Key: linearization of CES function, which represents the producer's behavior in GTAP model
Economic model: cost minimization problem
Economic solution: conditional factor demand z(w, q) and unit cost function c(w, q) / q
Set up producer's problem
The producer's production function is CES function, and the cost minimization problem is:

Where:
- x: inputs
- p: market prices of inputs
- α: efficiency parameter / factor productivity / scaling parameter, α > 0
- δ: distribution parameter / share parameter.
- ρ: exponent, ρ > -1 so the production function is concave (if ρ = -1. the production function is linear with perfect substitution)
- σ:
: constant elasticity of substitution
Property of the CES function:
- defined for positive level of inputs
- continuous
- differentiable
- monotonic
- strictly quasi-concave
- constant return to scale
- The constant return to scale (CRTS) feature is commonly adopted in CGE models with perfect competition
Solve producer's problem
Full steps to solve the producer's problem:
Write the problem in the standard form of optimization with constraint:

Write Lagrange of the problem:
Calculate FOC with respect to 
(1)
(2)
(3)
Divide (1) and (2):


(4)
Note: here equation (4) gives the relative relationship between x1 and x2, given the price p1 and p2. Regardless of the total output y, as long as x1 and x2 are positive, they will satisfy this relationship. However, the exact value of x1 and x2 depend on the condition
, which is equivalent with the utility constraint in hicksian demand of expenditure minimization problem of consumer.
That is to say:
- relative relationship of inputs depends on production function
- exact solution of inputs depend on production function and output constraint
If we solve the problem with profit maximization problem, we can still get the relative relationship between x1 and x2, but not the exact value. However, we can transform certain part of the solution into the form of output y, and get exact solution as x(p, y).
substitute (4) into (3):









(5)
Substitute (5) into (4), we have:
(6)
(5) and (6) are the conditional demand of input x(p, y) of the problem. When p1, p2 and y are given, we can have exact value for x1 and x2.
Properties of conditional input demand x1 and x2:
- non-decreasing and homogeneous of degree one with respect to production level
- homogeneous of degree zero with respect to input price p1, p2
- non-increasing with respect to own price
- non-decreasing with respect to other input price
To further simplify the conditional demand, consider the unit cost
, where
Property of the unit cost function:
- non-decreasing
- concave (It can be regarded as CES format if we regard
, and we have
the CES function is concave) - continuous with respect to each input price
- homogeneous of degree one with respect to input price
Property of total cost function
- homogeneous of degree one with respect to production level
Note: the unit cost function c(p1, p2) is not related with specific input use. Instead, it is an optimal variable, a feature of the specific production function.
To further relate conditional demand with unit cost, we can:.
Start with unit cost function

Denote
Note: the notion of A will be used in the linearization of unit cost function below.
Recall that the conditional demand of inputs is:

Then we have:

Recall in CES function, we have the constant elasticity of substitution as 
Similarly, we have:
Implication: here δ, α are parameters. When the unit cost relative to input i's own price (
) increases, the conditional demand of input i would also increase.
Question: Still, I am not sure why to convert the conditional demand in the form of unit cost function. The original conditional demand function is already a function of exogenous variables and parameters, but when converted in the form of unit cost function, we actually incorporate additional endogenous function
. Think about this question and keep reading.
Answer to question above: for the producer we have price equals to marginal cost, which further equals unit cost. So by putting
into conditional demand function can model the impact of output price change on input use. So there the
is not the endogenous variable
anymore, it equals to the exogenous variable of output price.
Linearization of conditional demand
Here I use A and B to refer to common variables (note: here A is not the notion above
)
Consider the proportional change of A as
,
Operation of proportional changed term:




(Refer to Linearization in Economic Models)
Use rules here to linearize x1(p1, y, cy) and x2(p2, y, cy)

By doing this, it decompose the proportional change of input demand xi into:
- expansion effect:
, change in output level - substitution effect:
: change in relative price (output price and own input price pi) - biased technical change:
: factor biased technical change - neutral technical change
: neutral technical change
- This is called "neutral" because it is not related with any specific input
Here I have:
- x1, y, cy, p1: quantity and price variables, should be considered as variable in linearization
- α: technical parameter, but I also wonder how its change can influence the solution, so it is also considered as variable in linearization
- σ: we do not change the substitution elasticity, so it is regarded as constant in linearization
- δ: I wonder how its change influences results, so it is variable in linearization.
- share parameter
Recall the original CES production function:

The original demand function

And the unit cost function:
Linearization of unit cost function
Two ways to linearize the unit cost function:
1) "Brute force" method
From the step in solving producer's problem:

We have:
Total differentiation of the equation above, regarding to α,
,
and
(they are variables in the decomposition of conditional demand function, while ρ or σ are parameters)
Recall that for variable A, we have
, then we can simplify the equation above as:





Recall we have:
Then we can get

And

Then we have:

So we can write the function above as:
Denote cost share of input i as 
Then we can write the function above as

Implication: the change of unit cost is a function of three technical change (netural technical change
and biased technical change
) and two price change
.
2) Based on homogeneous of degrees one in input price
Use Euler theorem and Shephard's lemma.
Linearization of cost share
from the cost share equation:
We have the linearized form
Recall we have the linearization of x:
And linearization of cy:

And recall 
Substitute these two functions into the linearization of θ (take θ1 as example), we have:


Note here:
So we have:
Denote
as the proportional change of the "effective price" of input i.
Then we have 
Feature of this expression:
- proportional change (responsiveness with respect to price change) of cost share of input 1 relates with size of cost share of input 2
- When the CES becomes C-D function, σ = 1, then the proportional change θ1 and θ2 are all zero. That is to say, in C-D function cost share does not change from exogenous shocks.
- the proportional change of θ1 with respect to to ratio of effective price change
depends on whether σ > 1 or not. If σ > 1, cost share of θ1 is decreasing in its relative effective price.
In GTAP model, the CES function is used for producer on
- substitution between intermediate inputs and value-added
- primary factors between value-added nest
GTAP model
- output, quantity: qo, price: ps
- Elasticity of substitution: ESUBT
- intermediate input, quantity: qf, price: pf
- Value added, quantity qva, price pva
- Only biased technical change term are specified
- Elasticity of substitution: ESUBVA
- primary input, quantity: qfe, price: pfe
- cost share of primary input: SVA
in the form before, we have for the value added CES:
- qfe = x
- pfe = p
- qva = y
- pva = cy
- afe =
: biased technical parameter - ESUBVA = σ
- SVA = θ
= 0
Then the linearized conditional demand of primary factors
now becomes
qfe = qva + ESUBVA( (ESUBVA-1) / ESUBVA afe + pva - pfe )
Rearrange it, we get the ENDWDEMAND (endowment demand) equation of GTAP model:
qfe = -afe + qva - ESUBVA(pfe - afe - pva)
Recall the demand function for land input in SIMPLE-G:
E_QLANDgl (all,g,GRID)(all,l,LTYPE)
p_QLANDgl(g,l)
- EIRRIGgl(g,l) * [p_PLANDgl(g,l) - p_AFLAND(g,l) - p_PLANDWTRgl(g,l) ] ;
Note: note the variable of "p_AFLAND" is the afe in GTAP model. The key here is to understand the definition of afe (biased technical parameter and its relationship with δ: afe =
, so we can get the equation in model from the linearization of conditional demand from solving cost minimization problem of CES production function.
unit cost of value added can be obtained from the linearization of unit cost

pva = Σ(sva(pfe - afe))
Recall: afe =
:
This is the equation VAPRICE in the GTAP model
Note:
To emphasis it here:
From producer's cost minimization function (given output level and input price), but no information from other functions
We solve for
- conditional demand of inputs
- unit cost of output, which equals the price of output
And linearize them with respect to output level, input price, cost share, substitution elasticity and other parameters in production function.
Similarly, we can have the equation of value added and intermediate as input of qo production:
Note: here ao is the unbiased technical parameter, not the biased technical parameter of qo (it does not exist since qo is already the top-level output in the economy)
qva = qo + ESUBT(
ava + ps - pva) + (ESUBT-1)ao
qf = qo + ESUBT(
af + ps - pf) + (ESUBT -1)ao
Rearrange them, we have the equation VADEMAND:
qva = -ava + qo - ao - ESUBT(pva - ava - ps -ao)
And equation INTDEMAND
qf = -af + qo - ao - ESUBT(pf - af - ps - ao)
Also, the price = unit cost would be the zero-profit condition in GTAP, which is given by linearization of unit cost function.
ps + ao = θf (pf - af) + θva (pva - ava)
The value of pva is replaced by the unit-cost equation from value-added production function:
pva = Σ(sva(pfe - afe))
Actually this function is the same as the one above it.
Note:
For each production function, it takes:
- output level from higher level production's demand (or consumer's demand)
- price of inputs from lower level production function
And solves:
- demand of inputs in lower level production
- price of the output in its own level.
Reference: Gohin, Alexandre, and Thomas Hertel. "A note on the CES functional form and its use in the GTAP model." Center for Global Trade Analysis, Department of Agricultural Economics, Purdue University (2003). https://www.gtap.agecon.purdue.edu/AgEc618/modules/readings/RM2.pdf