Productivity Review of Egypt

Annex B. Outputs and metadata of MultiProd Egypt

Ideal data for MultiProd

For OECD countries, the MultiProd module uses administrative firm-level data (e.g. tax records) or production survey data that ideally cover the full non-financial sector, including manufacturing as well as non-financial services and construction.

The data must be longitudinal and need to contain a panel dimension: the units need to be identified by a unique longitudinal identifier (id) that must be constant over time. If an active unit exits the production survey or business register, its identifier must not be re-assigned to another firm.

At a minimum, the data should include:

• A unique firm identifier.

• A year to which the time-varying variables refer.

• A three- or four‑digit sector identifying the main economic activity of the unit.

• The firm’s labour input, either as a headcount measure of persons engaged (HC) or full-time equivalent (FTE).

• The gross output (GO) of the firm. If the GO is unavailable, sales can be used.

• The firm’s value added (VA).

• Intermediate inputs – the code will still run in the absence of intermediate data, but many core outputs (including those covering markups and some MFP measures) will not be produced.

• Either investment or capital stock.

• The total labour costs of the firm – the code will run in the absence of labour cost data, but statistics on wages and the labour share of value added will not be calculated.

Additional data, if available, will be used by the programme:

• The birth year of the unit, used to identify new entrants and produce statistics by age class.

• Intangible capital and/or intangible investment, used to generate descriptive statistics and inform later analysis.

• Membership of a business group; if available, separate descriptive statistics will be calculated for independent enterprises and members of business groups.

• Foreign ownership; if available, separate descriptive statistics will be calculated for foreign and domestically owned firms.

Set of variables used in MultiProd

The variables used in the module are summarised in Annex Table B.1. , which shows, for each variable and its description (Columns 1 and 2), the original name in the Egyptian Economic Census 2022/2023 (Column 3) and the name assigned in the macro-aggregated MultiProd dataset (Column 4), after statistics have been computed and aggregated through the MultiProd STATA^® code.

The MultiProd module uses the following variables on establishments information from the census:

• Number of workers (L) expressed as persons engaged; and number of paid workers (Labour_WageBill).

• Performance variables: gross value added (VA), total revenue (GO), intermediates (II), total fixed assets (K) and wage bill (WL).

• Additional variables include average wage (W=WL/Labour_WageBill), labour share (L_sh=WL/VA) and industry concentration measures. In the absence of an FTE variable in the microdata, an additional variable has been included to account for hours worked in Egypt and OECD in the MultiProd dataset, which is calculated as the product of the number of workers multiplied by a factor indicating differences in hours worked in Egypt and the OECD average at the industry level. This factor is the ratio of average weekly hours worked in Egypt relative to European average at the two-digit industry level (L_FTE).1

• Economic activity at the level of the macro sector (ind_a7, for example Agriculture, Manufacturing) and A38 industry-level following OECD Structural Analysis (STAN) database classification (ind_a38), retrieved from aggregating some of the two-digit ISIC Rev.4 information in the microdata.

• The establishment’s age, corresponding to the starting year of the activity.

• Whether the establishment is part of a business group. This information is used to calculate aggregate moments separately for business groups or for single establishments.

The firm-id, sampling weights and public/private sector dummy have been used to compute aggregate moments (e.g. public/private variable has been used to calculate deflators).2

Annex Table B.1. EC variables used in MultiProd

Note: Monetary variables in the original Egyptian Economic Census are in EGP, while they have been transformed in MultiProd into real 2005 USD.

(.): missing variable in the MultiProd dataset.

*: industry follows A38 STAN classification based on ISIC Rev.4.

Egyptian deflators

To allow cross-country comparison, the MultiProd code transforms every nominal monetary variable into real 2005 USD.3 To convert nominal variables into real ones, it uses deflators for gross output, value added, intermediates and gross fixed capital formation at the A38 industry-level, usually available from OECD STAN, complemented with country-specific information not available in STAN.

For Egypt only, the value added deflator is currently available from the Ministry of Planning and Economic Development (MPED) national accounts and at one-digit sectoral level. In the absence of other variable-specific and more disaggregated deflators, this value added deflator has been applied also to output, intermediates and fixed assets. 4 This choice comes with some limitations for the analysis. First, deflators might be heterogeneous within broadly defined sectors. Indeed, the standard MultiProd code would use two‑digit deflators, when available, as opposite to one‑digit information used for Egypt. Second, applying the value added deflators to capital, output and intermediates works under the assumption that the price of capital, intermediates and outputs has experienced the same increase/decrease than the value added price within each macro sector (one digit), but that might not necessary be the case. However, this is a solution widely used in the literature in the absence of variable-specific deflators.

For Egypt, the MultiProd routine has been adapted to apply two sets of deflators to account for differences between public and private sectors, as those sectors differ in the time of reference (fiscal year for the public sector and calendar year for the private sector) and in the national accounts.5

MPED does not directly provide the deflators by economic activity, but rather data on value added at current prices and the annual growth rate of real value added by economic activity ($j$) at the one-digit level. From these data, deflators growth rate $g_{Djt}$ have been calculated as:

$g_{Djt}=\frac{1+g_{nomina{l}_{jt}}}{1+g_{rea{l}_{jt}}}-1$,

from which it has been possible to obtain deflators relative to 2007 prices (or 2006/07 for the public sector). The choice of 2007 is driven by the fact that the “information sector” has missing values before 2007 in national accounts. The standard MultiProd code transforms every monetary variable in real 2005 local currency, and then applies 2005 PPP adjustments to convert it into USD (PPPs are taken from the Eurostat-OECD PPP Programme).

Since for Egypt the base year for deflators is 2007 instead of 2005, deflation and price level adjustments have been made as follows to convert them to 2005 levels: i) local currency values have been deflated to 2007 local currency values; ii) values have been adjusted using the PPP-adjusted 2007 USD value of the local currency; 6 and iii) a second adjustment has been applied using the ratio of the PPP-adjusted USD value of the local currency in 2007 and 2005 respectively, as follows.

For each industry the deflator ($D$) for the year 2017 with base year 2005 is thus constructed as:

${D_{base\mathrm{ }2005}^{\frac{2022}{23}}\approx D}_{base\mathrm{ }2007}^{\frac{2022}{23}}\mathrm{*}PP{P}_{2007}\mathrm{*}\left(\frac{PP{P}_{2007}}{PP{P}_{2005}}\right)$.

where PPP is purchasing power parities.

Wages have then been deflated using the Consumer Price Index (CPI) at the national level, from World Bank World Development Indicators.7

Productivity measures

The MultiProd code – adapted for the Egyptian data – produces multiple measures of productivity summarised in Annex Table B.2. : labour productivity (LP), MFP estimated using ordinary least squares regression (MFP_OLS) and MFP using Solow residuals (MFP_S and MFP_SW). All these measures are calculated using as output variable either value added or revenues and are discussed in detail below.

Labour productivity calculated using firms’ revenues as gross output (LP_GO) captures the amount of output produced by an establishment for a given amount of labour input. Yet, this measure does not control for the role of intermediates. As a result, a company with high reselling activity will probably rank very high in this measure. A measure of productivity that takes into account the role of intermediates is thus labour productivity calculated using value added as output (LP_VA), as value added is the difference between output and intermediates.

However, labour productivity does not necessarily reflect a firm’s efficiency as it does not account for capital input and thus does not measure differences in capital intensities across firms. For this reason, the programme calculates two measures of MFP: two index measures of Solow residuals (MFP_S and MFP_SW) and MFP estimated using OLS regression (MFP_OLS). Note that the standard MultiProd code does not usually calculate MFP_OLS, as it uses instead more advanced estimates of MFP such as the one proposed by Wooldridge (2009[1]) or the one from Ackerberg, Caves and Frazer (2015[2]), which allow to account for simultaneity and endogeneity issues (see Annex Box 2.B.1). Yet, as already discussed, those measures cannot be computed with the available data for Egypt.

MFP_SW is estimated using external inputs shares from OECD STAN, where the shares are given by the median across OECD countries and years for each A38 industry.8 As suggested by Syverson (2011[3]), labour and intermediate shares are taken as constant over time. Constant returns to scale are assumed and, thus, the share of capital is calculated as one minus the other input shares. This measure is particularly suitable for cross-country comparisons.

MFP_OLS is estimated using a standard OLS regression, run separately for each industry (at the STAN A38 level) assuming a standard Cobb-Douglas production function. Within each industry at least 20 observations are required for the estimation to be performed.

Annex Table B.2. Productivity measures in MultiProd, Egypt

Note: Variables in lower case letters refer to natural logarithms.

Annex Box B.1. Productivity using MFP estimated from Wooldridge (2009), OECD countries

The lack of a panel dimension in Egyptian data also has implications for the MFP calculations based on production function estimations. Indeed, for Egypt, the MFP has been estimated as a residual from a standard ordinary least squares regression (MFP-OLS, see Annex B). However, the literature has shown that estimates based on MFP-OLS might be biased and inconsistent, as they are affected by an endogeneity bias.9 The MultiProd module usually considers, as preferred measures for MFP, the residuals from either Wooldridge (2009[1]) or Ackerberg, Caves and Frazer (2015[2]) approaches to estimating production function. However, to be estimated these measures require that the unit of analysis is observed over at least two consecutive periods.

Among the MFP measures based on a production function approach, the preferred measure in the MultiProd module is the MFP estimated using the Wooldridge (2009[1]) methodology on a production function with two inputs (capital and labour) and value added as output.10 Value added is preferred to gross output due to identification problems with the latter (see Bond and Söderbom (2005[4])).

Woodridge (2009[1]) uses a semi-parametric approach that mitigates the endogeneity problem of input choices, by using intermediate inputs as proxy variables for productivity and (twice) lagged values of labour as instruments. Wooldridge developed a technique that builds upon the Levinsohn and Petrin method (2003[5]) but uses a quicker one-step procedure (instead of a two-step approach) with consistent standard errors without bootstrapping. It also allows one to overcome the critique of Ackerberg et al. (2006[6]) on the identification of the labour coefficient.

Annex Figure B.1. shows the MFP estimated relying on the Wooldridge methodology by size class in the OECD benchmark. MFP is increasing in size and the pattern is steeper in manufacturing than non‑financial market services, in line with previous OECD findings (Berlingieri, Calligaris and Criscuolo, 2018[7]).

Annex Figure B.1. Multifactor productivity calculated with Wooldridge (2009) methodology is increasing by firms’ size class in OECD countries

MFP-Wooldridge (2009) by size classes, 2022

Note: The graph plots the average log value added MFP_W by size class with respect to base category (1-4 employees), where coefficients come from the following regression: ${LogMFP}_{cjs}= {\mathit{D}}_{cjs}^{\text{'}} \beta +m_{cjs}{\mathit{D}}_{cjs}^{\text{'}} { \gamma + z}_j+{ϵ}_{cjs }\left[3\right]$, where ${LogMFP}_{cjs}$ is the log labour MFP mean in country $c$, industry $j$ and size class $s$. $D$ is a vector of categorical variables for each size class (1-4, 5-9, 10-19, 20-49, 50-99, 100-240, 250+). $m$ is he macro sector indicating 1 for manufacturing and 2 for non-financial services. Regressions include analytical weights, as share of value added in each cell. Regressions control for industry-country fixed effects ($z\_cj)$. OECD countries include Belgium, Canada, Croatia, Estonia, Finland, France, Hungary, Italy, Lithuania, Portugal, Slovenia and Spain.

Sources: OECD (n.d.[8]), MultiProd V2 (database), https://www.oecd.org/en/about/projects/multifactor-productivity.html.

Sources (box): Bond, S. and M. Söderbom (2005[4]), “Adjustment costs and the identification of Cobb Douglas production functions”, https://doi.org/10.1920/wp.ifs.2005.0504; Levinsohn, J. and A. Petrin (2003[5]), “Estimating production functions using inputs to control for unobservables”, https://www.jstor.org/stable/3648636; Ackerberg, D., K. Caves and G. Frazer (2006[6]), “Structural identification of production functions”, https://mpra.ub.uni-muenchen.de/id/eprint/38349; Berlingieri, G., S. Calligaris and C. Criscuolo (2018[7]), “The productivity-wage premium: Does size still matter in a service economy?”, https://doi.org/10.1787/04e36c29-en.

Capital stock measure

For production inputs, the baseline case of the MultiProd code defines the capital stock variable through the widely used perpetual inventory method (PIM) (Berlingieri et al., 2017[9]) to increase the comparability of results across countries. In the event an alternative measure of capital stock is directly available in the microdata, the programme uses it to initialise the PIM and carry out a robustness analysis.

The perpetual inventory method

The perpetual inventory method requires firm-level investment data, preferably over a sufficiently long time period. As is common in the literature (e.g. Olley and Pakes (1996[10]), Levinsohn and Petrin (2003[5])), the initial value is set to the capital stock reported by the firm in the initial year, whenever this is available. If, the capital stock is not available (or missing in the first year), the code calculates the initial capital stock as the average of two proxies described below:

• Average firm investment in all years divided by the depreciation rate (from STAN), borrowing a similar idea from Mueller (2008[11]).

• Country-industry-year specific capital-labour ratio (K/L) from STAN multiplied by firm-level employment (L).

The two proxies for the initial capital stock have a correlation of 0.7 in the microdata used for testing, and, as expected, the capital stock obtained from the industry average capital-labour ratio is usually larger since it tends to overestimate capital for smaller firms. The code has an option to set the first n years of the calculated capital stock to missing, to exclude observations affected by the initial noise due to the PIM.

The code simply treats consecutive observations as consecutive years; this implies that whenever the firm is missing in some years, the PIM essentially assumes that capital is constant in the years in between. A similar situation arises when only investment is missing in some years: in fact the code sets investment to zero whenever missing and the capital stock evolution will depend only on depreciation in those years (unless it is missing for all years for that firm, in which case investment is set to missing). 11

The case of MultiProd Egypt

To use the above methodology, firm-level investment data, preferably over a sufficiently long time period, are needed. In the absence of this yearly information for Egypt, due to lack of panel dimension in the Egyptian Economic Census, the stock has been proxied with the book value of capital, using total fixed assets. This different methodology lowers the full comparability with OECD countries.

MultiProd outputs and metadata for Egypt

Several statistics are collected by the MultiProd code. The statistics are never reported at the level of the individual establishment: instead, the programme splits the establishment-level dataset along various dimensions, into cells, and for each cell calculates aggregate data. Each output file produced by MultiProd ends with the list of variables defining what constitutes a cell in the output file (Annex Table B.6. ).

The basic level of aggregation is at the year-industry level, either the macro sector (ind_a7) or the more detailed industry level (A38). Then cells can be further broken down into size class, age class and business group (group, equal to zero if single branch), or a combination of those dimensions. Annex Table B.4 and Annex Table B.5 show the different levels of aggregation used for industry, size and age classes.

Datasets are stored as .dta files, as reported in Annex Table B.6. . The name “plant” refers to the unit of analysis (plant = establishment) while “W” stands for weighted statistics.

The code produces a set of statistics including:

1. Basic statistics: including mean, median, standard deviation and number of non-missing values for main variables, collected and defined in the _statistics.ado file. Correlations are also computed. Statistics are calculated using sampling weights and the sum of the inverse probability weights is collected as [Variable]_sum_w, corresponding to the number of establishments in the population in each cell (Annex Table B.3. ). Those statistics are collected in the Main_Stat datasets.

Annex Table B.3. MultiProd Statistics

Name	Statistics
_N	Number of observations for which statistics are computed
_sum_w	Sum of the weights to each observation
_av	Mean
_sd	Standard deviation
_corr_[X]_[Y]	Correlation between X and Y
_pXY	Percentile of the variable (e.g. p50 median)
_pareto_b ; _pareto_se	Estimated Pareto coefficient (and sd) for the distribution of the variable

2. Measure of allocative efficiency: The MultiProd code implements, for each measure of productivity (LP and MFP), the OP gap ( _opgap ) from Olley and Pakes (1996[10]) static productivity decomposition. This variable reflects the gap between the weighted average and the unweighted average of productivity. A large OP gap implies that firms which are more productive capture a larger share of inputs (or a larger share of the market), hence it is a good proxy for static allocative efficiency. As discussed above, due to the cross-sectional nature of the data, only this static measure of allocative efficiency is calculated for Egypt.

3. Statistics by percentiles of the distributions: the code collects establishments’ characteristics and main statistics by quantiles of the productivity distribution, and by quantiles of the revenue distributions (Moments_Quant datasets).

4. Concentration ratios: statistics are computed for the output share (GO and VA) of the top 8 and top 20 firms in each industry (collected in the Concentration datasets).

5. Basic tabulations of establishment with negative value added, including their number and characteristics (collected in the SumStats_VA_Neg dataset).

6. Characteristics of firms at the productivity frontier (Frontiers dataset). In the case of Egypt, this only collects information on the number of firms at the frontier of the productivity distribution (90^th and 100^th percentile of log productivity) by industry. This lacks, however, the standard MultiProd information on persistence of frontiers firms at the top of the distribution that would require longitudinal information.

Annex Table B.4. Sectoral aggregation

STAN A38 aggregation based on ISIC Rev.4 classification	STAN A7 aggregation
01 to 03 Agriculture, forestry and fishing	Agriculture
05 to 09 Mining and quarrying	Mining
10 to 12 Food products, beverages and tobacco	Manufacturing
13 to 15 Textiles, wearing apparel, leather and related products
16 to 18 Wood and paper products, and printing
19 Coke and refined petroleum products
20 Chemicals and chemical products
21 Basic pharmaceutical products and pharmaceutical preparations
22 to 23 Rubber and plastics products, and other non-metallic mineral products
24 to 25 Basic metals and fabricated metal products, except machinery and equipment
26 Computer, electronic and optical products
27 Electrical equipment
28 Machinery and equipment n.e.c.
29 to 30 Transport equipment
31 to 33 Furniture; other manufacturing; repair and installation of machinery and equipment
35 Electricity, gas, steam and air conditioning supply	Utilities
36 to 39 Water supply; sewerage, waste management and remediation activities
41 to 43 Construction	Construction
45 to 47 Wholesale and retail trade, repair of motor vehicles and motorcycles	Market services
49 to 53 Transportation and storage
55 to 56 Accommodation and food service activities
58 to 60 Publishing, audiovisual and broadcasting activities
61 Telecommunications
62 to 63 IT and other information services
64 to 66 Financial and insurance activities	Excluded
68 Real estate activities	Market services (continued)
69 to 71 Legal and accounting activities; activities of head offices; management consultancy activities; architecture and engineering activities; technical testing and analysis
72 Scientific research and development
73 to 75 Advertising and market research; other professional, scientific and technical activities; veterinary activities
77 to 82 Administrative and support service activities
84 Public administration and defence	Excluded
85 Education	Non-market services
86 to 88 Human health and social work activities
90 to 93 Arts, entertainment and recreation
94 to 96 Other service activities

Annex Table B.5. Size and age classes

Size class/ age class	Description
Size Class 22*	Two size classes with threshold at 10 employees: Small (<10), MediumLarge (>=10)
Size Class 2222	Two size classes with threshold at 10 employees but statistics not reported for the first group: MediumLarge (>=5)
Size Class 52*	Five size classes with thresholds at 10, 20, 50 and 250 employees: Micro (1-9), Small (10-19), MediumSmall (20-49), Medium (50-249), Large (250+)
Size Class 5222*	Five size classes with thresholds at 10, 20, 50 and 250 employees but statistics not reported for the first group: Small (10-19), MediumSmall (20-49), Medium (50-249), Large (250+)
Size Class 71	Seven size classes with thresholds at 5, 10, 20, 50, 100 and 250 employees: Micro (<5), Micro (5-9), Small (10-19), MediumSmall (20-49), Medium (50-99), MediumLarge (100-249), Large (250+)
Age Class 21*	Two age classes with threshold at 5 years: Young (0-5), Old (6+)

Annex Table B.6. Output datasets MultiProd, Egypt

Dataset name	Description
Concentration_W_by_year_ind_a7_plant.dta	Concentration ratios by macro industry-year
Concentration_W_by_year_ind_a38_plant	Concentration ratios by industry A38-year
Frontier_W_by_year_ind_a38_plant.dta	Number of establishments at the frontiers by industry A38-year
MainStats_W_10plus_by_year_ind_a38_sizeclass2222_plant.dta	Basic moments and allocative efficiency measure for establishments with more than 10 workers by industry A38-year
MainStats_W_by_year_ind_a38_plant.dta	Basic moments and allocative efficiency measure for all establishments by industry A38-year
Moments_Quant_W_by_year_ind_a7_LogLP_VA_plant.dta	Basic moments and allocative efficiency measure for quantile of the labour productivity distribution and macro industry
Moments_Quant_W_by_year_ind_a7_LogMFP_OLS_plant.dta	Basic moments and allocative efficiency measure for quantile of the MFP_OLS productivity distribution and macro industry
Moments_Quant_W_by_year_ind_a7_LogMFP_SW_plant.dta	Basic moments and allocative efficiency measure for quantile of the MFP_SW productivity distribution and macro industry
Moments_Quant_W_by_year_ind_a38_GO_plant.dta	Basic moments and allocative efficiency measure for quantile of the revenue distribution and A38 industry
Moments_Quant_W_by_year_ind_a38_LogLP_VA_plant.dta	Basic moments and allocative efficiency measure for quantile of the labour productivity distribution and A38 industry
Moments_Quant_W_by_year_ind_a38_LogMFP_OLS_plant.dta	Basic moments and allocative efficiency measure for quantile of the MFP_OLS productivity distribution and A38 industry
Moments_Quant_W_by_year_ind_a38_LogMFP_SW_plant.dta	Basic moments and allocative efficiency measure for quantile of the MFP_SW productivity distribution and A38 industry
Moments_W_by_year_ind_a7_sizeclass71_plant.dta	Basic moments and allocative efficiency measure for all establishments by macro industry and size class at seven categories
Moments_W_by_year_ind_a38_ageclass21_plant.dta	Basic moments and allocative efficiency measure for all establishments by industry A38-year and age class at two categories
Moments_W_by_year_ind_a38_ageclass21_sizeclass22_plant.dta	Basic moments and allocative efficiency measure for all establishments by industry A38-year and size class at two categories and age class at two categories
Moments_W_by_year_ind_a38_group_plant.dta	Basic moments and allocative efficiency measure for all establishments by industry A38-year and if part of a business group or not
Moments_W_by_year_ind_a38_group_sizeclass5222_plant.dta	Basic moments and allocative efficiency measure for all establishments by industry A38-year and size class at five categories (but excluded 1-9)
Moments_W_by_year_ind_a38_sizeclass52_plant.dta	Basic moments and allocative efficiency measure for all establishments by industry A38-year and size class at five categories
Moments_W_by_year_ind_a38_sizeclass71_plant.dta	Basic moments and allocative efficiency measure for all establishments by industry A38-year and size class at seven categories

Source: Berlingieri et al. (2017[12]).

References

[2] Ackerberg, D., K. Caves and G. Frazer (2015), “Identification Properties of Recent Production Function Estimators”, Econometrica, Vol. 83/6, pp. 2411-2451, https://doi.org/10.3982/ecta13408.

[6] Ackerberg, D., K. Caves and G. Frazer (2006), “Structural identification of production functions”, https://mpra.ub.uni-muenchen.de/id/eprint/38349.

[9] Berlingieri, G. et al. (2017), “The Multiprod project: A comprehensive overview”, OECD Science, Technology and Industry Working Papers, No. 2017/04, OECD Publishing, Paris, https://doi.org/10.1787/2069b6a3-en.

[12] Berlingieri, G. et al. (2017), “The Multiprod project: A comprehensive overview”, OECD Science, Technology and Industry Working Papers, No. 2017/04, OECD Publishing, Paris, https://doi.org/10.1787/2069b6a3-en.

[7] Berlingieri, G., S. Calligaris and C. Criscuolo (2018), “The productivity-wage premium: Does size still matter in a service economy?”, OECD Science, Technology and Industry Working Papers, No. 2018/13, OECD Publishing, Paris, https://doi.org/10.1787/04e36c29-en.

[4] Bond, S. and M. Söderbom (2005), “Adjustment costs and the identification of Cobb Douglas production functions”, IFS Working Papers, No. 05/04, Institute for Fiscal Studies, London, https://doi.org/10.1920/wp.ifs.2005.0504.

[5] Levinsohn, J. and A. Petrin (2003), “Estimating production functions using inputs to control for unobservables”, Review of Economic Studies, Vol. 70/2, pp. 317-341, https://www.jstor.org/stable/3648636.

[11] Mueller, S. (2008), “Capital stock approximation using firm level panel data”, Jahrbücher für Nationalökonomie und Statistik, Vol. 228/4, pp. 357-371.

[13] OECD (2024), OECD Compendium of Productivity Indicators 2024, OECD Publishing, Paris, https://doi.org/10.1787/b96cd88a-en.

[8] OECD (n.d.), MultiProd (database), OECD, Paris, https://www.oecd.org/en/about/projects/multifactor-productivity.html.

[10] Olley, G. and A. Pakes (1996), “The dynamics of productivity in the telecommunications equipment industry”, Econometrica, Vol. 64/6, p. 1263, https://doi.org/10.2307/2171831.

[3] Syverson, C. (2011), “What determines productivity?”, Journal of Economic Literature, Vol. 49/2, pp. 326-265, https://www.aeaweb.org/articles?id=10.1257/jel.49.2.326.

[1] Wooldridge, J. (2009), “On estimating firm-level production functions using proxy variables to control for unobservables”, Economics Letters, Vol. 104/3, pp. 112-114, https://doi.org/10.1016/j.econlet.2009.04.026.

Notes

← 1.  For Egypt, information on weekly average hours worked is retrieved from the ILO database, while for European countries, the information comes from Eurostat.

← 2.  As of now, the standard MultiProd module does not allow the use of survey weights but relies instead on the use of business registries to calculate weights in the sample and report on the population of firms. The code has thus been adapted to account for sampling weights (Adjusted_weights). Due to rounding issues, the number of firms in the population differs slightly, respectively 3.739.823 in MultiProd and 3.740.013 in the Egyptian Economic Census (excluding ISIC Rev.4 Section K).

← 3.  In the original data, all numeric variables should be in single units (e.g. not in thousands or millions). All monetary variables should be in nominal values and recorded in local currency.

← 4.  See https://mped.gov.eg/Analytics?lang=en.

← 5.  MPED distinguishes between the private and public sectors in national accounts data.

← 6.  Information on PPPs come from World Bank data.

← 7.  See https://data.worldbank.org/indicator/FP.CPI.TOTL?locations=EG.

← 8.  Labour share is either LABS_VA = LABR/VALU (wage bill/value added in nominal prices) or LABS_GO = LABR/PROD (share of gross output); intermediates share is INTS_GO = INTI/PROD. The external aggregate labour and intermediate shares are taken from the OECD STAN database. Whenever labour and intermediates shares are missing (or their sum larger than 1), the value is taken from sectors at a more aggregated level, as for deflators (see below). Note that the aggregate labour share might be underestimated because the income of self-employed workers is missing. The share of labour input costs published in the OECD Productivity database is adjusted for self-employed income but, unfortunately, the database does not have the country-industry-time coverage needed. Adjustments like the one suggested in the OECD Productivity Manual (OECD, 2024[13]) have not been carried out because this might end up overestimating the labour share; the microdata are also more likely to underestimate it (e.g. temporary workers from employment agencies are not always included). Egypt is not included in the median.

← 9.  Indeed, productivity changes directly affect both input (e.g. firms will hire more workers) and output (value added or gross output).

← 10.  MultiProd also uses the Ackerberg, Caves and Fraser (2015[2]) methodology to calculate MFP.

← 11.  Alternative measure: for countries whose microdata already include a measure of capital stock (from book values or other measures of capital stock obtained by country researchers), the code creates a second measure of capital (K_Rob). The code fills in missing values by interpolating the provided information, which makes K_Rob coherent with the measure of capital (K) built using the PIM. In the microdata used for testing, the two capital stock measures have a correlation of 0.96. For the countries that have the capital stock information but not the investment data, K_Rob becomes the main (and only) measure of capital. This is the case for Egypt.