OECD Economic Surveys: Poland 2025

Annex A. Micro-data approach to productivity analysis - experimental research by Statistics Poland

Statistics Poland has developed an experimental analysis of the productivity developments in the Polish economy using a firm-level panel dataset covering small, medium, and large non-financial enterprises. The dataset has been developed from an annual enterprise survey covering non-financial enterprises employing more than nine full-time workers for a wide selection of NACE sections (Statistics Poland, 2019). Drawing on cooperation with the IMF and the World Bank, Statistics Poland defined analogues of macroeconomic concepts, including gross value added, capital, investments, and labour productivity. Preliminary data processing procedures, including cleaning data from incomplete observations, were carried out. More detailed descriptions can be found in Statistics Poland (2024), Górajski and Błażej (2020), World Bank (2021) and International Monetary Fund (2019).

The panel data set, used for the analysis included in the OECD’s 2025 Economic Survey consists of over 970 000 observations, accounting for 87% of all observations recorded in survey over 2008–2022 and includes on average over 65 000 enterprises annually. The enterprises recorded an average employment of 5.14 million full-time equivalents per year and average annual revenues of PLN 3,222 billion. The final sample constitutes the majority of the non-financial enterprise sector in Poland. However, it is not selected using a representative method and cannot be fully generalised to the whole population.

Labour productivity (LP) is a calculated as a simple ratio of gross value added to the number of full-time employees employed using deflators of the relevant two-digit NACE sectors. Firms’ employment shares are used to aggregate firm-level labour productivity indices. The measurement of the unobserved firm-level TFP is possible by determining the Solow residual from the production function equation applied at the firm level (van Beveren, 2012). Gross value-added ${GVA}_{it}$ for enterprise $i$ from the two-digit NACE Rev. 2 division $G_d\mathrm{ }$in period $t$ is determined by the Cobb-Douglas function:

(1)

${GVA}_{it}=TF{P}_{it}{K}_{it}^{{\beta }_{k,d}}{L}_{it}^{{\beta }_{l,d}}{e}^{{ϵ}_t},\mathrm{ }i\in G_d,\mathrm{ }$

in which ${ϵ}_t\mathrm{ }$is i.i.d output shock, $TF{P}_{it}$ is the unadjusted total factor productivity; $L_{it},K_{it}$ are the quantities of labour and capital. The parameters ${\beta }_{k,d}$ and ${\mathrm{ }\beta }_{l,d}$ denote capital and labour's gross value-added sectoral elasticities, respectively.

The firm-level production function from (1) is estimated using control function methods, such as Olley and Pakes’ (1996) model (OP model) and Levinsohn and Petrin’s (2003) model (LP model), both of which have been enhanced by the correction made by Ackerberg, Caves, and Frazer (2015). Within this framework, the productivity coefficient $TF{P}_{it}$ is a state variable in the company-decision problem, which involves selecting production factors. The two models produce consistent estimates of output elasticities that solve the endogeneity problems of explanatory variables and attrition (van Beveren, 2012). Due to data availability restrictions, Statistics Poland estimated the production function using the LP model with attrition, regional and time effects, and energy and materials expenditures as a proxy for unobserved TFP indices. Consequently, the following estimate of the company’s productivity coefficient ${\widehat{\omega }}_{it}$, the unadjusted TFP ${\widehat{TFP}}_{it}$, and labour productivity ${\widehat{LP}}_{it}$ are defined directly:

(2)

${\widehat{LP}}_{it}=\frac{{GVA}_{it}}{L_{it}},$

(3)

${\widehat{\omega }}_{it}={gva}_{it}-{\widehat{\beta }}_{k,d}{k}_{it}-{\widehat{\beta }}_{l,d}{l}_{it},$

where ${\widehat{\omega }}_{it}=\log TF{P}_{it},{gva}_{it}=\log GV{A}_{it}$, $k_{it}=\log K_{it}$, and $l_{it}=\log L_{it}$.

The firm-level labour and total factor productivity can be aggregated to sector, region, or economy level using their input-based shares.

Figure A.1. Main features and structure of the firm-level panel dataset

2013-22

Note: The calculations are part of experimental statistics conducted by Statistics Poland. The OECD is solely responsible for the interpretation of the results, conclusions, and policy recommendations.

Source: Statistics Poland.

Statlink