douglas form Y = A X1^b1 X2^b2 ... Xn^bn is the most estimated function in agricultural economics. Taking logarithms gives ln Y = ln A + b1 ln X1 + ... + bn ln Xn, which is linear in parameters and fitted by ordinary least squares. Each bi is the elasticity of production of input i and is constant; the sum of the bi is the returns-to-scale parameter; the marginal product of input i is bi Y / Xi; and the elasticity of substitution between any pair of inputs is exactly one. Its weaknesses follow from the same simplicity: it never reaches a maximum, a single estimate describes only one stage (stage II when 0 < bi < 1), and it forces unit substitution elasticity. The constant elasticity of substitution (CES) function of Arrow, Chenery, Minhas and Solow (1961), Y = A[d X1^-r + (1 - d) X2^-r]^(-v/r), frees substitution: the elasticity of substitution is 1/(1 + r), which tends to one (Cobb-Douglas) as r tends to zero, to infinity (perfect substitutes) as r tends to -1, and to zero (fixed proportions) as r grows large. The transcendental function of Halter, Carter and Hocking (1957), Y = A X^b e^(cX), has elasticity b + cX, which falls with X when c < 0, so it can show all three stages. The transcendental logarithmic or translog function of Christensen, Jorgenson and Lau (1973) adds squared and cross-product log terms to the Cobb-Douglas; it is a flexible form that can approximate any smooth technology at a point, and it is the usual choice for cost and profit functions.
Production functions are estimated from experimental data (fertiliser or irrigation trials, where input levels are set by design), from cross-section farm surveys, from time series, or from panels. Experimental data give clean response curves but on research-station conditions; survey data show real farmer behaviour but mix input levels chosen by the farmers themselves. Four problems recur. First, multicollinearity: farm inputs move together, so individual elasticities are imprecise even when their sum is well estimated; aggregating inputs or using restricted estimation is the usual remedy. Second, simultaneity: Marschak and Andrews (1944) showed that if farmers choose inputs knowing their own productivity, inputs are correlated with the error term and ordinary least squares is biased; Zellner, Kmenta and Dreze (1966) showed that if farmers maximise expected profit and the disturbance is not known when inputs are chosen, OLS on a Cobb-Douglas remains consistent. Third, omitted management: better managers use more inputs and get more output, biasing input elasticities upwards; panel fixed effects are one answer. Fourth, specification of the error: a multiplicative error in the Cobb-Douglas becomes additive in logs. In practice, marginal products from a Cobb-Douglas are evaluated at the geometric means of output and inputs, because the logarithmic regression passes through the point of means of the logs.
The factor-product relationship holds all inputs but one constant and traces how output responds to the variable input. Total physical product (TPP) is the output at each level of the variable input. Average physical product (APP) is TPP divided by the input level, Y/X: output per unit of input. Marginal physical product (MPP) is the change in output from one more unit of input, dY/dX, or with discrete data the change in TPP divided by the change in X between two observations. With discrete data the MPP belongs to the interval between two levels, not to either level, a detail examiners like to test. Geometrically, APP at a point is the slope of a ray from the origin to that point on the TPP curve and MPP is the slope of the curve itself. Three fixed relationships follow: MPP is zero where TPP is at its maximum; MPP equals APP where APP is at its maximum, because when the marginal is above the average it pulls the average up and when below it pulls it down; and MPP is at its maximum at the inflection point of TPP, where TPP changes from increasing to decreasing rate of increase.
The law of variable proportions, or law of diminishing marginal returns, states that if equal increments of one input are added while the quantities of other inputs are held constant, the resulting increments of output will eventually decrease. It is an empirical law rather than a theorem: it holds because the fixed inputs (land area, plant number, animal capacity) become progressively scarcer relative to the variable one. It has three conditions, which students must state: technology is unchanged, at least one input is fixed, and the units of the variable input are homogeneous. Three types of returns are recognised along a response curve: increasing returns (each unit adds more than the last), constant returns (each unit adds the same) and decreasing returns (each unit adds less). Most agricultural response lies in the diminishing zone over the range farmers use. Diminishing returns should not be confused with decreasing returns to scale, which concern all inputs changing together.
The elasticity of production, Ep, is the percentage change in output divided by the percentage change in the input, which reduces to MPP/APP. It divides the response curve into three stages. Stage I runs from zero input to the point where APP is maximum and equal to MPP; here Ep is greater than one, so every extra unit raises the average product of the variable input and the fixed input is being used too intensively relative to the variable one. Stage II runs from maximum APP to maximum TPP; here Ep lies between one and zero, both APP and MPP are falling but positive. Stage III lies beyond maximum TPP; MPP and Ep are negative, so more input reduces output. A rational producer never operates in stage III, because the same output can be had with less input, nor in stage I, because as long as output rises faster than input the return per unit of input is still increasing, so if it pays to use any input it pays to use more. Stage II is therefore called the rational stage or the stage of economic relevance. Where in stage II the farmer should be depends on prices, which is the subject of the next chapter. A symmetric point: stage I for the variable input corresponds to stage III for the fixed input when the function has constant returns to scale, which is a second reason stage I is irrational. Take the cubic response Y = 6X^2 - 0.2X^3, where X is units of irrigation water and Y is quintals of output on a fixed plot. APP = 6X - 0.2X^2, which is maximum where its derivative 6 - 0.4X is zero, at X = 15. MPP = 12X - 0.6X^2, which is zero at X = 20. So stage I runs from 0 to 15 units, stage II from 15 to 20 units and stage III beyond 20. At X = 15 both APP and MPP equal 45, confirming that MPP cuts APP at its maximum. At X = 18, APP = 43.2 and MPP = 21.6, so Ep = 0.5, inside the rational range. MPP itself peaks at X = 10, the inflection point, where MPP = 60.
Physical relationships say where the rational stage is; prices say where in it to stop. Multiply MPP by the output price Py and the result is the value of the marginal product, MVP = MPP x Py, the extra revenue from one more unit of input. The marginal factor cost, MFC, is the extra cost of that unit; for a price-taking farm it is simply the input price Px. Profit is maximised where MVP = MFC, provided MVP is falling at that point (the second-order condition), which is true in stage II. Dividing both sides by Py gives the equivalent physical rule MPP = Px/Py: use the input until its marginal physical product falls to the inverse price ratio. Before this point each unit returns more than it costs; beyond it each unit costs more than it returns. The rule is also written as the ratio MVP/MFC: a ratio above one means the input is underused and profit rises with more of it; below one it is overused; equal to one it is at the optimum. Resource-use-efficiency studies across Asia and Africa routinely report such ratios for fertiliser, labour, irrigation and seed.
The optimum can equally be found on the output axis. The marginal cost of output, MC, is the cost of the extra input needed to produce one more unit of output, which equals Px/MPP. Marginal revenue, MR, for a price taker is the output price Py. Profit is maximised where MC = MR. Since MC = Px/MPP, setting it equal to Py gives MPP = Px/Py, the same condition. This is why an examiner can ask the factor-product question as 'MVP = MFC' or as 'MC = MR' and expect the same answer; the ICAR Master's entrance syllabus names both. When MR is below average variable cost the farm shuts down in the short run, and the part of the MC curve above minimum average variable cost is the short-run supply curve of the firm. For a single variable input, average variable cost is Px/APP, so the shutdown point, minimum AVC, lies exactly at the boundary between stages I and II.
Because the optimum depends only on the ratio Px/Py, a rise in the output price or a fall in the input price lowers the ratio and moves the optimum to a higher input level, further along stage II towards the yield maximum; a rise in the input price does the reverse. Doubling both prices changes nothing. The profit-maximising input levels traced out as Px changes form the farm's demand curve for the input, which is the downward-sloping part of the MVP curve within stage II. An input subsidy therefore raises input use by lowering Px, and the size of the response depends on the curvature of the production function. The maximum-yield dose (MPP = 0) is the economic optimum only when the input is free; any positive price puts the optimum below maximum yield. That distinction, between the agronomic recommendation 'maximum yield' and the economic optimum, is the most practical lesson in the subject.
The MVP = MFC rule assumes the farmer can buy as much input as is profitable. When working capital is limited, the question becomes how to spread a fixed budget across several uses, crops or inputs. The equimarginal principle states that returns are maximised when the last unit of money spent earns the same marginal return in every use: MVPa/Pa = MVPb/Pb = ... . If one use returns more at the margin, money should be moved to it until the margins are equal. Under a binding budget the common ratio will be above one, so each input is used below its unconstrained optimum. The same principle underlies opportunity cost: the cost of using a unit of capital on wheat is the marginal return it would have earned on the best alternative use. Worked example: with a quadratic response Y = 20 + 12X - 0.5X^2, output price 10 and input price 20, MPP = 12 - X must equal 20/10 = 2, so X* = 10 units against a maximum-yield level of 12 units.
The factor-factor relationship varies two inputs while holding output constant and asks how one can replace the other. An isoquant (iso-product curve) is the locus of all combinations of two inputs that produce the same output. The isoquant map is the production surface seen from above, with higher isoquants for higher outputs. The slope of an isoquant is the marginal rate of technical substitution. MRTS of X1 for X2 is -dX2/dX1, the amount of X2 that one extra unit of X1 can replace with output unchanged, and it equals MPP1/MPP2. Isoquants have four standard properties in the economic region: they slope downwards, they are convex to the origin because the MRTS diminishes as X1 replaces X2, they do not intersect, and higher isoquants represent higher output. The diminishing MRTS reflects diminishing marginal products: as X2 becomes scarce relative to X1, each unit of X2 becomes harder to replace. In farming, isoquants are used for feed rations, for labour against machinery, and for water against nitrogen.
Three cases are taught. Perfect substitutes, such as two sources of the same nutrient in equivalent units or two grains in a ration valued only for energy, substitute at a constant rate, so isoquants are straight lines and MRTS is constant. Perfect complements, such as a tractor and its driver, or the fixed ratio of plough to bullock pair, combine in fixed proportions, so isoquants are right angles, MRTS is undefined at the corner and there is no substitution. Imperfect substitutes, such as labour and herbicide for weed control or grain and roughage in a dairy ration, give smooth convex isoquants with a diminishing MRTS; this is the general case. The degree of substitutability is measured by the elasticity of substitution, the percentage change in the input ratio X2/X1 divided by the percentage change in the MRTS. It is infinite for perfect substitutes, zero for fixed proportions and one for Cobb-Douglas. For the Cobb-Douglas Y = A X1^b1 X2^b2 the MRTS is (b1/b2)(X2/X1), which depends only on the input ratio.
An isocline is a line joining points on successive isoquants that have the same MRTS. Isoclines show how the input mix must change as output expands if the rate of substitution is to stay constant. For homogeneous and homothetic functions, such as the Cobb-Douglas, isoclines are straight lines through the origin, so the optimal input ratio does not depend on the output level; for other functions they curve. Two special isoclines are the ridge lines. On one, the MPP of X1 is zero (the isoquants become vertical); on the other, the MPP of X2 is zero (isoquants become horizontal). Between the ridge lines both marginal products are positive and isoquants slope downward; this is the economic region, the two-input counterpart of stage II. Outside the ridge lines one input has a negative marginal product and more of it would require more, not less, of the other to hold output, so no rational farmer operates there. The expansion path, taken up in the next chapter, is the particular isocline along which MRTS equals the input price ratio.
An iso-cost line shows all input combinations that cost the same total outlay C: P1X1 + P2X2 = C. Its slope is -P1/P2. The least-cost combination for a given output is the point where the isoquant is tangent to the lowest attainable iso-cost line, where the MRTS equals the price ratio: MPP1/MPP2 = P1/P2, or rearranged, MPP1/P1 = MPP2/P2. In words, the last unit of money spent on each input adds the same output. With discrete data the rule becomes: substitute X1 for X2 as long as the amount of X2 replaced is worth more than the X1 added, that is, while dX2/dX1 x P2 exceeds P1. If the MRTS exceeds the price ratio, X1 is relatively cheap for what it does, and the farmer should use more X1 and less X2. The same tangency answers the dual problem, maximum output for a given budget, which is why the least-cost combination is also called the maximum-output combination.
Formally, minimise C = P1X1 + P2X2 subject to f(X1, X2) = Y0. The Lagrangian L = P1X1 + P2X2 + l[Y0 - f(X1, X2)] gives first-order conditions P1 = l f1 and P2 = l f2, so f1/f2 = P1/P2, and l = P1/f1 = P2/f2 is the marginal cost of output. The second-order condition requires the isoquant to be convex at the solution. For a Cobb-Douglas Y = A X1^b1 X2^b2 the tangency condition (b1/b2)(X2/X1) = P1/P2 gives the optimal input ratio X2/X1 = (b2/b1)(P1/P2). Because this ratio does not depend on Y, the expansion path is a straight line through the origin. For example, with b1 = 0.4, b2 = 0.2, P1 = 50 and P2 = 100, the optimal X2/X1 = 0.5 x 0.5 = 0.25: one unit of X2 for every four of X1, whatever the output level. The expansion path is the isocline that joins the least-cost points on successive isoquants for fixed input prices. Every point on it is cost-efficient, but only one point is profit-maximising: the output where marginal cost equals the output price. So profit maximisation implies cost minimisation, but not the reverse. The two steps can be run in either order: first find the cost-minimising path and then move along it to MC = Py, or solve MVP1 = P1 and MVP2 = P2 simultaneously. For a Cobb-Douglas with increasing returns to scale (sum of elasticities above one) the profit maximum does not exist, because marginal cost keeps falling, which is one reason why estimated farm elasticities that sum above one must be interpreted carefully.
In the short run, total cost is total fixed cost (land rent, depreciation, interest on fixed capital) plus total variable cost (seed, fertiliser, hired labour, fuel). Average fixed cost falls continuously; average variable cost and average total cost are U-shaped, reflecting first increasing and then diminishing returns to the variable input; marginal cost cuts both AVC and ATC at their minima. The short-run cost curves are mirror images of the product curves: when APP is at its maximum, AVC is at its minimum (AVC = Px/APP), and when MPP is at its maximum, MC is at its minimum (MC = Px/MPP). In the long run every input is variable and the long-run average cost curve is the envelope of the short-run average cost curves, touching each at one point. Its shape depends on returns to scale and economies of size, the topic of chapter 8. Fixed costs do not affect the short-run optimum, which depends only on marginal quantities.
The product-product relationship holds resources fixed and asks how they should be divided between two products. The production possibility curve (PPC), also called the product transformation curve or opportunity curve, shows the combinations of two outputs that a given bundle of resources can produce. Its slope is the marginal rate of product substitution (MRPS), also called the marginal rate of product transformation: MRPS of Y1 for Y2 = -dY2/dY1, the amount of Y2 given up to produce one more unit of Y1. For competitive products, where resources are specialised to some degree, the MRPS increases as more Y1 is produced, so the PPC is concave to the origin. If resources are equally productive in both uses the MRPS is constant and the PPC is a straight line. An iso-revenue line joins product combinations that bring the same total revenue: Py1Y1 + Py2Y2 = R. Its slope is -Py1/Py2. Revenue from the fixed resources is maximised where the PPC is tangent to the highest iso-revenue line, which is where MRPS = Py1/Py2. With discrete data the rule is: substitute Y1 for Y2 as long as the value of Y1 gained exceeds the value of Y2 lost, that is, as long as dY2/dY1 is less than Py1/Py2. If the price of Y1 rises relative to Y2 the iso-revenue line becomes steeper and the optimum moves towards more Y1. When costs of producing the two products differ, the comparison is made on net returns rather than gross revenue. The principle is the theoretical basis of enterprise choice, and linear programming is its practical, many-enterprise, many-constraint version.
Five relationships are distinguished. Joint products come from the same process in more or less fixed proportions: grain and straw, cotton lint and cottonseed, wool and mutton from sheep, milk and calves. They are planned as one enterprise because one cannot be produced without the other. Complementary enterprises are those where increasing one increases the other over some range with resources fixed: a legume in rotation that adds nitrogen for the following cereal, or livestock whose manure raises crop yields while using crop by-products. Supplementary enterprises use resources that would otherwise be idle, so one can be expanded without reducing the other: a small poultry or dairy unit using family labour in the slack season, or a short-duration catch crop in a fallow period. Competitive enterprises compete for the same resources, so more of one means less of the other; this is the usual case once complementary and supplementary ranges are used up. Antagonistic enterprises are those where an increase in one reduces the other and the total output of both falls, for example two crops that share a pest or disease that builds up when both are grown; they are avoided. A farmer should always expand complementary and supplementary enterprises to the limit of those ranges before the competitive zone, where the price ratio decides.
Returns to scale describe what happens to output when all inputs are increased in the same proportion. If output increases in the same proportion, returns are constant; if more than in proportion, increasing; if less, decreasing. A function is homogeneous of degree k if multiplying every input by t multiplies output by t^k: k = 1 is constant, k > 1 increasing and k < 1 decreasing returns to scale. The Cobb-Douglas is homogeneous of degree equal to the sum of its elasticities. The function coefficient, a term from Carlson's early theory of production, is the sum of the individual elasticities of production and measures returns to scale at a point for any function. Euler's theorem states that for a function homogeneous of degree k, the sum of Xi times MPPi equals kY. With constant returns and every input paid its marginal product, the product is exactly exhausted; with increasing returns marginal-product payments would exceed output, with decreasing returns there is a residual, often interpreted as the return to an unmeasured fixed factor such as management or land.
Farm economists separate two ideas that textbooks sometimes run together. Returns to scale is a purely physical concept on the production function with all inputs scaled in the same proportion. Economies of size refer to how average cost per unit of output changes as the farm grows, whatever the input mix. Real farms do not expand by scaling every input: a larger farm uses more machinery and less labour per hectare, buys inputs at discounts and sells in bulk. So a farm can show economies of size (falling average cost) even with constant returns to scale, through lower input prices or by adopting indivisible machinery that a small farm cannot use fully. Economies of scale in the strict sense are cost savings arising from returns to scale. The sources of economies of size in agriculture are indivisible inputs (a combine, a milking parlour, the manager's own time), better specialisation of labour, bulk purchase and sale, and access to credit; diseconomies arise from the difficulty of supervising hired labour spread over a biological, weather-dependent, geographically dispersed process, which is the standard explanation for the persistence of family farms.
The long-run average cost curve (LRAC) is the envelope of short-run curves. Studies of farm costs in several countries find a characteristic 'L' shape: average cost falls steeply as size increases from very small units, then flattens over a wide range, with no clear upturn within observed sizes. The implication is that beyond a moderate size many sizes are about equally efficient, and other factors, such as management, off-farm income, land market rules and access to technology, decide which size survives. The inverse farm-size productivity relationship, the observation that output per hectare often falls as farm size rises, was prominent in the Indian Farm Management Studies of the 1950s and has been reported in many developing countries since. Its explanation is contested: family labour applied more intensively on small farms, land quality differences, measurement error in area, and imperfect labour and credit markets are all proposed. A student should present it as an empirical regularity with competing explanations, not as settled law.
Production theory describes the best practice; real farms fall short of it. Farrell (1957) proposed measuring the shortfall against a frontier and splitting it into two parts. Technical efficiency (TE) is the ability to obtain maximum output from given inputs, or equivalently to use minimum inputs for given output. Allocative efficiency (AE), also called price efficiency, is the ability to use inputs in proportions that minimise cost at the prevailing prices. Economic efficiency (EE), or overall cost efficiency, is their product: EE = TE x AE. In the input-oriented diagram with a unit isoquant, if a farm uses input bundle P, the ray from the origin to P crosses the frontier isoquant at Q and the iso-cost line through the cost-minimising point at R. Then TE = OQ/OP, AE = OR/OQ, and EE = OR/OP. A farm with TE = 0.8 and AE = 0.9 has EE = 0.72, meaning it could produce the same output at 72 percent of its present cost if it were both technically and allocatively efficient. Output-oriented measures ask instead by how much output could rise with the same inputs; under constant returns to scale the two orientations give the same TE score.
Two families of method estimate the frontier from data. The parametric stochastic production frontier, proposed independently by Aigner, Lovell and Schmidt (1977) and Meeusen and van den Broeck (1977), writes ln Y = f(X; b) + v - u. The symmetric error v captures noise such as weather, luck and measurement error; the one-sided error u (u >= 0) captures inefficiency. Technical efficiency of a farm is exp(-u), the ratio of observed output to frontier output for its inputs. Estimation is by maximum likelihood with an assumed distribution for u, typically half-normal, truncated-normal or exponential. The share of inefficiency in the total error variance is often reported as a parameter gamma; a gamma near zero means deviations are mostly noise and an ordinary average function would do. Battese and Coelli's models allow the mean of u to depend on farm characteristics such as education, extension contact and credit access, which is how studies identify drivers of inefficiency.
The non-parametric alternative, data envelopment analysis (DEA), uses linear programming to wrap a piecewise linear frontier around the observed farms, without assuming a functional form. Charnes, Cooper and Rhodes (1978) introduced the constant-returns model (CCR); Banker, Charnes and Cooper (1984) the variable-returns model (BCC). The ratio of a farm's CRS technical efficiency to its VRS technical efficiency gives its scale efficiency, so overall technical efficiency is split into pure technical efficiency and scale efficiency. DEA handles many outputs and inputs easily and needs no distributional assumption, but it attributes all deviation from the frontier to inefficiency, so noise, outliers and measurement error in the data bias it. SFA handles noise but needs a functional form and a distribution. Neither is right in every setting; many studies report both. Efficiency scores are relative to the sample: a farm that scores 1.0 is best among those observed, not efficient in any absolute sense.
Technological change is a change in the production function itself: more output from the same inputs, or the same output from fewer inputs. On the single-input diagram the TPP curve shifts upward; on the isoquant map a given isoquant moves towards the origin. New varieties, improved breeds, irrigation, fertiliser-responsive seeds, mechanisation and better practices all act this way. The distinction from a movement along a function is fundamental: using more fertiliser on the old variety is a movement along the curve, switching to a variety that gives more grain per kilogram of nitrogen is a shift. The high-yielding varieties of the Green Revolution illustrate both, since they shifted the response curve upwards and also raised the economic optimum input level because their MPP of fertiliser was higher. Adoption of a new technology is itself an economic decision: it pays only if the gain in revenue from the shift exceeds the extra cost of the new inputs, learning and risk, which is why adoption is gradual and uneven across farms of different size, education and access to credit and information.
Hicks (1932) classified technical change by its effect on the marginal rate of technical substitution at a constant input ratio. Neutral technical change raises the marginal products of both inputs in the same proportion, so the MRTS at any input ratio is unchanged and the isoquants shift inward without changing shape. Labour-saving (capital-using) change raises the marginal product of capital relative to labour, so at unchanged prices the farm substitutes capital for labour; mechanisation is the usual example. Capital-saving (labour-using) change does the opposite. In agriculture the useful pair is land-saving biological and chemical technology (improved seed, fertiliser, which raise output per hectare) and labour-saving mechanical technology (tractors, harvesters, which raise output per worker). Hayami and Ruttan's induced innovation hypothesis holds that the direction of technical change responds to relative factor scarcity and prices: land-scarce Japan developed land-saving biological technology, while land-abundant, labour-scarce United States agriculture developed labour-saving machinery. The idea is a hypothesis supported by historical comparison, not a law.
Partial productivity measures, yield per hectare and output per worker, are easy to compute but confuse input substitution with technical change. Total factor productivity (TFP) is an index of total output divided by an index of total input, with inputs weighted by their cost shares. Its growth rate equals the growth of output minus the share-weighted growth of inputs; this is the growth-accounting method associated with Solow (1957), and the residual is attributed to technical change and efficiency gains. For example, if output grows at 3.0 percent a year and the share-weighted input index at 1.8 percent, TFP grows at 1.2 percent. The Malmquist index, computed from DEA or SFA frontiers, splits TFP change into technical change (the frontier shifting) and efficiency change (farms catching up with the frontier), and the latter can be split again into pure technical and scale components. TFP growth depends on the weights, the treatment of input quality, and whether environmental costs are counted; results for the same country can differ between studies for these reasons.
Knight (1921) distinguished risk, where outcomes are unknown but their probabilities are known or can be estimated, from uncertainty, where the probabilities themselves are unknown. Modern decision analysis mostly treats both with subjective probabilities, but the distinction survives in exams. Farming faces six commonly listed types. Production or yield risk comes from weather, pests, diseases and the biological nature of crops and animals. Price or market risk arises because output prices are unknown when production begins, and input prices may change. Institutional risk comes from changes in policy, regulation, support prices, trade rules and tax. Financial risk arises from borrowing: variable interest rates, credit availability and leverage that magnify losses. Human or personal risk covers death, illness, injury, divorce or dispute in the farm family or workforce. Technological risk is the chance that a new technology does not perform or becomes obsolete. Some texts, following Hardaker and colleagues, group these into business risk (production, price, institutional and human risk) and financial risk.
The expected monetary value (EMV) criterion chooses the action with the highest probability-weighted mean return; it is correct only for a risk-neutral decision-maker. Most farmers are risk averse and give up some expected income for lower variability. Expected utility theory captures this: a risk-averse person has a concave utility function, and the certainty equivalent (CE) of a risky prospect is the sure sum giving the same utility, which is below the expected value; the difference is the risk premium. With U = W^0.5 and a gamble of 100 or 400 at equal chance, expected utility is 15, so the CE is 225 while the expected value is 250, a risk premium of 25. The Arrow-Pratt coefficient of absolute risk aversion, -U''/U', measures the degree of aversion. Where probabilities are not used, the payoff-matrix rules (maximin, maximax, Hurwicz, Laplace, minimax regret), covered in farm management, apply. The mean-variance (E-V) approach of Markowitz (1952) ranks plans by their mean and variance; the efficient set contains plans for which no other plan has a higher mean with the same or lower variance. The coefficient of variation, standard deviation divided by mean, compares relative variability between enterprises of different sizes. Roy's (1952) safety-first rule minimises the probability that income falls below a disaster level.
Diversification reduces income variance because the variance of a combination depends on the covariances between enterprises. For a plan with shares a and b of two enterprises, the variance of income is a^2 s1^2 + b^2 s2^2 + 2ab r s1 s2, where s1 and s2 are standard deviations and r is the correlation. With r below one, the combined standard deviation is below the weighted average of the separate standard deviations; with r negative the reduction is large; with r = +1 there is no gain. So enterprises whose incomes move in opposite directions, such as a rainfed crop and irrigated vegetables, or crops and dairy, stabilise farm income. Under risk, the product-product optimum moves away from the pure-profit tangency towards a more diversified mix, trading expected income for stability. Operational methods include quadratic risk programming (Freund 1956), which maximises expected income minus a risk-aversion coefficient times variance; MOTAD (Hazell 1971), a linear-programming approximation that minimises mean absolute deviation; and target MOTAD and chance-constrained programming. Other responses to risk are flexibility (short-duration crops, rented rather than owned machinery), liquidity reserves, forward contracts and futures, and insurance. Diversification is not free: it sacrifices economies of specialisation, so the right amount depends on the farmer's risk aversion and the cost of the alternatives.
Most of the world's farms are family farms, where the same household supplies labour and management, consumes part of the output and owns or rents the land. Chayanov's The Theory of Peasant Economy (Russian original 1925) argued that such a farm does not maximise profit in the capitalist sense because it pays no wages to itself. Instead, the family works until the marginal drudgery of extra labour equals the marginal utility of the extra consumption it brings. The level of activity depends on the consumer-worker ratio: a family with many dependants per working member works harder and farms more intensively. As the family ages and the ratio changes, the farm expands and contracts, a pattern Chayanov called demographic differentiation. Schultz's Transforming Traditional Agriculture (1964) argued the opposite emphasis: traditional farmers are 'poor but efficient', allocating their resources well given the technology they have, so growth needs new technology rather than better allocation of old inputs. Sen (1966) analysed when family farms apply labour beyond the point where its marginal product equals the market wage, which is one explanation for higher output per hectare on small farms.
The agricultural household model, set out by Singh, Squire and Strauss (1986), treats the farm household as a producer, a consumer and a supplier of labour at once. It maximises utility from consumption and leisure subject to a production function, a time constraint and a budget. The key result is separability. If markets for all goods and labour exist and the household is a price taker, the production decision can be made first, as ordinary profit maximisation at market prices, and consumption and labour supply are then chosen given that profit; preferences and family size do not affect input use. When markets are missing or imperfect, for example when family labour cannot be sold off-farm or hired labour is costly to supervise, separability breaks down: production decisions then depend on the household's preferences and composition, which is Chayanov's world. The model explains why a rise in the price of a food crop may raise its production yet increase household consumption of it, so marketed surplus responds less than production.
Duality theory shows that, under standard regularity conditions, a technology can be described equally well by its production function, its cost function or its profit function, and each can be recovered from the others. The cost function C(w, y) gives the minimum cost of producing output y at input prices w. It is non-decreasing in input prices, homogeneous of degree one in them (doubling all input prices doubles cost), and concave in them. Shephard's lemma states that the partial derivative of the cost function with respect to an input price is the cost-minimising (conditional) demand for that input: dC/dwi = xi(w, y). The profit function p(p, w) gives maximum profit at output price p and input prices w. It is non-decreasing in p, non-increasing in w, homogeneous of degree one in all prices together, and convex in prices. Hotelling's lemma states that its derivative with respect to output price is the supply function, dp/dp = y(p, w), and its derivative with respect to an input price is minus the input demand, dp/dwi = -xi(p, w).
The practical value is that supply and input demand systems can be estimated directly from prices, which are usually easier to observe than quantities of every input and are more plausibly exogenous to the farm, so the simultaneity problem of chapter 2 is reduced. A translog cost or profit function is estimated with its share equations; elasticities of input demand, output supply and substitution follow from the parameters. Lau and Yotopoulos (1971) introduced the normalised restricted profit function, in which prices are divided by the output price and fixed inputs such as land enter as quantities; they used it to test the relative economic efficiency of large and small farms. A simple check of the lemmas: with cost function C = 2y(w1 w2)^0.5, the conditional demand for input 1 is y(w2/w1)^0.5; with profit function p^2/(4w) for a single input, supply is p/(2w). The self-dual Cobb-Douglas, whose cost function is also Cobb-Douglas in prices, is the textbook case.
Learn to move between the three diagrams (response curve, isoquant map, production possibility curve) and their three optimum rules without hesitation: MPP = Px/Py, MRTS = P1/P2, MRPS = Py1/Py2. Given a response schedule you should compute MPP, APP and Ep and mark the stage boundaries in under two minutes. Given a Cobb-Douglas you should read off elasticities, returns to scale, MPP at the means and the optimal input ratio. Memorise the dates and authors that recur: Farrell 1957, Aigner-Lovell-Schmidt 1977, Charnes-Cooper-Rhodes 1978, Hicks 1932, Knight 1921, Chayanov, Singh-Squire-Strauss 1986, Shephard and Hotelling lemmas.
An extension economist, agronomist or bank field officer five years in uses this every season. You turn a fertiliser trial into an economic dose by fitting a quadratic and setting MPP equal to the price ratio, and you redo it when the fertiliser price moves. You compare a farmer's actual input use with that optimum and explain whether profit is being lost through under- or over-use. You judge whether a proposed dairy or poultry unit is supplementary to the crop plan or competes with it for labour and cash. In a credit appraisal you ask whether the loan will push the farm along its expansion path or simply buy input it cannot use profitably.
With twenty-five years in a ministry, development bank, input company or farm business, you use production economics to set and defend policy and strategy. You read resource-use-efficiency and frontier studies critically, asking whether low efficiency scores are really inefficiency or omitted land quality and weather. You design input-subsidy or price-support schemes knowing that the response depends on the curvature of the function and on the price ratio, not on the subsidy alone. You judge claims about farm size and productivity, consolidation and mechanisation against the evidence on economies of size and on the inverse relationship. You plan research priorities with induced-innovation reasoning about which factor is becoming scarce.
A senior researcher works at the frontier of method and uses its limits. Typical tasks are estimating flexible cost or profit systems with curvature imposed, choosing between SFA and DEA and testing the sensitivity of efficiency rankings, separating technical change from efficiency change in Malmquist TFP indices, correcting for endogenous input choice with panel or control-function methods, and building agricultural household models where missing labour or credit markets break separability. You supervise students through these, referee papers that misuse Cobb-Douglas elasticities or report DEA scores without bias correction, and advise governments on whether measured productivity gaps justify intervention.
This manual does not yet work through: dynamic production economics (optimal control of livestock replacement, soil capital and groundwater mining), multi-output distance functions and directional distance functions, bootstrapped DEA and panel stochastic frontiers beyond a mention, damage-control specifications for pesticides, and the econometrics of supply response (Nerlove partial adjustment belongs with price analysis). Numerical examples use round illustrative figures, not estimates for any country or crop; published elasticities and efficiency scores vary widely by region, crop, data set and method, so none are quoted as typical. Risk programming models (quadratic programming, MOTAD) are described but not solved step by step.
Original practice questions written to the syllabus and pattern of the exams named in each question's examTags. They are not past papers and do not reproduce any board's questions. Numerical answers were computed by script; figures are illustrative, not data for any country or crop.
These notes are free to read here, with no account.
What the Laisha app adds:
Free for ten days, then a subscription. We would rather write that here than have you find it later.