The Impact of Overeducation on Wages of Recent Economic Sciences Graduates

According to human capital theory, wages are determined by workers' productivity which, in its crudest form, implies that return to education does not depend on how workers' skills are used. However, after controlling for other differences, the empirical evidence shows that workers with identical education are paid differently. The literature evidences that young people are more likely to experience a mismatch between their formal education and the one required for their jobs. While there is no consensus on the reasons for the mismatch phenomenon, one consequence is clear, in terms of wages overeducation means a penalty. Our evidence shows that overeducated graduates of the Facultad de Ciencias Económicas of the Universidad Nacional de Córdoba suffer a wage penalty compared to those working in a job that requires a university degree. The results are robust to different specifications and the use of alternative estimators. Even though is not statistically significant, the penalty of a severe level of overeducation is higher than one for a mild level of overeducation; having had work experience while studying at university helps to reduce the cost of overeducation; women exhibit a similar penalty to men. While on average overeducation means a wage penalty, there is great heterogeneity among overeducated graduates, with those at the top end of the wage distribution experiencing a much lower penalty, or even a premium in some cases. Finally, while in the case of overeducation we find statistically significant effects, the same is not true of the horizontal mismatch in terms of knowledge.


Introduction
Traditionally, the standard characterisation of the demand side of the labour market does not consider specific qualitative aspects of the job. However, jobs are quite different in many characteristics that impact on labour productivity. In this context, job requirements attract attention because comprising not only a level of schooling but also other dimensions of skills, abilities, and attitudes (Hartog, 2000). Empirical studies have found that a substantial proportion of young people experience a mismatch between their educational background and the requirements of their jobs. This mismatch can be vertical, meaning that the worker's level of formal education differs to that required by the job, or horizontal in the sense that there is a difference between the field of study (or the knowledge it provided) of the worker and the one needed for the job. One of the most relevant consequences of overeducation is related to the wage penalty that mismatched workers suffer in comparison with workers that can be considered to be matched correctly. This problem is more relevant for university graduates, since, among other reasons, investment in superior education is typically the highest per capita amongst all education categories and usually implies an important share of public funds, with overeducation representing a poor return on this investment for both the graduate and the society (Carroll & Tani, 2013).
In the specific case of recent university graduates, the analysis of the mismatch -overeducation and field mismatch-is particularly interesting. For instance, it has been argued that over-and undereducation may help to explain wage dispersion among university graduates in OECD countries (Ordine & Rose, 2015). Also, there is the hypothesis that the mismatch could be a temporary phenomenon related to imperfect information about the labour market, but also it may be the result from a deliberate choice due to a low-level job is a good investment opportunity (Rubb, 2006). In any case, with the empirical evidence suggesting that the first job can have significant long-term impacts in a person labour history, be overeducated can influence a person's future work opportunities. In this regards, Battu et al. (1999), who analyse the labour history of two cohorts of university graduates in the United Kingdom over a period of 11 years, find that overeducation was not a temporary phenomenon, however their results suggest a pattern from graduation into a job which requires a degree, and then into work where the degree's importance is denuded.
As it is well known, and the literature has remarked, the job mismatch episode is more likely to affect young professionals, due to the fact that they are relatively new participants in the labour market and without working experience. For instance, Dekker et al. (2002), using Dutch data, find that the percentage of overeducated individuals decreases as the range of age increases. If the mismatch phenomenon takes place mostly during a transitional period in which recent graduates exchange knowledge for other sorts of human capital, it could be that the transitional period is not so long. Sicherman & Galor (1990) remark that individuals may accept jobs with low returns to education if the odds of being upgraded is high. For instance, Dolton & Vignoles (2000) report that the wage penalty associated with overeducation falls during the first six years after graduation. However, if genuinely mismatch is permanent, the effects of such a phenomenon are multidimensional, and in this respect, the empirical literature is vast.
This paper aims to provide evidence if the overeducation phenomenon is present for the recent graduates of economic sciences, which obtained their bachelor degree from the Facultad de Ciencias Económicas of the Universidad Nacional de Córdoba (FCE-UNC). As Zhu (2014) points out, in developing countries there is some void in the literature regarding how the occupation-education mismatch affects college graduates' earnings. This paper 1 3 gives some piece of evidence which helps to fill this gap since, to the best of our knowledge, our study is the first attempt to deal with this topic for the case of Argentina. It is also important to remark that, as the data set required to carry out the analysis was not available, we built a data set specially designed, through which we followed four cohorts of graduates during the first year after graduation. Hence, the paper contributes in this respect also and differs from other studies that rely on datasets that were not intended originally for studying the topic. Besides, we are part of a limited group of studies that use a panel-structured data set to control for unobserved individual heterogeneity, unlike most empirical applications that are based on cross-sectional data.
The study inquires on the relationship between wages and overeducation as well as other explanatory variables usually used in the wage equation literature. Particularly, we look at the penalty associated with the status of overeducation for the graduates of the FCE-UNC during the first year after finishing their studies. With this aim, we estimate different pooled and panel data models in which personal characteristics explain the level of income for salaried employees and variables aimed at identifying the status of overeducation as well as the degree of horizontal match in terms of the knowledge acquired in the university and those required by the job. 1 Also, we estimate a nonparametric model to incorporate the heterogeneity that exists among overeducated graduates.
In the framework of the Verdugo-Verdugo model (Verdugo & Verdugo, 1989), overeducation means a penalty in income when compared with workers with a similar level of formal education but who are well matched. According to our empirical results, overeducated graduates of the FCE-UNC suffer a wage penalty when compared to those working in a job requiring a university degree. This result is robust to different specifications and the use of alternative estimators. Also, while the difference is not statistically significant, the penalty for those severely overeducated is larger than for those with a mild level of overeducation. Interestingly, having working experience while at the university helps to reduce the cost of overeducation. In terms of gender, it appears there is not much difference between female and male overeducated graduates. When we move beyond the mean effects, independently of the categorisations we work with, in all cases it is possible to observe some heterogeneity among overeducated graduates, with a not minor share of them experiencing a wage premium instead of a penalty, which is more likely as we move upward along the wage distribution. Finally, while for the case of overeducation we find statistically significant effects, the same is not the case for the level of horizontal mismatch in terms of knowledge.
The rest of the paper is structured as follow. Section 2 briefly reviews the related literature. Section 3 presents and describes the data, while in Sect. 4 we laid out the empirical approach. Section 5 presents and discusses the results of different econometrics models. Finally, Sect. 6 is of summary and conclusions.

Related Literature
An overeducated individual can be defined as an employee with more schooling than required by the worker's occupation, while an undereducated individual has less schooling than required (Rubb, 2006). In both cases, there is a mismatch between the level of education needed for the job and the worker's education. One topic, among others, which have attracted the attention of the mismatch literature, is the impact that such event has on pay. Theoretically, there are at least three main models through which one can inquire into the phenomenon and its effect on wages, namely, the human capital theory (supply side explanations), the job competition theory (demand side explanations) and the assignment theory (supply and demand side explanations).
As it is well known, the insight of human capital theory (Becker, 1964) is that education plays a key role in determining wages, since as individuals has faced costs in studying, education could be viewed as an investment that will rise individuals' productivity and will give returns in the form of higher wages. In this background, overeducation may arise as a temporary status of an individual, but in the long run, as the labour market adjusts, individuals with the same stock of human capital and skills are rewarded equally. Alternatively, for the job competition theory (Thurow, 1975), job characteristics determine productivity and wages. From the point of view of this branch, individuals compete in the labour market for those jobs that pay higher wages; such competition creates a job queue, in which jobs are ranked by earnings. Employers competes for high productivity workers and generate a labour queue; in such queue workers are ranked by their educational level; hence highly educated individuals are matched to jobs that reward high wages. As pointed out by Thurow (1975), the overeducation phenomenon is a consequence of an improvement in the educational achievement of individuals, because of such improvement provokes a shift in the distribution of individuals in the labour queue that push individuals to accept jobs that are lower in the labour queue. 2 The result is a penalty in terms of incomes, in the sense that returns to education diminishes. Finally, the assignment theory (Sattinger, 1993) considers as a key determinant of wages, the quality of the matching between individuals with different characteristics and jobs with different requirements that emerges from the labour market; if the outcome of the market is a good matching, impacts positively on productivity and hence on wages, but a bad matching turns to translate into a lower productivity and wage. Therefore, from the lens of the assignment theory, wages depend on the worker characteristics as well the characteristics of his/her job.
The empirical literature about overeducation traditionally considers the standard earning function, in which, the wage rate is explained by Over-Required (O), Required (R) and Under-Required (U) education, widely known as the ORU model. Following Hartog (2000), the econometric specification can be written as follows: where w it is the individual's wage in the job, E r it are the years of school required for the job, E o it and E u it are the number of years of over or under the required schooling, it includes other explanatory variables affecting the wage rate, and it is a random error term. This function differs from the typical mincerian wage equation because it introduces characteristics of the demand side of the labour market. While in the former an education mismatch would not have an impact on earnings, in the ORU model these are also determined by the job characteristics. In general, the empirical evidence suggests that overeducation (undereducation) impacts negatively (positively) on wages, being the effect of undereducation stronger (Allen & van der Velden, 2001). In this regard, using data from various annual demographic supplements of the Current Population Surveys for 1994-2000, Rubb (2006) finds that overeducated individuals earn less than similarly educated individuals who are at a job match but more than workers doing the same job but that are not mismatched. On the other hand, undereducated earn more than others with the same level of schooling but well matched, but less than those with the same job and the required education. This author also concludes that overeducated workers have higher probabilities of upward occupational mobility. Also Campos-Soria et al. (2015) find that overeducation involves a wage penalty for the case of hospitality sector in Andalusia, but conclude that has a limited impact on promotion opportunities within the firm.
Badillo-Amador & Vila (2013) analyse the consequences of both education and skill mismatches on job satisfaction and wages using Spanish data from the European Community Household Panel (ECHP) survey for the year 2001. Authors capture the employees' status in terms of both matches by taking into account two workers' answers: i) whether the studies or the training provide them the skills needed at the present type of work and ii) the workers' feeling about if the skill or personal capacities would allow them to do a more demanding job than the one that they have at that moment. The statistical analysis shows that educational and skill mismatches are weakly related in the Spanish labour market and while skill mismatches appear as a key determinant of workers' job satisfaction, educational mismatches have much weaker impacts, if any, on workers' job satisfaction. However, both skill and education mismatches have negative impacts on wages. In a similar vein, Nieto & Ramos (2017) test the hypothesis that lower skill level of overeducated workers explains the wage penalty of overeducation in Spain. Results confirm authors' hypothesis only at a certain extent, as workers' skills only explain partially the wage penalty of overeducation. Also, Wu & Wang (2018) evidence that overeducation produces a wage penalty among workers that reached a tertiary education level in China after controlling by different types of skills.
As in our case, some studies focus on the consequences that educational mismatch have on specific group of workers such as university graduates. Allen & van der Velden (2001) analyse the relationship between skill mismatch and educational mismatch and their effects on wages and job satisfaction. 3 The study considers a sample of Dutch individuals who graduated from tertiary education seven years before the survey and were in paid employment for at least 12 h per week. Authors use a worker's self-rating of the level of education most suitable for the current job and determine whether individuals are working below or above their own level of study by comparing employees' highest attained degree with their self-response. 4 Results confirm the negative effect of overeducation on wages is stronger than the premium of being undereducated and do not support the assignment theory since skill mismatches account for only a small proportion of the wage effects of educational mismatches. Mavromaras et al. (2013) estimate the effects of being overeducated, overskilled or both on wages, job mobility and job satisfaction, with a panel data from the HILDA Survey, which began in 2010. It comprises all working-age male paid employees holding a university degree or equivalent qualification in full-time employment. To define whether workers are in a situation of job mismatch, authors take into account the mark that responders give to the sentence: I use many of my skills and abilities in my current job; for which a mark of one means a strong disagreement with the sentence, while a mark of seven denotes a strong agreement with it. The study finds that overeducation and overskilling are distinct phenomena and they have different effects on different labour market outcomes such as pay and job satisfaction. The negative effect of being simultaneously overeducated and overskilled is larger than when the person experiences just one of those states. Authors highlight they use panel data allowing to control for unobserved heterogeneity, being their econometric outcomes more reliable than those from cross sections. Carroll & Tani (2013) analyse the evolution of overeducation and their impact on pay with data from Australian graduates with bachelor degrees who left college in 2007 and were followed up in 2008, 2009 and 2010. In order to construct the overeducation variable, authors implement the Job Analysis (JA) method, which measures overeducation on the basis of occupational definitions. Besides, the study uses the Worker Self-Assessment (WA) method in order to code manually the graduates' occupations in 2007 and 2010. 5 Authors follow this mix of methodological strategies to address one of the main criticisms of JA method, which is related to the assumption that employees with the same occupational title do tasks with the same degree of difficulty. The analysis shows a notably reduction of the rate of overeducation by 2010, especially for young graduates who were more likely to be overeducated initially and that the penalty of overeducation on young graduates' pay is not significant in comparison with their well-matched peers. Sellami et al. (2017) analyse the effects of both vertical and horizontal educational mismatches on the pay of individuals with a higher education degree in Flanders. In this study, vertical educational mismatch is measured by the gap between the years of education that the job requires and the ones a person coursed in his/her study program. They use a panel data from SONAR with the cohort of those born in 1978, surveyed at age 23 for the first time and followed up at ages 26 and 29. These authors estimate a wage equation and control for the measurement error in educational mismatch and unobserved heterogeneity. Their results consistently show that overeducated individuals without a field of study mismatch earn less than adequate educated workers with a similar educational background, and that for those individuals who are working outside their field of study such a penalty is not always observed.
Alternatively, Zhu (2014) applies a nonparametric technique to account for the effect of each individual's major-job mismatch on wage for recent graduates in China. Like other studies, the major mismatch variable is constructed by using employees' self-assessment about their current job; they have to choose between two alternatives: i) I am now employed, and the job is related to my major and ii) I am now employed, but the job is unrelated to my major. The mismatch variable equals to 1 if the person answers alternative ii) and 0 otherwise. Interestingly, the author finds that though the mean impact is negative, there are more or less 32% of individuals that present a positive coefficient. This result is in line with Robst (2007), which argues that individuals with a major-occupation mismatch may earn more than those individuals which show a well major-occupation match because mismatched individuals may accept such a situation for career opportunity reasons. The study also finds support for the assignment theory; the level of wages is explained by college education, job characteristics and the major-job matching.
The literature reviewed here finds that different variables of control are significant. According to Rubb (2006), the experience of overeducated workers is rewarded at lower rates than the experience of undereducated workers and tends to increase the wages of young overeducated workers without necessarily increasing occupational mobility. In contrast, experience tends to increase the occupational mobility of older workers without necessarily having a beneficial impact on their wages. For the mismatched groups, Zhu (2014) estimates that for one more month of experience in the current job, on average, the monthly income increases by almost 2%.
The age of individuals seems to be relevant in the analysis. As remarked above, the overeducation phenomenon seems to affect different vulnerable groups such as young persons. Dekker et al. (2002) find in the Dutch sample that the percentage of overeducated individuals decreases as the range of age increases, i.e. 41.7% for the 15-19 age interval, 27% for the 30-40 age interval and 18% for the 49-64 age range. The mismatch phenomenon may appear to be transitional, since recent graduates may exchange knowledge for other sorts of human capital during a transitional period. Sicherman & Galor (1990) note that individuals may accept jobs with low returns to education if the odds of being upgraded is high. In this line, Sicherman (1991) finds support to the hypothesis that overeducated individuals have a greater probability of obtaining promotion than those who are well matched in the United States. Reflecting that overeducation can be a stepping-stone into appropriate employment, Carroll & Tani (2013) found the rate of overeducation fell notably three years after graduation, especially for young graduates. Remarkable, about the effect of overeducation on earnings, authors evidence that payment of young overeducated graduates does not differ significantly to those of their well-matched peers; however, overeducation penalises older overeducated graduates. Zhu (2014) evidences that age has little effect on average incomes of the Chinese graduates, but it affects wages negatively for two subgroups of interest, i.e. for the 25th and 75th percentiles of the estimated distribution for mismatched persons.
An interesting inquest is whether gender may play an important role when one analyses the job mismatch phenomenon. At least, three questions may arise: i) if there is one, which is the gender more vulnerable to be overeducated? ii) does the impact of the mismatch on wages differ between males and women? and iii) are the reasons for accepting to be mismatched different for men and women? Groot & Massen van den Brink (2000) suggest that overeducation is more frequent among women than men. On the contrary, by defining job mismatch in relation with the field of study, Robst (2007) finds that men are more likely to be mismatched than women; such a difference is statistically significant but relatively small (2%). The impact of job mismatch is different between genders, mismatched women earn 8.9% less than well-matched women, while mismatched males earn 10.2% less than wellmatched males, though such difference is statistically significant at the 10% level. About the reasons for accepting to be mismatched, there are significant gender differences also. The results suggest that women are more likely to report being mismatched because of amenity/constraint-related reasons, while men are more likely than women to report being mismatched due to career-oriented reasons. 6 As it is expected, such reasons affect differently on wages; for the amenities/constraints and demand-side reasons, the wage losses range between 18/29% and 17/21% for men and women respectively. In contrast, workers of both genders that accept to be mismatched due to pay or promotion opportunities earn more than correctly matched workers. The results also show that when men are mismatched, they suffer greater wage penalties, while woman workers gain more when they accept to be mismatched due to pay and promotion opportunities. Women also have wage gains when the mismatch is because of a change in career interests while men have wage losses. Zhu (2014) finds that male graduates have a lower proportion of mismatch than women, and the econometric results show that the variable gender, which identifies males, is statistically significant and positively associated with the average income, as well as for all percentiles of the estimated distribution for mismatched individuals. The nonparametric model indicates that on average, mismatched males earn 5.25% more than mismatched females. Hence, the evidence about how gender plays a role presents mixed results.
Skill mismatch also seems to be relevant to explain salaries. Mavromaras et al. (2013) find that when controlling for unobserved heterogeneity, graduate men who change status from a well-matched job to an overeducated job or an overskilled job, do not suffer a wage penalty. It is only well-matched graduate men who change status to a job where they are both overeducated and overskilled that suffer an approximate 5.9% wage penalty.
As pointed out before, in addition to the vertical mismatch in terms of the level of required education and the one held by the worker, Sellami et al. (2017) include a measure of horizontal mismatch (defined in term of the match between field of study and competencies required for every occupation) and its interaction with overeducation. Interestingly, their results indicate that it is not associated with a wage penalty and, on the contrary, it even is associated with a wage bonus, in cases where these individuals are employed in labour market segments that face labour shortages, resulting in upward wage pressure.
As mentioned before, different arguments have been proposed to explain the phenomenon of educational mismatch. Following Rubb (2006), the existence of overeducation can be explained by the human capital theory, since overeducated workers may substitute weaknesses in other areas of human capital by having more schooling than required. Such weaknesses include lower quality schooling (Robst, 1995), less experience due to career interruptions (Albrecht et al., 1999;Mincer & Polachek, 1974), less on-the-job training (Sicherman, 1991), and a variety of other possibilities. Garcia-Mainar & Montuenga (2019) apply three different models to the case of Spanish workers providing evidence that educational mismatch plays a clear signalling role, allowing workers to compensate for the lack of certain other skills, or to gain access to the labour market. Conversely, undereducated workers may substitute their lack of schooling with strengths in other areas of the human capital. Hartog (2000) suggests that from the human capital perspective, overeducation may result from a deliberate choice because the low-level job is a good investment opportunity; but, at the same time points out that a mismatch can be the result of job search in an imperfect information context, especially in the early career development so, as it was mentioned above, the mismatch status is likely to be temporary. Allen & van del Velden (2001) and Hartog (2000) notice that the assignment theory can be a good explanation. According to this theory, the allocation is optimal when workers are allocated top-down according to their skills, whereby the most competent worker is assigned to the most complex job, and the least competent worker is assigned to the simplest job. Thus, the incidence of educational mismatches can be explained by differences in the shares of complex jobs and skilled workers. In this vein, Caroleo & Pastore (2018) contribute to the empirical literature by proposing a statistical test to discriminate among alternative theoretical interpretations of the determinants of overeducation through the Heckman sample selection procedure. Their model, using AlmaLaurea data for Italian graduates five years from graduation, supports not only the job competition and job assignment models, but also the human capital model. Deželan & Hafner (2014) study the success of political science graduates during the transition from higher education to the employment market in Slovenia. Though the authors do not inquire into the relationship between job matching and wages, they investigate the education-job matching of graduates in the first job by analysing the educational and skill matches. Based on human capital, credentialist, assignment predictions as well contextual characteristics, disciplinary idiosyncrasies, and period effects, they regress binary logistic econometric specifications in which dependent variables are the appropriate level of education for the first job, overall educational matches (horizontal and vertical ones) and good skill utilisation. Individuals respond about the suitability of the field of study for the first job, which gives an indication of the horizontal educational matching and the appropriate level of education required by the first job, which provides an indicator of the vertical educational matching. Also individuals answer about the usefulness of the knowledge they acquired during the university studies for the job, which hand an indication of the skill matching. Estimations corroborate many of the theoretical hypotheses from different backgrounds of the related literature. Particularly, job satisfaction increases the odds of being well educational matched. As expected, the sector in which graduates work is relevant in predicting a good match in education and skills; those individuals that work in the public sectors are more likely to have an adequate matching. Also, there is evidence for the credentialist premise which states that employers perceive the differences in the type of degrees as a signal trainability. Though gender and the human capital hypotheses are not corroborated in all estimations; the outcomes present weak evidence for gender discrimination against women and for the fact that graduates with higher degrees increase the odds of being educational matched.
Finally, Liu et al. (2021) analyse the phenomenon of overeducation in the process of talent cultivation in 25 universities of China in 2016. They estimate the effects of being overeducated on wages controlling by the quality of education received using a mincerian wage equation. The results show a negative effect of being overeducated but conclude that this impact can be partially compensated when the individuals received a better quality education in terms of applicability and practicality of knowledge acquired as well as in individual discipline in their studies.

Data and Descriptive Statistics
To carry out the study here proposed we need information which for the case of Argentina is not available, at least from Official Statistical Offices. Thus, we generated our dataset, which besides requiring a great deal of effort it also demands important financial resources. In light of these restrictions, we limited our analysis to the case of the FCE-UNC. The UNC, also being the oldest university of Argentina, is the second largest after the Universidad de Buenos Aires, with around 115 thousand students. In the particular case of the FCE, it is also among the largest in the country in terms of the number of students, with an area of influence that includes not only the Province of Córdoba, in whose capital it is located, but also the centre and the north-west of the country. 7 The population subject of our study are the graduates of the three undergraduate degrees granted by the FCE-UNC; these are Bachelor of Science in Economics, Bachelor of Arts in Administration, and Public Accountant. Every year, the FCE celebrates four graduation ceremonies, in which approximately 700 students graduate. Our sample covers half of that population for the years 2016 and 2017. More specifically, we include those who registered for the third and fourth graduation ceremonies in each of the two years. By large the main number of graduates corresponds to the degree of Accountancy, followed by Administration, and then a small number of BSc in Economics.
In our dataset, graduates were interviewed at the time of registering for the graduation and then four additional times, one every three months, on aspects related to their job performance, as well on some piece of information about personal characteristics. The main reason why we choose the beginning of the survey to be the moment graduates register for the ceremony was it allowed making the survey compulsory since it was included as a requisite by the FCE-UNC. For the follow-up surveys we depended on the goodwill of the graduates, however we managed to achieve very high rates of responses (see Table 1). All surveys were carried out online using the tool LimeSurvey. Except for objective variables, such as age, gender, civil status, and other of a similar nature, answers given by the respondents are self-reported perceptions.
Another important challenge for this type of surveys is to keep as low as possible the attrition of the original sample. As reported in Table 2, we were quite successful in this regards. Just above half of graduates completed the four follow-up surveys, with the percentage reaching 68.3% if we also consider those who responded three out of the four surveys. Table 3 shows some descriptive statistics about the variables we use in the econometric exercises, distinguishing between the base and follow-up surveys. The aim of Table 3 is twofold. Firstly, it gives a summary picture of the personal characteristics of the population under study. Secondly, it helps to gather an idea if the patterns of attrition reported in Table 2 may be of concern in terms of our results being biased by a problem of self-selection. Let take a look at this second issue first. In Table 3, variables identified with a (*) refer to questions made only for the base survey, so the figures reported for these variables in the columns corresponding to the follow-up surveys are for the answers given on the occasion of the base survey but considering only graduates that responded the follow-up questionnaires. If the attrition of the original sample would mean a self-selection problem, we could expect that the summary of the figures reported for the base and follow-up surveys showing important differences; however, this is not the case.
With regards to the characteristics of the graduates that constitute our sample, some interesting results are worth mentioning: • Women constitute 60% of graduates.
• 83/87% declare their civil status to be single. • The share whose parents have a university degree is around 30%, both for the case of the father or the mother. • The age at which graduates finish their studies is well beyond the expected age of 22/24 years old. • A well-known problem, closely related to the length of time to finish the studies, is the low average mark with which students graduate, just above 5 on a scale from 0 to 10. • One of the main features of university students in Argentina is that a large percentage of them start working before they graduate. Among the reasons behind this behaviour is the lack of enough funding to support their studies as well as a mean of gaining experience during the transitional period before they finish their studies when they will look to enter into the labour market. This pattern emerges more clearly when looking  Table 3, with almost three out of four students having declared they had a working experience, excluding the job they may have held at the moment of graduation. • Of those who declared a working experience while studying, 75% declared their job was somehow related to their field of study. • Considering that our period of study covers the first year after graduation, and related to the two previous points, the average tenure of around 2.6 years also reflects that a large proportion of students start working well before they finish their studies. • At the moment of graduation those who declare having formal employment, which we approximate by employers complying with contributions to social security, represent about two-thirds of the sample, increasing to almost 72% in the follow-up surveys. These figures, especially for the follow-up surveys, mean a slight increase relative to the average of the Argentinean labour market, in which about 35% are informal workers. • Most graduates, almost half, work in organisations with at least 50 employees, followed by those with between 6 and 20. • 60% of the people surveyed work more than 40 h./week, followed by those who work between 30 and 40 h., which represent almost 20%.
Before we take a look at the variables we are most interested in, it is important to state as clear as possible the definitions of the different mismatches we use hereafter. Educational 1 3 mismatch is understood to arise when the level of formal education of the person is not the same as the one required by the job he/she is in employed in. Since we work with university graduates, only overeducation can arise. In regards to what in the literature is referred as skill mismatch, we control for the degree of match between the knowledge acquired during the university studies and those required by the job. Since in the case of overeducation we are working with formal levels of education that have a natural order, in the literature this mismatch is often referred to as vertical. In contrast, in the case of knowledge mismatch, our comparison does not focus on whether the graduate has more or less knowledge than required by the job, but on the extent to which he or she makes use of the knowledge acquired at university in his or her job, which is why this mismatch is hereafter referred to as horizontal. 8 For the case of overeducation, we consider a person to fall into that category if he/she declares that his/her job requires a tertiary non-university degree or less. Additionally, we distinguish two categories among overeducated people: moderately overeducated are those whose job requires a tertiary non-university degree, while those in a job which does not require a tertiary/university degree are classified as severely overeducated. Thus, it is important to stress that the status assigned to each person is the results of his/her selfassessment, as opposed to the alternative of using a systematic evaluation of the characteristics of each job, usually referred as "objective measure" of overeducation, or the so-called "empirical or statistical method" in which a person is compared to a group of his/her peers using the mean or modal values of formal education, usually measured in years, as point of reference. 9 Each of the three alternatives has its advantages and disadvantages. Figure 1 reports that at the moment of graduation around a one-third of those working as salaried employees defined themselves as overeducated; this proportion rises in the first follow-up survey, and then it falls continuously, reaching a 26.3% in the fourth follow-up survey. 10 When distinguishing between severe and mild overeducation, the first category shows a time pattern similar to that of overall overeducation, while mild overeducation shows a more unstable behaviour. Interestingly, as time passes on, severely overeducated people explain a larger share of those classified as overeducated.
While overeducation is the reflection of a vertical mismatch, in the sense that there is a difference between the level of formal education held by the person and the one required by the job, the second mismatch we work with looks at comparing the knowledge acquired during the studies and those required by the job. This second mismatch we refer to it as a horizontal. Thus, to identify the existence of a horizontal mismatch, we make use of the following question, that like in the case of overeducation, correspond to a self-assessment each person makes of his/her situation. The question asks the graduates to rank, in a range from 1 (the worst match) to 10 (the best match), to what extent he/she uses in the job the knowledge learnt during their undergraduate studies at the FCE-UNC.
As reported in Fig. 2, the knowledge match is quite stable over the first year after graduation, showing also a clear negative relationship with the status of overeducation, with the proportion of those who declare themselves as overeducated decreasing as the level of horizontal match increases (see Fig. 3).

Empirical Approach
As it was pointed out before, this research aims to look at the effects on salaried income of the vertical and horizontal mismatches between formal education and the requirements of the job. The vertical mismatch is approximated by the relationship between the person's level of formal education, undergraduate studies in the case of our sample, and the level of education required by the job. In the case of the horizontal mismatch we measure it by the degree of correspondence between the knowledge acquired during the undergraduate studies at the FCE-UNC and the ones required by the job.
Before introducing the different specifications we estimate, it is necessary to make some observations on our dependent variable. At the time of carrying out the different surveys, those who declared to be employed were asked to declare the level of income earned in their main occupation, having two options to respond: to declare a specific income or identify the range in which their income falls into. 11 As the majority of respondents chose the second alternative, we had to define a criterion to assign a certain income level to each individual. In particular, we work with two options. Firstly, as in Preston (1997) and Kler (2005), for each individual we assigned him/her an income equal to the middle point of the interval he/she declared. 12 Secondly, instead of allocating a particular income, the interval declared by the respondent is used. Due to the important increase in prices that occurred during the collection of the data, nominal values were deflated using the consumer price index, which as a side result means an increase in the possible values taken by our dependent variable, rendering it almost continuous. For the first option, we use both a pool and a random effect linear estimator, while for the second option we used an interval regression estimator in its pool and random effect versions. We estimate also some fixed effect models, as well as a nonparametric estimator following Zhu (2014).
The empirical literature branch that study the effect of mismatch hands two main econometric specifications that are quite used and well accepted, the so-called ORU model based on Duncan & Hoffman (1981) and the dummy variable approach due to Verdugo & Verdugo (1989). The starting point of the ORU model is the comparison of individuals who have the same job (same required years of schooling) but have different levels of education by decomposing actual years of schooling E it into required years of schooling (E r ) , years of overschooling (E o ) , and years of underschooling (E u ) (see Eq. (1) above). Alternatively, under the dummy variable approach, the comparison is between workers with the same level of education (equal E it ) but with one having the required level of education for the job and the other being either overeducated (OV) or undereducated (UN). Under this model, the equation to be estimated takes the following form: with: In Eq. (2) the coefficient 1 meassures, as in the standard mincerian wage equation, the returns to each year of scholling, so it expected to be positive. However, when comparing two workers with the same education as well as other characteristics as reflected by it , one of them being overeducated OV it = 1 while the other is neither overeducated OV it = 0 nor undereducated UN it = 0 , the existence of a wage penalty associated with overeducation means we expect 2 < 0 , which reflects that overeducated workers earn less than others with the same level of education that are correctly matched. In a similar vein, when one of the workers is undereducated UN it = 1 while the other is well matched OV it = 0 and UN it = 0 , the bonus associated with undereducation means we expect 3 > 0 , which reflects that undereducated workers earn more than others with the same level of education that are correctly matched. Finally, in the case of 2 = 3 = 0 , the evidence would support the capital human theory, in which the mismatch phenomenon is not present and workers gain according to their productivity as measured by their education level E it . (2) For the lowest interval we use the upper limit of it, while for the highest interval we use its lower limit.
In our study, and because of the population under analysis has all the same level of actual education, we cannot estimate the ORU model as expressed by Eq. (2), but we are left with a variation of it. Our equation is specified as follows: where OV it is defined as in (2) and, since there are not undereducated workers, the variable UN it is excluded. 13 The variable actual years of education E it is also excluded since it takes the same value for all graduates. Thus, under the usual hypothesis that an overeducated worker would earn less than another worker with the same level of education but correctly matched, we expect coefficient 1 to be negative.
Also, in Eq.
(3) we introduce the variable HMK it , which denotes the horizontal match of workers in skills. As Mavromaras et al. (2013) point out, mismatches in terms of education and skills may reflect different phenomena that can have different effects on different labour market outcomes, such as wages and job satisfaction. Hence, the inclusion of variable HMK it aims at controlling for the potential effects of the second phenomenon and is measured in terms of the match between the knowledge acquired in the university and the ones effectively used in the job. With regards to the influence of the degree of horizontal match, arguments can be made in favour of 1 negative or positive.

Results and Discussion
In Table 4 we report the results for the pooled sample for each of the two dependent variables. The main result that emerges quite clearly is the significant and negative effect associated with the overeducation dummy. This outcome means that for any two graduates with the same characteristics than their overeducation status, the one for whom there is not a match between his/her level of education and the required by the job earns a lower income.  area of study, the effect on wages is positive but the estimates are less robust; and finally viii) a negative impact of training activities on wages, which is an odd result, though the coefficients are not significant. 16 The use of cross-section data, or as in the previous results a pooled one, raises the possibility that the results are biased due to unobserved heterogeneity. As pointed out by Bauer (2002), controlling for unobserved heterogeneity might be important if individuals with lower innate ability need more education to attain a job for which they are formally overeducated. If this argument is true, we could expect that the coefficient for the overeducation status be lower in absolute value (since the unobserved ability and the probability of being overeducated are negatively correlated). In the extreme case, overeducation is only a problem of measurement error, with apparently overeducated workers being, in reality, less able than others on other dimensions. Thus, when all relevant differences in abilities are taken into account, the returns to education should become independent of the skill requirements of the job (Korpi & Tahlin, 2009). In light of these arguments, in Table 5 we report the results of estimating different random effect models.
As we can observe from the reading of Table 5, the results of making use of the panel structure of our data are qualitatively similar to the ones reported in Table 4. However, some differences are worth pointing out. As in Table 4, the coefficients for the variable measuring the status of overeducation are in all cases negative and statistically significant at the 1%, however the (absolute) values of the estimates are lower than the obtained for the pooled regressions, meaning that as expected unobserved ability is correlated negatively with the probability of being overeducated. 17 The penalty associated with the status of overeducation now varies between 7.1/7.7% compared with the 8.7/10.5% range reported in Table 4. Interestingly, these values are close to the ones found by Di Pietro & Urwin (2006), who also apply the Verdugo-Verdugo model to Italian university graduates three years after their graduation. The reduction in the wage penalty associated with overeducation when unobserved heterogeneity is controlled for, is in line with the finding in the literature as reported for example in Chevalier (2000), Allen & van der Velden (2001), Korpi &Tahlin (2009), andPark &Jang (2019).
The results for the horizontal match, are now not significant. This result could mean that the significant coefficients obtained with the pooled data was not real but simply resulted from unobserved characteristics. Sellami et al. (2017) obtain a similar result.
In Table 6 we run once again the random effect model, but now we allow for the wage penalty to be different between those who are severely overeducated and those who are mildly overeducated. As we could have expected, the wage penalty associated with severe overeducation is larger than for mild overeducation, 9.2/9.8% compared to 5.9/6%. However, as reported at the bottom of Table 6, we cannot reject the null hypothesis that the penalties are statistically the same. Once again, the coefficients for the horizontal match 16 Also, we run our different models including other control variables, such as knowledge of foreign language and of software packages, sector of activity, having people economically dependent, average grade at university, and the degree obtained. In all cases we did not find significant estimates, and since its exclusion did not affect the results for the remaining variables here reported, we choose to exclude them with the aim of simplicity and easy of presentation. These results are available upon request. 17 Gaeta et al. (2021) propose a heteroscedasticity-based instrumental variable estimation approach to deal with the potential bias from the use of cross-section data. Applying the estimator to a sample of Italian PhD holders, the results show that previous studies provided slightly upward estimates of the impact of vertical mismatch on wages. Alternatively, to deal with the bias arising from the use of cross-section data, Park (2021) uses a propensity score matching estimator on a sample of Korean doctorate holders.
are not significant. Nordin et al. (2010), who look at the wage effects of field of educationoccupation mismatches, find that, relative to well-matched workers, the penalty of being weakly matched is lower than that of being mismatched.
In Table 7 we allow for the effect of overeducation varying in terms of some personal characteristics: working experience during the time as student and gender. In all cases, the coefficients are statistically significant at 1%. As previously noted, the penalty for those (3) Robust standard errors ***p < 0.01, **p < 0.05, *p < 0.1 with working experience is lower than for those without it; hence, as in Nordin et al. (2010), this outcome would suggest that experience could partially compensate the lack of some education and/or specific skills. Also, the penalty is lower for men than form women. Nonetheless, we cannot reject the null that such penalties are statistically the same. A result that emerges from the panel data models is that when we control for the horizontal match, the magnitudes of the coefficients for the overeducation variable are quite similar to those obtained when that control is not included. Di Pietro & Urwin (2006) point out that a reduction in the penalty associated with overeducation would suggest that the assignment theory of overeducation would fit the data better than the alternatives. This appears not to be the case in our study. 18 All previous panel data models assumed that the individual effects are random instead of fixed. In Table 8 we compare the results of the pooled OLS estimator with the random and fixed effects alternatives. Before looking at the results, two points need to be highlighted. Firstly, we need to exclude three variables from the analysis since they show no variability across time for each individual and so become perfectly collinear with the fixed effects: gender, and working experience while studying, either in jobs related or not to the field of study. Secondly, and more important, is the issue pointed out by Bauer (2002) regarding if the status of overeducation has enough variation within each individual to identify the effects of an educational mismatch on wages, luckily this is the case in our dataset. For whom we have more than one observation, around 35% experienced at least one change in their status when we distinguish between overeducated and not overeducated, and the percentage rises to 42% when overeducation is further divided into severe and mild.
Regarding the results, the Breusch-Pagan test favours in all cases the random effect model over the pooled OLS, while the Hausman test points out that the orthogonality assumption between the individual effects and the explanatory variables is rejected, so the fixed effect estimator is favoured over the random one. When looking at the magnitudes of the penalty associated with overeducation a clear pattern emerges. The penalty is lower for the random effect models than for the pooled OLS, and it is further reduced when using the fixed effect estimator. Thus, as expected, when we account for the unobserved heterogeneity the effect of overeducation changes in the right direction. Still, the results are qualitatively quite similar to the ones reported before. The wage penalty is larger for those who are severely overeducated, and lower for those with work experience while studying at university. Instead, women exhibit now a larger penalty than men. However, if we consider the fixed-effect model as the most appropriate, the differences are not statistically significant.

Nonparametric Estimates
All the models reported previously provided us with a single estimate of the mean wage penalty associated to overeducation for each of the different subgroups we considered, (3) Robust standard errors ***p < 0.01, **p < 0.05, *p < 0.1 Table 8 Pooled OLS, random and fixed effect models      which, as pointed out by Zhu (2014), may conceal a heterogeneity in the results. As it is shown in Figs. 4 and 5, there is some degree of heterogeneity within those who are categorised as overeducated. For instance, in Figure 4 it is possible to observe that not a small share of overeducated graduates has incomes well above those of their well-matched peers. While for men the income distribution of the non-overeducated graduates is slighted shifted to the right of their overeducated peers, for the case of female graduates the density functions are quite similar by most part. Figure 5 confirms the intuition that for many overeducated graduates there appears not to be a wage penalty.
To lift the constraint imposed by just estimating the mean effects, we follow Zhu (2014), who looks at the case of recent college graduates in the Shandong province in China and estimate a nonparametric model using the local linear kernel estimation developed by Li & Racine (2004). 19 Unlike with our previous estimates, now no functional form or distribution assumptions are imposed, allowing to obtain for each overeducated graduate an individual wage effect, which can then be used to look at the heterogeneity across different groups. 20 In Table 9 we present the outcomes from the nonparametric model, which, on average, give us similar results as the one reported above. The mean wage penalty associated with the status of overeducation is around 10.3%. As mentioned before, an advantage of the nonparametric estimation is that we obtain a single estimate of the wage penalty/premium for each overeducated graduate. Using this set of estimates, we can calculate the mean values for different groups of graduates. Working this way, we obtain that the average penalty for men is slightly larger than for women, although the difference is not statistically significant. Also, having working experience during the time as student, either related or not to  In Table 10 we further test alternative hypotheses regarding marginal effects on wages conditional on some graduate's characteristics. The results show that in despite that even when overeducated men and women earn on average less than non-overeducated graduates, the mean penalties for the two groups are not statistically different between them. The same result emerges when comparing those with and without working experiences. Instead, when the working experience is related to the field of study, the marginal effect of having such experience relative to not having it is positive and significant at 5%.
As mentioned in Sect. 2, Robst (2007) argues that individuals with an occupation mismatch may earn more than those individuals that experience well occupation match because mismatched individuals may accept such a situation for career opportunity reasons. In this regards, we obtain that even when on average there still exists a penalty of being overeducated, this falls with the level of income. As reported in Table 9, while for the first quartile of the salary level the penalty averages around 16%, for the fourth quartile the figure falls just below 3%. These figures are similar to the values found by Ordine & Rose (2015) for the case of young Italian university graduates. Using a quintile regression approach, the authors find the wage penalty diminishes from a 14.6% for the first quartile to a 6.6% for the fifth one. As we can appreciate from Fig. 6, this result is robust to dividing the sample of overeducated graduates in terms of different dimensions. Overall, 13.1% of overeducated graduates does not suffer a wage penalty. In terms of women and men, the proportions are 12.5% and 14.3% respectively, with larger differences when we consider having or not working experience, 15.3/15.7% for those without and 10.6/12.6% for those with it. In all cases, when the average monthly wage is above $9.500, we have that some overeducated graduates exhibit a wage premium instead of a penalty. 21

Conclusions
Under the human capital theory, wages are determined by worker productivity, which is among other things, influenced by the level of education. As put clearly by Sloane (2003), in its crudest form the return to education is not contingent on how the worker's skills are used in the labour market. However, jobs are quite different in many characteristics that impact on labour productivity, and so in pay. In this context, job requirements attract attention because comprising not only a level of schooling but also other dimensions of skills, abilities, and attitudes (Hartog, 2000). In this regards, empirical studies have found that a substantial proportion of young people experience a mismatch between their educational background and the requirements of the job. As a response to this stylised finding, the literature has proposed different explanations, as well it has studied its effects over different outcomes of the labour market. As summarised in Sect. 2, alternative theoretical explanations have been proposed to explain the existence of over and undereducation in the labour market. While the empirical analysis has yet not so far reached a consensus over which of these different explanations is more likely to be behind the phenomenon, there is instead a clear message on the consequences of over and undereducation in terms of wages. In the framework of the Verdugo-Verdugo model, overeducation means a penalty in income when compared with workers with a similar level of formal education but which are well matched, while under the ORU model of Duncan & Hoffman (1981), the years of overeducation show a lower rate of return than the required years. According to the empirical evidence presented in Sect. 4, the proportion of overeducated graduates from the FCE-UNC is not negligible, though decreasing after the first year of being graduated. Unfortunately, our finding show they suffer a wage penalty when compared to those working in a job requiring a university degree. Interestingly, and in despite that no direct comparison is possible, our estimates are within the figures usually found in the literature, from 7.4% to 27% according to McGuinness (2006). Not least important, our results are robust to different specifications and the use of alternative estimators. The robustness of our results are not only qualitative but also in terms of the magnitude of the (average) wage penalty associated with the status of overeducation. While the difference is not statistically significant, the penalty for those severely overeducated is larger than for those with a mild level of overeducation. Having had working experience while in the university helps reduce the cost of overeducation. The penalty for women appears to be similar to that for men. From the nonparametric models, we find evidence that even when in average the results are similar to the one reported previously, there appears to be heterogeneity among overeducated graduates, in particular in terms of their level of income, with about a 13% experiencing a wage premium. These graduates belong to the upper part of the wage distribution. Finally, while for the case of overeducation we find statistically significant effects, the same is not the case for the level of horizontal mismatch, measured in terms of the use in the job of the knowledge acquired in the university.
Finally, our results have important implications from a policy point of view. While in most universities in Argentina, as is the case with the FCE-UNC, is not uncommon to assist students to obtain some training and internships during the time of studying, there are no similar efforts in helping to find jobs once the students graduate. It would be interesting to look into the information and advice that young people receive at the moment of choosing their university degree, with the aim of future graduates be able to take a more informed decision, for example in terms of what are expected to be the future demands of the labour market. In this sense, coordinated educational and labour policies could provide relevant objective information on the professional labour market to strengthen decision-making. In this regards, there is room for the design and implementation of programs aimed at facilitating a better match between the supply of the graduates and the demands of the labour market. An example of these programs are the so-called "job fairs", through which higher education institutions seek to put their future graduates in contact with companies that may be interested in employing them. In Argentina, these activities are more widespread in privately funded universities which, as in developed countries, are interested in the employment rates of their graduates as a mechanism for attracting potential students, especially those with better performance at the high school level. 22 This issue becomes even more important in the context of the FCE-UNC which, like all other schools belonging to public universities in Argentina, is fully funded with public resources, and so it becomes important that the investment made by the society as a whole translates into graduates having jobs for which they have prepared for, which in line with the literature on the topic would translate into higher productivity. Also, and in particular for an economy such as Argentina's, where a high proportion of jobs are still associated with low skill requirements, public policies aimed at improving firms' productivity should be strengthened, allowing them to increase the offer of high-qualified positions, which would result in lower rates of overeducation if, as expected, the number of graduates entering the labour market continue to grow. Also, the relevance of this analysis relies on shedding light on the wage penalty associated to the phenomenon of overeducation for graduates that face different backgrounds than those prevailing in developed countries. Thus, more research on this topic is welcome and indeed required to confirm our findings, also in expanding the population under study, as well as extending the period after students graduate. In this regard, studies looking at the evidence on Argentinean university graduates are almost non-existent.
Funding This research has benefited from the financial support of the Fondo para la Investigación Científica y Tecnológica (FONCyT) under grant PICT 2015-1771. All opinions, as any remaining errors, are the sole responsibility of the authors.

Data Availability
After an embargo period the data used in this study will be available upon request.

Conflict of interest
There is no research-related conflict of interest.

Ethical Approval
The surveys conducted for this study did not require prior approval by an ethical body according to the rules of the funding institution or the Facultad de Ciencias Económicas of the Universidad Nacional de Córdoba.

Consent to Participate
Not applicable. Due to the nature of the research, the informed consent of the respondents was not required.

Consent for Publication
Not applicable. Due to the nature of the research, the informed consent of the respondents was not required.