Egwald Economics - Cost Functions: Normalized Quadratic Cost Function

Egwald Web Services
Domain Names
Web Site Design

Egwald Economics: Microeconomics

Cost Functions

by

Elmer G. Wiens

Egwald's popular web pages are provided without cost to users.
Follow Elmer Wiens on Twitter:

J. Normalized Quadratic Cost Function

Linear Least Squares | Nonlinear Least Squares

Suppose we have a data set relating output quantities, q, to (cost minimizing) factor inputs, L, K, M, and input prices, wL, wK, and wM, and consequently data on the total cost of producing specific levels of outputs.

The three factor Normalized Quadratic (Total) Cost Function is:

C(q;wL,wK,wM) = h(q) * c(wL,wK,wM) (**)

where the returns to scale function is:

h(q) = q^^(1/nu1)

a continuous, increasing function of q (q >= 1), with h(0) = 0 and h(1) = 1,

and the unit cost function is:

c(wL,wK,wM) = cL * wL + cK * wK + cM * wM

+ (1/2) * [dLL * (wL*wL) + dLK*(wL*wK) + dLM * (wL*wM)
+ dKL * (wK*wL) + dKK*(wK*wK) + dKM * (wK*wM)
+ dML * (wM*wL) + dMK*(wM*wK) + dMM * (wM*wM)] * (wL + wK + wM)^^-1

linear in its twelve parameters, cL, cK, cM, dLL, dLK, dLM, dKL, dKK, dKM, dML, dMK, and dMM. Dividing the non-linear in variables, wL, wK, and wM, portion of the quadratic unit cost function by (wL + wK + wM) ensures that the unit cost function is homogeneous of degree one in prices, since c(t*wL, t*wK, t*wM) = t * c(wL,wK,wM) (Diewert and Wales, (1988), 327-342). Multiplying the unit cost function by the returns to scale function to obtain the total cost function presumes that the production technology is homothetic.

It is convenient to express the normalized unit cost function in terms of vectors and matrices.

Define the vectors:

c = [cL, cK, cM]^T, w = [wL, wK, wM]^T, 1 = [1, 1, 1]^T, and 0 = [0, 0, 0]^T.

Define the matrix D by :

D =	dLL	dLK	dLM
	dKL	dKK	dKM
	dML	dMK	dMM

The unit cost function becomes:

c(w) = c^T * w + (1/2) * w^T * D * w / (1^T * w),

a linear function in its parameters, c, and D.

Restrictions:

D = D^T, and D * w* = 0,

where the reference vector w* = [wL*, wK*, wM*]^T for the base prices, wL*, wK*, and wM*. The symmetry restriction D = D^T ensures that dLK = dKL, dLM = dML, and dKM = dMK (Young's Theorem).

Consequently, as specified the Normalized Quadratic cost function apparently has 9 free parameters (actually only six free parameters — see below).

Using conventional matrix calculus (Intriligator, (1971), 497-500), the first and second order partial derivatives (Diewert and Fox, (2009), 158-164) of the unit cost function, c(w), are:

∇c(w) = c + D * w / ( 1^T * w) - (1/2) * w^T * D * w * 1 / (1^T * w)², and

∇²c(w) = D / ( 1^T * w) - D * w * 1^T / (1^T * w)² - 1 * w^T * D / (1^T * w)² + w^T * D * w * (1 * 1^T) / (1^T * w)³.

At the reference vector w*:

∇c(w*) = c,

∇²c(w*) = D / ( 1^T * w*).

Shephard's lemma provides that the gradient vector of the unit cost function is the vector of unit input demand functions:

∇c(w) = [l(w), k(w), m(w)]^T.

Re-write the factor prices as:

v = [vL, vK, vM] = [wL, wM, wK] / (wL + wM + wK).

The factor demands as functions of q, vL, vK, and vM are:

L(q; vL, vK, vM) / q^^(1/nu1) =	l(vL, vK, vM) =	cL + dLL * vL + dLK * vK + dLM * vM	- (1/2) * [dLL * vL * vL + dLK * vL * vK + dLM * vL * vM + dKL * vK * vL + dKK * vK * vK + dKM * vK * vM + dML * vM * vL + dMK * vM * vK + dMM * vM * vM]
K(q; vL, vK, vM) / q^^(1/nu1) =	k(vL, vK, vM) =	cK + dKL * vL + dKK * vK + dKM * vM	- (1/2) * [dLL * vL * vL + dLK * vL * vK + dLM * vL * vM + dKL * vK * vL + dKK * vK * vK + dKM * vK * vM + dML * vM * vL + dMK * vM * vK + dMM * vM * vM]
M(q; vL, vK, vM) / q^^(1/nu1) =	m(vL, vK, vM) =	cM + dML * vL + dMK * vK + dMM * vM	- (1/2) * [dLL * vL * vL + dLK * vL * vK + dLM * vL * vM + dKL * vK * vL + dKK * vK * vK + dKM * vK * vM + dML * vM * vL + dMK * vM * vK + dMM * vM * vM]

À1: Estimate the factor demand equations using linear least squares.

I. Linear least squares with restrictions on the parameter values.

For the purpose of estimating the factor demand equations, re-write these functions:

L(q; vL, vK, vM) / q^^(1/nu1) =	l(vL, vK, vM) =	cL + dLL * vL + dLK * vK + dLM * vM	+ [tdLL * vL * vL + tdLK * vL * vK + tdLM * vL * vM + tdKL * vK * vL + tdKK * vK * vK + tdKM * vK * vM + tdML * vM * vL + tdMK * vM * vK + tdMM * vM * vM]
K(q; vL, vK, vM) / q^^(1/nu1) =	k(vL, vK, vM) =	cK + dKL * vL + dKK * vK + dKM * vM	+ [tdLL * vL * vL + tdLK * vL * vK + tdLM * vL * vM + tdKL * vK * vL + tdKK * vK * vK + tdKM * vK * vM + tdML * vM * vL + tdMK * vM * vK + tdMM * vM * vM]
M(q; vL, vK, vM) / q^^(1/nu1) =	m(vL, vK, vM) =	cM + dML * vL + dMK * vK + dMM * vM	+ [tdLL * vL * vL + tdLK * vL * vK + tdLM * vL * vM + tdKL * vK * vL + tdKK * vK * vK + tdKM * vK * vM + tdML * vM * vL + tdMK * vM * vK + tdMM * vM * vM]

We can express the factor demands as functions of ten explanatory variables with their corresponding parameters:

	1	2	3	4	5	6	7	8	9	10
Variable	constant	vL	vK	vM	vL*vL	vL*vK	vL*vM	vK*vK	vK*vM	vM*vM
L	cL	dLL	dLK	dLM	tdLL	tdLK	tdLM	tdKK	tdKM	tdMM	β_L
K	cK	dLK	dKK	dKM	tdLL	tdLK	tdLM	tdKK	tdKM	tdMM	β_K
L	cL	dLM	dKM	dMM	tdLL	tdLK	tdLM	tdKK	tdKM	tdMM	β_M

The construction of the variables along with their restrictions described below ensures the symmetry of the estimated matrix D, i.e. D = D^T.

Suppose we have m observations on q, L , K, M, wL, wK, and wM. Let V be the m x 10 matrix of observations of the explanatory variables constructed from v = [vL,vK,vM]^T. Let L, K, and M be the m-component vectors of observed factor demands. Then we can estimate the factor demand equations as a group (Johnston, (1972), 238-241) along with the necessary constraints, D = D^T, D * w* = 0, and the accounting constraints among the "d__" and "td__" parameters within and between the demand equations.

This group of equations can be represented as:

W			*	β	=	F
V			*	β_L	=	L
	V		*	β_K	=	K
		V	*	β_M	=	M

Subject to the constraints on the thirty parameters, our objective is to find the values for the vector of parameters, β, that minimizes the distance between F and W * β, i.e. the norm || W * β - F ||.

Looking at the construction of the ten variables of the matrix V, one finds that the rank of V is six, i.e. V has six independent column vectors. Therefore, when one estimates the factor demand equations as a group, the 3*m by 30 matrix of observations, W, has 18 independent column vectors out of thirty.

Moreover, the matrix of restrictions, R, on the parameters has 24 rows, i.e. R is a 24 x 30 matrix, with a rank of 24. These restrictions can be expressed as:

R * β = r,

where r is a 24 component vector. The restrictions matrix, R, and restrictions vector, r:

	dLL	dLK	dLM	tdLL	tdLK	tdLM	tdKK	tdKM	tdMM	dLK	dKK	dKM	tdLL	tdLK	tdLM	tdKK	tdKM	tdMM	dLM	dKM	dMM	tdLL	tdLK	tdLM	tdKK	tdKM	tdMM
1	1	0	0	2	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
2	0	1	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
3	0	0	1	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
4	0	0	0	0	0	0	0	0	0	1	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0
5	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0	2	0	0	0	0	0	0	0	0	0	0	0
6	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0
7	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0	1	0	0	0
8	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0	1	0
9	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0	2
10	0	0	0	1	0	0	0	0	0	0	0	0	-1	0	0	0	0	0	0	0	0	0	0	0	0	0	0
11	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	-1	0	0	0	0	0
12	0	0	0	0	1	0	0	0	0	0	0	0	0	-1	0	0	0	0	0	0	0	0	0	0	0	0	0
13	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	-1	0	0	0	0
14	0	0	0	0	0	1	0	0	0	0	0	0	0	0	-1	0	0	0	0	0	0	0	0	0	0	0	0
15	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	-1	0	0	0
16	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	-1	0	0	0	0	0	0	0	0	0	0	0
17	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	-1	0	0
18	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	-1	0	0	0	0	0	0	0	0	0	0
19	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	-1	0
20	0	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	-1	0	0	0	0	0	0	0	0	0
21	0	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	-1
22	0.269	0.5	0.231	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
23	0	0	0	0	0	0	0	0	0	0.269	0.5	0.231	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
24	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.269	0.5	0.231	0	0	0	0	0	0

II. Objective: Solve the LSE problem:

Among all 30 component β vectors obeying:

R * β = r,

find the β vector that minimizes the norm:

|| W * β - F ||.

By construction the observation matrix W is rank deficient, making it infeasible to use the usual QR algorithm described on the web page, Linear and Restricted Multiple Regression, to solve the LS problem:

find the 30 component β vector that minimizes the norm:

|| W * β - F ||,

subject to the requirement:

R * β = r.

III. To determine the Normalized Quadratic cost function's efficacy in estimating the cost structure of a production technology, we shall use it to approximate cost data generated by a CES Production function. The estimated parameters of the Normalized Quadratic cost function will vary with the parameters sigma, nu, alpha, beta and gamma of the CES production function.

CES Production Function:

q = A * [alpha * (L^^-rho) + beta * (K^^-rho) + gamma *(M^^-rho)]^^(-nu/rho) = f(L,K,M).

where L = labour, K = capital, M = materials and supplies, and q = product. The parameter nu is a measure of the economies of scale, while the parameter rho yields the elasticity of substitution:

sigma = 1/(1 + rho).

Set the parameters below to re-run with your own CES parameters.

Restrictions: .7 < nu < 1.3; .5 < sigma < 1.5;
.25 < alpha < .45, .3 < beta < .5, .2 < gamma < .35
sigma = 1 → nu = alpha + beta + gamma (Cobb-Douglas)
sigma < 1 → inputs complements; sigma > 1 → inputs substitutes
.5 < nu1 < 2;
4 <= wL* <= 11, 7<= wK* <= 16, 4 <= wM* <= 10

CES Production Function Parameters

CES elasticity of scale parameter: nu

elasticity of substitution: sigma

alpha

beta

gamma

Normalized Quadratic elasticity of scale: nu1

Base Factor Prices

wL*

wK*

wM*

Distribution to Randomize Factor Prices

Use [-2, 2] Uniform distribution
Use .25 * Normal (μ = 0, σ² = 1)

The CES production function as specified:

q = 1 * [0.35 * (L^^{- 0.17647}) + 0.4 * (K^^{- 0.17647}) + 0.25 *(M^^{- 0.17647})]^^(-1/0.17647)

The factor prices are distributed about the base factor prices, wL*, wK*, wM*, by adding a random number distributed uniformly in the [-2, 2] domain.

IV. For these coefficients of the CES production function, I generated a sequence (displayed in the "Normalized Quadratic Cost Function" table) of factor prices, outputs, and the corresponding cost minimizing inputs. Then I used these data to estimate the coefficients of the factor demand equations as a group.

I used the following method to solve the LSE problem:

Among all 30 component β vectors obeying:

R * β = r,

find the β vector that minimizes the norm:

|| W * β - F ||.

1. Start by obtaining the Householder QR factorization of the transpose matrix, R^T, where the R^T matrix has 30 rows and 24 columns, with rank 24:

R^T = Q * U,

where Q is a 30 by 30 orthogonal matrix, and U is a 30 by 24 matrix.

2. Multiplying the matrix R by Q will produce a 24 by 24 lower triangular matrix, T, and a 24 by 6 zero matrix, 0:

R * Q = [T | 0].

3. Multiplying the matrix W by Q will produce a 3*m by 24 matrix, H1, and a 3*m by 6 matrix, H2:

W * Q = [H1 | H2].

4. Solve by forward substitution for the 24 component vector, Γ₁, the lower triangular problem:

T * Γ₁ = r.

5. Compute the 30 component vector, f:

f = F - H1 * Γ₁.

If the vector r = 0, f = F.

6. Solve for the 6 component vector, Γ₂, using Householder QR factorization, the least squares problem:

H₂ * Γ₂ = f.

7. After creating the 30 component vector, Γ = [Γ₁, Γ₂]^T, obtain the solution vector, β, by:

β = Q * Γ.

Note: The solution vector, β, is unique since the augemented matrix:

M =	\| R \| \| W \|

has full rank = 30, (Lawson and Hanson, (1974), 134-143).

V. The method described above obtains the unique solution vector, β.

I obtained the Covariance Matrix of the solution vector, β, as follows:

Form the matrix, tW, as:

tW = H2 * H2⁺ * W,

where H2⁺ is the pseudo-inverse matrix of H2.

The solution to the augmented least squares problem:

| R |
| tW |

* β =

| r |
| F |

is also a solution to the original least squares problem:

|| W * β - F ||,

subject to the requirement:

R * β = r,

(Lawson and Hanson, (1974), 143).

I used the Covariance Matrix from the augmented system to obtain the t-statistics that are displayed below in the solution to the LSE problem. The reported "Observation Matrix Rank" is the rank of the augmented matrix of the augmented least squares problem.

LSE Restricted Least Squares
Parameter Estimates
Parameter	Coefficient	std error	t-ratio
cL	1.231023	0.014943	82.380672
dLL	-2.710088	0.062457	-43.391358
dLK	0.902455	0.03229	27.948212
dLM	1.206449	0.047159	25.582776
tdLL	1.355044	0.032218	42.058954
tdLK	-0.902455	0.033616	-26.846213
tdLM	-1.206449	0.047243	-25.537354
tdKK	0.403433	0.021521	18.746221
tdKM	-0.695344	0.034792	-19.985944
tdMM	1.457051	0.035171	41.427756
cK	0.808053	0.012599	64.137907
dLK	0.902455	0.036827	24.505432
dKK	-0.806865	0.032897	-24.527072
dKM	0.695344	0.037981	18.307772
tdLL	1.355044	0.035219	38.47434
tdLK	-0.902455	0.035367	-25.516612
tdLM	-1.206449	0.049432	-24.406354
tdKK	0.403433	0.017786	22.682058
tdKM	-0.695344	0.036551	-19.023921
tdMM	1.457051	0.037999	38.343995
cM	1.064702	0.015862	67.12482
dLM	1.206449	0.047115	25.606578
dKM	0.695344	0.031859	21.825676
dMM	-2.914103	0.06284	-46.373373
tdLL	1.355044	0.035299	38.387942
tdLK	-0.902455	0.036686	-24.599388
tdLM	-1.206449	0.047235	-25.541394
tdKK	0.403433	0.025609	15.753798
tdKM	-0.695344	0.033221	-20.930978
tdMM	1.457051	0.032419	44.94447

R2 = 0.9971	R2b = 0.9956	# obs = 84
Observation Matrix Rank: 30

Check that the parameter estimates satisfy the restrictions matrix, R.

If the t-ratio = the parameter coefficient / std error seems large in absolute value
with nu = 1 & nu1 = 1 try setting nu = 1.3 & nu1 = 0.5.

VI. As estimated, the parameters of the Normalized Quadratic cost function are:

c = [cL, cK, cM]^T = [1.231023, 0.808053, 1.064702], and

D =	dLL	dLK	dLM
	dKL	dKK	dKM
	dML	dMK	dMM

-2.710088	0.902455	1.206449
0.902455	-0.806865	0.695344
1.206449	0.695344	-2.914103

The Normalized Quadratic cost function is:

C(q;vL,vK,vM) / q^^(1/1) = c(vL,vK,vM) = 1.231023 * vL + 0.808053 * vK + 1.064702 * vM

+ (1/2) *

[-2.710088 * (vL*vL) +	0.902455 * (vL*vK) +	1.206449 * (vL*vM)
+ 0.902455 * (vK*vL) +	-0.806865 * (vK*vK) +	0.695344 * (vK*vM)
+ 1.206449 * (vM*vL) +	0.695344 * (vM*vK) +	-2.914103 * (vM*vM)]

The factor demand functions are:

L(q;vL,vK,vM) / q^^(1/1) = l(vL,vK,vM) = 1.231023 + -2.710088 * vL + 0.902455 * vK + 1.206449 * vM

- (1/2) *

[-2.710088 * (vL*vL) +	0.902455 * (vL*vK) +	1.206449 * (vL*vM)
+ 0.902455 * (vK*vL) +	-0.806865 * (vK*vK) +	0.695344 * (vK*vM)
+ 1.206449 * (vM*vL) +	0.695344 * (vM*vK) +	-2.914103 * (vM*vM)]

K(q;vL,vK,vM) / q^^(1/1) = k(vL,vK,vM) = 0.808053 + 0.902455 * vL + -0.806865 * vK + 0.695344 * vM

- (1/2) *

[-2.710088 * (vL*vL) +	0.902455 * (vL*vK) +	1.206449 * (vL*vM)
+ 0.902455 * (vK*vL) +	-0.806865 * (vK*vK) +	0.695344 * (vK*vM)
+ 1.206449 * (vM*vL) +	0.695344 * (vM*vK) +	-2.914103 * (vM*vM)]

M(q;vL,vK,vM) / q^^(1/1) = m(vL,vK,vM) = 1.064702 + 1.206449 * vL + 0.695344 * vK + -2.914103 * vM

- (1/2) *

[-2.710088 * (vL*vL) +	0.902455 * (vL*vK) +	1.206449 * (vL*vM)
+ 0.902455 * (vK*vL) +	-0.806865 * (vK*vK) +	0.695344 * (vK*vM)
+ 1.206449 * (vM*vL) +	0.695344 * (vM*vK) +	-2.914103 * (vM*vM)]

The Uzawa Partial Elasticities of Substitution in terms of the w factor price vector:

u_LK = C(q;wL,wK,wM) * ∂L(q;wL,wK,wM)/∂wK / (L(q;wL,wK,wM) * K(q;wL,wK,wM)),

u_LM = C(q;wL,wK,wM) * ∂L(q;wL,wK,wM)/∂wM / (L(q;wL,wK,wM) * M(q;wL,wK,wM)),

u_KL = C(q;wL,wK,wM) * ∂K(q;wL,wK,wM)/∂wL / (K(q;wL,wK,wM) * L(q;wL,wK,wM)),

u_KM = C(q;wL,wK,wM) * ∂K(q;wL,wK,wM)/∂wM / (K(q;wL,wK,wM) * M(q;wL,wK,wM)),

u_ML = C(q;wL,wK,wM) * ∂M(q;wL,wK,wM)/∂wL / (M(q;wL,wK,wM) * L(q;wL,wK,wM)),

u_MK = C(q;wL,wK,wM) * ∂M(q;wL,wK,wM)/∂wK / (M(q;wL,wK,wM) * M(q;wL,wK,wM)),

VII. Table of Results: check that the costs, factor inputs, and Uzawa elasticities from the estimated Normalized Quadratic cost function at given levels of output agree with those obtained from the CES cost function.

Normalized Quadratic Cost Function CES Production Function: Returns to Scale = 1, Elasticity of Substitution = 0.85 Normalized Quadratic: Returns to Scale = 1; Base Prices: wL* = 7, wK* = 13, wM* = 6
					— — CES Cost Data — —													— Normalized Quadratic Cost Data —
		Factor Prices			Factor Inputs			Factor Shares			Uzawa Elasticities								Factor Inputs			Factor Shares			Uzawa Elasticities
obs #	q	wL	wK	wM	L	K	M	sL	sK	sM	u_LK	u_LM	u_KL	u_KM	u_ML	u_MK	cost	est cost	est L	est K	est M	sL	sK	sM	u_LK	u_LM	u_KL	u_KM	u_ML	u_MK
1	17	5.72	12.16	6.62	23.37	13.79	15.51	0.331	0.415	0.254	0.85	0.85	0.85	0.85	0.85	0.85	404.07	404.2	23.42	13.8	15.47	0.331	0.415	0.253	0.65	0.94	0.65	1.11	0.94	1.11
2	18	8.18	14.84	5.18	20.62	13.92	22.84	0.342	0.419	0.24	0.85	0.85	0.85	0.85	0.85	0.85	493.5	493.51	20.65	14.01	22.52	0.342	0.421	0.236	1.11	0.75	1.11	0.54	0.75	0.54
3	19	7.28	11.32	5.04	20.99	16.16	21.56	0.344	0.412	0.244	0.85	0.85	0.85	0.85	0.85	0.85	444.35	444.25	20.71	16.18	21.89	0.339	0.412	0.248	1.07	0.99	1.07	0.57	0.99	0.57
4	20	7.36	11.08	4.1	20.89	16.53	25.81	0.347	0.414	0.239	0.85	0.85	0.85	0.85	0.85	0.85	442.72	442.4	20.41	16.88	25.64	0.34	0.423	0.238	1.23	0.9	1.23	0.4	0.9	0.4
5	21	6.64	11.36	7.32	26.48	18.79	18.31	0.336	0.408	0.256	0.85	0.85	0.85	0.85	0.85	0.85	523.36	523.66	26.93	18.73	18.04	0.341	0.406	0.252	0.66	1.22	0.66	1.05	1.22	1.05
6	22	7.06	13.08	4.68	25.62	16.99	27.3	0.341	0.419	0.241	0.85	0.85	0.85	0.85	0.85	0.85	530.98	530.95	25.69	17.05	27.04	0.342	0.42	0.238	1.07	0.75	1.07	0.58	0.75	0.58
7	23	8.9	14.8	7.34	26.98	19.62	23.88	0.34	0.411	0.248	0.85	0.85	0.85	0.85	0.85	0.85	705.81	706.07	26.92	19.49	24.24	0.339	0.409	0.252	0.92	1.02	0.92	0.74	1.02	0.74
8	24	5.2	14.48	4.18	33.68	15.8	30.47	0.33	0.431	0.24	0.85	0.85	0.85	0.85	0.85	0.85	531.28	530.77	34.03	15.72	30.18	0.333	0.429	0.238	0.87	0.43	0.87	0.74	0.43	0.74
9	25	5.04	14.1	4.56	36.01	16.82	29.45	0.328	0.429	0.243	0.85	0.85	0.85	0.85	0.85	0.85	553.02	552.51	35.97	16.71	29.73	0.328	0.427	0.245	0.81	0.48	0.81	0.83	0.48	0.83
10	26	5.88	12.16	4.98	33.13	20.01	28.66	0.335	0.419	0.246	0.85	0.85	0.85	0.85	0.85	0.85	580.87	580.77	33.22	19.87	28.89	0.336	0.416	0.248	0.89	0.76	0.89	0.79	0.76	0.79
11	27	7.02	12.36	6.72	33.36	23.1	26.01	0.337	0.411	0.252	0.85	0.85	0.85	0.85	0.85	0.85	694.47	694.89	33.53	22.94	26.19	0.339	0.408	0.253	0.79	1.06	0.79	0.9	1.06	0.9
12	28	8.5	13.8	4.82	30.15	22.37	36.68	0.345	0.416	0.238	0.85	0.85	0.85	0.85	0.85	0.85	741.74	741.63	29.82	22.78	36.07	0.342	0.424	0.234	1.2	0.82	1.2	0.43	0.82	0.43
13	29	5.94	14.06	5.5	39.51	21.28	31.69	0.331	0.422	0.246	0.85	0.85	0.85	0.85	0.85	0.85	708.12	707.84	39.51	21.12	32.05	0.332	0.419	0.249	0.82	0.66	0.82	0.87	0.66	0.87
14	30	7	13	6	36.89	24.42	31.59	0.337	0.415	0.248	0.85	0.85	0.85	0.85	0.85	0.85	765.17	765.3	36.93	24.24	31.94	0.338	0.412	0.25	0.89	0.9	0.89	0.79	0.9	0.79
15	31	5.26	14.02	6.48	46.82	22.79	29.46	0.325	0.422	0.252	0.85	0.85	0.85	0.85	0.85	0.85	756.75	757.23	45.97	22.9	29.98	0.319	0.424	0.257	0.62	0.7	0.62	1.14	0.7	1.14
16	32	8.08	13.96	4.68	35.37	24.9	42.27	0.344	0.418	0.238	0.85	0.85	0.85	0.85	0.85	0.85	831.18	831.34	35.29	25.27	41.31	0.343	0.424	0.233	1.17	0.76	1.17	0.46	0.76	0.46
17	33	6.2	11.22	7.72	43.58	29.48	27.17	0.333	0.408	0.259	0.85	0.85	0.85	0.85	0.85	0.85	810.74	810.54	44.55	29.64	26.13	0.341	0.41	0.249	0.56	1.24	0.56	1.21	1.24	1.21
18	34	6.64	12.32	4.6	39.98	26.48	41.04	0.34	0.418	0.242	0.85	0.85	0.85	0.85	0.85	0.85	780.52	780.42	40.05	26.5	40.88	0.341	0.418	0.241	1.04	0.77	1.04	0.61	0.77	0.61
19	35	6.44	12.6	5.3	43.46	27.52	38.53	0.337	0.417	0.246	0.85	0.85	0.85	0.85	0.85	0.85	830.91	830.85	43.55	27.34	38.85	0.338	0.415	0.248	0.92	0.81	0.92	0.76	0.81	0.76
20	36	5	14	4.56	51.95	24.26	42.21	0.328	0.429	0.243	0.85	0.85	0.85	0.85	0.85	0.85	791.85	791.13	51.85	24.1	42.65	0.328	0.426	0.246	0.8	0.48	0.8	0.84	0.48	0.84
21	37	8.86	14.8	5.94	41.64	30.16	43.94	0.343	0.415	0.243	0.85	0.85	0.85	0.85	0.85	0.85	1076.3	1076.06	41.32	30.26	44.12	0.34	0.416	0.244	1.08	0.88	1.08	0.56	0.88	0.56
22	38	5.8	11.18	5.02	47.44	30.42	40.3	0.337	0.416	0.247	0.85	0.85	0.85	0.85	0.85	0.85	817.55	817.61	47.52	30.2	40.71	0.337	0.413	0.25	0.88	0.86	0.88	0.81	0.86	0.81
23	39	5.98	12.26	6.66	52.5	31.95	35.99	0.332	0.414	0.254	0.85	0.85	0.85	0.85	0.85	0.85	945.39	945.72	52.71	31.9	35.95	0.333	0.414	0.253	0.68	0.96	0.68	1.07	0.96	1.07
24	40	5.02	14.96	5.26	60.78	26.91	43.89	0.325	0.429	0.246	0.85	0.85	0.85	0.85	0.85	0.85	938.61	938.25	59.79	26.82	45.03	0.32	0.428	0.252	0.72	0.49	0.72	0.95	0.49	0.95
25	41	6.88	12.14	7.62	52.31	36.16	36.03	0.335	0.409	0.256	0.85	0.85	0.85	0.85	0.85	0.85	1073.45	1074	53.09	36.07	35.55	0.34	0.408	0.252	0.66	1.17	0.66	1.06	1.17	1.06
26	42	7	12.64	4.32	47.77	32.38	54.09	0.342	0.419	0.239	0.85	0.85	0.85	0.85	0.85	0.85	977.42	977.52	47.85	32.67	53.13	0.343	0.422	0.235	1.12	0.74	1.12	0.52	0.74	0.52
27	43	7.6	13.24	6.98	52.46	36.66	42.37	0.338	0.411	0.251	0.85	0.85	0.85	0.85	0.85	0.85	1179.84	1180.5	52.63	36.39	42.8	0.339	0.408	0.253	0.83	1.04	0.83	0.85	1.04	0.85
28	44	8.32	14.56	7.1	52.93	36.85	45.51	0.339	0.413	0.249	0.85	0.85	0.85	0.85	0.85	0.85	1300.05	1300.49	52.94	36.59	46.09	0.339	0.41	0.252	0.89	0.98	0.89	0.78	0.98	0.78
AVE:	30.5	6.74	13.09	5.69	38.46	24.15	33.32	0.337	0.417	0.246	0.85	0.85	0.85	0.85	0.85	0.85	758.21	758.23	38.46	24.15	33.32	0.337	0.417	0.246	0.89	0.84	0.89	0.78	0.84	0.78

VIII. The Normalized Quadratic cost function as a flexible functional form.

Diewert (1976) defines a flexible function form as follows. Write c(w) as the unit Normalized Quadratic cost function, and c*(w) as the unit CES cost function. Then the unit Normalized Quadratic cost function, c(w), is a flexible functional form if it has enough free parameters to satisfy:

c(w) = c*(w),

∇c(w) = ∇c*(w), and

∇²c(w) = ∇²c*(w),

at any factor price vector w = [wL, wK, wM]^T in the domain of definition of c(w).

Since both c(w) and c*(w) are linear homogeneous functions (Diewert and Fox, (2009), 155), the 6 free parameters of c(w) classify the Normalized Quadratic cost function as a "parsimonious flexible functional form" within the present context.

However, when we estimate the unit Normalized Quadratic cost function, c(w), we choose these free parameters to minimize the distance between ∇c(w) and ∇c*(w) at the set of data points {q, wL, wK, wM, ∇c*(w)}, subject to the augmented restrictions matrix, [R | r].

Consequently, ∇²c*(w) and ∇²c(w) can differ, as seen by comparing the Uzawa elasticities in the table above, and the second order partial derivatives of c*(w) and c(w) in the table below.

IX. Table of Results: check that second order partial derivatives of the estimated Normalized Quadratic cost function for a given level of output agree with the second order partial derivatives of the CES cost function. If all of the eigenvalues of ∇²c*(w) and ∇²c(w) are nonpositive, the cost functions c*(w*) and c(w*) are concave in factor prices.

(Write: ∂(∂c(wL,wK,wM)/∂wL)/∂wL = c_LL, ∂(∂c(wL,wK,wM)/∂wL)/∂wK = c_LK, etc.)

Normalized Quadratic Cost Function CES Production Function: Returns to Scale = 1, Elasticity of Substitution = 0.85 Normalized Quadratic: Returns to Scale = 1; Base Prices: wL* = 7, wK* = 13, wM* = 6
					— — CES Cost Data — —									— — Normalized Quadratic Cost Data — —
		Factor Prices			∇²c*(w)						Eigenvalues			∇²c(w)						Eigenvalues
obs #	q	wL	wK	wM	c_LL	c_LK	c_LM	c_KK	c_KM	c_MM	e₁	e₂	e₃	c_LL	c_LK	c_LM	c_KK	c_KM	c_MM	e₁	e₂	e₃
1	17	5.72	12.16	6.62	-0.137	0.04	0.045	-0.033	0.026	-0.087	-0.167	-0.09	-0	-0.123	0.031	0.05	-0.033	0.035	-0.106	-0.165	-0.097	-0
2	18	8.18	14.84	5.18	-0.078	0.027	0.045	-0.026	0.03	-0.158	-0.18	-0.082	-0	-0.09	0.036	0.039	-0.027	0.019	-0.117	-0.145	-0.088	-0
3	19	7.28	11.32	5.04	-0.085	0.034	0.046	-0.038	0.035	-0.145	-0.171	-0.096	0	-0.103	0.042	0.053	-0.038	0.024	-0.131	-0.172	-0.099	-0
4	20	7.36	11.08	4.1	-0.079	0.033	0.052	-0.037	0.041	-0.204	-0.226	-0.093	-0	-0.102	0.048	0.053	-0.039	0.02	-0.149	-0.184	-0.105	-0
5	21	6.64	11.36	7.32	-0.107	0.038	0.038	-0.04	0.027	-0.075	-0.136	-0.086	-0	-0.111	0.03	0.054	-0.039	0.032	-0.099	-0.159	-0.089	-0
6	22	7.06	13.08	4.68	-0.092	0.032	0.051	-0.029	0.034	-0.171	-0.198	-0.095	-0	-0.104	0.04	0.045	-0.03	0.023	-0.131	-0.164	-0.1	-0
7	23	8.9	14.8	7.34	-0.074	0.028	0.034	-0.029	0.025	-0.09	-0.117	-0.076	0	-0.083	0.03	0.041	-0.029	0.021	-0.093	-0.13	-0.075	-0
8	24	5.2	14.48	4.18	-0.154	0.035	0.068	-0.022	0.032	-0.196	-0.247	-0.125	0	-0.129	0.036	0.035	-0.021	0.027	-0.138	-0.169	-0.12	0
9	25	5.04	14.1	4.56	-0.163	0.037	0.065	-0.023	0.03	-0.166	-0.23	-0.123	-0	-0.132	0.035	0.037	-0.022	0.03	-0.133	-0.17	-0.118	-0
10	26	5.88	12.16	4.98	-0.122	0.037	0.053	-0.031	0.032	-0.142	-0.187	-0.109	-0	-0.122	0.039	0.048	-0.031	0.03	-0.131	-0.175	-0.109	0
11	27	7.02	12.36	6.72	-0.099	0.035	0.039	-0.035	0.027	-0.091	-0.135	-0.09	0	-0.105	0.033	0.049	-0.034	0.029	-0.104	-0.154	-0.089	-0
12	28	8.5	13.8	4.82	-0.07	0.028	0.045	-0.029	0.034	-0.176	-0.196	-0.079	0	-0.088	0.039	0.042	-0.03	0.017	-0.124	-0.152	-0.089	-0
13	29	5.94	14.06	5.5	-0.13	0.035	0.052	-0.026	0.028	-0.127	-0.181	-0.102	-0	-0.117	0.033	0.041	-0.025	0.029	-0.117	-0.158	-0.102	-0
14	30	7	13	6	-0.099	0.033	0.043	-0.031	0.029	-0.112	-0.149	-0.093	-0	-0.104	0.035	0.046	-0.031	0.027	-0.112	-0.155	-0.092	-0
15	31	5.26	14.02	6.48	-0.165	0.039	0.05	-0.026	0.024	-0.093	-0.194	-0.09	0	-0.125	0.028	0.041	-0.026	0.033	-0.106	-0.157	-0.099	0
16	32	8.08	13.96	4.68	-0.076	0.028	0.048	-0.028	0.034	-0.183	-0.204	-0.083	-0	-0.092	0.039	0.041	-0.029	0.018	-0.126	-0.154	-0.092	0
17	33	6.2	11.22	7.72	-0.121	0.041	0.038	-0.04	0.025	-0.067	-0.147	-0.081	-0	-0.117	0.028	0.054	-0.039	0.035	-0.094	-0.161	-0.09	-0
18	34	6.64	12.32	4.6	-0.099	0.034	0.053	-0.031	0.035	-0.169	-0.199	-0.101	0	-0.111	0.042	0.048	-0.032	0.025	-0.135	-0.172	-0.105	-0
19	35	6.44	12.6	5.3	-0.109	0.035	0.049	-0.031	0.031	-0.133	-0.171	-0.101	0	-0.112	0.038	0.047	-0.031	0.028	-0.123	-0.166	-0.101	0
20	36	5	14	4.56	-0.165	0.038	0.065	-0.023	0.031	-0.165	-0.231	-0.123	0	-0.133	0.035	0.037	-0.022	0.03	-0.134	-0.171	-0.118	0
21	37	8.86	14.8	5.94	-0.071	0.027	0.039	-0.027	0.028	-0.129	-0.15	-0.077	0	-0.084	0.034	0.04	-0.028	0.019	-0.107	-0.138	-0.081	-0
22	38	5.8	11.18	5.02	-0.121	0.039	0.052	-0.036	0.034	-0.135	-0.181	-0.111	0	-0.125	0.041	0.054	-0.035	0.032	-0.133	-0.183	-0.111	0
23	39	5.98	12.26	6.66	-0.128	0.039	0.044	-0.033	0.027	-0.088	-0.159	-0.09	0	-0.119	0.031	0.049	-0.033	0.033	-0.106	-0.162	-0.095	-0
24	40	5.02	14.96	5.26	-0.174	0.037	0.06	-0.022	0.027	-0.134	-0.218	-0.111	0	-0.128	0.031	0.035	-0.021	0.031	-0.12	-0.159	-0.11	-0
25	41	6.88	12.14	7.62	-0.105	0.037	0.036	-0.036	0.025	-0.073	-0.132	-0.082	-0	-0.107	0.029	0.05	-0.036	0.031	-0.095	-0.151	-0.086	0
26	42	7	12.64	4.32	-0.091	0.032	0.054	-0.03	0.036	-0.193	-0.218	-0.096	0	-0.105	0.043	0.046	-0.031	0.022	-0.138	-0.171	-0.104	-0
27	43	7.6	13.24	6.98	-0.09	0.032	0.037	-0.032	0.026	-0.09	-0.128	-0.085	-0	-0.097	0.031	0.046	-0.032	0.026	-0.1	-0.145	-0.084	-0
28	44	8.32	14.56	7.1	-0.081	0.029	0.036	-0.029	0.025	-0.093	-0.123	-0.08	-0	-0.089	0.03	0.042	-0.028	0.023	-0.096	-0.134	-0.079	-0
AVE:	30.5	6.74	13.09	5.69	-0.11	0.034	0.048	-0.03	0.03	-0.132	-0.178	-0.095	-0	-0.109	0.035	0.045	-0.03	0.027	-0.118	-0.16	-0.097	-0

X. The Factor Demand Price Elasticities of the Normalized Quadratic cost function

The estimated factor demand elasticities are obtained by:

ε_L,wL = ∂ln(l(wL,wK,wM))/∂ln(wL) = wL * c_LL / l(wL,wK,wM),

ε_L,wK = ∂ln(l(wL,wK,wM))/∂ln(wK) = wK * c_LK / l(wL,wK,wM),

ε_L,wM = ∂ln(l(wL,wK,wM))/∂ln(wM) = wM * c_LM / l(wL,wK,wM),

ε_K,wL = ∂ln(k(wL,wK,wM))/∂ln(wL) = wL * c_KL / k(wL,wK,wM), etc.

XI. Table of Results: check that factor demand price elasticities of the estimated Normalized Quadratic cost function for a given level of output and factor prices agree with the factor demand price elasticities of the CES cost function.

Normalized Quadratic Cost Function CES Production Function: Returns to Scale = 1, Elasticity of Substitution = 0.85 Normalized Quadratic: Returns to Scale = 1; Base Prices: wL* = 7, wK* = 13, wM* = 6
					— — CES Cost Data — —						— — Normalized Quadratic Cost Data — —
		Factor Prices			Factor Demand Price Elasticities						Factor Demand Price Elasticities
obs #	q	wL	wK	wM	ε_L,wL	ε_L,wK	ε_L,wM	ε_K,wK	ε_K,wM	ε_M,wM	ε_L,wL	ε_L,wK	ε_L,wM	ε_K,wK	ε_K,wM	ε_M,wM
1	17	5.72	12.16	6.62	-0.569	0.353	0.216	-0.497	0.216	-0.634	-0.509	0.271	0.238	-0.293	0.166	-0.511
2	18	8.18	14.84	5.18	-0.56	0.356	0.204	-0.494	0.204	-0.646	-0.643	0.466	0.177	-0.343	0.086	-0.526
3	19	7.28	11.32	5.04	-0.558	0.35	0.208	-0.5	0.208	-0.642	-0.686	0.439	0.247	-0.393	0.11	-0.604
4	20	7.36	11.08	4.1	-0.555	0.352	0.203	-0.498	0.203	-0.647	-0.732	0.519	0.214	-0.423	0.079	-0.597
5	21	6.64	11.36	7.32	-0.564	0.347	0.218	-0.503	0.218	-0.632	-0.575	0.268	0.307	-0.341	0.184	-0.564
6	22	7.06	13.08	4.68	-0.56	0.356	0.205	-0.494	0.205	-0.645	-0.629	0.451	0.178	-0.334	0.091	-0.524
7	23	8.9	14.8	7.34	-0.561	0.35	0.211	-0.5	0.211	-0.639	-0.634	0.376	0.258	-0.361	0.135	-0.584
8	24	5.2	14.48	4.18	-0.57	0.366	0.204	-0.484	0.204	-0.646	-0.474	0.372	0.102	-0.215	0.081	-0.408
9	25	5.04	14.1	4.56	-0.571	0.365	0.206	-0.485	0.206	-0.644	-0.462	0.345	0.117	-0.218	0.095	-0.423
10	26	5.88	12.16	4.98	-0.565	0.356	0.209	-0.494	0.209	-0.641	-0.56	0.372	0.189	-0.297	0.117	-0.509
11	27	7.02	12.36	6.72	-0.563	0.349	0.214	-0.501	0.214	-0.636	-0.592	0.324	0.267	-0.339	0.155	-0.565
12	28	8.5	13.8	4.82	-0.556	0.354	0.203	-0.496	0.203	-0.647	-0.7	0.508	0.192	-0.391	0.078	-0.561
13	29	5.94	14.06	5.5	-0.568	0.359	0.209	-0.491	0.209	-0.641	-0.508	0.344	0.164	-0.262	0.116	-0.474
14	30	7	13	6	-0.563	0.353	0.211	-0.497	0.211	-0.639	-0.593	0.367	0.226	-0.328	0.13	-0.546
15	31	5.26	14.02	6.48	-0.573	0.359	0.214	-0.491	0.214	-0.636	-0.443	0.264	0.178	-0.245	0.146	-0.461
16	32	8.08	13.96	4.68	-0.558	0.355	0.202	-0.495	0.202	-0.648	-0.673	0.497	0.176	-0.365	0.077	-0.535
17	33	6.2	11.22	7.72	-0.567	0.347	0.22	-0.503	0.22	-0.63	-0.539	0.229	0.309	-0.326	0.2	-0.539
18	34	6.64	12.32	4.6	-0.561	0.355	0.206	-0.495	0.206	-0.644	-0.623	0.437	0.186	-0.333	0.097	-0.529
19	35	6.44	12.6	5.3	-0.564	0.355	0.209	-0.495	0.209	-0.641	-0.582	0.381	0.201	-0.313	0.118	-0.526
20	36	5	14	4.56	-0.571	0.365	0.207	-0.485	0.207	-0.643	-0.461	0.343	0.118	-0.218	0.096	-0.424
21	37	8.86	14.8	5.94	-0.559	0.352	0.206	-0.498	0.206	-0.644	-0.665	0.451	0.214	-0.37	0.1	-0.57
22	38	5.8	11.18	5.02	-0.564	0.354	0.21	-0.496	0.21	-0.64	-0.58	0.364	0.215	-0.317	0.128	-0.534
23	39	5.98	12.26	6.66	-0.568	0.352	0.216	-0.498	0.216	-0.634	-0.524	0.281	0.243	-0.301	0.164	-0.52
24	40	5.02	14.96	5.26	-0.574	0.365	0.209	-0.485	0.209	-0.641	-0.431	0.308	0.123	-0.211	0.108	-0.423
25	41	6.88	12.14	7.62	-0.565	0.348	0.217	-0.502	0.217	-0.633	-0.566	0.27	0.296	-0.334	0.181	-0.556
26	42	7	12.64	4.32	-0.559	0.356	0.203	-0.494	0.203	-0.647	-0.648	0.474	0.174	-0.346	0.083	-0.525
27	43	7.6	13.24	6.98	-0.563	0.35	0.213	-0.5	0.213	-0.637	-0.602	0.339	0.263	-0.343	0.149	-0.569
28	44	8.32	14.56	7.1	-0.562	0.351	0.211	-0.499	0.211	-0.639	-0.612	0.366	0.246	-0.345	0.136	-0.567
AVE:	30.5	6.74	13.09	5.69	-0.564	0.355	0.209	-0.495	0.209	-0.641	-0.58	0.372	0.208	-0.318	0.122	-0.524

XII. Curvature of the Normalized Quadratic cost function.

Approximating a CES cost function with a Normalized Quadratic cost function provides a "best of all possible worlds" test for the effectiveness of the Normalized Quadratic cost function. Diewert and Wales (1987) prove that a necessary and sufficient condition for the estimated cost function, c(w), to be concave in factor prices is that the estimated matrix:

D =	dLL	dLK	dLM
	dKL	dKK	dKM
	dML	dMK	dMM

-2.710088	0.902455	1.206449
0.902455	-0.806865	0.695344
1.206449	0.695344	-2.914103

be a symmetric, negative semidefinite matrix.

The eigenvalues of the matrix D are:

e1 = -4.026, e2 = -2.405, e3 = -0.

The matrix D is negative semidefinite, since its three eigenvalues are nonpositive.

Furthermore, at the (constant) reference vector, v*, of base prices:

v* = w* / (wL* + wK* + wM*) = [vL*, vK*, vM*]^T = [0.2692, 0.5, 0.2308]^T = [7, 13, 6]^T / 26, and

D * v* = 0,

making v* the eigenvector corresponding to the zero eigenvalue of the D matrix.

If D is a symmetric, negative semidefinite matrix, then -D is a symmetric, positive semidefinite matrix that can be decomposed into the product of a lower triangular matrix L and its transpose L^T:

-D = L * L^T, where

L =	bLL	0	0
	bKL	bKK	0
	bML	bMK	bMM

L^T =	bLL	bKL	bML
	0	bKK	bMK
	0	0	bMM

Applying the reference vector, v*, to L^T, we get the equations:

bLL * vL* + bKL * vK* + bML * vM* = 0,

bKK * vK* + bMK * vM* = 0,

bMM * vM* = 0.

bMM = 0,

bMK = - bKK * vK* / vM*,

bML = -(bLL * vL* + bKL * vK*) / vM*.

Consequently, the matrices L, L^T, -D, and D can be expressed in terms of three parameters, bLL, bKL, and bKK.

-D = B =	bLL^²	bLL*bKL	bLL(-bLLvL* - bKLvK)/vM*
	bLL*bKL	bKL^² + bKK^²	bKL(-bLLvL-bKLvK)/vM - bKK^² * vK/vM
	bLL(-bLLvL* - bKLvK)/vM*	bKL(-bLLvL* - bKLvK)/vM* - bKK^² * vK/vM	(bLLvL + bKLvK)^² /vM^² + bKK^² vK^² /vM^²

bLL	0	0
bKL	bKK	0
-(bLLvL+bKLvK)/vM*	-bKKvK/vM*	0

bLL	bKL	-(bLLvL+bKLvK)/vM*
0	bKK	-bKKvK/vM*
0	0	0

The -B matrix is a reparameterization of the of the D matrix. By construction -B contains all the restrictions inherent in the restrictions matrix and vector, R, r, of the original parameters, and by construction the -B matrix is a symmetric, negative semidefinite matrix.

The reparameterized unit normalized quadratic cost function can be written as:

ç(w) = b^T * w + (1/2) * w^T * (-B) * w / (1^T * w),

a nonlinear function in its parameters, b = [bL, bK, bM]^T, and (-B), i.e. parameters bL, bK, bM, bLL, bKL, and bKK.

If c*(w) represents the unit cost function that we want to approximate at the set of data points {q, wL, wK, wM, ∇c*(w)}, our objective is to use a nonlinear regression algorithm to obtain values for these six free parameters that minimize the distance between ∇ç(w) and ∇c*(w) at the set of data points.

Of course, if the symmetric matrix D as estimated is already a negative semidefinite matrix, the reparameterization of the unit cost function, c(w), and the reestimation of ç(w) with a nonlinear regression algorithm are unnecessary.

Equating the parameters of the normalized quadratic cost function as estimated, c(w), and its reparameterization, ç(w), we determine:

bL = cL, bK = cK, bM = cM, and

bLL^² = -dLL,

bLL*bKL = - dKL, and

bKL^² + bKK^² = -dKK.

bLL = (-dLL)^^1/2,

bKL = - dKL / bLL, and

bKK = (-dKK - bKL^²)^^1/2.

If restricted to using real numbers, we can always transform the estimated b and -B parameters into c and D parameters, and the estimated c and D parameters into b and -B parameters if the matrix D is negative semidefinite. If the matrix D is not negative semidefinite, the transformation from the estimated c and D parameters into b and -B parameters will require the use of complex numbers.

Transforming the estimated c and D parameters into b and B parameters, we obtain:

bL = 1.231023, bK = 0.808053, bM = 1.064702, and

bLL = (-dLL)^^1/2 = 1.6462345 + 0*i,

bKL = - dKL / bLL = -0.5481936 + 0*i, and

bKK = (-dKK - bKL^²)^^1/2 = (0.5063491 + -0*i)^^1/2 = 0.7115821 + 0*i.

À2: Estimate the factor demand equations using nonlinear least squares.

XIII. Nonlinear Estimation of the Normalized Quadratic cost function.

The reparameterized unit normalized quadratic cost function::

ç(w) = b^T * w + (1/2) * w^T * (-B) * w / (1^T * w),

is a nonlinear function in its parameters, b = [bL, bK, bM]^T, and (-B), i.e. in its free parameters bL, bK, bM, bLL, bKL, and bKK.

We will approximate the CES unit cost function, c*(w), at the set of data points {q, wL, wK, wM, ∇c*(w)}, with the normalized quadratic cost function, ç(w), by using a nonlinear regression algorithm to obtain values for the b,and (-B) parameters that minimize the distance between ∇ç(w) and ∇c*(w) at the set of data points. Having obtained estimates of the six free parameters, bL, bK, bM, bLL, bKL, and bKK, we will transform them into estimates of the c, and D parameters using c = b, and D = (-B).

Shephard's lemma provides that the gradient vector of the unit cost function is the vector of unit input demand functions:

∇ç(w) = [l(w), k(w), m(w)]^T =b + (-B) * w / ( 1^T * w) - (1/2) * w^T * (-B) * w * 1 / (1^T * w)².

The reparameterized factor demands as functions of q, vL, vK, and vM are:

L(q; vL, vK, vM) / q^^(1/nu1) =	l(vL, vK, vM) =	bL + (-bLL^2 * vL - bLLbKL vK - (bLL(-bLLvL-bKLvK)/vM) * vM	- (1/2) * v^T * (-B) * v
K(q; vL, vK, vM) / q^^(1/nu1) =	k(vL, vK, vM) =	bK + (-bLLbKL vL + (-bKL^2-bKK^2) * vK + ((-bKL(-bLLvL-bKLvK)/vM)+bKK^2vK/vM) vM	- (1/2) * v^T * (-B) * v
M(q; vL, vK, vM) / q^^(1/nu1) =	m(vL, vK, vM) =	bM + ((-bLL(-bLLvL-bKLvK)/vM) * vL + ((-bKL(-bLLvL-bKLvK)/vM)+(bKK^2vK)/vM) vK + (-(-bLLvL-bKLvK)^2/vM^2-bKK^2vK^2/vM^2) * vM)	- (1/2) * v^T * (-B) * v

where the (constant) reference vector, v*, of base prices is:

v* = w* / (wL* + wK* + wM*) = [vL*, vK*, vM*]^T = [0.2692, 0.5, 0.2308]^T = [7, 13, 6]^T / 26.

XIV. Newton's Nonlinear Least Squares Method.

Newton's nonlinear least squares method is a generalization of Newton's method for finding a root of the equation, g(x) = 0, where g is a function of the 1-dimensional real variable x. Newton's method calculates a sequence of approximations {xⁿ: n = 0, 1, 2, ...} to g(xⁿ) ~= 0, beginning with a starting value, x⁰. At each iteration, the new approximation, xⁿ⁺¹, is calculated according to the formula:

xⁿ⁺¹ = xⁿ - g(xⁿ) / g'(xⁿ).

where g'(x) is the derivative of the function g evaluated at the value of the variable x.

Newton's method generalizes for a system of n equations:

G(x) = [g₁(x), g₂(x), ... , g_n(x)]^T = 0

where G, and thus g₁, g₂, ... , g_n are functions of the n-dimension real variable (vector) x = [x₁, x₂, ..., x_n]^T. Newton's method calculates a sequence of vector approximations {xⁿ: n = 0, 1, 2, ...} to G(xⁿ) ~= 0, beginning with a vector of starting values, x⁰. At each iteration, the new vector, xⁿ⁺¹, is calculated according to the formula:

xⁿ⁺¹ = xⁿ - J(xⁿ)^^-1 * G(xⁿ)

where J(x) is the nxn Jacobian matrix of G(x) defined by:

J(x) = [∂g_i / ∂x_j], i, j = 1, ..., n.

At each iteration, Newton's algorithm solves the system of linear equations for the vector y:

J(xⁿ) * y = - G(xⁿ),

and computes the new vector, xⁿ⁺¹, by:

xⁿ⁺¹ = xⁿ + y.

The algorithm stops when the norm of y is less than some tolerance value, TOL, i.e, ||y|| < TOL.

We will use Newton's algorithm to obtain the "least squares fit" between the gradient of the CES unit cost function, c*(w), and the gradient of the normalized quadratic unit cost function, ç(w) at the set of data points {q, wL, wK, wM, ∇c*(w)}.

Let {[l_i, k_i, m_i]^T = ∇c*(w_i)} be the CES factor demands at the data set {q_i, w_i = [wL_i, wK_i, wM_i]^T}.

Objective: Choose parameters bL, bK bM, bLL, bKL, bKK that minimize:

S(bL, bK, bM, bLL, bKL, bKK) = ∑_i [ (l_i - l(w_i))^² + (k_i - k(w_i))^² + (m_i - m(w_i))^²]

over the data set {q_i, w_i}.

The first order conditions are the system of six equations:

[∂S/∂bL, ∂S/∂bK, ∂S/∂bM, ∂S/∂bLL, ∂S/∂bKL, ∂S/∂bKK]^T = 0.

For example:

∂S/∂bL = 2 * ∑_i [ (l_i - l(w_i)) * ∂l(w_i)/∂bL + (k_i - k(w_i)) * ∂k(w_i)/∂bL + (m_i - m(w_i)) * ∂m(w_i)/∂bL] = 0.

The Jacobian of the first order conditions, also the Hessian matrix of the function S(bL, bK, bM, bLL, bKL, bKK), is the 6 by 6 matrix of second order partial derivatives:

J(bL, bK, bM, bLL, bKL, bKK) = [ ∂²S / ∂b_∂b_ ]

for all b_ ∈ {bL, bK, bM, bLL, bKL, bKK}.

(In practice, only the first order partial derivatives of l(w_i), k(w_i), and m(w_i) of the Jacobian are used.)

Using Newton's nonlinear regression algorithm as discussed, we get the following estimates:

Nonlinear Least Squares: Newton's Method

Parameter Estimates

Iter # bL bK bM bLL bKL bKK

0 1 1 1 1 -1 1

1 1.229848 0.808985 1.063931 1.886453 -0.042914 1.364483

2 1.229848 0.808985 1.063931 1.678179 -0.497391 0.965846

3 1.229848 0.808985 1.063931 1.665255 -0.557625 0.745633

4 1.229848 0.808985 1.063931 1.665205 -0.55811 0.71032

5 1.229848 0.808985 1.063931 1.665205 -0.55811 0.709442

6 1.229848 0.808985 1.063931 1.665205 -0.55811 0.709442

7 1.229848 0.808985 1.063931 1.665205 -0.55811 0.709442

Next, we transform these estimates of the six free parameters, bL, bK, bM, bLL, bKL, and bKK into estimates of the c, and D parameters using c = b, and D = (-B). By construction, the estimated matrix D is negative semidefinite.

XV. As estimated using nonlinear least squares, the parameters of the Normalized Quadratic cost function are:

c = [cL, cK, cM]^T = [1.229848, 0.808985, 1.063931], and

D =	dLL	dLK	dLM
	dKL	dKK	dKM
	dML	dMK	dMM

-2.772907	0.929367	1.221429
0.929367	-0.814794	0.681125
1.221429	0.681125	-2.900772

The eigenvalues of the matrix D are:

e1 = -4.068, e2 = -2.421, e3 = -0.

The matrix D is negative semidefinite, since its three eigenvalues are nonpositive.

The Normalized Quadratic cost function is:

C(q;vL,vK,vM) / q^^(1/1) = c(vL,vK,vM) = 1.229848 * vL + 0.808985 * vK + 1.063931 * vM

+ (1/2) *

[-2.772907 * (vL*vL) +	0.929367 * (vL*vK) +	1.221429 * (vL*vM)
+ 0.929367 * (vK*vL) +	-0.814794 * (vK*vK) +	0.681125 * (vK*vM)
+ 1.221429 * (vM*vL) +	0.681125 * (vM*vK) +	-2.900772 * (vM*vM)]

The factor demand functions are:

L(q;vL,vK,vM) / q^^(1/1) = l(vL,vK,vM) = 1.229848 + -2.772907 * vL + 0.929367 * vK + 1.221429 * vM

- (1/2) *

[-2.772907 * (vL*vL) +	0.929367 * (vL*vK) +	1.221429 * (vL*vM)
+ 0.929367 * (vK*vL) +	-0.814794 * (vK*vK) +	0.681125 * (vK*vM)
+ 1.221429 * (vM*vL) +	0.681125 * (vM*vK) +	-2.900772 * (vM*vM)]

K(q;vL,vK,vM) / q^^(1/1) = k(vL,vK,vM) = 0.808985 + 0.929367 * vL + -0.814794 * vK + 0.681125 * vM

- (1/2) *

[-2.772907 * (vL*vL) +	0.929367 * (vL*vK) +	1.221429 * (vL*vM)
+ 0.929367 * (vK*vL) +	-0.814794 * (vK*vK) +	0.681125 * (vK*vM)
+ 1.221429 * (vM*vL) +	0.681125 * (vM*vK) +	-2.900772 * (vM*vM)]

M(q;vL,vK,vM) / q^^(1/1) = m(vL,vK,vM) = 1.063931 + 1.221429 * vL + 0.681125 * vK + -2.900772 * vM

- (1/2) *

[-2.772907 * (vL*vL) +	0.929367 * (vL*vK) +	1.221429 * (vL*vM)
+ 0.929367 * (vK*vL) +	-0.814794 * (vK*vK) +	0.681125 * (vK*vM)
+ 1.221429 * (vM*vL) +	0.681125 * (vM*vK) +	-2.900772 * (vM*vM)]

XVI. Table of Results: check that the costs, factor inputs, and Uzawa elasticities from the estimated Normalized Quadratic cost function at given levels of output agree with those obtained from the CES cost function.

Normalized Quadratic Cost Function CES Production Function: Returns to Scale = 1, Elasticity of Substitution = 0.85 Normalized Quadratic: Returns to Scale = 1; Base Prices: wL* = 7, wK* = 13, wM* = 6
					— — CES Cost Data — —													— Normalized Quadratic Cost Data —
		Factor Prices			Factor Inputs			Factor Shares			Uzawa Elasticities								Factor Inputs			Factor Shares			Uzawa Elasticities
obs #	q	wL	wK	wM	L	K	M	sL	sK	sM	u_LK	u_LM	u_KL	u_KM	u_ML	u_MK	cost	est cost	est L	est K	est M	sL	sK	sM	u_LK	u_LM	u_KL	u_KM	u_ML	u_MK
1	17	5.72	12.16	6.62	23.37	13.79	15.51	0.331	0.415	0.254	0.85	0.85	0.85	0.85	0.85	0.85	404.07	404.17	23.45	13.79	15.45	0.332	0.415	0.253	0.67	0.95	0.67	1.1	0.95	1.1
2	18	8.18	14.84	5.18	20.62	13.92	22.84	0.342	0.419	0.24	0.85	0.85	0.85	0.85	0.85	0.85	493.5	493.52	20.6	14.05	22.49	0.341	0.422	0.236	1.14	0.76	1.14	0.52	0.76	0.52
3	19	7.28	11.32	5.04	20.99	16.16	21.56	0.344	0.412	0.244	0.85	0.85	0.85	0.85	0.85	0.85	444.35	444.17	20.62	16.23	21.88	0.338	0.414	0.248	1.1	1.01	1.1	0.55	1.01	0.55
4	20	7.36	11.08	4.1	20.89	16.53	25.81	0.347	0.414	0.239	0.85	0.85	0.85	0.85	0.85	0.85	442.72	442.3	20.3	16.95	25.63	0.338	0.425	0.238	1.26	0.92	1.26	0.38	0.92	0.38
5	21	6.64	11.36	7.32	26.48	18.79	18.31	0.336	0.408	0.256	0.85	0.85	0.85	0.85	0.85	0.85	523.36	523.63	26.9	18.74	18.05	0.341	0.407	0.252	0.68	1.23	0.68	1.03	1.23	1.03
6	22	7.06	13.08	4.68	25.62	16.99	27.3	0.341	0.419	0.241	0.85	0.85	0.85	0.85	0.85	0.85	530.98	530.96	25.64	17.09	27.01	0.341	0.421	0.238	1.1	0.76	1.1	0.56	0.76	0.56
7	23	8.9	14.8	7.34	26.98	19.62	23.88	0.34	0.411	0.248	0.85	0.85	0.85	0.85	0.85	0.85	705.81	706	26.86	19.53	24.24	0.339	0.409	0.252	0.95	1.04	0.95	0.72	1.04	0.72
8	24	5.2	14.48	4.18	33.68	15.8	30.47	0.33	0.431	0.24	0.85	0.85	0.85	0.85	0.85	0.85	531.28	530.8	34.13	15.72	30.09	0.334	0.429	0.237	0.89	0.44	0.89	0.73	0.44	0.73
9	25	5.04	14.1	4.56	36.01	16.82	29.45	0.328	0.429	0.243	0.85	0.85	0.85	0.85	0.85	0.85	553.02	552.49	36.08	16.7	29.65	0.329	0.426	0.245	0.83	0.48	0.83	0.82	0.48	0.82
10	26	5.88	12.16	4.98	33.13	20.01	28.66	0.335	0.419	0.246	0.85	0.85	0.85	0.85	0.85	0.85	580.87	580.78	33.23	19.88	28.85	0.336	0.416	0.247	0.92	0.77	0.92	0.78	0.77	0.78
11	27	7.02	12.36	6.72	33.36	23.1	26.01	0.337	0.411	0.252	0.85	0.85	0.85	0.85	0.85	0.85	694.47	694.85	33.49	22.96	26.19	0.338	0.408	0.253	0.82	1.07	0.82	0.88	1.07	0.88
12	28	8.5	13.8	4.82	30.15	22.37	36.68	0.345	0.416	0.238	0.85	0.85	0.85	0.85	0.85	0.85	741.74	741.56	29.69	22.86	36.05	0.34	0.425	0.234	1.23	0.84	1.23	0.42	0.84	0.42
13	29	5.94	14.06	5.5	39.51	21.28	31.69	0.331	0.422	0.246	0.85	0.85	0.85	0.85	0.85	0.85	708.12	707.84	39.57	21.11	31.99	0.332	0.419	0.249	0.84	0.67	0.84	0.86	0.67	0.86
14	30	7	13	6	36.89	24.42	31.59	0.337	0.415	0.248	0.85	0.85	0.85	0.85	0.85	0.85	765.17	765.28	36.9	24.27	31.92	0.337	0.412	0.25	0.92	0.92	0.92	0.78	0.92	0.78
15	31	5.26	14.02	6.48	46.82	22.79	29.46	0.325	0.422	0.252	0.85	0.85	0.85	0.85	0.85	0.85	756.75	757.09	46.12	22.87	29.93	0.32	0.423	0.256	0.64	0.7	0.64	1.13	0.7	1.13
16	32	8.08	13.96	4.68	35.37	24.9	42.27	0.344	0.418	0.238	0.85	0.85	0.85	0.85	0.85	0.85	831.18	831.34	35.18	25.35	41.27	0.342	0.426	0.232	1.2	0.77	1.2	0.45	0.77	0.45
17	33	6.2	11.22	7.72	43.58	29.48	27.17	0.333	0.408	0.259	0.85	0.85	0.85	0.85	0.85	0.85	810.74	810.51	44.55	29.63	26.15	0.341	0.41	0.249	0.58	1.25	0.58	1.19	1.25	1.19
18	34	6.64	12.32	4.6	39.98	26.48	41.04	0.34	0.418	0.242	0.85	0.85	0.85	0.85	0.85	0.85	780.52	780.44	39.98	26.55	40.84	0.34	0.419	0.241	1.07	0.79	1.07	0.59	0.79	0.59
19	35	6.44	12.6	5.3	43.46	27.52	38.53	0.337	0.417	0.246	0.85	0.85	0.85	0.85	0.85	0.85	830.91	830.85	43.53	27.37	38.8	0.337	0.415	0.248	0.95	0.82	0.95	0.75	0.82	0.75
20	36	5	14	4.56	51.95	24.26	42.21	0.328	0.429	0.243	0.85	0.85	0.85	0.85	0.85	0.85	791.85	791.1	52.02	24.08	42.53	0.329	0.426	0.245	0.82	0.49	0.82	0.83	0.49	0.83
21	37	8.86	14.8	5.94	41.64	30.16	43.94	0.343	0.415	0.243	0.85	0.85	0.85	0.85	0.85	0.85	1076.3	1075.97	41.19	30.34	44.09	0.339	0.417	0.243	1.11	0.9	1.11	0.55	0.9	0.55
22	38	5.8	11.18	5.02	47.44	30.42	40.3	0.337	0.416	0.247	0.85	0.85	0.85	0.85	0.85	0.85	817.55	817.6	47.5	30.23	40.67	0.337	0.413	0.25	0.91	0.87	0.91	0.79	0.87	0.79
23	39	5.98	12.26	6.66	52.5	31.95	35.99	0.332	0.414	0.254	0.85	0.85	0.85	0.85	0.85	0.85	945.39	945.67	52.75	31.89	35.93	0.334	0.413	0.253	0.7	0.97	0.7	1.05	0.97	1.05
24	40	5.02	14.96	5.26	60.78	26.91	43.89	0.325	0.429	0.246	0.85	0.85	0.85	0.85	0.85	0.85	938.61	938.1	60.01	26.78	44.91	0.321	0.427	0.252	0.74	0.49	0.74	0.94	0.49	0.94
25	41	6.88	12.14	7.62	52.31	36.16	36.03	0.335	0.409	0.256	0.85	0.85	0.85	0.85	0.85	0.85	1073.45	1073.94	53.05	36.07	35.56	0.34	0.408	0.252	0.69	1.19	0.69	1.04	1.19	1.04
26	42	7	12.64	4.32	47.77	32.38	54.09	0.342	0.419	0.239	0.85	0.85	0.85	0.85	0.85	0.85	977.42	977.54	47.74	32.76	53.07	0.342	0.424	0.235	1.15	0.75	1.15	0.5	0.75	0.5
27	43	7.6	13.24	6.98	52.46	36.66	42.37	0.338	0.411	0.251	0.85	0.85	0.85	0.85	0.85	0.85	1179.84	1180.42	52.56	36.43	42.79	0.338	0.409	0.253	0.86	1.05	0.86	0.83	1.05	0.83
28	44	8.32	14.56	7.1	52.93	36.85	45.51	0.339	0.413	0.249	0.85	0.85	0.85	0.85	0.85	0.85	1300.05	1300.41	52.86	36.64	46.07	0.338	0.41	0.252	0.92	0.99	0.92	0.76	0.99	0.76
AVE:	30.5	6.74	13.09	5.69	38.45	24.17	33.29	0.337	0.417	0.246	0.85	0.85	0.85	0.85	0.85	0.85	758.21	758.19	38.45	24.17	33.29	0.336	0.418	0.246	0.92	0.85	0.92	0.77	0.85	0.77

XVII. Table of Results: check that second order partial derivatives of the estimated Normalized Quadratic cost function for a given level of output agree with the second order partial derivatives of the CES cost function. If all of the eigenvalues of ∇²c*(w) and ∇²c(w) are nonpositive, the cost functions c*(w*) and c(w*) are concave in factor prices.

Normalized Quadratic Cost Function CES Production Function: Returns to Scale = 1, Elasticity of Substitution = 0.85 Normalized Quadratic: Returns to Scale = 1; Base Prices: wL* = 7, wK* = 13, wM* = 6
					— — CES Cost Data — —									— — Normalized Quadratic Cost Data — —
		Factor Prices			∇²c*(w)						Eigenvalues			∇²c(w)						Eigenvalues
obs #	q	wL	wK	wM	c_LL	c_LK	c_LM	c_KK	c_KM	c_MM	e₁	e₂	e₃	c_LL	c_LK	c_LM	c_KK	c_KM	c_MM	e₁	e₂	e₃
1	17	5.72	12.16	6.62	-0.137	0.04	0.045	-0.033	0.026	-0.087	-0.167	-0.09	-0	-0.125	0.032	0.05	-0.033	0.034	-0.106	-0.167	-0.098	-0
2	18	8.18	14.84	5.18	-0.078	0.027	0.045	-0.026	0.03	-0.158	-0.18	-0.082	-0	-0.092	0.037	0.04	-0.027	0.019	-0.116	-0.146	-0.089	-0
3	19	7.28	11.32	5.04	-0.085	0.034	0.046	-0.038	0.035	-0.145	-0.171	-0.096	0	-0.105	0.044	0.054	-0.038	0.023	-0.13	-0.174	-0.1	0
4	20	7.36	11.08	4.1	-0.079	0.033	0.052	-0.037	0.041	-0.204	-0.226	-0.093	-0	-0.104	0.049	0.054	-0.04	0.019	-0.148	-0.185	-0.106	-0
5	21	6.64	11.36	7.32	-0.107	0.038	0.038	-0.04	0.027	-0.075	-0.136	-0.086	-0	-0.114	0.031	0.054	-0.039	0.032	-0.098	-0.161	-0.09	-0
6	22	7.06	13.08	4.68	-0.092	0.032	0.051	-0.029	0.034	-0.171	-0.198	-0.095	-0	-0.107	0.041	0.045	-0.03	0.022	-0.13	-0.166	-0.101	0
7	23	8.9	14.8	7.34	-0.074	0.028	0.034	-0.029	0.025	-0.09	-0.117	-0.076	0	-0.085	0.031	0.042	-0.029	0.021	-0.093	-0.131	-0.076	-0
8	24	5.2	14.48	4.18	-0.154	0.035	0.068	-0.022	0.032	-0.196	-0.247	-0.125	0	-0.132	0.037	0.035	-0.021	0.027	-0.137	-0.17	-0.121	-0
9	25	5.04	14.1	4.56	-0.163	0.037	0.065	-0.023	0.03	-0.166	-0.23	-0.123	-0	-0.135	0.036	0.037	-0.022	0.03	-0.133	-0.171	-0.119	-0
10	26	5.88	12.16	4.98	-0.122	0.037	0.053	-0.031	0.032	-0.142	-0.187	-0.109	-0	-0.125	0.04	0.049	-0.032	0.03	-0.13	-0.177	-0.109	0
11	27	7.02	12.36	6.72	-0.099	0.035	0.039	-0.035	0.027	-0.091	-0.135	-0.09	0	-0.107	0.034	0.05	-0.034	0.028	-0.104	-0.156	-0.09	0
12	28	8.5	13.8	4.82	-0.07	0.028	0.045	-0.029	0.034	-0.176	-0.196	-0.079	0	-0.09	0.04	0.043	-0.031	0.017	-0.123	-0.154	-0.09	0
13	29	5.94	14.06	5.5	-0.13	0.035	0.052	-0.026	0.028	-0.127	-0.181	-0.102	-0	-0.119	0.034	0.041	-0.026	0.028	-0.117	-0.159	-0.102	0
14	30	7	13	6	-0.099	0.033	0.043	-0.031	0.029	-0.112	-0.149	-0.093	-0	-0.107	0.036	0.047	-0.031	0.026	-0.112	-0.156	-0.093	0
15	31	5.26	14.02	6.48	-0.165	0.039	0.05	-0.026	0.024	-0.093	-0.194	-0.09	0	-0.128	0.029	0.041	-0.026	0.033	-0.105	-0.159	-0.1	0
16	32	8.08	13.96	4.68	-0.076	0.028	0.048	-0.028	0.034	-0.183	-0.204	-0.083	-0	-0.094	0.04	0.042	-0.029	0.018	-0.125	-0.155	-0.093	0
17	33	6.2	11.22	7.72	-0.121	0.041	0.038	-0.04	0.025	-0.067	-0.147	-0.081	-0	-0.12	0.029	0.055	-0.039	0.034	-0.094	-0.163	-0.09	-0
18	34	6.64	12.32	4.6	-0.099	0.034	0.053	-0.031	0.035	-0.169	-0.199	-0.101	0	-0.113	0.043	0.048	-0.032	0.024	-0.135	-0.174	-0.106	0
19	35	6.44	12.6	5.3	-0.109	0.035	0.049	-0.031	0.031	-0.133	-0.171	-0.101	0	-0.115	0.039	0.048	-0.031	0.027	-0.123	-0.167	-0.102	-0
20	36	5	14	4.56	-0.165	0.038	0.065	-0.023	0.031	-0.165	-0.231	-0.123	0	-0.136	0.036	0.038	-0.023	0.03	-0.133	-0.173	-0.119	0
21	37	8.86	14.8	5.94	-0.071	0.027	0.039	-0.027	0.028	-0.129	-0.15	-0.077	0	-0.086	0.035	0.041	-0.028	0.018	-0.107	-0.139	-0.082	-0
22	38	5.8	11.18	5.02	-0.121	0.039	0.052	-0.036	0.034	-0.135	-0.181	-0.111	0	-0.128	0.042	0.054	-0.036	0.031	-0.132	-0.185	-0.111	0
23	39	5.98	12.26	6.66	-0.128	0.039	0.044	-0.033	0.027	-0.088	-0.159	-0.09	0	-0.121	0.032	0.05	-0.033	0.033	-0.105	-0.164	-0.096	-0
24	40	5.02	14.96	5.26	-0.174	0.037	0.06	-0.022	0.027	-0.134	-0.218	-0.111	0	-0.131	0.032	0.035	-0.021	0.03	-0.12	-0.161	-0.111	-0
25	41	6.88	12.14	7.62	-0.105	0.037	0.036	-0.036	0.025	-0.073	-0.132	-0.082	-0	-0.109	0.03	0.051	-0.036	0.03	-0.094	-0.153	-0.086	-0
26	42	7	12.64	4.32	-0.091	0.032	0.054	-0.03	0.036	-0.193	-0.218	-0.096	0	-0.108	0.044	0.047	-0.032	0.021	-0.138	-0.172	-0.105	0
27	43	7.6	13.24	6.98	-0.09	0.032	0.037	-0.032	0.026	-0.09	-0.128	-0.085	-0	-0.099	0.032	0.047	-0.032	0.026	-0.099	-0.146	-0.084	0
28	44	8.32	14.56	7.1	-0.081	0.029	0.036	-0.029	0.025	-0.093	-0.123	-0.08	-0	-0.091	0.031	0.042	-0.029	0.022	-0.096	-0.136	-0.079	-0
AVE:	30.5	6.74	13.09	5.69	-0.11	0.034	0.048	-0.03	0.03	-0.132	-0.178	-0.095	-0	-0.112	0.036	0.046	-0.031	0.026	-0.117	-0.161	-0.098	-0

XVIII. Table of Results: check that factor demand price elasticities of the estimated Normalized Quadratic cost function for a given level of output and factor prices agree with the factor demand price elasticities of the CES cost function.

Normalized Quadratic Cost Function CES Production Function: Returns to Scale = 1, Elasticity of Substitution = 0.85 Normalized Quadratic: Returns to Scale = 1; Base Prices: wL* = 7, wK* = 13, wM* = 6
					— — CES Cost Data — —						— — Normalized Quadratic Cost Data — —
		Factor Prices			Factor Demand Price Elasticities						Factor Demand Price Elasticities
obs #	q	wL	wK	wM	ε_L,wL	ε_L,wK	ε_L,wM	ε_K,wK	ε_K,wM	ε_M,wM	ε_L,wL	ε_L,wK	ε_L,wM	ε_K,wK	ε_K,wM	ε_M,wM
1	17	5.72	12.16	6.62	-0.569	0.353	0.216	-0.497	0.216	-0.634	-0.52	0.28	0.24	-0.295	0.163	-0.508
2	18	8.18	14.84	5.18	-0.56	0.356	0.204	-0.494	0.204	-0.646	-0.66	0.48	0.18	-0.348	0.084	-0.525
3	19	7.28	11.32	5.04	-0.558	0.35	0.208	-0.5	0.208	-0.642	-0.705	0.454	0.251	-0.399	0.108	-0.604
4	20	7.36	11.08	4.1	-0.555	0.352	0.203	-0.498	0.203	-0.647	-0.754	0.535	0.218	-0.432	0.076	-0.598
5	21	6.64	11.36	7.32	-0.564	0.347	0.218	-0.503	0.218	-0.632	-0.589	0.278	0.31	-0.344	0.181	-0.562
6	22	7.06	13.08	4.68	-0.56	0.356	0.205	-0.494	0.205	-0.645	-0.645	0.464	0.181	-0.339	0.089	-0.522
7	23	8.9	14.8	7.34	-0.561	0.35	0.211	-0.5	0.211	-0.639	-0.65	0.388	0.262	-0.365	0.132	-0.583
8	24	5.2	14.48	4.18	-0.57	0.366	0.204	-0.484	0.204	-0.646	-0.484	0.38	0.103	-0.216	0.08	-0.404
9	25	5.04	14.1	4.56	-0.571	0.365	0.206	-0.485	0.206	-0.644	-0.471	0.353	0.118	-0.22	0.093	-0.419
10	26	5.88	12.16	4.98	-0.565	0.356	0.209	-0.494	0.209	-0.641	-0.573	0.383	0.191	-0.3	0.115	-0.506
11	27	7.02	12.36	6.72	-0.563	0.349	0.214	-0.501	0.214	-0.636	-0.606	0.335	0.271	-0.343	0.152	-0.563
12	28	8.5	13.8	4.82	-0.556	0.354	0.203	-0.496	0.203	-0.647	-0.719	0.524	0.196	-0.398	0.075	-0.561
13	29	5.94	14.06	5.5	-0.568	0.359	0.209	-0.491	0.209	-0.641	-0.519	0.354	0.166	-0.264	0.114	-0.471
14	30	7	13	6	-0.563	0.353	0.211	-0.497	0.211	-0.639	-0.607	0.378	0.229	-0.331	0.128	-0.544
15	31	5.26	14.02	6.48	-0.573	0.359	0.214	-0.491	0.214	-0.636	-0.451	0.272	0.18	-0.246	0.144	-0.457
16	32	8.08	13.96	4.68	-0.558	0.355	0.202	-0.495	0.202	-0.648	-0.691	0.512	0.179	-0.371	0.075	-0.534
17	33	6.2	11.22	7.72	-0.567	0.347	0.22	-0.503	0.22	-0.63	-0.55	0.238	0.312	-0.328	0.196	-0.536
18	34	6.64	12.32	4.6	-0.561	0.355	0.206	-0.495	0.206	-0.644	-0.639	0.45	0.189	-0.337	0.095	-0.527
19	35	6.44	12.6	5.3	-0.564	0.355	0.209	-0.495	0.209	-0.641	-0.596	0.392	0.204	-0.317	0.116	-0.523
20	36	5	14	4.56	-0.571	0.365	0.207	-0.485	0.207	-0.643	-0.47	0.351	0.119	-0.22	0.094	-0.42
21	37	8.86	14.8	5.94	-0.559	0.352	0.206	-0.498	0.206	-0.644	-0.683	0.465	0.218	-0.376	0.098	-0.569
22	38	5.8	11.18	5.02	-0.564	0.354	0.21	-0.496	0.21	-0.64	-0.593	0.375	0.218	-0.32	0.126	-0.532
23	39	5.98	12.26	6.66	-0.568	0.352	0.216	-0.498	0.216	-0.634	-0.536	0.29	0.245	-0.303	0.161	-0.517
24	40	5.02	14.96	5.26	-0.574	0.365	0.209	-0.485	0.209	-0.641	-0.439	0.315	0.124	-0.212	0.106	-0.419
25	41	6.88	12.14	7.62	-0.565	0.348	0.217	-0.502	0.217	-0.633	-0.579	0.28	0.299	-0.337	0.178	-0.554
26	42	7	12.64	4.32	-0.559	0.356	0.203	-0.494	0.203	-0.647	-0.665	0.488	0.177	-0.351	0.081	-0.523
27	43	7.6	13.24	6.98	-0.563	0.35	0.213	-0.5	0.213	-0.637	-0.616	0.35	0.266	-0.347	0.146	-0.567
28	44	8.32	14.56	7.1	-0.562	0.351	0.211	-0.499	0.211	-0.639	-0.627	0.378	0.25	-0.349	0.133	-0.565
AVE:	30.5	6.74	13.09	5.69	-0.564	0.355	0.209	-0.495	0.209	-0.641	-0.594	0.384	0.211	-0.322	0.119	-0.522

Mathematical Notes

Normalized Quadratic Cost Function

c(w) = c^T * w + (1/2) * w^T * D * w / (1^T * w),

a linear function in its parameters, c, and D.

Normalized Quadratic Unit Cost Function:

c(wL,wK,wM) = cL * wL + cK * wK + cM * wM

+ (1/2) * [dLL * (wL*wL) + dLK*(wL*wK) + dLM * (wL*wM)
+ dKL * (wK*wL) + dKK*(wK*wK) + dKM * (wK*wM)
+ dML * (wM*wL) + dMK*(wM*wK) + dMM * (wM*wM)] * (wL + wK + wM)^^-1

Returns to Scale Function:

h(q) = q^^(1/nu1)

a continuous, increasing function of q (q >= 1), with h(0) = 0 and h(1) = 1,

Normalized Quadratic (Total) Cost Function:

C(q;wL,wK,wM) = h(q) * c(wL,wK,wM)

Define the variable w as:

w = 1 / (wL + wM + wK) = 1 / (1^T * w).

Unit Factor Demand Functions:

l(wL, wK, wM) =	cL + (dLL * wL + dLK * wK + dLM * wM) * w	- (1/2) * [dLL * wL * wL + dLK * wvL * wvK + dLM * wL * wM + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w
k(wL, wK, wM) =	cK + (dKL * wL + dKK * wK + dKM * wM) * w	- (1/2) * [dLL * wL * wL + dLK * wL * wK + dLM * wL * wM + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w
m(wL, wK, wM) =	cM + (dML * wL + dMK * wK + dMM * wM) * w	- (1/2) * [dLL * wL * wL + dLK * wL * wK + dLM * wL * wM) * w + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w

Unit Factor Demand Functions' Partial Derivatives:

∂l/∂wL =	dLL * w	- (dLL * wL + dLK * wK + dLM * wM + dLL * wL + dKL * wK + dML * wM) * w * w	+ [dLL * wL * wL + dLK * wL * wK + dLM * wL * wM + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w * w
∂l/∂wK =	dLK * w	- (dLL * wL + dLK * wK + dLM * wM + dLK * wL + dKK * wK + dMK * wM) * w * w	+ [dLL * wL * wL + dLK * wL * wK + dLM * wL * wM + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w * w
∂l/∂wM =	dLM * w	- (dLL * wL + dLK * wK + dLM * wM + dLM * wL + dKM * wK + dMM * wM) * w * w	+ [dLL * wL * wL + dLK * wL * wK + dLM * wL * wM + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w * w
∂k/∂wL =	dKL * w	- (dKL * wL + dKK * wK + dKM * wM + dLL * wL + dKL * wK + dML * wM) * w * w	+ [dLL * wL * wL + dLK * wL * wK + dLM * wL * wM + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w * w
∂k/∂wK =	dKK * w	- (dKL * wL + dKK * wK + dKM * wM + dLK * wL + dKK * wK + dMK * wM) * w * w	+ [dLL * wL * wL + dLK * wL * wK + dLM * wL * wM + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w * w
∂k/∂wM =	dKM * w	- (dKL * wL + dKK * wK + dKM * wM + dLM * wL + dKM * wK + dMM * wM) * w * w	+ [dLL * wL * wL + dLK * wL * wK + dLM * wL * wM + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w * w
∂m/∂wL =	dML * w	- (dML * wL + dMK * wK + dMM * wM + dLL * wL + dKL * wK + dML * wM) * w * w	+ [dLL * wL * wL + dLK * wL * wK + dLM * wL * wM + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w * w
∂m/∂wK =	dMK * w	- (dML * wL + dMK * wK + dMM * wM + dLK * wL + dKK * wK + dMK * wM) * w * w	+ [dLL * wL * wL + dLK * wL * wK + dLM * wL * wM + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w * w
∂m/∂wM =	dMM * w	- (dML * wL + dMK * wK + dMM * wM + dLM * wL + dKM * wK + dMM * wM) * w * w	+ [dLL * wL * wL + dLK * wL * wK + dLM * wL * wM + dKL * wK * wL + dKK * wK * wK + dKM * wK * wM + dML * wM * wL + dMK * wM * wK + dMM * wM * wM] * w * w * w

Normalized Quadratic Cost Function's Uzawa Elasticities:

u_LK = c(wL,wK,wM) * ∂l/∂wK / (l * k);
u_LM = c(wL,wK,wM) * ∂l/∂wM / (l * m);
u_KL = c(wL,wK,wM) * ∂k/∂wL / (k * l);
u_KM = c(wL,wK,wM) * ∂k/∂wM / (k * m);
u_ML = c(wL,wK,wM) * ∂m/∂wL / (m * l);
u_MK = c(wL,wK,wM) * ∂m/∂wK / (m * k);

Reparameterized Normalized Quadratic Unit Cost Function:

ç(w) = b^T * w + (1/2) * w^T * (-B) * w / (1^T * w),

is a nonlinear function in its parameters, b = [bL, bK, bM]^T, and (-B), i.e. in its free parameters bL, bK, bM, bLL, bKL, and bKK.

Reparameterized Normalized Quadratic Unit Cost Function:

ç(wL,wK,wM) =	bL * wL + bK * wK + bM * wM
	+ .5 * (-wLbLL^2-wKbLLbKL-wMbLL(-bLLuL-bKLuK)/uM) wL * w + .5 * (-wLbLLbKL+wK(-bKL^2-bKK^2)+wM(-bKL(-bLLuL-bKLuK)/uM+bKK^2uK/uM))* wK * w + .5 * (-wLbLL(-bLLuL-bKLuK)/uM+wK(-bKL(-bLLuL-bKLuK)/uM+bKK^2uK/uM)+wM(-(-bLLuL-bKLuK)^2/uM^2-bKK^2uK^2/uM^2)) wM * w);

where u = w* / (wL* + wK* + wM*) = [uL, uK, uM]^T = [vL*, vK*, vM*] (redefined), the vector of base prices.

Shephard's lemma provides that the gradient vector of the unit cost function is the vector of unit input demand functions:

∇ç(w) = [l(w), k(w), m(w)]^T =b + (-B) * w / ( 1^T * w) - (1/2) * w^T * (-B) * w * 1 / (1^T * w)².

Reparameterized Normalized Quadratic Unit Factor Demand Functions:

l(wL,wK,wM) =	bL + (-bLL^2wL - bLLbKLwK - bLL(-bLLuL-bKLuK)/uMwM) w
	- .5 * (-wLbLL^2-wKbLLbKL-wMbLL(-bLLuL-bKLuK)/uM) wL * w * w - .5 * (-wLbLLbKL+wK(-bKL^2-bKK^2)+wM(-bKL(-bLLuL-bKLuK)/uM+bKK^2uK/uM))* wK * w * w - .5 * (-wLbLL(-bLLuL-bKLuK)/uM+wK(-bKL(-bLLuL-bKLuK)/uM+bKK^2uK/uM)+wM(-(-bLLuL-bKLuK)^2/uM^2-bKK^2uK^2/uM^2)) wM * w * w );
k(wL,wK,wM) =	bK + (-bLLbKLwL + (-bKL^2-bKK^2)wK + (-bKL(-bLLuL-bKLuK)/uM+bKK^2uK/uM) wM) * w
	- .5 * (-wLbLL^2-wKbLLbKL-wMbLL(-bLLuL-bKLuK)/uM) wL * w * w - .5 * (-wLbLLbKL+wK(-bKL^2-bKK^2)+wM(-bKL(-bLLuL-bKLuK)/uM+bKK^2uK/uM))* wK * w * w - .5 * (-wLbLL(-bLLuL-bKLuK)/uM+wK(-bKL(-bLLuL-bKLuK)/uM+bKK^2uK/uM)+wM(-(-bLLuL-bKLuK)^2/uM^2-bKK^2uK^2/uM^2)) wM * w * w );
m(wL,wK,wM) =	bM + (-bLL(-bLLuL-bKLuK)/uMwL + (-bKL(-bLLuL-bKLuK)/uM+bKK^2uK/uM)wK + (-(-bLLuL-bKLuK)^2/uM^2-bKK^2uK^2/uM^2)wM) w
	- .5 * (-wLbLL^2-wKbLLbKL-wMbLL(-bLLuL-bKLuK)/uM) wL * w * w - .5 * (-wLbLLbKL+wK(-bKL^2-bKK^2)+wM(-bKL(-bLLuL-bKLuK)/uM+bKK^2uK/uM))* wK * w * w - .5 * (-wLbLL(-bLLuL-bKLuK)/uM+wK(-bKL(-bLLuL-bKLuK)/uM+bKK^2uK/uM)+wM(-(-bLLuL-bKLuK)^2/uM^2-bKK^2uK^2/uM^2)) wM * w * w );

Reparameterized Unit Labour Demand Function Partial Derivatives:

∂l/∂bL =	1
∂l/∂bK =	0
∂l/∂bM =	0

∂l/∂bLL =	(-2wLbLL-wKbKL-wM(-bLLuL-bKLuK)/uM+wMbLLuL/uM) * w
	-((-wLbLL-1/2wKbKL-1/2wM(-bLLuL-bKLuK)/uM+1/2wMbLLuL/uM)wL + (-1/2wLbKL+1/2wMbKLuL/uM)wK + (-1/2wL(-bLLuL-bKLuK)/uM+1/2wLbLLuL/uM+1/2wKbKLuL/uM+wM(-bLLuL-bKLuK)/uM^2uL)wM)* w * w
∂l/∂bKL =	(-wKbLL+wMbLLuK/uM) w
	-((-1/2wKbLL+1/2wMbLLuK/uM)wL + (-1/2wLbLL-wKbKL+1/2wM(-(-bLLuL-bKLuK)/uM+bKLuK/uM))wK + (1/2wLbLLuK/uM+1/2wK(-(-bLLuL-bKLuK)/uM+bKLuK/uM)+wM(-bLLuL-bKLuK)/uM^2uK)wM)* w * w
∂l/∂bKK =	-((-wKbKK+wMbKKuK/uM)wK + (wKbKKuK/uM-wMbKKuK^2/uM^2)wM) w * w

Reparameterized Unit Capital Demand Function Partial Derivatives:

∂k/∂bL =	0
∂k/∂bK =	1
∂k/∂bM =	0

∂k/∂bLL =	(-wLbKL+wMbKLuL/uM) w
	-((-wLbLL-1/2wKbKL-1/2wM(-bLLuL-bKLuK)/uM+1/2wMbLLuL/uM)wL + (-1/2wLbKL+1/2wMbKLuL/uM)wK + (-1/2wL(-bLLuL-bKLuK)/uM+1/2wLbLLuL/uM+1/2wKbKLuL/uM+wM(-bLLuL-bKLuK)/uM^2uL)wM)* w * w
∂k/∂bKL =	(-wLbLL-2wKbKL+wM(-(-bLLuL-bKLuK)/uM+bKLuK/uM)) w
	-((-1/2wKbLL+1/2wMbLLuK/uM)wL + (-1/2wLbLL-wKbKL+1/2wM(-(-bLLuL-bKLuK)/uM+bKLuK/uM))wK + (1/2wLbLLuK/uM+1/2wK(-(-bLLuL-bKLuK)/uM+bKLuK/uM)+wM(-bLLuL-bKLuK)/uM^2uK)wM)* w * w
∂k/∂bKK =	(-2wKbKK+2wMbKKuK/uM) w
	-((-wKbKK+wMbKKuK/uM)wK + (wKbKKuK/uM-wMbKKuK^2/uM^2)wM) w * w

Reparameterized Unit Materials and Supplies Demand Function Partial Derivatives:

∂m/∂bL =	0
∂m/∂bK =	0
∂k/∂bM =	1

∂m/∂bLL =	(-wL(-bLLuL-bKLuK)/uM+wLbLLuL/uM+wKbKLuL/uM+2wM(-bLLuL-bKLuK)/uM^2uL) * w
	-((-wLbLL-1/2wKbKL-1/2wM(-bLLuL-bKLuK)/uM+1/2wMbLLuL/uM)wL + (-1/2wLbKL+1/2wMbKLuL/uM)wK + (-1/2wL(-bLLuL-bKLuK)/uM+1/2wLbLLuL/uM+1/2wKbKLuL/uM+wM(-bLLuL-bKLuK)/uM^2uL)wM)* w * w
∂m/∂bKL =	(wLbLLuK/uM+wK(-(-bLLuL-bKLuK)/uM+bKLuK/uM)+2wM(-bLLuL-bKLuK)/uM^2uK) w
	-((-1/2wKbLL+1/2wMbLLuK/uM)wL + (-1/2wLbLL-wKbKL+1/2wM(-(-bLLuL-bKLuK)/uM+bKLuK/uM))wK + (1/2wLbLLuK/uM+1/2wK(-(-bLLuL-bKLuK)/uM+bKLuK/uM)+wM(-bLLuL-bKLuK)/uM^2uK)wM)* w * w
∂m/∂bKK =	(2wKbKKuK/uM-2wMbKKuK^2/uM^2) * w
	-((-wKbKK+wMbKKuK/uM)wK + (wKbKKuK/uM-wMbKKuK^2/uM^2)wM) w * w

CES Production/Cost Functions:

CES Production Function:

q = A * [alpha * (L^^-rho) + beta * (K^^-rho) + gamma *(M^^-rho)]^^(-nu/rho) = f(L,K,M).

where L = labour, K = capital, M = materials and supplies, and q = product. The parameter nu is a measure of the economies of scale, while the parameter rho yields the elasticity of substitution:

sigma = 1/(1 + rho).

The efficiency parameter, A, changes output proportionally for changes in factor inputs, while the distribution parameters, alpha, beta, and gamma, determine the relative shares of the factors in the total cost of producing levels of outputs.

CES Cost Function:

C(q;wL,wK,wM) = h(q) * c(wL,wK,wM)

where the returns to scale function is:

h(q) = (q/A)^^1/nu

a continuous, increasing function of q (q >= 1), with h(0) = 0 and h(1)=1.

and the unit cost function is:

c(wL,wK,wM) = [alpha^^(1/(1+rho)) * wL^^{(rho/(1+rho))} + beta^^(1/(1+rho)) * wK^^{(rho/(1+rho))} + gamma^^(1/(1+rho)) * wM^^{(rho/(1+rho))}]^^{((1+rho)/rho)}

CES Unit Demand Functions:

l(wL,wK,wM) =	[(alpha / wL) * c(wL,wK,wM)]^^(1/(1+rho))
k(wL,wK,wM) =	[(beta / wK) * c(wL,wK,wM)]^^(1/(1+rho))
m(wL,wK,wM) =	[(gamma / wM) * c(wL,wK,wM)]^^(1/(1+rho))

Properties of the Unit CES Cost Function, c(wL,wK,wM).

a. c(wL,wK,wM) is linear homogeneous in factor prices.

c(t*wL,t*wK,t*wM) = [alpha^^(1/(1+rho)) * (t*wL)^^{(rho/(1+rho))} + beta^^(1/(1+rho)) * (t*wK)^^{(rho/(1+rho))} + gamma^^(1/(1+rho)) * (t*wM)^^{(rho/(1+rho))}]^^{((1+rho)/rho)},
= [t^^{(rho/(1+rho))} * (alpha^^(1/(1+rho)) * wL^^{(rho/(1+rho))} + beta^^(1/(1+rho)) * wK^^{(rho/(1+rho))} + gamma^^(1/(1+rho)) * wM^^{(rho/(1+rho))})]^^{((1+rho)/rho)},
= t * [alpha^^(1/(1+rho)) * wL^^{(rho/(1+rho))} + beta^^(1/(1+rho)) * wK^^{(rho/(1+rho))} + gamma^^(1/(1+rho)) * wM^^{(rho/(1+rho))})]^^{((1+rho)/rho)}, →
c(t*wL,t*wK,t*wM) = t * c(wL,wK,wM)

b. The matrix ∇²c of second order partial derivatives of the unit cost function c(wL,wK,wM) is symmetric.

∂(∂c(wL,wK,wM)/∂wL)/∂wK	= ∂([(alpha / wL) * c(wL,wK,wM)]^^(1/(1+rho)))/∂wK = (1/(1+rho)) * (alpha/wL) * ∂c(wL,wK,wM)/∂wK * [(alpha / wL) * c(wL,wK,wM)]^^(1/(1+rho)) / [(alpha / wL) * c(wL,wK,wM)] = (1/(1+rho)) * ∂c(wL,wK,wM)/∂wK * ∂c(wL,wK,wM)/∂wL / c(wL,wK,wM) = ∂(∂c(wL,wK,wM)/∂wK)/∂wL
∂(∂c(wL,wK,wM)/∂wL)/∂wM	= ∂([(alpha / wL) * c(wL,wK,wM)]^^(1/(1+rho)))/∂wM = (1/(1+rho)) * (alpha/wL) * ∂c(wL,wK,wM)/∂wM * [(alpha / wL) * c(wL,wK,wM)]^^(1/(1+rho)) / [(alpha / wL) * c(wL,wK,wM)] = (1/(1+rho)) * ∂c(wL,wK,wM)/∂wM * ∂c(wL,wK,wM)/∂wL / c(wL,wK,wM) = ∂(∂c(wL,wK,wM)/∂wM)/∂wL
etc.

c. c(wL,wK,wM) is concave in factor prices.

d. The CES production function is called homothetic, because the CES cost function can be separated (factored) into a function of output, q, times a function of input prices, wL, wK, and wM.