Higher Education Dropout and Youth Unemployment in Slovenia, 2011–2019: First Empirical Evidence

Abstract

This paper presents the first systematic attempt to quantify higher education dropout rates in Slovenia and to examine their relationship with youth unemployment. Using annual data for the period 2011–2019, dropout rates were estimated based on project V5-2360, while youth unemployment data were obtained from Eurostat. The study applies descriptive analysis, cointegration tests and a Bayesian Vector Autoregression (BVAR) with Minnesota priors, which is well suited for small samples. Three hypotheses are tested: that dropout and unemployment are inversely related, that higher unemployment reduces dropout, and that unemployment shocks are more persistent than dropout shocks. The empirical results broadly confirm these expectations. Dropout and unemployment move in opposite directions, unemployment shocks reduce dropout, and youth unemployment displays strong hysteresis. While the small sample requires cautious interpretation, the use of Bayesian VAR provides methodologically robust inference under data constraints. Beyond its national contribution, the Slovenian case carries broader implications for small post-socialist and open economies, where structural rigidities in higher education intersect with persistent labour market challenges. The findings underline the potential for higher education to act as a countercyclical buffer and highlight the importance of coordinated education and labour market policies. By linking dropout and unemployment dynamics for the first time in Slovenia, this study contributes new evidence to the international literature and opens avenues for comparative research across European higher education systems.

1. Introduction

The relationship between higher education (HE) dropout and youth unemployment has long been a subject of theoretical debate and empirical investigation. While international research has established that educational participation is typically countercyclical—rising when labor market opportunities deteriorate—the dynamics of dropout in small post-socialist economies remain underexplored. In particular, the case of Slovenia has received little attention despite persistent concerns regarding both tertiary education completion and youth labor market integration.

Previous research in Slovenia has primarily focused on enrollment trends or graduate employment outcomes, but dropout has never been quantified at the national level. To fill this gap, this paper constructs the first systematic national indicator of HE dropout and links it empirically to youth unemployment using data from Eurostat and the Targeted Research Project V5-2360.

The novelty of the study lies in two aspects. First, it provides a baseline quantification of dropout in Slovenia, thereby filling a major gap in the evidence base. Second, it integrates dropout and unemployment dynamics within an econometric framework, testing whether the countercyclical relationship observed elsewhere also holds in a small transition economy.

This study is anchored in three complementary theoretical frameworks: human capital theory, the countercyclical behaviour of education, and the hysteresis hypothesis in youth unemployment. Human capital theory [1,2] suggests that individuals adjust their educational decisions in response to opportunity costs and expected labour market returns, implying that higher unemployment reduces the incentive to drop out. The countercyclical education literature [3,4] similarly posits that participation in higher education rises when job prospects weaken. Finally, the concept of hysteresis [5] predicts that youth unemployment shocks have persistent effects over time.

By empirically testing these mechanisms in Slovenia—a small post-socialist and open economy with rigid education and labour market structures—this paper extends these established theories to a new institutional and demographic context. Methodologically, it applies a Bayesian Vector Autoregression (BVAR) with Minnesota priors to model the joint dynamics of dropout and youth unemployment under small-sample conditions, thereby demonstrating how Bayesian inference can yield robust insights even in data-limited environments. Through this dual theoretical and methodological framing, the study contributes to understanding how structural rigidities and limited policy buffers shape the interaction between higher education and the youth labour market in smaller European economies.

At the same time, the analysis faces significant challenges. The time series consists of only nine annual observations, which limits the statistical power of conventional methods. To address this, the study employs cointegration tests for completeness and relies primarily on a Bayesian Vector Autoregression (BVAR) with Minnesota priors, which is specifically designed for small samples. This approach allows us to draw cautious but meaningful insights from limited data. Nevertheless, the results should be interpreted as indicative rather than definitive, and the study is best understood as a pilot analysis that opens the field for further research once richer data become available.

The research is guided by three hypotheses derived from human capital theory, the literature on cyclical enrollment, and the concept of hysteresis in youth unemployment. These hypotheses are not only of theoretical relevance but also carry direct policy implications, as they highlight the potential for education to act as a buffer against labor market shocks and for labor market conditions to shape educational trajectories.

According to Eurydice [6] and World Bank [7] data, Slovenia has one of the highest tertiary participation rates in Central and Eastern Europe. Nearly half of all individuals aged 19–24 is enrolled in tertiary education, and around 74% of students do not pay tuition fees. Tertiary education in Slovenia consists of two interrelated subsystems: short-cycle higher vocational education and higher education. The former provides two-year, practice-oriented programmes closely aligned with labor-market needs, while the latter encompasses university and professional programmes at ISCED levels 6–8 under the Bologna framework. The Slovenian Quality Assurance Agency for Higher Education (SQAA), established in 2009 and registered in the European Quality Assurance Register (EQAR) in 2013, ensures the quality and accreditation of institutions and study programmes. These institutional features—high participation, public funding, and strong European integration—define Slovenia as a mature tertiary education system, making it a suitable context for analyzing the interaction between dropout and youth unemployment.

Slovenia can be seen as a representative case of small post-socialist economies, where structural rigidities in higher education and persistent youth unemployment jointly shape educational outcomes. Findings from Slovenia may therefore inform broader debates on how higher education systems in smaller European countries react to labor market shocks.

The paper is structured as follows. Section 2 reviews the relevant theoretical and empirical literature on dropout and youth unemployment. Section 3 formulates the research hypotheses. Section 4 describes the data, outlines the methodological framework, and explains the use of Bayesian VAR in a small-sample setting. Section 5 presents the empirical results, including descriptive statistics, cointegration analysis, and BVAR estimation. Section 6 discusses the findings in light of theory and previous research and derives policy implications. Section 7 concludes by summarizing the main contributions, highlighting limitations, and suggesting directions for future research.

Although the relationship between education and labor market outcomes has been widely studied, most analyses have relied on cross-country comparisons or large economies, typically using standard regression or VAR techniques. Far less is known about the joint dynamics of youth unemployment and higher education dropout in small transition economies. To our knowledge, this paper is among the first to apply a Bayesian VAR framework to these issues, providing systematic national-level evidence for Slovenia. By doing so, the study contributes not only to the national debate but also offers insights relevant for other small open economies facing similar challenges.

Overall, this study makes three contributions: it provides the first national estimation of higher education dropout in Slovenia, offers new empirical evidence on the interaction between dropout and youth unemployment, and demonstrates how Bayesian VAR methods can yield methodologically robust insights under small-sample conditions in post-socialist economies.

2. Literature Review

In recent decades, higher education institutions worldwide have been confronted with the problem of student dropout, as the median value for the entire European higher education area shows that almost one in three enrolled students does not complete their studies [8]. The problem is particularly pronounced in Slovenia, where the dropout rate among first-cycle students is exceptionally high. According to Education at a Glance [9], 40% of first-cycle students in Slovenia do not complete their studies or are no longer in education after the official duration of studies plus three years (the OECD average is 23%), while only 52% graduate in this period (the OECD average is 65%). Despite its policy salience, dropout is often insufficiently addressed in practice. The HEDOCE study [10] showed that academic success is high or very high on the political agenda in almost half of the 35 surveyed European countries, including Denmark, England, Finland, France, Greece, Hungary, Italy, the Netherlands, Norway, Slovenia and Sweden. Yet, in many countries, academic achievement goals rarely target students at higher risk of dropping out, and in half of European higher education systems, successful graduation or reduction in dropout rates has no impact on funding mechanisms [8]. Moreover, monitoring mechanisms for students who leave without a degree are largely absent.

The literature highlights that dropout is shaped by a complex interplay of national, institutional, and individual factors. At the national level, accessibility and selectivity of higher education, the flexibility of study systems, student support and funding frameworks, and broader socioeconomic conditions are decisive [10,11,12,13,14]. At the institutional level, important aspects include the diversity of the student population, the availability of financial and technical resources, pedagogical and didactic conditions, and the degree of student support offered. At the individual level, demographics, socioeconomic status, cognitive competencies, motivation, and previous educational trajectories play a central role. Aina et al. [15] identified four broad groups of determinants: student behaviour, skills and characteristics; family background and support; the features of higher education institutions and systems; and the characteristics of the labour market. Their findings echo earlier theoretical contributions by Tinto [16,17], Bean [18,19], and Pascarella and Terenzini [20], who emphasised that dissatisfaction with the study programme, lack of academic preparation, weak integration, and personal difficulties are critical drivers of dropout. Behr et al. [21] similarly stressed that non-completion is not a single event but the result of a long-term decision-making process shaped by interactions between students, institutions, and policy frameworks.

From a theoretical perspective, human capital theory Becker [1] provides the classical foundation for understanding dropout decisions, viewing education as an investment in productivity and future earnings. Within this framework, the choice to remain in or leave higher education depends on a cost–benefit calculation, where the opportunity cost of studying—foregone wages—plays a central role. Empirical evidence confirms that returns to education are positive though heterogeneous [22]. When labour market conditions are weak, the opportunity cost of continued study falls, which increases persistence in higher education. Research further shows that education participation is countercyclical. Dellas and Sakellaris [3] found that enrolment rises during recessions across countries, while Bedard and Herman [23] demonstrated that graduate enrolments are particularly sensitive to economic fluctuations. During the Great Recession, Clark [24] observed that higher youth unemployment increased post-compulsory education participation in England, Barr and Turner [4] found similar effects in the United States, and Kahn [25] showed that graduating in a downturn not only raises short-term unemployment risk but also depresses lifetime earnings, reinforcing incentives to remain in education when jobs are scarce. These findings confirm that participation in education is countercyclical, with unemployment associated with reduced dropout risk.

At the same time, hysteresis effects underline that youth unemployment is not only cyclical but also persistent. Blanchard and Summers [5] argued that negative labour market shocks can have long-lasting consequences by weakening skills, attachment, and employability. Bell and Blanchflower [26] documented how the Great Recession disproportionately scarred young workers, lowering wages and job stability long after recovery. Oreopoulos, von Wachter and Heisz [27] provided causal evidence that graduating in a recession reduces earnings for many years. This body of work shows that unemployment shocks shape not only short-term educational choices but also long-term trajectories, as young people often use education as a buffer against lasting disadvantage.

Institutional commitment can play a crucial role in counteracting dropout. Evidence from the Germany and the Netherlands [28,29] demonstrates that higher education institutions that foster student-centred learning and provide comprehensive support significantly increase student success. More recent studies expand this perspective by documenting how weak employment prospects may reduce dropout risk, as students choose to continue studying rather than enter adverse labour markets [9,30]. The OECD specifically highlighted for Slovenia that dropout is linked to programme design, limited flexibility, and weak labour market alignment.

New approaches to studying dropout include the application of data mining and machine learning. Noboa et al. [31] used classification methods to identify students at risk and found that access to library materials was a strong predictor of persistence. Segura et al. [32] showed that logistic regression can still serve as a robust baseline predictor, while more advanced models, such as random forests, can achieve accuracy levels of up to 99% in predicting dropout [33]. Da Cruz et al. [34] proposed a risk score model using k-nearest neighbours (KNN), which achieved 87% accuracy. These findings suggest that predictive analytics can complement traditional theoretical models by offering early-warning systems for institutions. In addition, the rise of online learning has added new challenges. Xu and Jaggars [35] found performance gaps between online and traditional learning, while Muljana and Luo [36] noted that completion rates in online programmes are significantly lower, with critical factors including institutional support, programme design, and a sense of belonging in digital environments.

Cross-national and comparative studies underline that dropout is shaped both by cyclical labour market dynamics and by structural institutional factors. In Sub-Saharan Africa, Njifen [37] documented that persistent youth unemployment correlates with higher dropout rates, while in Brazil, Finamor [38] showed that students strategically delay graduation during recessions. In the United States, Cortes et al. [39] demonstrated that the Great Recession significantly increased dropout risks, while Millenky [40] highlighted the long-term social and economic costs of early school leaving. Research on transition economies further stresses that country-specific institutional legacies matter: in Slovenia, youth unemployment remains a persistent concern despite high enrolment, and dropout rates remain elevated by systemic rigidity [9,30,41,42].

In summary, the literature shows that dropout is driven by a combination of economic incentives, institutional structures, and individual characteristics. While the countercyclical nature of education and the persistence of unemployment shocks are well established in larger economies, much less is known about how these mechanisms operate in smaller post-socialist states. Slovenia, with its exceptionally high dropout rates and rigid higher education system, represents a critical but understudied case, providing strong rationale for focused empirical investigation.

While much of the literature has focused on large economies or cross-national comparisons, little is known about dropout dynamics in small transition economies. This paper addresses this gap by providing the first systematic national-level evidence for Slovenia.

3. Research Hypotheses

The relationship between education and the labour market has long been emphasized in both economic theory and empirical research. Human capital theory [1] suggests that educational decisions are shaped by opportunity costs, which vary with labour market conditions. When youth unemployment is high, the relative benefit of remaining in education increases, discouraging dropout. Conversely, when employment opportunities are abundant, dropout rates tend to rise as students seize labour market prospects [43,44]. Empirical evidence for European countries indicates that dropout behaviour is often countercyclical, while youth unemployment exhibits strong persistence, or hysteresis [5,45]. Studies further suggest that while labour or market shocks can significantly influence education-related decisions, the reverse effect—dropout influencing unemployment—tends to be weaker and less persistent [46,47].

Based on these theoretical arguments and prior findings, we formulate the following hypotheses:

H1: 

There is a negative relationship between HE dropout rates and youth unemployment, meaning that periods of higher unemployment coincide with lower dropout, and vice versa.

H2: 

An increase in youth unemployment reduces dropout from higher education, while dropout has only a weaker and short-term effect on unemployment.

H3: 

Youth unemployment exhibits persistence (hysteresis), whereas dropout adjustments are primarily short-term.

These hypotheses are not only of theoretical interest but also relevant for education and labour market policy, since they shed light on how labour market shocks influence educational attainment and vice versa.

4. Methodology and Data

4.1. Data and Period of Analysis

The empirical analysis relies on two annual time series covering the period 2011–2019. The first variable, youth unemployment, is obtained from Eurostat [48] and refers to the unemployment rate of individuals aged 15–29. Although the standard measure of youth unemployment often focuses on the 15–24 cohort, the broader 15–29 definition was chosen to better approximate the population potentially enrolled in higher education. A limitation of this measure is that individuals aged 15–17 are not directly relevant for HE dropout, as most students enter tertiary education at age 18 or later. Nevertheless, the broader age band ensures that both undergraduate and postgraduate students are covered.

The second variable, HE dropout, was constructed from longitudinal student-level data processed within the Targeted Research Project V5-2360 (Analysis of Student Dropout in Higher Education). Each student was uniquely identified and tracked across the observation period, allowing precise classification of outcomes.

In this framework, a student is defined as a dropout if they did not complete a program within its nominal duration plus a two-year tolerance period. The denominator of the indicator includes all students whose study outcome was determined by a given year (on-time completions, delayed completions, and dropouts), while the numerator consists only of those exceeding the dropout threshold. Transfers between programs or institutions are not classified as dropouts, as these students retain active enrollment status.

The annual dropout rate is calculated as:

$${\mathit{dropout} \mathit{rate}}_t= {\displaystyle \frac{{\mathit{dropouts}}_t}{{\mathit{dropouts}}_t+{\mathit{on} \mathit{time} \mathit{completions}}_t+{\mathit{delayed} \mathit{completions}}_t}}$$

(1)

where

${\mathit{dropout} \mathit{rate}}_t$—annual proportion of students classified as dropouts in year t;

${\mathit{dropouts}}_t$—number of students who did not complete their study program within its nominal duration plus a two-year tolerance period;

${\mathit{on} \mathit{time} \mathit{completions}}_t$—number of students who completed their studies within the nominal duration;

${\mathit{delayed} \mathit{completions}}_t$—number of students who completed their studies after the nominal duration but within the tolerance period;

$t$—calendar year of observation.

The period of analysis, 2011–2019, was selected to focus on the “normal years” between two major shocks: the global financial crisis of 2008–2009 and the COVID-19 pandemic beginning in 2020. This allows us to examine the underlying dynamics of youth unemployment and HE dropout without distortions from extraordinary crisis events.

The selection of the 2011–2019 period is also driven by data consistency and definitional constraints. Official dropout data can be reliably determined only up to 2019 because a student is formally classified as a dropout after the nominal duration of study plus a two-year tolerance period. Hence, for cohorts enrolled after 2020, the observation window has not yet elapsed, and complete dropout information will only become available around 2025. Furthermore, the student-level data supplied by the Ministry of Education within the Targeted Research Project V5-2360 are methodologically consistent and comparable only from 2011 onward. Earlier records were compiled using different reporting procedures across institutions, which prevents harmonized longitudinal estimation. While the available period captures a relatively calm macroeconomic phase, this focus is intentional: it enables the identification of the baseline relationship between youth unemployment and higher education dropout in a stable environment, providing a reference point for future analyses that include crisis episodes.

Table 1. Dropout rates and youth unemployment in Slovenia, 2011–2019.

Year	Dropout Rate (%)	Youth Unemployment (%)
2011	45.7	8.2
2012	45.9	9.2
2013	45.8	10.1
2014	45.1	9.9
2015	45.6	8.8
2016	47.7	7.8
2017	49.9	6.1
2018	52.4	5.0
2019	48.8	4.1

Source: Eurostat (youth unemployment, 15–29 years) and V5-2360 project data (dropout rates).

presents the dropout rates in HE and youth unemployment rates in Slovenia for the period 2011–2019 [23]. The data show a relatively stable dropout rate at around 45–46% in the first half of the period, followed by an increase after 2016, peaking at 52.4% in 2018. In contrast, youth unemployment rose sharply during the aftermath of the global financial crisis, reaching more than 10% in 2013–2014, and then declined steadily to 4.1% by 2019. The table thus highlights an inverse pattern between the two variables: while dropout increased toward the end of the period, unemployment moved in the opposite direction. This descriptive evidence already points to the potential negative relationship between the two variables, which will be tested more formally in the subsequent econometric analysis.

4.2. Econometric Methodology

We begin with descriptive statistics (mean, standard deviation, minimum, maximum, skewness, and kurtosis) and graphical representations of both series in levels and first differences. This step provides a first view of distributional properties and potential asymmetries, as well as the visual co-movement of dropout and youth unemployment. In line with Hypothesis 1, we expect to observe an inverse relationship: years of declining youth unemployment should coincide with higher dropout, and vice versa.

To establish the time-series properties, we apply Augmented Dickey–Fuller (ADF) and Kwiatkowski–Phillips–Schmidt–Shin (KPSS) tests. As the two tests assume opposite null hypotheses, using both strengthens inference. The results indicate whether differencing is required and whether cointegration analysis is meaningful. With only nine annual observations, test power is limited, so outcomes must be interpreted cautiously.

If the variables are non-stationary in levels but stationary in first differences, we test for cointegration. We apply both the Engle–Granger procedure and Johansen’s tests to ensure robustness. Detecting cointegration would support Hypothesis 1, as it implies a stable long-run relationship between dropout and youth unemployment. Moreover, adjustment coefficients in a Vector Error Correction Model (VECM) help assess Hypothesis 3 by indicating which variable corrects deviations from equilibrium more strongly. Given the short sample, results serve mainly as a robustness check rather than definitive evidence.

The main framework is a Bayesian Vector Autoregression, which is particularly suitable for small samples. Both variables—dropout and youth unemployment—are treated as endogenous within this system, allowing their joint dynamics to be estimated simultaneously. This structure inherently accounts for potential simultaneity and endogeneity without requiring external instruments or a two-stage estimation. Using the Minnesota prior, own lags are assumed more informative than cross-variable lags, reducing overfitting while preserving essential dynamics. For Hypothesis 1, the sign and magnitude of coefficients and the variance–covariance structure reveal whether dropout and unemployment are negatively related. For Hypothesis 2, impulse response functions show how shocks propagate between the variables. For Hypothesis 3, autoregressive patterns indicate whether unemployment displays persistence while dropout adjusts in the short term.

We validate the BVAR with standard MCMC diagnostics (trace and density plots, acceptance rates) to confirm convergence. In addition, we generate a five-year out-of-sample forecast to illustrate the implied trajectories of both variables. These forecasts are not intended as precise predictions but as an additional check on model consistency and policy relevance. The Bayesian VAR framework was selected because it performs well under small-sample conditions, where classical frequentist estimators such as OLS-based VARs tend to suffer from parameter proliferation and overfitting. By incorporating Minnesota priors, the model constrains non-own lags toward zero while allowing own lags to remain dominant, thereby reflecting economic persistence without imposing rigid structure. Compared with alternative approaches such as ARDL, the BVAR enables the joint endogenous treatment of dropout and unemployment, which is theoretically consistent with human capital and hysteresis frameworks that emphasize reciprocal adjustment between education and labour market outcomes.

For completeness, we also apply Granger causality tests on the VAR representation to assess whether the relationship between the two variables is bidirectional. These tests serve as a diagnostic complement to the BVAR framework.

The Bayesian VAR framework is particularly suitable for small-sample macroeconomic applications, as it allows for parameter shrinkage through prior distributions and yields stable estimates even with limited time series. In this study, the Minnesota prior was adopted to regularize the estimation by shrinking cross-variable coefficients toward zero and own lags toward a random walk, thus reducing parameter uncertainty while preserving dynamic flexibility. The approach has been widely applied in contexts characterized by short time spans, strong persistence, or limited data availability, providing reliable inference under conditions similar to those faced in this analysis [49,50]. This provides a theoretically grounded justification for using a Bayesian VAR in the Slovenian case, where only a short annual series is available.

Although the sample is limited to nine annual observations, the Bayesian VAR framework with Minnesota priors has been explicitly chosen for its suitability in small-sample contexts, ensuring that the results are cautious but methodologically robust.

Finally, to assess the robustness of the results, we perform a sensitivity analysis using alternative specifications, including a classical VAR(1), an ARDL(1,1), and Bayesian VARs with different Minnesota prior tightness levels. In addition, a robustness test was performed using the alternative Eurostat definition of youth unemployment for the 15–24 age group. The model specification and priors were identical to the baseline (15–29) estimation. This additional check confirmed that differences in results stem from demographic scope rather than model instability, further supporting the robustness of the baseline findings.

5. Empirical Results

5.1. Descriptive Statistics and Graphical Analysis

Table 2. Descriptive statistics (2011–2019).

Variable	Mean	SD	Min	Max	Skewness	Kurtosis
Dropout (%)	47.43	2.49	45.1	52.4	0.93	2.61
Unemployment (%)	7.69	2.15	4.1	10.1	−0.53	1.89

reports the descriptive statistics for HE dropout and youth unemployment over the period 2011–2019.

The average annual dropout rate was 47.4 percent, with a standard deviation of 2.5 percentage points and a range from 45.1 to 52.4 percent. The distribution is moderately right-skewed (skewness = 0.93) and close to mesokurtic (kurtosis = 2.61). Youth unemployment (ages 15–29) averaged 7.7 percent, with a minimum of 4.1 percent and a maximum of 10.1 percent. In contrast to dropout, the unemployment series is slightly left-skewed (skewness = −0.53) and somewhat platykurtic (kurtosis = 1.89).

Figure 1 illustrates the two series in levels, while Figure 2 presents their first differences. The graphical evidence suggests that dropout and youth unemployment tend to move in opposite directions. Dropout rates gradually increased over the period, while youth unemployment declined, particularly after 2013. The differenced series fluctuate around zero, without long-term drift, confirming the need for further stationarity testing.

Taken together, the descriptive statistics and graphical inspection provide preliminary support for Hypothesis 1, which posits a negative relationship between HE dropout and youth unemployment. Although visual inspection alone is not conclusive, the inverse co-movement of the two variables is clearly visible.

5.2. Stationarity Tests

Stationarity was examined using Augmented Dickey–Fuller (ADF) and KPSS tests. The ADF tests on the differenced series yielded very high p-values, while the KPSS tests on the level series produced statistics below the 5% critical values. This provides weak and partly conflicting evidence, which is not unexpected given the small sample of nine annual observations. Results are reported in

Table 3. Results of ADF and KPSS tests.

Variable	ADF Test Statistic	ADF p-Value	KPSS Statistic	KPSS 5% Crit.
Dropout	−0.36	0.98	0.33	0.46
Unemployment	−0.05	0.99	0.34	0.46

.

For Hypothesis 1, the inconclusive results imply that long-run relationships cannot be assessed solely on the basis of stationarity and require cointegration analysis. Regarding Hypothesis 3, the mixed outcomes suggest that youth unemployment may be more persistent than dropout, though the evidence remains tentative.

5.3. Cointegration Analysis

To examine whether dropout and youth unemployment are linked in the long run, we applied Phillips–Ouliaris residual-based tests and Johansen’s system-based procedure.

The Phillips–Ouliaris test statistic (2.25) was far below the 5% critical value (33.7), so the null of no cointegration could not be rejected. By contrast, Johansen’s trace test indicated one cointegrating relation: the test statistic for r = 0 (29.17) exceeded the 5% critical value (19.96), while the statistic for r ≤ 1 (8.24) did not exceed its 5% threshold. The maximum eigenvalue test gave similar results, with a statistic of 20.93 above the 5% critical value (15.67), again suggesting a single cointegrating vector. Results are reported in

Table 4. Cointegration test results.

Test	Statistic	5% Critical Value	Decision (5%)
Phillips–Ouliaris	2.25	33.71	No cointegration
Johansen trace (r = 0)	29.17	19.96	1 cointegrating vector
Johansen trace (r ≤ 1)	8.24	9.24	Fail to reject (r = 1)
Johansen eigen (r = 0)	20.93	15.67	1 cointegrating vector
Johansen eigen (r ≤ 1)	8.24	9.24	Fail to reject (r = 1)

.

These mixed results highlight the limitations of working with very small samples. Nevertheless, Johansen’s system-based approach provides tentative support for the existence of a stable long-run relationship between dropout and youth unemployment, consistent with Hypothesis 1.

In addition, the adjustment coefficients from the vector error correction representation indicate how the two variables respond to disequilibria. Dropout appears only weakly related to the error correction term, while youth unemployment shows a stronger and significant adjustment. This is in line with Hypothesis 3, suggesting that unemployment exhibits persistence while dropout responds primarily in the short term.

5.4. Vector Error Correction Model (For Completeness)

Given the Johansen test results, we estimated a Vector Error Correction Model (VECM) with one cointegrating relation. The model combines short-run dynamics with an error correction term (ECM) that measures the speed of adjustment to long-run equilibrium. Results are in

Table 5. Selected VECM estimates (cointegration rank = 1).

Equation	ECM Coeff.	t-Stat	Significance	Interpretation
Dropout (Δ)	+0.62	1.56	n.s.	Weak/insignificant adjustment
Unemployment (Δ)	−0.30	−3.56	p < 0.05	Adjustment to long-run relation

bellow.

The estimates show that the ECM coefficient in the unemployment equation is negative and statistically significant (−0.30, p < 0.05), indicating that deviations from the long-run relation are corrected mainly through adjustments in youth unemployment. In contrast, the ECM coefficient in the dropout equation is positive but insignificant, suggesting that dropout does not systematically adjust to disequilibria.

Regarding the lagged short-run dynamics, the unemployment equation shows that lagged dropout has a significant negative effect on unemployment (−0.87, p < 0.05), while lagged unemployment is insignificant. In the dropout equation, none of the short-run coefficients are statistically different from zero.

These results, though based on very limited data, are consistent with Hypothesis 3: youth unemployment exhibits persistence and adjusts to restore equilibrium, whereas dropout reacts only weakly in the long run and appears primarily driven by short-term factors.

Nevertheless, the robustness of the VECM estimates is limited by the extremely small sample size (nine annual observations). For this reason, the VECM is presented for completeness, while the main conclusions rely on the Bayesian VAR analysis.

To verify whether the relationship between dropout and youth unemployment is bidirectional, Granger causality tests were conducted using the VAR representation. The null hypothesis that dropout does not Granger-cause youth unemployment could not be rejected (F = 0.41, p = 0.54), and the reverse direction likewise showed no significant effect (F = 1.39, p = 0.27). No instantaneous causality was detected (χ² = 0.01, p = 0.92). These findings indicate the absence of short-run feedback between the two variables. As will be shown in the subsequent impulse response analysis, dropout responds modestly to unemployment shocks, while unemployment remains highly persistent. Taken together, these results suggest a primarily unidirectional adjustment process, running from youth unemployment to dropout behaviour.

Moreover, because both variables are modelled endogenously within the Bayesian VAR framework, the approach already accounts for potential simultaneity in their short-run dynamics. The Granger causality tests presented here therefore serve as an additional diagnostic tool, reinforcing the conclusion that no significant feedback exists between the two series.

5.5. Results of Bayesian VAR

The Bayesian VAR (BVAR) was estimated on the differenced series with one lag. Although information criteria (AIC, HQ, SC, FPE) consistently suggested two lags as optimal, the very limited sample size (seven effective observations after differencing) necessitated a more parsimonious specification. Accordingly, a lag length of one was chosen to preserve degrees of freedom. To ensure that these results are not model-specific, Section 5.9 later presents a sensitivity analysis using alternative specifications.

Estimation was conducted with a Minnesota prior, which shrinks own lags toward a random walk and cross-variable lags toward zero. The prior hyperparameter controlling overall tightness (λ) was optimized during estimation, converging to approximately 0.31. This choice is particularly appropriate for small samples, as it stabilizes estimation while retaining flexibility for cross-dynamics.

5.5.1. Posterior Diagnostics

Diagnostic checks confirm the reliability of the BVAR estimates. The MCMC chains show good convergence, with stable trace plots and unimodal posterior densities for all parameters. The acceptance rate of 98% indicates efficient exploration of the posterior distribution, while the marginal likelihood (−43.9) is consistent with stable model performance given the limited data. Figure 3 illustrates fitted versus actual values, showing that the BVAR tracks the observed dynamics of both dropout and youth unemployment reasonably well.

5.5.2. Posterior Estimates

Table 6. BVAR coefficient estimates (posterior medians).

Equation	Constant	Dropout (t − 1)	Unemployment (t − 1)
Dropout (Δ)	−0.06	0.45	−0.20
Unemployment (Δ)	−0.43	0.07	0.85

reports posterior median coefficient estimates. Dropout displays moderate own-lag persistence (0.45), while the effect of lagged unemployment is negative but small (−0.20). Youth unemployment is highly persistent (0.85), with only a negligible effect of lagged dropout (0.07).

Table 7. BVAR variance–covariance matrix (posterior medians).

	Dropout Shocks	Unemployment Shocks
Dropout	3.78	−0.11
Unemployment	−0.11	0.26

presents the posterior variance–covariance matrix. Dropout innovations are far more volatile than unemployment shocks (3.78 vs. 0.26). The covariance term is negative (−0.11), suggesting that shocks to unemployment tend to reduce dropout, in line with Hypothesis 1.

5.5.3. Interpretation

These results carry two main implications. First, the persistence of youth unemployment (autoregressive coefficient 0.85) is much stronger than that of dropout, consistent with Hypothesis 3 that unemployment exhibits hysteresis while dropout reacts more flexibly. Second, the negative covariance between shocks supports Hypothesis 1, indicating an inverse association between dropout and unemployment. The Bayesian framework, with its Minnesota prior, thus provides stable inference despite the limited sample and confirms key theoretical expectations.

The trace plots demonstrate that the MCMC chains mix well and fluctuate around stable regions, while the density plots are unimodal, indicating satisfactory convergence. Some instability in the trace of the hyperparameter λ and the asymmetric shape of the marginal likelihood distribution reflect the very small sample size, but do not suggest major convergence problems.

The comparison of actual and fitted values shows that the BVAR captures the general trends in dropout and unemployment, but smooths over short-term fluctuations. This pattern is consistent with the use of a Minnesota prior, which shrinks estimates toward persistence and reduces noise at the cost of underestimating sudden changes.

Regarding limitations. The Bayesian VAR analysis is subject to several important limitations. First, the effective sample consists of only seven observations in first differences, which restricts the ability of the model to capture more complex dynamics and increases the sensitivity of the results. Second, the fitted values (Figure 3) show that the BVAR smooths over short-term fluctuations, reflecting the influence of the Minnesota prior. While this protects against overfitting, it also means that sudden shifts in dropout or unemployment are not well reproduced. Third, the posterior diagnostics (Figure 4) confirm general convergence of the Markov chains but also reveal some instability in the hyperparameter λ and an asymmetric posterior distribution of the marginal likelihood, both of which reflect the small-sample environment.

Despite these constraints, the results remain meaningful for two reasons. First, the Minnesota prior is specifically designed to stabilize inference in small samples, which makes BVAR a more suitable choice than classical VAR in this context. Second, the direction of the estimated relationships—negative association between dropout and unemployment, and strong persistence in unemployment—is consistent with both theory and previous empirical findings. This consistency increases confidence that the model, while limited in statistical power, captures the essential dynamics of the relationship between dropout and youth unemployment.

5.6. Impulse Response Functions

Impulse response functions (IRFs) trace the dynamic effects of shocks in one variable on the other within the BVAR framework. Figure 5 displays the estimated responses over a five-year horizon with 95% credible intervals.

The results show several consistent patterns. First, a shock to dropout has a strong immediate positive effect on itself that quickly decays (top left panel), confirming the short-run persistence of dropout. Second, a shock to unemployment slightly reduces dropout in the following years (top right panel), although the response is small and imprecisely estimated. This pattern supports Hypothesis 2, which expects higher unemployment to reduce dropout, as individuals are less likely to leave education when labour market prospects are weak.

Third, a dropout shock has a weak but positive effect on unemployment (bottom left panel), consistent with the notion that higher dropout may contribute to higher youth unemployment. Finally, the unemployment shock on itself (bottom right panel) shows strong persistence, with the effect decaying only gradually. This finding is in line with Hypothesis 3, which emphasizes hysteresis in youth unemployment.

Taken together, the IRFs broadly confirm the three research hypotheses. Hypothesis 1 (negative association) is supported by the negative covariance of shocks and by the direction of responses between dropout and unemployment. Hypothesis 2 (higher unemployment lowers dropout) is directly visible in the top-right panel. Hypothesis 3 (unemployment is more persistent than dropout) is strongly confirmed in the bottom-right panel, where unemployment shocks display slow mean reversion.

5.7. Forecasts

To further evaluate the robustness of the Bayesian VAR results, we generated five-year-ahead forecasts for dropout and youth unemployment (Figure 6). Forecasting in this context is not intended to provide precise numerical predictions, given the limited sample size, but rather to test the internal consistency of the estimated dynamics and to illustrate potential policy relevance.

The forecasts suggest a gradual stabilization of dropout rates around their historical average, with no indication of explosive behaviour. Youth unemployment shows a slow but persistent decline, consistent with the strong autoregressive component estimated in the BVAR. These patterns are robust to uncertainty, as reflected in the relatively wide but stable credible intervals.

From a policy perspective, the forecasts underline two key messages. First, the relative stability of dropout suggests that policy interventions may need to focus more on structural rather than cyclical determinants of educational attainment. Second, the persistence of youth unemployment reinforces Hypothesis 3 and indicates that labour market shocks can have long-lasting consequences for young cohorts. This highlights the importance of countercyclical policies that buffer youth unemployment during downturns in order to indirectly support lower dropout rates as well.

The forecast graphs show that dropout stabilizes close to its historical mean, while youth unemployment follows a gradual downward trajectory. Both forecasts display relatively wide but stable uncertainty bands, which highlight that the model does not predict explosive behaviour and is internally consistent despite the small sample size.

5.8. Synthesis of Empirical Results

The empirical analysis proceeded in several steps. Descriptive statistics and preliminary tests confirmed that both dropout and youth unemployment required differencing to ensure stationarity. Cointegration tests indicated the possibility of a long-run relationship, but given the small sample size the main focus was placed on a parsimonious Bayesian VAR with Minnesota priors.

The BVAR estimates revealed three consistent patterns. First, youth unemployment is highly persistent, whereas dropout shows lower persistence. Second, the variance–covariance structure indicated a negative co-movement of shocks, consistent with an inverse relationship between the two variables. Third, impulse response functions confirmed that unemployment shocks tend to reduce dropout in subsequent periods, while dropout shocks have weaker but positive effects on unemployment. The forecasts further showed stable dropout dynamics and a gradual decline in unemployment, supporting the robustness of the model’s implications.

Taken together, these findings are broadly consistent with the three research hypotheses. Hypothesis 1 is supported by the negative covariance between dropout and unemployment. Hypothesis 2 receives confirmation from the impulse responses showing that higher unemployment reduces dropout. Hypothesis 3 is strongly confirmed by the persistence of unemployment shocks, which dominate the dynamics compared to dropout shocks.

Overall, while the small sample size limits the statistical strength of the results, the direction of effects is consistent with theoretical expectations and previous empirical work. The analysis thus provides indicative evidence that labour market conditions and educational attainment are jointly determined, with important implications for both education and labour market policy.

While the small sample limits statistical power, the direction and consistency of the estimated effects—namely the negative association between dropout and unemployment and the strong persistence of unemployment shocks—remain robust across specifications. The results should therefore be interpreted as indicative of underlying structural mechanisms rather than as precise quantitative estimates. This interpretation emphasizes that, despite limited data, the analysis provides valuable insights into how cyclical and structural factors jointly shape education–labor market dynamics in small post-socialist economies.

5.9. Sensitivity Analysis

To verify that the baseline BVAR findings are not sensitive to model choice or prior assumptions, additional estimations were performed using a classical VAR(1), an ARDL(1,1), and Bayesian VARs with alternative prior tightness values (λ = 0.1, 0.3, 0.6, 1.0).

The baseline vector autoregression VAR(1) confirmed a negative, though statistically weak, short-run response of dropout to unemployment shocks. The impulse response analysis showed that higher unemployment tends to reduce dropout in the following period, consistent with countercyclical education incentives.

The autoregressive distributed lag model ARDL(1,1) yielded similar qualitative results. The estimated long-run multiplier (−0.78) implies that a one-point increase in unemployment reduces the dropout rate by nearly 0.8 percentage points in equilibrium. Short-run dynamic multipliers oscillate in sign, indicating temporary adjustments around a stable long-run relationship.

To ensure that results are not driven by prior selection, BVAR models were re-estimated with different levels of Minnesota prior tightness. Across λ values ranging from 0.1 to 1.0, the posterior medians of unemployment coefficients remained negative (between −0.09 and −0.89) and consistent in sign and magnitude. The posterior covariance matrices showed stable negative co-movement between dropout and unemployment shocks.

Taken together, the VAR, ARDL, and BVAR sensitivity checks confirm that the observed inverse relationship between youth unemployment and higher-education dropout is robust across alternative econometric frameworks and prior assumptions. This reinforces the main empirical conclusion that periods of high unemployment tend to discourage early exits from education, reflecting the countercyclical nature of human-capital investment decisions.

5.10. Robustness Check: Alternative Definition of Youth Unemployment (15–24)

To assess whether the main findings are sensitive to the definition of youth unemployment, a robustness check was performed using the Eurostat indicator for individuals aged 15–24 instead of the broader 15–29 cohort used in the baseline model. The time period (2011–2019) and the dropout series remained unchanged, ensuring that only the definition of the unemployment variable differed. The narrower 15–24 indicator covers young individuals at the transition from secondary to tertiary education, and thus only partly overlaps with the population at risk of higher education dropout.

The Bayesian VAR model was re-estimated using identical specifications, lag structure, and Minnesota priors as in the baseline analysis. The optimized hyperparameter for the Minnesota prior converged to λ ≈ 0.52, slightly tighter than in the 15–29 model (λ ≈ 0.31), reflecting the lower variance of the narrower unemployment series.

Posterior median estimates are reported in

Table 8. Comparison of posterior medians in baseline and robustness models (BVAR, 15–29 vs. 15–24).

Equation	Parameter	Baseline (15–29)	Robustness (15–24)	Interpretation
ΔDropout	Constant	−0.06	0.14	Slightly higher mean level in the 15–24 model
ΔDropout	Dropout (t − 1)	0.45	0.03	Own-lag persistence largely disappears
ΔDropout	Unemployment (t − 1)	−0.20	0.05	Effect changes sign; weaker inverse link
ΔUnemployment	Constant	−0.43	−0.12	Both indicate gradual decline in unemployment
ΔUnemployment	Dropout (t − 1)	0.07	−0.23	Effect reverses but remains small
ΔUnemployment	Unemployment (t − 1)	0.85	0.47	Persistence of unemployment remains strong
Covariance (dropout, unemployment shocks)		−0.11	0.32	Covariance becomes positive but remains modest

Note: Posterior medians estimated using Minnesota priors with optimized hyperparameter λ ≈ 0.31 (baseline) and λ ≈ 0.52 (robustness model). Youth unemployment data (15–24 and 15–29) sourced from Eurostat.

. The results reveal several differences between the two specifications. The coefficient of lagged unemployment in the dropout equation changes from negative (−0.20) to slightly positive (+0.05), while the covariance between dropout and unemployment shocks turns from weakly negative to positive. Moreover, unemployment persistence declines from 0.85 to 0.47, whereas dropout persistence largely disappears (0.03 vs. 0.45). These changes indicate that the relationship between dropout and youth unemployment weakens and becomes less clearly inverse when the narrower age group is used.

This result is consistent with demographic and behavioural reasoning. The 15–24 cohort includes a large proportion of upper-secondary students who are not yet at risk of higher education dropout, thereby diluting the direct cyclical relationship between unemployment and dropout. In contrast, the 15–29 definition aligns more closely with the active student population in tertiary education and thus captures the countercyclical link more effectively. Despite these definitional differences, youth unemployment remains more persistent than dropout in both specifications, confirming the hysteresis property found in the baseline model.

Overall, the robustness analysis shows that while the magnitude and sign of short-run coefficients differ, the fundamental dynamic pattern—high persistence of youth unemployment and weaker short-term adjustment of dropout—remains consistent. The findings therefore highlight definitional sensitivity rather than substantive instability, reinforcing the structural interpretation of the baseline results. Results are reported in Table 8.

6. Discussion

6.1. Alignment with Theory and Previous Research

The empirical analysis confirms all three hypotheses formulated in this study. First, the results provide evidence for a negative association between dropout and youth unemployment (H1). The covariance structure and descriptive dynamics point to an inverse relationship: periods of falling unemployment coincide with rising dropout rates. Second, impulse response functions show that higher unemployment reduces dropout (H2), albeit with relatively modest effect sizes, which is consistent with the expectation that weak labour market prospects incentivize students to remain in education. Third, the persistence of unemployment shocks is much stronger than that of dropout shocks (H3). While dropout shocks dissipate quickly, unemployment exhibits significant hysteresis, highlighting structural inertia in the youth labour market.

These core results were further subjected to a sensitivity analysis using alternative econometric frameworks (VAR, ARDL, and Bayesian VARs with varying prior tightness), which confirmed the stability of the estimated relationships across specifications, albeit within the limits imposed by the short time series.

Findings are consistent with the logic of human capital theory, which frames educational investment as a rational response to labour market conditions [2]. The inverse relationship observed in Slovenia mirrors international evidence that enrolment in tertiary education increases when labour market opportunities deteriorate, as individuals substitute education for employment [3,4]. Similar countercyclical effects of education have been documented in the European context by Clark [24], while Card [22] stresses the importance of opportunity costs in shaping schooling decisions.

The persistence of unemployment shocks observed in the BVAR results strongly resonates with the hysteresis hypothesis [5], which argues that temporary shocks to unemployment can have long-lasting effects. More recent studies provide supporting evidence of scarring effects among young cohorts, both in Europe [26] and in North America [27]. The Slovenian case, where unemployment shocks dominate the dynamics, fits well within this broader body of evidence.

Contemporary analyses also confirm the continuing relevance of these mechanisms. Recent OECD and Eurofound reports emphasize that educational attainment and labour market resilience are closely linked, particularly for young people [30,42]. ILO [51] highlights that global crises, such as the COVID-19 pandemic, magnified existing vulnerabilities of youth employment, reinforcing the importance of countercyclical educational responses. More recent economic research points to the dual mechanisms through which unemployment influences dropout, namely via opportunity costs and expected returns to education [52,53]. Studies of the Slovenian HE system also stress relatively favourable employment outcomes for graduates, but acknowledge persistent challenges related to dropout and skills mismatch [42,54,55,56].

Taken together, the Slovenian results for 2011–2019 are in line with both classical theoretical expectations and modern empirical findings. Despite the small sample, the alignment of the results with well-established patterns in the international literature lends credibility to the conclusion that dropout and youth unemployment are closely and inversely related, and that youth unemployment exhibits strong persistence across time.

While the overall findings of this study align with the main theoretical expectations and empirical literature, several points of divergence should be noted.

First, the unit root and stationarity tests produced less clear results than typically reported in the literature. Augmented Dickey–Fuller tests did not reject the null hypothesis of a unit root even after differencing, whereas KPSS statistics suggested trend stationarity. This inconsistency reflects the very short time series available (nine annual observations) rather than substantive deviations in the data-generating process; however, it means that the empirical analysis cannot replicate the statistical robustness typically found in studies using longer series [24].

Second, the magnitude of the unemployment effect on dropout (H2) is weaker than in most international studies. While the direction of the relationship is consistent with theory and evidence [3,4], the impulse responses in the Slovenian case suggest only a modest decline in dropout following increases in unemployment. This likely reflects the particular institutional context of Slovenia, including relatively strong financial support mechanisms for students and a small, open labour market.

Third, dropout rates in Slovenia display a long-term upward trend, peaking in 2018, which cannot be explained by cyclical dynamics alone. Theoretical models grounded in human capital theory generally treat dropout as a countercyclical decision [6]; however, the Slovenian case suggests that strong structural influences, such as study program design, support services, and labour market mismatch, also play a significant role. This interpretation is consistent with OECD (2022), which highlights institutional challenges in tertiary education completion in Slovenia.

Consistent with findings from other Central and Eastern European contexts (e.g., Romania), some students may drop out to enter employment earlier, suggesting that perceived labour-market returns to work experience and language skills can, in specific cases, outweigh those associated with degree completion.

Although youth unemployment in Slovenia remains among the lowest in the EU, the strong social and institutional emphasis on tertiary education sustains high enrolment levels. This combination produces high absolute dropout numbers even when employment conditions are favourable, suggesting that tertiary expansion has outpaced the labour market’s absorptive capacity. This “degree generation” effect, observed in several European contexts, reflects growing perceptions that a bachelor’s degree represents a minimum labour-market credential rather than a differentiating advantage [10,26].

Similar dynamics have been documented in other Central and Eastern European countries. For instance, studies on Romania show that some students voluntarily withdraw from higher education to enter employment earlier, perceiving work experience and foreign language proficiency as more valuable for labour-market success than formal degree completion [15]. These behavioural patterns suggest that in mature tertiary systems with high participation and relatively rigid labour markets, dropout decisions may increasingly reflect rational adaptations to perceived returns on education rather than academic failure or disengagement.

These divergences do not undermine the validity of the findings but rather emphasize that the Slovenian case combines cyclical and structural determinants. Recognizing this dual nature enriches the analysis by showing that while classical theories capture broad cyclical patterns, country-specific institutional contexts also play a decisive role in shaping dropout behaviour. A robustness test using the narrower 15–24 age group confirmed that the main findings are definition-sensitive rather than unstable. The weaker or reversed effects in this model reflect the inclusion of secondary-school cohorts not yet exposed to tertiary dropout risk, while the inverse relationship remains clear within the 15–29 population relevant to higher education.

These findings extend beyond Slovenia by showing that even in small open economies, higher education can serve as a countercyclical buffer, while labour market hysteresis reinforces the need for coordinated policy design.

While the aggregate dropout rate provides valuable macro-level insights, future research should explore heterogeneity across academic disciplines and institutional contexts. Differences in program structure, workload, and labor market relevance may explain part of the post-2016 increase in dropout rates. At present, such disaggregation is not yet feasible due to the incomplete harmonization of program-level metadata in the V5-2360 database, but future extensions will enable a more detailed analysis of disciplinary dynamics. Future extensions could incorporate field-level fixed effects or discipline-specific dropout functions once metadata harmonization allows consistent disaggregation.

6.2. Policy Implications

The empirical findings of this study also carry clear implications for education and labour market policy. Although the evidence is based on a relatively short time series, the alignment with established theoretical and empirical literature enhances the credibility of the conclusions and enables cautious policy recommendations.

First, the persistence of youth unemployment shocks (H3) highlights the need for timely and proactive labour market interventions. Active labour market policies (ALMPs), including youth guarantees, subsidized employment, and targeted training programs, should be prioritized during downturns to mitigate long-term scarring effects. This is consistent with the policy emphasis in the European Union and ILO recommendations, which stress rapid responses to labour market shocks to prevent structural unemployment among young cohorts [30,51].

Second, the inverse relationship between dropout and unemployment (H1 and H2) suggests that the education system can serve as a buffer in times of economic weakness. When labour market opportunities deteriorate, students are more likely to remain in education. Policymakers can reinforce this countercyclical mechanism by ensuring sufficient financial support through scholarships, tuition waivers, and flexible study arrangements. In this way, HE becomes a stabilizing factor that absorbs part of the labour market shock and improves long-term human capital accumulation.

Third, the evidence points to the importance of addressing structural drivers of dropout in Slovenia. Despite the countercyclical dynamics observed, dropout also reflects institutional factors such as study program design, student support, and alignment with labour market needs. OECD [9] underlines that better integration between curricula and employment opportunities is crucial for reducing dropout and improving transition outcomes. Early-warning systems and tailored academic support services for at-risk students could further enhance completion rates.

Finally, the joint dynamics of dropout and unemployment emphasize the need for closer coordination between HE and labour market policy. Ministries and agencies should develop integrated strategies that recognize the interdependence of the two domains, for example by promoting work–study pathways and creating bridges between HEIs and employers. At the same time, systematic collection of longitudinal data on dropout and graduate employment would enhance the ability to monitor trends and evaluate the effectiveness of policy interventions.

In sum, the results suggest that policy responses should combine countercyclical labour market interventions with structural improvements in HE. By addressing both short-term shocks and long-term institutional challenges, policymakers can mitigate the risks of youth unemployment while fostering higher educational attainment.

The robustness analysis suggests that policies linking education and labor markets should primarily target young adults aged 18–29, where countercyclical effects are most evident, rather than the broader 15–24 group that partly falls outside tertiary education.

This study also opens a new line of inquiry. While previous research has explored either youth unemployment or dropout in isolation, few studies have examined their joint dynamics, and even fewer have done so in small transition economies. By applying a Bayesian VAR framework to the Slovenian case, this paper provides novel empirical evidence and demonstrates the potential of Bayesian methods for studying education–labour market interactions under data constraints. Future research could extend this approach to other countries and longer datasets, enabling comparative insights and a deeper understanding of how macroeconomic shocks influence educational outcomes.

Finally, it is important to note that the analysis concludes in 2019 because complete cohort-level dropout data beyond that year are not yet available; students enrolled in 2020 would only complete their studies around 2025. Nevertheless, international evidence suggests that the mechanisms identified for the pre-pandemic period likely persisted during the COVID-19 crisis. According to the European University Association [57], more than half of European Higher Education Area institutions reported no increase in dropout, and several countries even experienced improved retention and higher domestic enrolment. These patterns are consistent with the countercyclical behavior observed in our analysis, where higher unemployment coincides with greater participation in tertiary education. Thus, while the pandemic period could not be formally included, available survey data indicate continuity rather than reversal of the long-run relationship between youth unemployment and dropout behavior.

7. Conclusions

This paper has investigated the relationship between higher education dropout and youth unemployment in Slovenia during the period 2011–2019. Using Bayesian VAR methods, complemented by descriptive and cointegration analyses, the study tested three hypotheses: that dropout and youth unemployment are inversely related (H1), that higher unemployment reduces dropout (H2), and that unemployment shocks are more persistent than dropout shocks (H3). The empirical results support all three hypotheses. Descriptive evidence showed that rising dropout coincided with declining unemployment, while the BVAR results confirmed a modest but consistent negative effect of unemployment on dropout and strong persistence of unemployment shocks.

A key contribution of this study lies in quantifying dropout rates for Slovenia and linking them empirically to youth unemployment, making this the first systematic attempt to analyze these dynamics in the Slovenian context. While dropout and unemployment have both been studied separately, their joint behaviour has not previously been examined quantitatively. By doing so, this study extends the international literature to a small Central European economy and provides a baseline for further research and policy debate. Slovenia can be seen as a representative case of small post-socialist economies, where structural rigidities in higher education and persistent youth unemployment jointly shape educational outcomes. Findings from Slovenia may therefore inform broader debates on how higher education systems in smaller European countries react to labour market shocks.

The analysis has also shown where the Slovenian case diverges from established theory and prior research. As discussed in Section 6, the effects of unemployment on dropout are weaker than in most international studies, and dropout rates display a structural upward trend that cannot be fully explained by cyclical dynamics. These divergences highlight the dual nature of the Slovenian case, where cyclical labour market forces and institutional features of higher education jointly shape dropout behaviour. Future research should therefore explicitly integrate both dimensions, combining macroeconomic shocks with institutional indicators such as student support systems, program design, and labour market alignment.

The study has clear limitations. The time series covers only nine annual observations, which reduces the statistical power of econometric tests and prevents the use of more advanced methodologies. In addition, the mismatch between the definition of youth unemployment (ages 15–29) and the typical age of higher education students introduces potential measurement challenges. These limitations must be acknowledged, but the consistency of the findings with theory and international evidence suggests that the results are nonetheless informative. Although the sample is limited to nine annual observations, the Bayesian VAR framework with Minnesota priors has been explicitly chosen for its suitability in small-sample contexts, ensuring that the results are cautious but methodologically robust. To further verify that the main findings are not driven by model choice or prior assumptions, an additional sensitivity analysis was conducted using alternative specifications, including a classical VAR(1), an ARDL(1,1), and Bayesian VARs with different Minnesota prior tightness levels. A brief robustness check using the 15–24 unemployment indicator yielded weaker effects but confirmed that differences arise from demographic scope, not from model instability. The core mechanism linking higher unemployment with lower dropout thus remains valid for the tertiary-age population.

The consistency of direction and qualitative magnitude across all models reinforces confidence that the estimated inverse relationship between dropout and youth unemployment reflects a genuine structural mechanism rather than a modelling artefact.

From a policy perspective, the findings confirm the importance of countercyclical education and labour market policies. Dropout responds to labour market conditions, while youth unemployment displays hysteresis. Policymakers should therefore strengthen mechanisms that keep students in education during downturns and implement timely interventions to prevent persistent unemployment among young people. These findings extend beyond Slovenia by showing that even in small open economies, higher education can serve as a countercyclical buffer, while labour market hysteresis reinforces the need for coordinated policy design. For policymakers in comparable European countries, the Slovenian case illustrates how targeted support for students during downturns and integrated labour market–education strategies can reduce both dropout and youth unemployment.

The restriction of the analysis to the 2011–2019 period reflects both methodological and data-related considerations. Dropout status can only be determined after the nominal study duration plus a two-year tolerance period, meaning that complete information for cohorts enrolled after 2020 is not yet available. Moreover, reliable and harmonized dropout data exist only from 2011 onward, as earlier institutional reporting practices were not standardized. While this time frame excludes major crises, focusing on a stable pre-crisis period allows the estimation of a baseline relationship between youth unemployment and higher education dropout, providing a clear benchmark for identifying potential structural breaks once post-2020 data become available.

In conclusion, this paper contributes to the literature by demonstrating that dropout and youth unemployment in Slovenia are closely connected and jointly determined. The results are both theoretically consistent and policy-relevant, underlining the need for integrated strategies that address short-term labour market shocks as well as long-term structural challenges in higher education. Framed within the broader European context, the Slovenian evidence illustrates that small post-socialist economies experience similar education–labor market interactions as larger EU members, yet are constrained by stronger institutional rigidities and narrower policy buffers. By documenting these dynamics, the study contributes to comparative debates on how higher education can mitigate youth unemployment across diverse economic settings.