The role of EU cohesion funds in Romanian labour productivity: Insights from machine learning and econometric modelling

Figure 1

Estimation of the impact of ESIF on labour productivity_

Variable	Coefficients
log (EAFRDEMFFpc)	7.3573***
log (ERDFpc)	3.2260***
log (GFCF)	9.5602**
EduLow	−1.0459***
PrimGVA	1.4592*
CONS	49.6282**
N of observations	104
N of groups	8
F(5,91)	25.68
Prob > F	0.0000
F (FE are jointly zero)	38.71
Prob > F	0.0000

Empirical results of dynamic Dif-GMM estimation_

Variable	Coefficients
LP(−1)	0.5863***
log (EAFRDEMFFpc)	3.1287***
log (GFCF)	7.6210**
EduLow	−0.0400**
PrimGVA	1.1370**
CONS	13.3223*
N of observations	96
N of groups	8
Wald χ 2 (5)	5725.69
Prob > χ 2	0.0000

Estimated variable coefficients for modelling labour productivity_

Region	Intercept	log (EAFRDpc)	EduLow	log (GFCF)	PrimGVA	LP(−1)
RO11	−9.614	3.826	−0.215	12.690	0.484	0.657
RO12	−15.345	9.533	0.337	−6.408	1.164	0.644
RO21	−18.865	−0.723	0.337	−6.408	1.164	0.826
RO22	−20.162	8.677	0.619	−5.458	0.212	0.660
RO31	55.856	2.299	0.067	1.976	0.009	0.272
RO32	−26.214	2.647	0.578	33.340	4.731	0.386
RO41	77.121	5.999	−1.500	11.051	−2.604	0.446
RO42	45.884	2.392	−0.925	6.661	3.807	0.143

Data sources_

Indicator	Source
Labour productivity	Eurostat
EAFRD per capita	Directorate-General for Regional and Urban Policy, EU Payments History – Regionalized and Modeled dataset
EMFF per capita	Directorate-General for Regional and Urban Policy, EU Payments History – Regionalized and Modeled dataset
ERDF per capita	Directorate-General for Regional and Urban Policy, EU Payments History – Regionalized and Modeled dataset
GFCF	Eurostat
Gross Domestic Expenditure on R&D (GERD)	Eurostat
European Quality of Government Index (EQI)	European Quality of Government Index (EQI)
Proportion of Population with Low Education (EduLow)	Eurostat
Share of Primary Sector in Gross Value Added (PrimGVA)	Eurostat
Initial GDP per capita (2007)	Eurostat
Population size	Eurostat
Road accessibility	DG Regio Regional Competitiveness Index – 2022 edition
Air accessibility	DG Regio Regional Competitiveness Index – 2022 edition
Regions analysed: 8 NUTS2 regions in Romania
Period covered: 2007–2020
Total observations: 112 observations (8 regions × 14 years)

Testing the hypothesis on random effects in modelling labour productivity_

FE test
Statistic	p-value	Alternative
2.9508	0.0078	Significant effects
Hausman endogeneity test
Statistic	p-value	Alternative
24.7177	0.0001	One model is inconsistent

LASSO’s results regarding the most important determinants_

Variable	LASSO	Post-estimation OLS
log (EAFRDEMFFpc)	10.0348	10.3199**
log (ERDFpc)	2.9831	3.0211**
log (GFCF)	29.2896	29.4165**
log (GERD)	4.9112	4.9793**
EQI	−15.7593	−16.1696**
EduLow	−0.5370	−0.5364**
PrimGVA	−0.8517	−0.8623**
CONS	0.4388006	1.846910**

Empirical results of the LASSO method for the labour productivity indicator_

Lambda	L1-Norm	EBIC	R-square
5241.22405	0.00000	705.66556	0.0000
4775.60815	3.48932	696.13488	0.1274
2268.80077	21.77004	616.51606	0.6119
1883.59826	23.47559	607.44592	0.6599
1563.79637	24.66088	601.04308	0.6941
742.92997	27.86920	582.66685	0.7549
561.99928	29.55684	582.16015	0.7667
425.131884	33.10793	580.57758	0.7803