Closed rickecon closed 1 year ago
I found this South African central bank paper (Pain, et al, 2020) estimating TFP in South Africa by industry. It looks very helpful in that they estimate TFP growth rates for South African industries from 2009- 2020. However, it is not clear to me how to back out the Z values rather than the growth rates (log difference).
Furthermore, I can't tell what method Pain, et al (2020) are using to account for the various types of endogeneity described in Beveren (2012; see below for full reference). But I might just need to read Pain, et al (2020) more carefully.
cc: @SeaCelo @sarvonz @jdebacker
On TFP, just to add that a possible alternative is the South African Reserve Bank Quarterly Bulletin that you shared. It has most (all?) the data needed to compute the TFP or an approximation (Output, labor, capital) by industry. I'm not clear on how to convert these to the Z_m
variable and the documentation doesn't make clear what the units/base should be.
(edit to remove information on capital share of income. This is updated with new data source below)
@rickecon @sarvonz @jdebacker
We have done some TFP calculations. Given the industry-level data on value added, fixed capital stock, people employed and employee compensation from the South African Reserve Bank, we used Y=AF(K,L)
, with F(K,L)=(alpha*(K^rho)+(1-alpha)*(L^rho))^(1/rho)
, to back out A. This specification of F(K,L)
was estimated in the SARB paper Pain et al (2020). Rho is (sigma-1)/sigma
, and sigma is set as 0.9 following the paper.
Marcelo reached out to SARB and they were very helpful with providing the data.
For 2021:
| TFP (=GVA/F(K,L)) | TFP (Secondary=1) -- | -- | -- Primary | 396.80 | 0.17 Secondary | 2377.11 | 1.00 Tertiary | 6604.17 | 2.78
We want to calibrate multiple industries in OG-ZAF.
What industries to calibrate? The first step is to choose what industries to calibrate. I recommend the following, which come from the South African Reserve Bank Quarterly Bulletin:
M
th industry because it will be the industry that produces goods that can be used as capital): Manufacturing + Electricity, gas and water + Construction (contractors)Another industry we could include is the informal sector, which is a production sector that does not pay corporate income taxes.
What parameters and what data are needed to calibrate these 3 or 4 industries? Let
m
be the index of them
th industry with the total number of industries beingM
. Leti
be the index of thei
th consumption good with the total number of consumption goods beingI
. The theory for the parameters listed below is in the OG-Core documentation for the theory of the Firm and Households.M
: Number of industries is set inogzaf_default_parameters.json
as the parameter scalarM
.I
: The choice of the number of consumption goods depends on the research questions you want to answer and the quality of the input output data you have for the country. The cheapest calibration is to assume that each production industry is a good that households consume. In this caseI = M
. ButI
does not have to equalM
in general.Z_m
: TFP is represented in the theory asZ_m
and its value is given by the list of lists parameterZ
inogzaf_default_parameters.json
. For example, ifM=3
, a corresponding value forZ
isZ = [[1.0., 1.2, 0.9]]
. TheZ
value for a given industry is estimated by a standard log regression of industry output on industry capital stock and industry labor hours. Because the model units are endogenous, the important thing about calibrating theZ
value for each industry is getting its relative size right. As such, it makes sense to normalize one of the industries TFP to 1.0 and set the other TPF values so that the relative size of the industry matches the output data.epsilon
. A Cobb-Douglas specification forM=3
industries would beepsilon = [1.0, 1.0, 1.0]
.K_g
from the production function using thegamma_g
parameter. If the number of industries isM=3
, the correct specification would begamma_g = [0.0, 0.0, 0.0]
. This parameter allows us to account for government infrastructure investment that goes into the production function of each industry. Although this is sufficient, we should setinitial_Kg_ratio
to zero as well.gamma_m
. In theogzaf_default_parameters.json
, the parameter isgamma
and is a list of scalars strictly between 0 and 1. For example, ifM=3
a valid specification would begamma = [0.2, 0.5, 0.3]
. Because labor cost data for industries is usually more accurate than capital cost data, we calibrate the capital share of income by calculating the labor share of income. Then the capital share is 1 - labor share. Labor share = total labor compensation / output, where total labor compensation is employee compensation plus proprietor compensation.io_matrix
: This parameter maps production industry output to consumption categories. The easiest calibration for a model withM=3
andI=3
is a 3 x 3 identity matrix input intoogzaf_default_parameters.json
in the following way.alpha_c
: In the documentation, the consumption share is defined asalpha_i
(see equation 10). In OG-Core, the parameter isalpha_c
. A valid calibration for a model withI=3
consumption goods would bealpha_c = [0.3, 0.4, 0.3]
.cit_rate
tau_c
cc: @SeaCelo @sarvonz @jdebacker