TaskBasedProduction is a Julia package that provides functions for calculating unit labor demands, marginal products of labor, output values, assignment thresholds, and elasticities of substitution / complementarity for the model of production in Supply, Demand, Institutions, and Firms: A Theory of Labor Market Sorting and the Wage Distribution by Daniel Haanwinckel (NBER Working Paper No. 31318). The package includes utilities both for the Exponential-gamma parameterization described in the paper as well as for arbitrary functional forms for blueprints and efficiency functions. This package was developed by Daniel Haanwinckel and Luca Lorenzini.
using Pkg
Pkg.add("TaskBasedProduction")using TaskBasedProduction
θ = 1.0
κ = 0.5
z = 1.2
αVec = [0.1, 0.2, 0.3]
labor_input=[0.5; 0.04; 0.19;;]
initial_guess=getStartGuess_xT(θ, κ, z, αVec)
q, xT = prodFun(labor_input, θ, κ, z, αVec; initial_guess=initial_guess)
# note this is the same as not including initial guess since it will computed within the function
q, xT = prodFun(labor_input, θ, κ, z, αVec)
println("Quantity Produced: ", q)
println("Task Thresholds: ", xT)
# Call unitInputDemand and print the output
labor_input2 = unitInputDemand( xT, q, θ, κ, z, αVec)
println("Labor Demand: ", labor_input2)
# Call margProdLabor with labor demand and print the output
MPL= margProdLabor(labor_input, θ, κ, z, αVec)
println("Marginal Products of Labor (with labor demand): ",MPL)
# Call elasticity_substitution with labor demand, MPL, xT and parameters of the gamma function
ϵ_sub, ϵ_compl=elasticitySubComp(labor_input, θ, κ, z, αVec, MPL, xT)
println("Allen partial elasticity of substitution:", ϵ_sub)
println("Hicks partial elasticity of substitution:", ϵ_compl)
# Call elasticity_substitution with labor demand, MPL, xT and parameters of the gamma function
ϵ_sub, ϵ_compl=elasticitySubComp(labor_input, θ, κ, z, αVec, MPL, xT, q)
println("Allen partial elasticity of substitution:", ϵ_sub)
println("Hicks partial elasticity of substitution:", ϵ_compl)
## GENERAL PARAMETERIZATION OF FUNCTIONS
# Define the density function b_g(x)
using SpecialFunctions # Note: this package is needed only to compare the output of the general parameterization with the output of the Gamma parameterization. It is used in the next line to obtain the PDF of a Gamma distribution.
using LeastSquaresOptim # Package needed for optimization
b_g(x) = (x^(κ - 1) * exp(-x / θ)) / (θ^κ * gamma(κ)) # Gamma PDF with the same parameterization as above for comparability
e_h1(x)=exp(0.1*x)
e_h2(x)=exp(0.2*x)
e_h3(x)=exp(0.3*x)
e_h = [e_h1, e_h2, e_h3] # Example e_h functions
q_gen, xT_gen= prodFunGeneral(labor_input,z,b_g, e_h)
labor_input_general = unitInputDemandGeneral(xT_gen, q_gen, z, b_g, e_h)
println("Labor Demand: ", labor_input_general)
isapprox(labor_input, labor_input_general, atol=1e-6)
println("Quantity Produced: ", q_gen)
println("Task Thresholds: ", xT_gen)
MPL_gen=margProdLaborGeneral(labor_input_general, z, b_g, e_h, xT_gen, q_gen)
ϵ_sub_gen, ϵ_compl_gen=elasticitySubCompGeneral(nothing, z, b_g, e_h, MPL_gen, xT_gen, q_gen)This package provides models and functions for analyzing labor markets, task thresholds, and elasticity in both competitive and monopsonistic settings. It includes functions for both the Exponential-gamma parameterization from the paper (Section 4.4 of the paper) and for arbitrary efficiency functions and blueprints. The first examples demonstrate the Exponential-gamma parameterization, with general examples for arbitrary efficiency functions and blueprints provided later.
θ = 1.0
κ = 0.5
z = 1.2
αVec = [0.1, 0.2, 0.3] In a competitive labor market, wages are taken as given and must equate the marginal product of labor. Optimality requires that marginal product ratios equal wage ratios, a relation that is used to obtain the task thresholds xT. Once the task thresholds are known, the function unitInputDemand is used to obtain the labor per output, setting q = 1.
q = 1
wage = [0.1; 0.2; 0.7]
# Compute thresholds xT suing the functional form
diff_alpha = diff(αVec)
log_wage_ratios = log.(wage[2:end] ./ wage[1:end-1])
xT = (1 ./ diff_alpha) .* log_wage_ratios
# Calculate labor unit input requirements
labor_input_1 = unitInputDemand(xT, q, θ, κ, z, αVec)If labor inputs per each type are known, they can be given to the function prodFun or prodFunGeneral to compute the task thresholds and the total output produced.
labor_input=[0.5; 0.04; 0.19;;]
q, xT = prodFun(labor_input, θ, κ, z, αVec)Use the function elasticitySubComp to obtain the elasticities of substitution and complementarity. Precompiled values for marginal products (MPL), task thresholds (xT), and total output (q) can be used for efficiency, but they are computed within the function if not provided.
# Call elasticity_substitution with labor demand given
ϵ_sub, ϵ_compl=elasticitySubComp(labor_input, θ, κ, z, αVec)# Call elasticity_substitution with labor demand given
ϵ_sub, ϵ_compl=elasticitySubComp(nothing, θ, κ, z, αVec, MPL, xT, q)First, define the parameter governing the elasticity of labor supply:
β=4Then define general equilibrium objects (total labor supply and inclusive value) and the price of the good sold by the firm:
# Define general equilibrium objects
L=[1 ; 1 ; 1] # Total labor force
p=1 # Price for the good
w_inclusive=[0.4; 0.9; 2] # Inclusive value of wages Define an objective function that will be minimized to find the solution in total output (q) and task thresholds (xT). This function takes a guess for q and xT as input, computes the labor input required by the guess, and the marginal product of labor. It then computes the wage as a constant markdown applied to the marginal product of labor and the implied labor supply given the firm wage. Finally, the function returns an error that is the discrepancy from the labor supply to the labor required implied by the guess xT and q.
# Define the objective function for optimization derived from the firm problem
function objective_to_minimize(initial_guess)
# Compute q and xT from the initial guess
q = exp(initial_guess[1])
xT = cumsum(exp.(initial_guess[2:end]))
# Calculate labor input demand
labor_input = unitInputDemand(xT, q, θ, κ, z, αVec)
# Calculate MPL
MPL = margProdLabor(labor_input, θ, κ, z, αVec, xT)
# Calculate wages
w = p * (β / (β + 1)) * MPL
# Calculate labor supply
labor_supply = (w ./ w_inclusive) .^ β .* L
# Objective to minimize: sum of squared errors
return log.(labor_input ./ labor_supply)
endTo obtain the result, first a good initial guess is obtained and then the above function is minimized.
# Initial guess
initial_guess = getStartGuess_xT(θ, κ, z, αVec; threshold=1e-2)
using LeastSquaresOptim # Need also this package to solve optimizations
result = optimize(objective_to_minimize, initial_guess, LevenbergMarquardt())
# Extract results
optimal_initial_guess = result.minimizer
q_opt = exp(optimal_initial_guess[1])
xT_opt = cumsum(exp.(optimal_initial_guess[2:end]))
# Print results
println("Optimal q: ", q_opt)
println("Optimal xT: ", xT_opt)using SpecialFunctions
b_g(x) = (x^(κ - 1) * exp(-x / θ)) / (θ^κ * gamma(κ)) # Gamma PDF with the same parameterization as above for comparability
e_h1(x)=exp(0.1*x)
e_h2(x)=exp(0.2*x)
e_h3(x)=exp(0.3*x)
e_h = [e_h1, e_h2, e_h3] In a competitive labor market, wages are taken as given and must equate the marginal product of labor. Optimality requires that marginal product ratios equal wage ratios, a relation that is used to obtain the task thresholds xT. Once the task thresholds are known, the function unitInputDemand is used to obtain the labor per output, setting q = 1.
# Objective function for optimization
function objective(x, h)
x=x[1]
return (e_h[h+1](x) / e_h[h](x) - wage[h+1] / wage[h])^2
end
# Find the solutions for xT_h
H=length(wage)
xT = Vector{Float64}(undef, H-1)
# Initial guess: recall that the first element is q
initial_guess=getStartGuessGen_xT(z, b_g, e_h)
for h in 1:H-1
x0=[initial_guess[h+1]]
# Perform the optimization
res = optimize(x ->objective(x, h), x0, LevenbergMarquardt())
# Extract the optimized value
xT[h]=res.minimizer[1]
end
labor_input_2 = unitInputDemandGeneral(xT, q, z, b_g, e_h)If labor inputs per each type are known, they can be given to the function prod_fun or prod_fun_general to compute the task thresholds and the total output produced.
# General parameterization
q, xT=prodFunGeneral(labor_input,z,b_g, e_h)Use the function elasticity_sub_comp to obtain the elasticities of substitution and complementarity. Precompiled values for marginal products (MPL), task thresholds (xT), and total output (q) can be used for efficiency, but they are computed within the function if not provided.
# General case
ϵ_sub_gen, ϵ_compl_gen=elasticitySubCompGeneral(labor_input, z, b_g, e_h)First, define the parameter governing the elasticity of labor supply:
β=4Then define general equilibrium objects (total labor supply and inclusive value) and the price of the good sold by the firm:
# Define general equilibrium objects
L=[1 ; 1 ; 1] # Total labor force
p=1 # Price for the good
w_inclusive=[0.4; 0.9; 2] # Inclusive value of wages Define an objective function that will be minimized to find the solution in total output (q) and task thresholds (xT). This function takes a guess for q and xT as input, computes the labor input required by the guess, and the marginal product of labor. It then computes the wage as a constant markdown applied to the marginal product of labor and the implied labor supply given the firm wage. Finally, the function returns an error that is the discrepancy from the labor supply to the labor required implied by the guess xT and q.
# Same thing but with the general functions
function objective_to_minimize(initial_guess)
# Compute q and xT from the initial guess
q = exp(initial_guess[1])
xT = cumsum(exp.(initial_guess[2:end]))
# Calculate labor input demand
labor_input = unitInputDemandGeneral(xT, q, z, b_g, e_h)
# Calculate MPL
MPL= margProdLaborGeneral(labor_input, z, b_g, e_h, xT, q)
# Calculate wages
w = p * (β / (β + 1)) * MPL
# Calculate labor supply
labor_supply = (w ./ w_inclusive) .^ β .* L
# Objective to minimize: sum of squared errors
return log.(labor_input ./ labor_supply)
end
using LeastSquaresOptim
result = optimize(objective_to_minimize, initial_guess, LevenbergMarquardt())
# Extract results
optimal_initial_guess = result.minimizer
q_opt = exp(optimal_initial_guess[1])
xT_opt = cumsum(exp.(optimal_initial_guess[2:end]))
# Print results
println("Optimal q: ", q_opt)
println("Optimal xT: ", xT_opt)Calculates unit labor demands given blueprint scale θ, blueprint shape κ, productivity z, an array of comparative advantage values αVec with H elements (one for each worker type), and an array xT of H-1 thresholds in task space.
xT: An array of H-1 thresholds in task space.θ: Blueprint scale parameter.κ: Blueprint shape parameter.z: Productivity parameter.αVec: An array of comparative advantage values with H elements.skipParamChecks: A boolean indicating whether to skip parameter checks (default is false).
- An array representing the labor demand for each labor type.
unitInputDemand(xT::AbstractArray{<:Real}, θ::Real, κ::Real, z::Real, αVec::AbstractArray{<:Real}, skipParamChecks::Bool = false) -> AbstractArray{<:Real}Calculates the marginal productivity of labor for each worker type given the input parameters.
labor_input: An array of labor demand values.θ: Blueprint scale parameter.κ: Blueprint shape parameter.z: Productivity parameter.αVec: An array of comparative advantage values.xT: (optional) An array representing the precomputed task thresholds. If not provided, it will be computed within the function.q: (optional) A scalar representing the precomputed quantity produced. If not provided, it will be computed within the function.
- An array representing the marginal productivity of labor for each worker type.
margProdLabor(labor_input::AbstractArray{<:Real}, θ::Real, κ::Real, z::Real, αVec::AbstractArray{<:Real}; xT=nothing) -> AbstractArray{<:Real}Calculates the quantity produced (q), and task thresholds (xT) given labor inputs (l), blueprint scale θ, blueprint shape κ, productivity z, and an array of comparative advantage values αVec with H elements (one for each worker type).
Inputs:
labor_input: Array of labor inputs of different types.θ: Blueprint scale parameter.κ: Blueprint shape parameter.z: Productivity parameter.αVec: Array of comparative advantage values with H elements.initial_guess: (optional) Initial guess for optimization. If not provided, sensible starting point is compute within the function.optim_options: (optional) Optimization options. If not provided, defaults to high tolerance values.
Returns:
q: Quantity produced.xT: Array of task thresholds.
prodFun(labor_input::AbstractArray{<:Real}, θ::Real, κ::Real, z::Real, αVec::AbstractArray{<:Real}; initial_guess=nothing, optim_options=nothing)Calculates the elasticity of substitution and complementarity for a given set of parameters.
labor_input: An array of labor inputs of different types with H elements.θ: Blueprint scale parameter.κ: Blueprint shape parameter.z: Productivity parameter.αVec: An array of comparative advantage values with H elements.MPL: (optional) An array representing the marginal productivity of labor. If not provided, it will be computed within the function.xT: (optional) An array representing precomputed task thresholds. If not provided, it will be computed within the function.
ϵ_h_sub: Matrix of elasticity of substitution values for each worker type h (rows) relative to worker type h_prime (columns).ϵ_h_compl: Matrix of elasticity of complementarity values for each worker type h (rows) relative to worker type h_prime (columns).
elasticitySubComp(labor_input::AbstractArray{<:Real}, θ::Real, κ::Real, z::Real, αVec::AbstractArray{<:Real}; MPL=nothing, xT=nothing) -> (AbstractArray{<:Real}, AbstractArray{<:Real})Calculates unit labor demands given an array xT of H-1 thresholds in task space, a productivity value z,
a density function b_g for the task distribution, and an array e_h of H functions
representing the cost of each labor type as a function of task complexity.
The function first verifies that b_g is a valid density function. Then it computes
the labor demand for each labor type by numerically integrating the ratio b_g(x) / (z * e_h[h](x))
over the intervals defined by the thresholds in xT.
xT: A vector of H-1 thresholds in task space.z: Productivity value.b_g: A density function for the task distribution.e_h: A vector of H functions representing the cost of each labor type as a function of task complexity.
- A vector representing the labor demand for each labor type.
unitInputDemandGeneral(xT::Vector{Float64}, z::Real, b_g::Function, e_h::Vector{Function}) -> Vector{Float64}Calculates the marginal productivity of labor for each worker type given the input parameters.
labor_input: An array of labor inputs of different types with H elements.z: A productivity scalar.b_g: A task density function.e_h: A vector of comparative advantage functions.xT: (optional) An array representing the precomputed task thresholds. If not provided, it will be computed within the function.q: (optional) A scalar representing the precomputed quantity produced. If not provided, it will be computed within the function.
- An array representing the marginal productivity of labor for each worker type.
margProdLaborGeneral(labor_input::AbstractArray{<:Real}, z::Real, b_g::Function, e_h::Vector{Function}; xT=nothing) -> AbstractArray{<:Real}Calculates the quantity produced (q), and task thresholds (xT) given labor inputs (labor_input), productivity z, general blueprint density function (b_g), and a vector of efficiency functions (e_h), one for each labor type.
Inputs:
labor_input: Array of labor inputs of different types.z: Productivity parameter.b_g: Blueprint density function.e_h: Vector of efficiency functions, one for each type.initial_guess: (optional) Initial guess for optimization. If not provided, sensible starting point is compute within the function.x_tol: (optional) Tolerance for the solution vector. Default is 1e-12.f_tol: (optional) Tolerance for the function value. Default is 1e-12.g_tol: (optional) Tolerance for the gradient. Default is 1e-12.iterations: (optional) Maximum number of iterations for the optimization. Default is 1000.max_retries: (optional) Maximum number of retries if the optimization fails. Default is 5.
Returns:
q: Quantity produced.xT: Array of task thresholds.
prodFunGeneral(labor_input::AbstractArray{<:Real}, z::Real, b_g:: Function, e_h::Vector{Function}; initial_guess=nothing, x_tol=1e-12, f_tol=1e-12, g_tol=1e-12, iterations=1000, max_retries=5)Calculates the elasticity of substitution and complementarity for a given set of parameters.
labor_input: An array of labor inputs of different types with H elements.z: Productivity parameter.b_g: General task density function.e_h: Vector of comparative advantage functions.MPL: (optional) An array representing the marginal productivity of labor. If not provided, it will be computed within the function.xT: (optional) An array representing precomputed task thresholds. If not provided, it will be computed within the function.
ϵ_h_sub: Matrix of elasticity of substitution values for each worker type h (rows) relative to worker type h_prime (columns).ϵ_h_compl: Matrix of elasticity of complementarity values for each worker type h (rows) relative to worker type h_prime (columns).
elasticitySubCompGeneral(labor_input::AbstractArray{<:Real}, z::Real, b_g::Function, e_h::Vector{Function}; MPL=nothing, xT=nothing) -> (AbstractArray{<:Real}, AbstractArray{<:Real})Generate an initial guess for the optimization problem in prod_fun such that the implied labor demand is non-trivial.
labor_input::AbstractArray{<:Real}: The observed labor input for each task.θ::Real: The scale parameter of the gamma distribution.κ::Real: The shape parameter of the gamma distribution.z::Real: A scaling factor for the labor input.αVec::AbstractArray{<:Real}: An array of task-specific parameters.threshold::Real: The minimum acceptable labor demand for each task.
initial_guess::Array{<:Real}: A vector containing the initial guess for the optimization, including the log of the initial production quantityqand the initial task thresholdsxT.
getStartGuess_xT(labor_input::AbstractArray{<:Real}, θ::Real, κ::Real, z::Real, αVec::AbstractArray{<:Real}; threshold::Real=1e-2)Generate an initial guess for the optimization problem using a general density function such that the implied labor demand is non-trivial.
labor_input::AbstractArray{<:Real}: The observed labor input for each task.z::Real: A scaling factor for the labor input.αVec::AbstractArray{<:Real}: An array of task-specific parameters.pdf::Function: The general density function.threshold::Real: The minimum acceptable labor demand for each task.verbose::Bool: Optional boolean flag to enable or disable verbose output for debugging.
initial_guess::Array{<:Real}: A vector containing the initial guess for the optimization, including the log of the initial production quantityqand the initial task thresholdsxT.
getStartGuessGen_xT(labor_input::AbstractArray{<:Real}, z::Real, αVec::AbstractArray{<:Real}, pdf::Function; threshold::Real=1e-2, verbose::Bool=false)Contributions are welcome! Please feel free to submit a pull request or open an issue if you have any suggestions or find any bugs.
This project is licensed under the MIT License - see the LICENSE file for details.