1 Introduction and purpose

This document consolidates the full PKP2 analysis of investment in the European Union between 2000 and 2025. It provides a systematic, comparable and policy-relevant account of the level, dynamics, composition and macroeconomic correlates of investment across the EU member states, with particular attention to:

  • the post-global-financial-crisis investment weakness,
  • the shift toward intangible (knowledge-based) capital,
  • the effect of successive crises, and
  • Slovenia’s relative position against the EU, the euro area and its Central and Eastern European peers.

The analysis proceeds from data preparation through descriptive statistics, turning-point and convergence analysis, a cluster-based country typology, fixed-effects panel models of investment drivers, a Slovenia benchmarking exercise, and robustness checks.

Interpretation note. The panel-regression estimates in this report are conditional associations, not causal effects. Investment and its drivers are jointly determined. The report relies on the sign and the robustness of relationships across specifications rather than on precise magnitudes.

2 Data and sources

The dataset is a balanced country–year panel for the EU member states, 2000–2025, assembled in PKP2_DATABASE.xlsx (sheet MASTER PANEL). One row is one country in one year; one indicator is one column. The workbook also carries a variable dictionary, a country list, and the raw source extracts.

2.1 Source of each indicator

All sources are documented in the workbook’s variable dictionary and are reproduced verbatim below. In the figures, the following short source names are used:

Documented source Source note used in this report
Eurostat (national accounts, R&D, HICP, government finance, demography, education) Source: Eurostat.
Global INTAN-Invest Source: Global INTAN-Invest.
World Uncertainty Index Source: World Uncertainty Index.
DBnomics / AMECO Source: AMECO (European Commission).
World Bank – Worldwide Governance Indicators Source: World Bank, Worldwide Governance Indicators.
Constructed (institutional-quality index) combined WGI + author's calculations
Variable dictionary (from PKP2_DATABASE.xlsx)
Variable Indicator Category Source Dataset code Unit Coverage
total_gfcf_gdp Total GFCF / GDP Investment outcomes & composition Eurostat nama_10_gdp % of GDP 2000–2025
business_investment Business investment / GDP Investment outcomes & composition Eurostat nasa_10_ki % of GDP 2000–2025 where available
real_gfcf_growth Real GFCF growth Investment outcomes & composition Eurostat nama_10_an6 Annual % change Chain linked volumes, percentage change on previous period 2000–2025 where available
ipp_share_gfcf IPP share of GFCF Investment outcomes & composition Eurostat nama_10_an6 % of GFCF 2000–2025 where available
productive_nonres_inv_gdp Productive (non-residential) investment / GDP Investment outcomes & composition Eurostat nama_10_an6 + existing total_gfcf_gdp % of GDP 2000–2025 where available
broad_intangible_investment Broad intangible investment / GDP Investment outcomes & composition Global INTAN-Invest All Countries Annual database % of GDP 2010–2025 in the combined All Countries file; some country-year gaps may remain
ipp_inv_gdp IPP investment / GDP Investment outcomes & composition Eurostat nama_10_an6, IPP asset category % of GDP 2000–2025 where available
business_rd_gdp Business R&D expenditure / GDP Investment outcomes & composition / Human capital & innovation Eurostat rd_e_gerdtot % of GDP 2000–2024 where available
machinery_equipment_inv_gdp Investment in machinery and equipment / GDP Investment outcomes & composition Eurostat nama_10_an6, asset AN113_AN114 % of GDP 2000–2025 where available
buildings_structures_inv_gdp Investment in buildings and structures / GDP Investment outcomes & composition Eurostat nama_10_an6, asset AN112 % of GDP 2000–2025 where available
ict_equipment_inv_gdp Investment in ICT equipment / GDP Investment outcomes & composition Eurostat nama_10_an6, asset AN1132 % of GDP 2000–2025 where available
real_gdp_growth Real GDP growth Demand & macroeconomic conditions Eurostat National accounts Annual % change 2000–2025 where available
inflation_hicp Inflation (HICP) Demand & macroeconomic conditions Eurostat HICP Annual % change 2000–2025 where available
economic_uncertainty Economic uncertainty Demand & macroeconomic conditions World Uncertainty Index Country-level WUI Index Coverage varies
business_energy_prices Business energy prices Demand & macroeconomic conditions Eurostat nrg_pc_205 EUR per kWh Coverage varies
real_interest_rate Real interest rate Financing conditions DBnomics / AMECO AMECO/ILRV % 2000–2025 where available
loans_to_NFI Loans to non-financial corporations (% GDP) Financing conditions Eurostat tipspd25 % of GDP 2000–2025 where available
corporate_profitability Corporate profitability Financing conditions Eurostat nasa_10_ki % 2000–2025 where available
government_inv_gdp Government investment / GDP Fiscal & public-sector conditions Eurostat sdg_08_11 / institutional-sector GFCF % of GDP 2000–2025 where available
public_debt_gdp Public debt / GDP Fiscal & public-sector conditions Eurostat gov_10dd_edpt1 % of GDP 2000–2025 where available
regulatory_quality Regulatory Quality Institutional quality & governance World Bank Worldwide Governance Indicators (WGI) WGI estimate 2000–2025 where available
government_effectiveness Government Effectiveness Institutional quality & governance World Bank Worldwide Governance Indicators (WGI) WGI estimate 2000–2025 where available
control_corruption Control of Corruption Institutional quality & governance World Bank Worldwide Governance Indicators (WGI) WGI estimate 2000–2025 where available
rule_of_law Rule of Law Institutional quality & governance World Bank Worldwide Governance Indicators (WGI) WGI estimate 2000–2025 where available
political_stability Political Stability Institutional quality & governance World Bank Worldwide Governance Indicators (WGI) WGI estimate 2000–2025 where available
institutional_quality_index Institutional Quality Index Institutional quality & governance Constructed Selected WGI indicators Constructed index Depends on WGI coverage
manufacturing_share_gva Manufacturing share of GVA Structural characteristics Eurostat nama_10_a64 % of GVA 2000–2025 where available
gdp_per_capita_pps GDP per capita PPS Structural characteristics Eurostat tec00114 PPS / index (confirm exact Eurostat representation) 2000–2025 where available
trade_openness Trade openness Structural characteristics Eurostat P6 exports, P7 imports, B1GQ GDP % of GDP 2000–2025 where available
working_age_population_growth Working-age population growth Structural characteristics Eurostat demo_pjanbroad; age=Y15-64; sex=T; unit=NR Annual % change 2000–2025 where available
tertiary_education Human capital / tertiary education Human capital & innovation Eurostat edat_lfse_03 Confirm source unit / age group 2000–2025 where available

2.2 Coverage and known gaps

  • Broad intangible investment (Global INTAN-Invest) starts in 2010 and is entirely missing for Cyprus, Greece and Malta; the intangibles analysis covers about 24 countries.
  • The World Bank governance indicators end in 2024 (no 2025 value).
  • Business energy prices start in 2007; business investment has no 2025 value for most countries.
  • The World Uncertainty Index has no data for Cyprus, Estonia, Luxembourg or Malta; the real interest rate is missing for Czechia. The full 14-driver panel model is therefore identified on 21 countries.
  • Ireland and Luxembourg show mechanically extreme values (Ireland 2019: gross fixed capital formation 53 percent of GDP, intellectual-property share 74 percent) driven by multinational balance-sheet relocation. They are excluded from EU aggregates and cross-country analysis by default and re-introduced only for robustness.

3 Data preparation

Panel validation checks
check result
Countries 27
Years 2000–2025
Rows 702
Missing country-years 0
Duplicate country-years 0
Period label vs year rule mismatches 0
slovenia_flag errors 0
irl_lux_flag errors 0
Coverage of each indicator in the master panel
Variable Non-missing Percent First year Last year
broad_intangible_investment 364 51.9 2010 2025
business_energy_prices 512 72.9 2007 2025
economic_uncertainty 592 84.3 2000 2025
regulatory_quality 648 92.3 2000 2024
government_effectiveness 648 92.3 2000 2024
control_corruption 648 92.3 2000 2024
rule_of_law 648 92.3 2000 2024
political_stability 648 92.3 2000 2024
institutional_quality_index 648 92.3 2000 2024
real_interest_rate 649 92.5 2000 2025
business_rd_gdp 664 94.6 2000 2024
manufacturing_share_gva 680 96.9 2000 2025
business_investment 682 97.2 2000 2025
corporate_profitability 682 97.2 2000 2025
government_inv_gdp 682 97.2 2000 2025
tertiary_education 696 99.1 2000 2025
ict_equipment_inv_gdp 698 99.4 2000 2025
ipp_share_gfcf 699 99.6 2000 2025
ipp_inv_gdp 699 99.6 2000 2025
machinery_equipment_inv_gdp 699 99.6 2000 2025
working_age_population_growth 700 99.7 2000 2025
real_gfcf_growth 701 99.9 2000 2025
real_gdp_growth 701 99.9 2000 2025
year 702 100.0 2000 2025
slovenia_flag 702 100.0 2000 2025
irl_lux_flag 702 100.0 2000 2025
total_gfcf_gdp 702 100.0 2000 2025
productive_nonres_inv_gdp 702 100.0 2000 2025
buildings_structures_inv_gdp 702 100.0 2000 2025
inflation_hicp 702 100.0 2000 2025
loans_to_nfi 702 100.0 2000 2025
public_debt_gdp 702 100.0 2000 2025
gdp_per_capita_pps 702 100.0 2000 2025
trade_openness 702 100.0 2000 2025

4 Variable construction

4.1 Country groupings

Groupings used throughout:

  • EU aggregate – unweighted mean across countries, excluding Ireland and Luxembourg (multinational distortion). An EU-27 variant is shown for robustness.
  • EU15 vs member states that joined in 2004 or later.
  • Euro area – time-varying membership (a country enters in its accession year).
  • Central and Eastern Europe (CEE) – used for region flags and figures throughout; Slovenia’s own benchmark peers are the cluster-derived structural peers below, not a separately chosen CEE peer set.
  • Structural peers and aspiration countries for Slovenia – both are full cluster memberships from the cluster analysis (Section 7), not a hand-picked subset: structural peers are every other member of Slovenia’s own cluster (“CEE and Baltic”), and the aspiration group is every member of the “Western and Nordic core” cluster. Both are only defined once the clustering has run, and are listed in full in Section 8.2.
  • Regions – Nordic, Baltic, Southern, CEE (non-Baltic), Western/Core.

4.2 Constructed indicators

Institutional-quality index: rebuilt vs workbook
Correlation with workbook index Mean (workbook) Mean (rebuilt, z-score)
0.983 1.064 0

Two indicators are constructed here rather than taken directly:

  1. Institutional-quality index – the pooled z-score mean of the five World Bank governance pillars (regulatory quality, government effectiveness, control of corruption, rule of law, political stability). It correlates 0.98 with the workbook’s own institutional_quality_index; results are unchanged whichever is used.
  2. Real investment volume index (2007 = 100) – the cumulative product of each country’s chain-linked real gross-fixed-capital-formation growth, rebased to its 2007 level, used to compare the depth and length of the post-crisis fall.

5 Descriptive statistics

5.1 Investment intensity: level and comparison

Source: Eurostat.

EU investment intensity rose through the mid-2000s boom, collapsed in 2008–09, and then spent the 2010s well below its earlier level. The member states that joined in 2004 or later invested far more than the EU15 during the boom,driven by housing expansion and economic catch-up, but lost that lead entirely in the crash.

Source: Eurostat.

Relative to their own 2000–07 average, Greece, Spain, Estonia, Latvia, Slovakia and Slovenia show the largest shortfalls; Sweden, Belgium and Denmark are the main countries investing more now than before the crisis.

Source: Eurostat.

5.2 Investment dynamics and common turning points

Source: Eurostat.

2009 was a universal contraction (all countries); 2020 hit about three quarters of them. 2010–2013 was a run of broad-based declines rather than a single event, and a further broad contraction appears in 2024.

Source: Eurostat.

Core/Western and Nordic countries regained their pre-crisis real investment level by the mid-2010s; the South and Slovenia only around 2021–22. The post-crisis weakness was a persistent shortfall in one part of the EU, not a shared temporary dip.

5.3 Composition: tangible and intangible capital

Source: Eurostat.

The four national-accounts asset types (machinery and equipment, buildings and structures, intellectual-property products, ICT equipment) are tracked across the five sub-periods used throughout this report (the professor’s classification: 2000–07, 2008–12, 2013–19, 2020–21, 2022–25), first as the sample average and then across countries.

Source: Eurostat; author’s calculations.

Across the five periods the pattern is clear: machinery and equipment and buildings and structures both drift down (from about 8.7 and 7.6 percent of GDP before the crisis to about 7.1 and 6.5 in 2022–25), ICT equipment edges lower (partly a relative-price effect), and only intellectual-property products rise steadily (from about 2.4 to 3.9). The shift toward intangibles is as much a tangible decline as an intangible rise.

Countries ordered by full-sample average total investment rate; same order in every panel. Excludes Ireland and Luxembourg. Source: Eurostat; author’s calculations.

Panels B–F show that the average trajectory hides large cross-country differences. Machinery and buildings investment is consistently heaviest in the CEE and Baltic economies (top of each panel) and lightest in the South; the intellectual-property column darkens over time for the Western and Nordic core but much less for CEE, and ICT is a thin slice everywhere. Reading down a single column across the five panels shows how each country’s asset mix evolved.

5.3.1 Composition trajectories by asset type and country group

To read the change in the asset mix directly, the next figure follows all four asset types – machinery and equipment, buildings and structures, intellectual-property products and ICT equipment – through the five sub-periods for the three investment clusters identified in the cluster analysis (see Section 7), with Slovenia held out as a separate benchmark. Each panel has its own vertical scale, so trajectory shapes are comparable even though the asset levels are not.

Source: Eurostat; author’s calculations.

Machinery and intellectual-property investment by country group and sub-period, with change relative to 2000-07 and a compositional-shift indicator (percent of GDP; a positive shift means the group moved relatively toward intellectual property)
Regional group Period Machinery & equip. (% GDP) IPP (% GDP) Machinery chg. vs 2000-07 IPP chg. vs 2000-07 Compositional shift (IPP chg. - machinery chg.)
Slovenia 2000-07 10.46 2.72 0.00 0.00 0.00
Slovenia 2008-12 8.08 3.02 -2.38 0.29 2.68
Slovenia 2013-19 7.84 3.00 -2.62 0.27 2.89
Slovenia 2020-21 7.80 3.15 -2.66 0.43 3.09
Slovenia 2022-25 8.00 3.18 -2.46 0.45 2.91
Western and Nordic core 2000-07 7.15 3.98 0.00 0.00 0.00
Western and Nordic core 2008-12 6.38 4.37 -0.77 0.39 1.16
Western and Nordic core 2013-19 6.38 4.68 -0.78 0.70 1.48
Western and Nordic core 2020-21 6.20 5.16 -0.95 1.19 2.14
Western and Nordic core 2022-25 6.34 5.49 -0.81 1.51 2.32
CEE and Baltic 2000-07 11.02 1.62 0.00 0.00 0.00
CEE and Baltic 2008-12 8.79 1.90 -2.23 0.27 2.50
CEE and Baltic 2013-19 8.61 2.21 -2.41 0.58 2.99
CEE and Baltic 2020-21 8.19 2.99 -2.83 1.36 4.19
CEE and Baltic 2022-25 8.38 2.75 -2.63 1.12 3.76
Mediterranean 2000-07 7.32 1.60 0.00 0.00 0.00
Mediterranean 2008-12 6.21 2.14 -1.11 0.54 1.65
Mediterranean 2013-19 5.87 2.75 -1.45 1.14 2.60
Mediterranean 2020-21 5.88 3.42 -1.44 1.81 3.25
Mediterranean 2022-25 6.13 3.83 -1.19 2.22 3.41
Source: Eurostat; author’s calculations.
Heterogeneity within the CEE and Baltic cluster: machinery and intellectual-property investment, 2000-07 versus 2022-25, by country (percent of GDP; Slovenia included here for reference although it is plotted separately in the figure)
Country Machinery 2000-07 Machinery 2022-25 IPP 2000-07 IPP 2022-25 Machinery chg. IPP chg. Compositional shift
EST 13.80 6.58 1.26 3.92 -7.23 2.66 9.89
CZE 12.74 9.57 2.59 5.05 -3.16 2.46 5.63
LVA 13.79 9.55 1.26 2.38 -4.24 1.11 5.35
SVK 11.96 8.47 1.96 2.10 -3.49 0.14 3.63
BGR 10.90 8.25 1.36 2.05 -2.65 0.69 3.34
SVN 10.46 8.00 2.72 3.18 -2.46 0.45 2.91
LTU 8.55 7.43 1.29 2.68 -1.12 1.39 2.51
ROU 11.47 9.80 1.65 2.25 -1.67 0.60 2.27
HRV 8.81 8.25 1.44 2.88 -0.56 1.44 2.00
POL 8.29 6.68 1.26 1.50 -1.61 0.24 1.85
HUN 9.86 9.25 2.18 2.67 -0.61 0.50 1.11
Source: Eurostat; author’s calculations.

Every group has moved the same direction – machinery and equipment down, intellectual property up – so the compositional-shift indicator is positive for all four and rises through the five sub-periods. The mechanism differs, though. For the CEE and Baltic cluster and the Mediterranean cluster the shift is mostly a machinery decline (about –2.6 and –1.2 points of GDP since 2000-07) with only a modest intellectual-property rise; for the Western and Nordic core the machinery fall is smaller (about –0.8) and the intellectual-property gain larger (+1.5), from an already high base (4.0 to 5.5 percent of GDP). Slovenia tracks between the CEE average and the core on the machinery side but its intellectual-property investment is almost flat (2.7 to 3.2 percent of GDP), so its compositional shift is the smallest of the four. The country-level table shows the CEE and Baltic average hides a wide range: the shift is largest for Estonia, Czechia and Latvia (machinery-heavy economies that de-industrialised their investment mix fastest) and smallest for Hungary, Poland and Croatia.

The other two panels reinforce the picture. Buildings and structures investment falls in every group except the core (steepest in Slovenia, about 10.0 to 7.4 percent of GDP, and in the Mediterranean, 6.7 to 4.9), so the tangible decline is in construction as well as machinery. ICT equipment is a thin and gently declining slice everywhere – below 2 percent of GDP throughout and converging near 0.8 – so the rotation toward knowledge-based capital is almost entirely an intellectual-property phenomenon, not an ICT-hardware one.

Source: Eurostat; Global INTAN-Invest.

National accounts capture only part of intangible capital. Where the broader INTAN-Invest measure exists, broad intangibles exceed measured intellectual-property products in every country.

Investment composition by country, 2022-25 (percent of GDP; intellectual-property share in percent of gross fixed capital formation)
country Machinery Buildings ICT IPP Business R&D Broad intangible IPP share of GFCF
Ireland 5.1 3.3 0.6 10.3 1.3 14.1 45.8
Denmark 5.4 6.1 0.9 7.3 1.9 14.9 31.2
Sweden 7.5 6.6 1.2 7.2 2.6 17.1 28.7
Malta 6.4 3.9 0.5 6.0 0.4 30.2
Austria 7.4 5.4 1.0 5.8 2.2 9.3 23.7
Belgium 7.9 5.4 1.3 5.3 2.4 13.4 22.1
Finland 5.5 6.9 0.4 5.2 2.1 13.9 22.8
France 5.2 6.1 0.4 5.1 1.4 14.4 22.4
Czechia 9.6 6.6 1.1 5.0 1.2 10.8 18.6
Netherlands 5.5 5.3 0.8 4.1 1.6 12.7 20.3
Cyprus 4.5 4.2 0.5 4.0 0.3 19.1
Spain 5.7 4.9 0.7 4.0 0.8 8.0 19.3
Estonia 6.6 9.9 0.9 3.9 1.1 11.2 15.7
Germany 6.3 4.2 0.6 3.9 2.1 18.6
Portugal 5.8 7.0 0.8 3.3 1.1 9.4 16.3
Slovenia 8.0 7.4 0.8 3.2 1.5 10.0 14.7
Italy 7.5 5.7 0.8 3.0 0.8 8.4 13.4
Croatia 8.2 9.3 0.9 2.9 0.8 9.6 11.8
Greece 7.0 3.9 1.5 2.8 0.8 17.5
Lithuania 7.4 9.1 1.1 2.7 0.5 11.4 11.7
Hungary 9.2 8.8 0.6 2.7 1.0 8.7 11.0
Latvia 9.6 8.6 0.7 2.4 0.3 9.3 10.3
Romania 9.8 10.5 0.8 2.2 0.3 9.5 8.7
Slovakia 8.5 6.2 0.7 2.1 0.6 10.4 10.2
Bulgaria 8.2 4.7 0.2 2.0 0.5 11.2 11.2
Luxembourg 4.3 6.3 0.6 1.8 0.5 12.8 11.1
Poland 6.7 6.8 0.4 1.5 0.9 8.3 8.8
Source: Eurostat; Global INTAN-Invest.

6 Main analysis

6.1 Structural breaks and turning points

Source: Eurostat; AMECO; author’s calculations.

The interest-rate series is shown as a vertically aligned mini-panel rather than on a secondary axis: the two series measure different things and share only the time axis, so overlaying them (or a dual scale) would invite a spurious visual correlation. It is a euro-area average because the investment aggregate covers the broader EU.

Source: Eurostat; author’s calculations.

Source: Eurostat; author’s calculations.

Slovenia’s position among its peers inverts after the crisis. In the two pre-crisis regimes it ranked 2nd of six – just behind the Baltics, well above the CEE (non-Baltic) group it would otherwise belong to (about 26 percent of GDP versus 24 percent in 2000–05, and 29 versus 27 percent in 2006–08). From 2012 onward it falls to 5th of six, ahead of only the Southern group (19.0 versus 16.5 percent of GDP in 2012–18; 20.7 versus 19.4 in 2019–25) – consistently below the CEE (non-Baltic) group average it used to lead. The pattern reinforces the country-level view in Panels B–F: Slovenia’s investment weakness after the global financial crisis is not just an EU-wide phenomenon but a fall relative to its own regional peers.

Source: Eurostat; author’s calculations.

Panel A. Bai–Perron breakpoints split the EU mean into five regimes – roughly 23.4 percent of GDP (2000–05), 25.7 (2006–08), 21.5 (2009–11), 20.4 (2012–18, the trough), and 22.0 (2019–25). The euro-area real interest rate below tracks the same timeline: it was low in the mid-2000s boom, spiked in 2009 and stayed high through the 2010–13 sovereign-debt regime (the deepest investment trough), turned negative during the 2014–21 recovery as inflation picked up and policy eased, and fell sharply further in 2022–23 before normalising – consistent with financing conditions being tight exactly when investment was weakest.

Panels B–F. The regime means hide wide cross-country heterogeneity. In the 2006–08 boom the Baltic economies (filled triangles) sat far to the right – investment rates near or above 30 percent of GDP – while the Southern group (filled squares) was already at or below the EU mean; by 2012–18 the ranking had inverted, with the Baltics and much of the CEE group falling below the line and only a handful of Western/Core and Nordic countries clearly above it. The spread within each period is 10–15 percentage points of GDP throughout, and individual countries move across the distribution between regimes.

At country level, the global financial crisis is the dominant common break (most countries break in 2008–10), with a secondary cluster in 2015–17.

6.2 Convergence or divergence?

Source: Eurostat; author’s calculations.

Source: Eurostat; author’s calculations.

Beta-convergence in total investment / GDP, by window (level-change specification)
Window n Slope (beta) HC3 s.e. p R-squared
2000–2025 25 -1.020 0.227 0.00 0.641
2000–2007 25 -0.839 0.297 0.01 0.264
2007–2019 25 -0.838 0.112 0.00 0.593
2013–2025 25 -0.522 0.098 0.00 0.501
Source: Eurostat; author’s calculations.
  • Sigma-convergence is cyclical, not trending: dispersion widened in the 2000–08 boom, widened again in the divergent 2009–13 recovery, and only genuinely narrowed after 2014. Intellectual-property investment is the exception – its dispersion has risen since about 2013.
  • Beta-convergence is negative and significant in every window (full-period slope about –1.0): the countries that started with higher investment ratios fell most. This “catch-down” is driven by the boom–bust of the high-flyers, which is why beta-convergence coexists with weak sigma-convergence.
  • Slovenia sits essentially on the beta-convergence line – its fall of about 6.5 percentage points is close to what its high (about 27 percent) starting point predicts.

7 Cluster analysis

## *** : The Hubert index is a graphical method of determining the number of clusters.
##                 In the plot of Hubert index, we seek a significant knee that corresponds to a 
##                 significant increase of the value of the measure i.e the significant peak in Hubert
##                 index second differences plot. 
## 
## *** : The D index is a graphical method of determining the number of clusters. 
##                 In the plot of D index, we seek a significant knee (the significant peak in Dindex
##                 second differences plot) that corresponds to a significant increase of the value of
##                 the measure. 
##  
## ******************************************************************* 
## * Among all indices:                                                
## * 4 proposed 2 as the best number of clusters 
## * 9 proposed 3 as the best number of clusters 
## * 3 proposed 4 as the best number of clusters 
## * 2 proposed 5 as the best number of clusters 
## * 3 proposed 6 as the best number of clusters 
## * 6 proposed 7 as the best number of clusters 
## 
##                    ***** Conclusion *****                            
##  
## * According to the majority rule, the best number of clusters is  3 
##  
##  
## *******************************************************************
## quartz_off_screen 
##                 2

Ward hierarchical clustering and k-means are run on 13 standardised 2013–25 investment features (level, composition, dynamics, and structural controls), excluding Ireland and Luxembourg. NbClust and the dendrogram both point to 3 clusters; Ward and k-means agree (adjusted Rand index = 1).

Source: Eurostat; author’s calculations.

Source: Eurostat; author’s calculations.

Source: Eurostat; author’s calculations.

Cluster profiles (Ward linkage)
Cluster n Members Investment / GDP IPP share of GFCF Business R&D / GDP GDP per capita (PPS) Investment volatility
Mediterranean 6 CYP ESP GRC ITA MLT PRT 18.1 17.6 0.6 28443.6 9.9
CEE and Baltic 11 BGR CZE EST HRV HUN LTU LVA POL ROU SVK SVN 22.2 11.3 0.7 24218.3 6.5
Western and Nordic core 8 AUT BEL DEU DNK FIN FRA NLD SWE 22.5 22.2 1.9 38549.6 4.2
Source: Eurostat; author’s calculations.

The two axes of the biplot have a simple reading. Component 1 (34 percent of the variance) is a physical-versus-knowledge-capital axis: countries on the left invest heavily in buildings, machinery and public infrastructure (loadings around –0.4), those on the right have a higher R&D and intellectual-property share and higher income (loadings around +0.2 to +0.3). Component 2 (27 percent) is a stability axis: the bottom of the chart is volatile, fast-growing economies (investment growth and volatility load around –0.4), the top is the stable, high-R&D, high-investment core (R&D and investment level load around +0.4). Slovenia sits close to the centre – marginally on the physical-capital side, neutral on stability.

Three groups emerge, aligning with EU geography but sharpening it into an investment typology:

  1. Mediterranean – lowest investment intensity, low business R&D, high volatility.
  2. CEE and Baltic (including Slovenia) – decent investment intensity but tilted toward tangible/construction and public investment; the lowest intellectual-property share and below-average business R&D.
  3. Western and Nordic core – high intensity, high intangible share and R&D, high income, low volatility.

Source: Eurostat; author’s calculations.

Slovenia’s closest peers by Euclidean distance are Slovakia, Poland, Croatia, Lithuania, Hungary, Bulgaria and Czechia – all CEE. This nearest-neighbour ranking is a diagnostic, not the benchmark itself: Section 8.2 instead benchmarks Slovenia against the full “CEE and Baltic” cluster it belongs to (10 countries: BGR, CZE, EST, HRV, HUN, LTU, LVA, POL, ROU, SVK), so the comparison is not sensitive to where a manually chosen cutoff is drawn.

8 Additional analyses

8.1 Macroeconomic correlates of investment

Source: Eurostat; AMECO; World Uncertainty Index; World Bank Worldwide Governance Indicators; author’s calculations.

8.1.1 Panel data and one-year lags

The Section 8.1 analysis dataset has 650 country-years (25 countries, 2000–2025) with no duplicate country-year rows. Constructing the one-year lag removes the first observed year for each country – 25 observations lost – leaving 625 country-years with a valid \(t-1\) regressor. The full 14-variable model is then identified on fewer rows still, because Cyprus, Estonia and Malta have no World Uncertainty Index and Czechia no real interest rate (listwise deletion; see below).

8.1.2 Standardising the (lagged) explanatory variables

Because the explanatory variables are standardised, each estimated coefficient is the percentage-point change in total investment (as a share of GDP) that is associated with a one-standard-deviation increase in the lagged explanatory variable, conditional on the other regressors and the fixed effects. The dependent variable is not standardised (it is already a percentage of GDP).

8.1.3 Baseline two-way fixed-effects model

The baseline is a two-way (country and year) fixed-effects panel regression of the general form

\[ \Big(\tfrac{I}{Y}\Big)_{i,t}=\alpha_i+\lambda_t +\beta_1\,\widetilde{\text{GDPgrowth}}_{i,t-1} +\beta_2\,\widetilde{\text{RealRate}}_{i,t-1} +\beta_3\,\widetilde{\text{Credit}}_{i,t-1} +\beta_4\,\widetilde{\text{GovInv}}_{i,t-1} +\beta_5\,\widetilde{\text{PublicDebt}}_{i,t-1} +\gamma'\widetilde{\text{Controls}}_{i,t-1}+\varepsilon_{i,t}, \]

where \(i\) indexes country and \(t\) year; \(\alpha_i\) (country fixed effects) absorb time-invariant country characteristics; \(\lambda_t\) (year fixed effects) absorb shocks common to all countries in a given year; a tilde denotes a standardised variable; and every right-hand-side variable enters at \(t-1\). Standard errors are clustered by country. The symbols map to these dataset variables (all lagged one year and standardised):

Symbol Dataset variable
\(\widetilde{\text{GDPgrowth}}\) real_gdp_growth
\(\widetilde{\text{RealRate}}\) real_interest_rate (real long-term rate, AMECO)
\(\widetilde{\text{Credit}}\) loans_to_nfi (loans to non-financial corporations, % of GDP)
\(\widetilde{\text{GovInv}}\) government_inv_gdp
\(\widetilde{\text{PublicDebt}}\) public_debt_gdp
\(\gamma'\widetilde{\text{Controls}}\) inflation_hicp, economic_uncertainty, corporate_profitability, institutional_quality_index, manufacturing_share_gva, trade_openness, working_age_population_growth, tertiary_education, gdp_per_capita_pps

The baseline is identified on 479 country-years across 21 countries (2001–2025): EU-27 minus Ireland and Luxembourg (multinational-distorted), minus Cyprus, Estonia, Malta (no uncertainty index) and Czechia (no real interest rate), all dropped by listwise deletion on the lagged regressors.

These are conditional associations, not causal effects. Lagging the regressors one year and including country and year fixed effects mitigates contemporaneous simultaneity – in particular the mechanical link between investment and GDP growth, since investment is a component of GDP – and time-invariant omitted heterogeneity. They do not establish causal identification: persistent shocks, longer-horizon reverse causality and time-varying confounders remain possible. The estimates describe within-country co-movement, net of shocks common to all countries in a year.

8.1.3.1 Simultaneity diagnostic: contemporaneous versus one-year-lagged

Source: Eurostat; AMECO; World Uncertainty Index; World Bank Worldwide Governance Indicators; author’s calculations.

The GDP-growth coefficient is 0.89 with the variable measured contemporaneously and 0.98 (95% CI [0.28, 1.67]) when lagged one year – essentially unchanged. Had the contemporaneous coefficient been driven by the accounting identity, lagging would have collapsed it; instead the positive within-country association between past demand growth and current investment persists. This is consistent with an accelerator mechanism but is not an estimated causal effect.

8.1.4 Regression table

Baseline two-way fixed-effects model of total investment (% of GDP) on one-year-lagged, standardised macroeconomic variables, 21 EU countries, 2000-2025
Main variables Baseline (+ controls)
Real GDP growth 1.122*** 0.977***
(0.332) (0.332)
Real interest rate -0.694*** -0.660***
(0.185) (0.231)
Corporate credit 0.809 1.316*
(0.619) (0.736)
Government investment 0.730*** 0.808***
(0.208) (0.227)
Public debt -1.888*** -1.797**
(0.650) (0.797)
Inflation (HICP) 0.518
(0.340)
Economic uncertainty -0.009
(0.089)
Corporate profitability 0.776*
(0.392)
Institutional quality 0.588
(0.944)
Manufacturing share 0.335
(0.582)
Trade openness -2.000**
(0.926)
Working-age population growth 0.830***
(0.253)
Tertiary education 0.280
(0.812)
GDP per capita -0.713
(0.912)
Num.Obs. 570 479
R2 Within 0.432 0.537
Dependent variable Total investment, % of GDP Total investment, % of GDP
Country fixed effects Yes Yes
Year fixed effects Yes Yes
Number of countries 24 21
Standard errors Clustered by country Clustered by country
* p < 0.1, ** p < 0.05, *** p < 0.01
Coefficients are percentage-point changes in total investment / GDP associated with a one-standard-deviation increase in the lagged explanatory variable. Country and year fixed effects included; standard errors clustered by country. Estimates are conditional associations, not causal effects.
Source: Eurostat; AMECO; World Uncertainty Index; World Bank Worldwide Governance Indicators; author’s calculations.

8.1.5 Coefficient plot

Notes. Dependent variable: total investment as % of GDP. Explanatory variables are standardised before estimation, so coefficients represent percentage-point changes in investment associated with a one-standard-deviation change in the lagged explanatory variable. Country and year fixed effects are included; standard errors are clustered by country. The six structural / institutional controls (institutional quality, manufacturing share, trade openness, working-age population growth, tertiary education, GDP per capita) are in the model but are not shown. The estimates are conditional (within-country) associations, not causal effects.

8.1.6 Subperiod robustness: 2000-2008, 2009-2019, 2020-2025

The supervisor asked whether a single coefficient over 25 years is meaningful given the major macroeconomic regimes in the sample. The baseline (main variables, one-year lags) is therefore re-estimated separately within three regime windows – the pre-GFC expansion (2000–2008), the post-GFC adjustment and recovery (2009–2019), and the pandemic / energy-price shock (2020–2025) – alongside the pooled estimate. Only the five main variables are used, because the slow-moving structural controls cannot be identified in windows this short. This is a stability check, not a formal structural-break test.

Source: Eurostat; AMECO; World Uncertainty Index; author’s calculations.

Main-variable coefficients by regime window (pp of GDP per 1 s.d. of the lagged variable; * p<0.05, ^ p<0.10)
Variable Pooled 2000-2025 2000-2008 2009-2019 2020-2025
Real GDP growth 1.12* 0.02 0.73^ 0.25
Real interest rate -0.69* -1.21* -0.12 -0.16
Corporate credit 0.81 -0.49 0.69 0.76
Government investment 0.73* 1.02* 0.45^ 1.19*
Public debt -1.89* -3.35* -2.13* -2.62*
Two-way fixed effects; SEs clustered by country. Conditional associations, not causal effects.

The principal associations are reasonably stable across the three regimes. Government investment is positively and significantly associated with total investment in every window, and the public-debt association is negative and significant in every window (largest, around –3 pp per s.d., in the debt-crisis years 2000–2008). Lagged GDP growth is positively signed in the post-2008 windows but small and imprecise in 2000–2008 and 2020–2025, where there is little within-country demand variation to identify it. The real-interest-rate and credit coefficients are the least stable – signed as in the pooled model but significant only in individual windows (the real rate in 2000–2008, when rate dispersion was widest). No window reverses a principal sign, so the pooled estimates are not an artefact of one macroeconomic regime; a formal structural-break test would be needed to say more.

8.1.7 Economic uncertainty: with and without year fixed effects

Economic uncertainty is retained in the baseline even though it is statistically insignificant. To see why it carries no independent within-country signal, the uncertainty coefficient is compared with and without year fixed effects, both on its own and inside the full baseline.

Economic uncertainty (t-1): estimated coefficient with and without year fixed effects
Specification Coefficient Clustered SE t-value p-value 95% CI
Uncertainty only, country FE -0.503 0.192 -2.63 0.016 [-0.90, -0.10]
Uncertainty only, country + year FE 0.002 0.154 0.01 0.992 [-0.32, 0.32]
Full baseline, country FE 0.006 0.081 0.07 0.945 [-0.16, 0.18]
Full baseline, country + year FE -0.009 0.089 -0.10 0.923 [-0.19, 0.18]
Dependent variable: total investment, % of GDP. SEs clustered by country. Conditional association, not a causal effect.

On its own, lagged uncertainty has a sizeable negative within-country association with investment (about −0.5 pp of GDP per standard deviation) when only country fixed effects are included; adding year fixed effects collapses it to essentially zero. Because year effects absorb whatever is common to all countries in a given year, this shows that the apparent uncertainty–investment relationship is a common-shock phenomenon – the 2009, 2020 and 2022 spikes hit every country at once – rather than a within-country association that identifies uncertainty separately from the global cycle. Inside the full baseline the coefficient is already near zero even without year effects, because the other cyclical variables (notably lagged GDP growth) absorb the same content.

8.1.8 Structural and institutional factors (between-country)

Slow-moving structural variables cannot be identified within countries (GDP per capita, education and trade openness are collinear at r around 0.7–0.9). A cross-section on 2013–25 country means:

Structural and institutional correlates of investment: cross-section on 2013-25 country means (drivers standardised)
Total investment / GDP Business investment / GDP &nbsp;IPP investment / GDP
Institutional quality 3.309** 2.333* 0.507
(1.462) (1.194) (0.318)
GDP per capita (PPS) -0.840 -0.560 0.726**
(1.525) (1.245) (0.332)
Manufacturing share of GVA -0.504 -0.610 -0.407**
(0.745) (0.608) (0.162)
Trade openness 0.855 0.535 -0.243
(0.686) (0.560) (0.149)
Tertiary education -1.517* -1.191* -0.110
(0.826) (0.675) (0.180)
Working-age population growth -0.698 -0.418 0.313
(0.865) (0.706) (0.188)
Num.Obs. 25 25 25
R2 0.345 0.296 0.858
R2 Adj. 0.127 0.061 0.811
F 1.584 1.259 18.138
* p < 0.1, ** p < 0.05, *** p < 0.01
Source: Eurostat; World Bank Worldwide Governance Indicators; author’s calculations.

Institutional quality is the main structural correlate of the investment level (about +3.3 percentage points of GDP per standard deviation for total investment). Intellectual-property investment is almost entirely structural: the cross-country R-squared is about 0.86, driven by GDP per capita and a smaller manufacturing share.

8.2 Slovenia benchmarked

8.2.1 Slovenia against structural peers and aspiration countries

Structural peers describe the countries Slovenia most resembles; they are not necessarily where it should aim. Both benchmark groups below are full cluster memberships from the Section 7 cluster analysis (feat$cluster) – no country is hand-picked, and neither list is chosen or trimmed to make a point. Structural peers are every other member of Slovenia’s own cluster, “CEE and Baltic”: 10 countries – BGR, CZE, EST, HRV, HUN, LTU, LVA, POL, ROU, SVK. The aspiration group is every member of the “Western and Nordic core” cluster – the most developed of the three clusters on investment level, intangible share and income (Section 7): 8 countries – AUT, BEL, DEU, DNK, FIN, FRA, NLD, SWE. All five indicators are existing project series expressed as a percent of GDP (total_gfcf_gdp, business_investment, ipp_inv_gdp, government_inv_gdp, business_rd_gdp).

Data coverage for the Slovenia / structural-peer / aspiration comparison, before plotting
Indicator Years used SVN obs Structural-peer obs Aspiration obs Structural peers complete Aspiration complete Missing cells, 2022-2025
Total investment 2022-2025 4 40 32 Yes Yes 0
Business investment 2022-2025* 3 30 29 Yes Yes 14
IPP investment 2022-2025 4 40 32 Yes Yes 0
Government investment 2022-2025* 3 30 29 Yes Yes 14
Business R&D 2022-2024 3 30 24 Yes Yes 19
Source: Eurostat; author’s calculations. * = fallback window (each country’s own available years; coverage differs by country – see prose).

Every structural-peer and aspiration-group member contributes at least one observation to every indicator (both “complete” columns read “Yes” throughout). Total investment and IPP investment have a genuine common window across all 19 countries, 2022-2025 and 2022-2025, and business R&D a common 2022-2024. Business investment and government investment, however, cannot be forced onto a shared window once the larger clusters are used: Bulgaria (a structural peer) last reports both in 2022, while several aspiration countries (Germany, Denmark, Finland, the Netherlands, Sweden) already report 2025. Requiring every country to agree would collapse those two indicators to a single year, which is too thin to average, so they instead use each country’s own available 2022-2025 observations (marked * in the table) – most countries contribute 2022-2024 or 2022-2025, only Bulgaria contributes 2022 alone. This is documented here rather than silently averaged over an inconsistent period.

Source: Eurostat (nama_10_gdp, nasa_10_ki, nama_10_an6, rd_e_gerdtot, sdg_08_11); author’s calculations. Total investment, IPP investment and business R&D use a common window across all countries; business investment and government investment use each country’s own available 2022-2025 observations, because coverage differs by country once the full clusters are used (exact years and the reason are in the coverage table and text above).

Slovenia, structural peers and aspiration countries: investment indicators, percent of GDP (group averages of the pooled country-year observations; observation counts are in the coverage table above; * = fallback window, coverage differs by country)
Indicator Aspiration countries Slovenia Structural peers Period
Total investment 23.0 21.4 22.9 2022-2025
Business investment 14.0 11.9 13.6 2022-2025*
IPP investment 5.5 3.2 2.7 2022-2025
Government investment 3.8 5.4 4.8 2022-2025*
Business R&D 2.0 1.5 0.7 2022-2024
Source: Eurostat; author’s calculations.

Slovenia sits below both benchmark groups on total investment (21.4% of GDP, against 22.9% for the structural peers and 23.0% for the aspiration group – gaps of 1.5 and 1.6 points respectively) and further below both on business investment (11.9% versus 13.6% for the peers and 14.0% for the aspiration group). The composition, however, differs sharply by benchmark. Slovenia’s government investment (5.4%) is the highest of the three, above the structural peers (4.8%) and well above the aspiration group (3.8%) – public capital is doing work that private investment is not. On the knowledge side Slovenia is ahead of its structural peers but behind the aspiration group: intellectual-property investment is 3.2% of GDP (peers 2.7%, aspiration 5.5%) and business R&D is 1.5% (peers 0.7%, aspiration 2.0%). The pattern is consistent throughout Section 8.2 – Slovenia’s level gap with its ten structural peers is modest (a little over a percentage point of GDP on total and business investment), while the gap with the eight-country aspiration group is similar in size on total investment but wider on business investment and on knowledge-based capital. Private and intangible investment remain Slovenia’s weak margins, even against its own peer cluster on business investment. These are descriptive group averages, not estimates of any convergence mechanism.

Source: Eurostat.

Source: Eurostat; AMECO; World Uncertainty Index; World Bank Worldwide Governance Indicators; author’s calculations.

Slovenia versus benchmark groups, trough and latest period (percent of GDP)
Window Group Total inv./GDP Business inv./GDP IPP inv./GDP Business R&D/GDP Govt inv./GDP Public debt/GDP
2013-19 (trough) EU (excl. IE, LU) 20.5 12.5 3.2 1.0 3.6 70.9
2013-19 (trough) Euro area 20.4 12.6 3.9 1.0 3.3 79.3
2013-19 (trough) Slovenia 19.1 11.5 3.0 1.6 4.0 75.2
2013-19 (trough) Structural peers 21.9 13.6 2.2 0.5 4.2 43.9
2022-25 (latest) EU (excl. IE, LU) 22.2 13.2 3.9 1.2 4.0 68.2
2022-25 (latest) Euro area 21.5 12.6 4.1 1.1 3.8 72.9
2022-25 (latest) Slovenia 21.4 11.9 3.2 1.5 5.4 68.3
2022-25 (latest) Structural peers 22.9 13.6 2.7 0.7 4.8 47.3
Source: Eurostat.

Source: Eurostat; AMECO; author’s calculations.

Source: Eurostat.

Slovenia’s total investment intensity is back to roughly the EU average, but public investment is doing the work while business investment sits near the EU’s 25th percentile and corporate profitability near the bottom. The accounting decomposition attributes the gap to high public investment (raising it) offset by thin corporate credit (lowering it), in both the trough and the latest period. Slovenia does R&D like France or the Netherlands but converts it into intangible investment like Italy or the CEE average.

8.3 Sample robustness: excluding Malta

Source: Eurostat; author’s calculations.

Headline results under three country samples
Sample Trough mean (2013-19) Dispersion trend 2014-25 (%/yr) Beta-convergence (full period)
EU-27 20.80 -2.27 -0.95
EU-25 (excl. IE, LU) 20.55 -1.99 -1.02
EU-24 (excl. IE, LU, MT) 20.59 -2.26 -1.02
Source: Eurostat; author’s calculations.

Dropping Malta changes nothing material: the EU mean moves by at most 0.1 percentage point in any period, the post-2014 dispersion narrowing if anything strengthens, beta-convergence is unchanged, the three clusters are identical, and the fixed-effects models are numerically unchanged because Malta is already outside their estimation sample. The Ireland/Luxembourg exclusion is the one that matters – visible as the EU-27 dispersion spike in 2019.

9 Conclusions and interpretation

Level. EU investment intensity fell from about 24 percent of GDP before the crisis to a trough near 20.5 percent (2013–19) and has only partly recovered. In the latest period 15 of 27 countries still invest less, relative to GDP, than they did before the crisis – concentrated in the Mediterranean and CEE/Baltic clusters.

Turning points. 2009 was a universal contraction and 2020 near-universal, but the lasting damage was a persistent shortfall in the South and CEE, not a shared dip: core and Nordic countries regained their 2007 real investment level years earlier. The EU aggregate has clear structural breaks at 2005, 2008, 2011 and 2018.

Convergence. Countries show beta-convergence (high-investment starters fell most) but only weak, non-monotonic sigma-convergence, and they are actively diverging in intangible investment.

Country groups. Three robust clusters – Mediterranean, CEE/Baltic, and Western/Nordic core. Slovenia is firmly in the CEE/Baltic group; on the shape of its crisis experience it resembles the deep-fall Mediterranean cluster.

Correlates. The baseline is a two-way (country and year) fixed-effects model with every explanatory variable lagged one year and standardised, so no regressor is contemporaneous with the year-t investment shock (this also mitigates the mechanical GDP-growth/investment overlap – the growth coefficient is unchanged when lagged) and each coefficient reads as the percentage-point change in investment / GDP associated with a one-standard-deviation change in the lagged variable. These are conditional within-country associations, not causal effects. Within countries, investment is positively associated with lagged public investment, lagged demand growth (the accelerator) and corporate credit, and negatively with the real interest rate and public debt; uncertainty carries no independent signal once common year shocks are absorbed. The principal signs are stable across the three regime windows (2000–08, 2009–19, 2020–25) – no window reverses them – and intangible investment is almost purely structural.

Slovenia. Back to average total investment intensity, but public investment is compensating for weak business investment (near the EU 25th percentile) and thin corporate credit; and Slovenia does R&D like the core while converting it into intangible investment like the CEE average.

9.1 Policy implications

Derived from the analysis above; the panel evidence is correlational, so these follow the most robust and repeated patterns, not identified causal effects.

  1. Public investment is the most reliable lever – the one correlate positive and significant in every specification, and still binding after 2013. Slovenia already runs high public investment; the priority is to protect its stability and productive composition through the cycle and use it to crowd in private projects, not to raise the level further.
  2. The binding gap is private/business investment, not public. Instruments: de-risking finance for mid-sized firms, predictable permitting, equity and quasi-equity for scale-ups.
  3. The knowledge-capital deficit is specific. Business R&D is a Slovenian strength; the shortfall is in the broader intangibles – software, data, design, organisational capital – and the gap has widened since about 2017. Policy: digital-adoption and management-capability programmes for incumbent firms, not only R&D tax credits.
  4. Financing conditions and firm balance sheets matter. Thin corporate credit is the single biggest measurable drag on Slovenia’s investment gap; corporate profitability is weak. Deepening non-bank finance and the capital-markets-union agenda is the EU-level response.
  5. Institutional quality is the main structural correlate across countries. Regulatory predictability and public-administration effectiveness are slow but high-return levers.

What the evidence does not support: treating uncertainty as a national target (it is a common EU-wide shock), or reading any coefficient as a causal multiplier.

10 Limitations

  • Not causal. The panel regressions describe co-movement net of fixed country traits and common shocks. The baseline lags every driver one year, which removes contemporaneous simultaneity (including the mechanical investment-in-GDP overlap with GDP growth), but persistent shocks still preclude a clean causal estimate; there is no instrument or natural experiment here. Coefficients are conditional associations, not multipliers.
  • Short sub-period windows. The by-period regressions use 2–8 years each; the 2020–21 pandemic window (two years) cannot support the model and its estimates are shown only for completeness.
  • Intangibles are partly measured. National accounts capture intellectual-property products only; the broader INTAN-Invest series is unavailable for Cyprus, Greece and Malta and starts in 2010.
  • Ireland and Luxembourg are excluded for multinational balance-sheet distortion; a further check dropping Malta leaves the conclusions unchanged.
  • The full fixed-effects model covers 21 countries (no World Uncertainty Index for Cyprus, Estonia, Malta; no real interest rate for Czechia). The lighter models retain more countries with the same signs.
  • Aggregation. EU aggregates are unweighted country means (the “typical country”); a GDP-weighted view is not computed.
  • Structural variables under fixed effects move little within countries and are mutually collinear, so the report relies on the between-country cross-section for them.