Introduction

The literature on migration in Mexico has primarily focused on studying the effects of various determining factors on a scenario of stability in these effects. Broadly speaking, studies on migration flows within Mexico analyze differences of these determining factors both in destination and origin, including variables such as salary (^{Flores, Zey, & Hoque, 2013}; ^{Villarreal & Hamilton, 2012}), GDP levels (^{Soloaga, Lara, & Wendelspiess, 2010}), unemployment rates (^{Quintana & Salgado, 2016}), foreign direct investment level (^{Flores et al., 2013}), the border with the United States (^{Villarreal & Hamilton, 2012}), unemployment in the manufacturing sector, the proportion of migrant population and distance (^{Peeters, 2012}), education levels (^{Aguayo-Téllez & Martínez-Navarro, 2013}), population size (^{Soloaga et al., 2010}), and violence levels (^{Ybáñez & Alarcón, 2014}). Since these factors relate to various macroeconomic variables, it can be expected that these differences could be modified or aggravated by negative external impacts, specifically due to financial crises. However, not much analysis has been carried out on the influence of structural changes caused by these events, a matter re-gaining relevance given the recent global recession state of things and expectations.

For its part, the literature that examines the effects of financial crises shows that these events can influence various migration determinants, and therefore their effects, to an important extent. In general terms, ^{Reinhart and Rogoff (2009}) state that financial crises affect various macroeconomic variables, including growth —through public debt (higher taxes)—, the asset market, as well as production and employment. Likewise, ^{Colombo, Menna, and Tirelli (2019}) argue that financial crises are different (more severe) than “regular” economic recessions. The authors stress that these are important asymmetric impacts causing a quite noticeable reallocation of productive factors from the formal to the informal sector, specifically when it comes to labor, which in turn causes downward trends in production and investment, increasing unemployment. Based on the Mexican experience and the tequila crisis, ^{Pratap and Quintin (2011}) argue that financial crises decrease factual productivity, given how the 1995 crisis impacted the labor market (movements between industries and occupations), and therefore labor productivity.

These macroeconomic affects —primarily those related to employment— have been examined in the context of the recent 2008 financial crisis. For example, in an analysis from 61 countries, ^{Oulton and Sebastiá-Barriel (2017}) found that the financial sector crisis-affected total factor productivity (TFP) and GDP through the effect of low real interest rates and high unemployment rates. For their part, ^{Fernandes and Ferreira (2017}) argued that the crisis imposed financial restrictions that affected employment in Portugal. ^{Popov and Rocholl (2018}) examined the characteristics of companies and thus found that the financial crisis indeed affected employment in Germany; they specifically observed how employment was affected depending on the size of the company, and whether it had credit relationships with banks in good or bad standing.

Thus, we can see how the literature has provided evidence reinforcing the need to consider the effects of this crisis in the analysis of internal migration, which is the main hypothesis outlined in this paper.

As for the studies addressing the financial crisis of 2008 and its effects in Mexico, these focus mainly on the labor market, as well as on socioeconomic variables such as income, poverty levels, and education. In this regard, ^{Kemme and Koleyni (2017}) argued that the financial crisis would have impacted productivity in Mexico, with temporary effects on various macroeconomic variables that depend on the exchange rate regime. Likewise, ^{Cabral and Mollick (2017}) found that the financial crisis changed the effects of U.S. education and real GDP on real wages in Mexico.

Because of the 2008 financial crisis on unemployment, and therefore on migration flows, the goal of this paper is to analyze the effect of unemployment duration on the probability of migrating within Mexico or abroad. It is worth mentioning that some previous studies have found no significant relationship in this regard, which is primarily due to the context of the country where such analyses are carried out (e.g. whether there is unemployment insurance or benefits). For example, ^{Ahn, de La Rica, and Ugidos (1999}) made use of the inhabitants of Spain's willingness to migrate for work (not full migration) as a dependent variable. These authors found that family responsibility, age, and level of education are some important determining factors on the willingness to migrate, but they do not find unemployment duration to be significant (possibly due to cultural or lifestyle reasons) unless unemployment benefits are cut off or other family members become unemployed.

In terms of the problem presented previously, this analysis aims at filling a gap in the literature, and so it will analyze the structural changes that the 2008 financial crisis caused in the relationship between unemployment duration and the probability of migrating for the population of Mexico. For this purpose, data from the National Occupation and Employment Survey of 2007, 2008, and 2010 (^{INEGI, 2021}) were used; the effects of unemployment duration on the probability of migrating (internally or externally) are also estimated through a qualitative response model. Likewise, a Chow-type test is used to estimate structural breaks in the effect of unemployment duration after the 2008 financial crisis.

The next section addresses relevant theoretical aspects on microeconomic determining factors of the decision to migrate and unemployment duration. Subsequently, the proposed methodological strategy to estimate structural changes in the effect of unemployment duration on the probability of migrating is presented, as well as the description of the data. The following section will present the results of the empirical analysis; finally, the last section provides some conclusions.

Theoretical aspects

Methodology

To test the presence of a structural break in the relationship between unemployment duration and migration caused by the 2008 financial crisis, this study considers the econometric qualitative response model expressed in equation 1.

𝑚i𝑔𝑟𝑎𝑡𝑒𝑠_i𝑡 = 𝛼 + 𝛽𝑑𝑢𝑟𝑎𝑡i𝑜𝑛_i𝑡 + 𝐶_i′3 + 𝑢_i𝑡 1

Where migrates is a qualitative response variable equal to one if individual i is a household member in quarter t but becomes permanently absent (emigrates) in quarter t + 1. Given that the regressors data are collected in quarter t, the specification of the migrates variable controls the endogeneity of unemployment duration highlighted in the literature. The duration variable is the days that person i stated to be unemployed and looking for a job, C is a vector of control variables of characteristics of individual i, u is the error, and α, β, and γ are parameters to be estimated. Specifically, vector C includes age, a dichotomy equal to one if female (female), a dichotomy equal to one if the person is married (married), age (age), the person's monthly income (ocupinc), schooling (educ), the monthly minimum wage of the area of residence (salary), and the region of residence of the individual i. To this end, state dichotomies (state), and border region dichotomies (border) are included if they reside in the states of Baja California, Baja California Sur, Sonora, Chihuahua, Coahuila, Nuevo León, or Tamaulipas; or from the south of the country (south) if they reside in Chiapas, Oaxaca, Guerrero, or Michoacán.

To estimate structural changes, a Chow test is performed on the coefficients of the model of equation (1) under the context of a natural experiment. Specifically, the significance of the interactive variables of the duration of unemployment and additional dichotomous time (post- crisis) variables are tested in model 1. To collect pre-and post-crisis information, data from three independent samples corresponding to three periods are used: before, during, and after the financial crisis. Furthermore, the validity of a natural experiment requires the correct specification of the control and treatment groups before and after the event; in this case, the treatment group is the migrant population and the event is the financial crisis.

Therefore, to obtain robust results, the periods are chosen considering two aspects. On the one hand, given that the most significant stage of the financial crisis spans from the third quarter of 2007 to the first quarter of 2009, the samples for the periods before and after the crisis are taken from outside of this stage. On the other hand, so as the interactive variables capture as little as possible the influence of the war on drugs (an event that increased the levels of violence in the country), information is analyzed from the beginning of such war, i.e. the year of 2006.

Results

Table 1 shows that the proportion of migrants (in terms of the total population between ages 15 and 65) decreased between the first quarters of 2007 and 2010, going from 2.5 to 2%. It is worth mentioning that this negative trend was present in general since the end of the 1990s in the case of international migration, and since the 2000s for internal migration, as shown in Graph 1. In relation to the above, this drop in migration flows was considerably greater among those people who changed residence for working reasons (Graph 2).

Source: Own elaboration with information on international migration taken from the National Population Council (^{CONAPO, acronym in Spanish, 2016}), and with information on migration within Mexico taken from the 1990, 2000, and 2010 population censuses (^{INEGI, 2020a}), and the Intercensal Survey 2015 (^{INEGI, 2020b}).

Graph 1 Migration Flows Within Mexico and Abroad 

Note: The periods are P1: 2007-I, P2: 2008-III, P3: 2010-I.

Source: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

Graph 2 Number of Migrants by Reason and Period 

On the other hand, when accounting for the total population, the mean duration of unemployment (ID duration) increased by approximately one day (although if only the population group with a positive duration is considered, the mean ID duration goes from 55 days in 2007-I to 74 in 2010-I), and the maximum values increase by more than 28% (698 days) for the 2008-III quarter and more than 50% (1,292 days) for the 2010-I quarter. It is worth mentioning that the number of unemployed people (those who presented positive unemployment duration) increased along the three periods in absolute terms and with respect to the number of employed people, representing 2.58% in the first period, and subsequently 2.79% and 3.51% in periods two and three (see Graph 3).

Note: These values represent the total estimates of individuals.

Source: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

Graph 3 Number of People by Occupational Status and Period 

In the same way, the means of age and education increase, and the proportion of women and married people decreases slightly in the observed period. Finally, the monthly minimum wage means increases, reflecting the increases in the minimum wage in the period. Also, the average monthly income increased slightly between 2007-I and 2008-III, but decreased for 2010-I; interestingly, the maximum value was lower in the periods following 2007-I.

Graph 4 shows that on average there is a lower proportion of married migrants than unmarried ones, in addition to the fact that men migrate more than women. It also shows that the proportion of migrant married women almost did not change compared to 2007-I, but the rest of the subgroups had the largest drop in 2008-III.

Note: UM indicates unmarried persons and M indicates married persons.

Source: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

Graph 4 Proportion of Migrants by Marital Status, Sex, and Period in Mexico 

On the other hand, Graph 5 (quadrant I) shows that in the first period (2007-I) unemployment duration increased with educational level up to the undergraduate level, and then decreased for the postgraduate levels, being practically nil at the Ph.D. level; these findings being similar to those of ^{Hernández and García (2017}). However, the graph shows diverse pre-and post-crisis changes: for the second period, unemployment duration remained almost the same for the population with secondary and high school, decreased for those with a master’s degree, and increased for those with primary, undergraduate, and Doctorate education. In the third period, unemployment duration increased again for all educational levels, in addition to being higher on average at higher educational levels (except master’s degree). This may support the argument that educational attainment raised the expected reservation wage as the crisis increased unemployment duration for the best-prepared.

Note: Quadrants I and II show data for the periods 2007-I (1), 2008-III (2), and 2010-I (3), and the primary (P), secondary (S), high school (MS), professional (L), master’s (M), and Doctorate (D) educational levels.

Source: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

Graph 5 Proportion of Migrants and Duration of Unemployment by Education and Age in Mexico 

On the other hand, quadrant II of Graph 5 indicates that the intermediate educational levels (secondary and high school) had, on average, the highest percentages of migrants. In this case, there were also post-crisis changes: for example, in 2008-III and 2010-I the proportions of migrants decreased for all levels (except master’s degree in 2010-I); then the proportion of migrants decreased in 2010-I for the primary and secondary levels, but increased for high school, undergraduate and graduate. Finally, quadrants III and IV of Graph 5 show that as age increased (from ages 15 to approximately 25) the proportion of migrants and unemployment duration increased, and subsequently began to decrease.

Graph 6 (quadrant I) indicates that unemployment duration on average was greater for migrants than non-migrants, and it increased over time. On the other hand, although the non- migrant population reported higher average income than migrants (quadrant II), there were interesting changes in the post-crisis periods: average income decreased for migrants, but for those who did not migrate firstly it increased by 2008-III and then decreased in 2010-I. Finally, quadrant III of Graph 6 shows that the level of education on average increased over time for both groups (migrants and non-migrants), and migrants presented higher education levels than non- migrants, although in 2008-III the education level of migrants was lower than in the previous period.

Note: The graph shows values for the migrants (M) and non-migrants (NM) subgroups.

Source: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

Graph 6 Duration of Unemployment, Average Income, and Years of Education by Migration Status in Mexico 

Map 1 confirms that unemployment duration increased in the period analyzed. Furthermore, unemployment duration appears to have increased in the northern states and remained low in the states of the southern region. This suggests particular effects by region and/or state, which justifies including variables that capture these characteristics in model (1). Moreover, the proportion of migrants by state does not seem to reflect a regional pattern, except for some central states that show approximately constant proportions below the mean.

Note: The range of the box plots (0.01-0.05) includes the min-max values of the migrates variable and is crossed by the mean (0.02).

Source: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

Map 1 Duration of Unemployment and Migration in Mexican States 

Table 2 shows the estimation of model 1 with pooled data and the three periods separately. Overall, the results in all the estimates show that unemployment duration and the minimum wage in the area of residence had a positive effect on the probability that a person would migrate; on the contrary, this probability was lower if the person was married, female, or with a higher income. The estimates of unemployment duration and being a woman are quite different for the three periods; given such low standard errors, these coefficients seem to indicate that their effect on the probability of migrating decreased after the crisis, although this is subsequently proven. On the other hand, the coefficients of periods 2 and 3 in the pooled model show a decrease in the probability of migrating after the crisis.

The differences in the signs of the other covariates can be explained by the restriction of the pooled model on the regressors to be constant in the three subsamples, allowing us to identify the direction and significance of changes in the probability of migrating over time. On the other hand, the estimates for each period relax the restriction imposed by the pooled regression, yet provide limited information on the significance of the changes in the effects caused by the crisis. Even so, these differences show that in addition to changes in the magnitudes, there were some changes in the direction of the effects of some determining factors after the crisis. For instance, the pooled estimate and that of the third period show an inverted u-shape relationship between education and the probability of migrating, i.e. at the beginning higher education increases the probability of migrating, but upon reaching 15 years of education, such probability decreases. However, in the estimates of period 1, education has an u-shape relationship, indicating that the probability of migrating increased with education until approximately secondary education; yet the effect was entirely negative for the period in which the crisis began.

Finally, the coefficients of border and south show that after the crisis, the probability of migrating increased among residents of these regions, considerably for those from the border in period 2, and for those from the south in period 3. It is worth noting that the low level of adjustment of the models reflects the low proportion of individuals who decided to migrate (less than 3% of the total observations); still, this result does not disregard the goal of this study since we do not intend to explain migration as such, but rather evaluating how the financial crisis affected the causal relationship between unemployment duration and migration, for which it suffices to examine the statistical significance of the estimates.

To prove that changes in the effects of the explanatory variables are significant, the model is estimated with interactions of the explanatory variables with the temporal dichotomies P2 (corresponding to the 2008-III quarter) and P3 (corresponding to the 2010-I quarter). The results in Table 3 show that the effect of unemployment duration on the probability of migrating changed as of the financial crisis, since in the 2008-III and 2010-I quarters the interactive coefficients were negative and significant, indicating a lesser effect. When only the unemployed are included, the coefficient of the interactive term of duration and period 2 is positive (although less statistically significant), suggesting that in period 2 the effect on the probability of migrating was greater for these individuals (these results are not shown, but are available upon request).

Among other results, the age coefficients show that there was no significant change in the effects of education on the probability of migrating between the first and second period, such effect being primarily negative; yet, for the third period, the inverted u-shape relationship was modified in such a way that age had a positive effect on the probability of migrating up to approximately age 35. On the other hand, after the crisis, women continued to be less likely to migrate than men, although the gap narrowed. The probability of a married person migrating further decreased after the crisis, and the effect of income became more negative upon the third period. The effect of the monthly minimum wage became almost zero in the second period (it remained positive) but increased considerably in the third period. Finally, after the crisis, the probability of migrating changed direction (became positive) for border and southern states residents, but in the third period, it turned negative again for those living in border states.

The above results support the argument that the effect of duration changed after the crisis: it remained positive but decreased in magnitude. To observe the detail of this change, the marginal effects for a representative individual were calculated, obtained from two models: one estimated from the pooled column of Table 2, which assumes constant coefficients for the regressors in the period analyzed, but which now allows changes in the effect of unemployment duration; and the estimate in Table 3 that assumes changes in all explanatory variables. Based on some national averages, the representative individual was chosen as one who is 35 years old and has 9 years of education, is a married man, lives in an area with an average minimum wage of 1,405 MXN per month, has been unemployed for two thousand days, and does not live in the border region or the south.

Table 4 shows several important aspects. On the one hand, given that the values of the temporal dichotomies were set at zero, the marginal effects and the expected probability correspond to the representative individual in the first period; these individuals, with average values matching the representative individual, and a high level of unemployment duration (approximately five and a half years) have a relatively low probability of migrating (14%), which decreases in the following periods. It should be noted that the disproportionate magnitude of the marginal effects of P2 and P3 in the I-T Pooled column should be interpreted as the change in the probability of migrating if a representative individual in period 1 changes into an individual with values of zero in all covariates, which makes no sense. However, although the characteristic variables of the individual with the greatest marginal effects are marital status and sex, the dichotomies of the periods shown in the I-D Pooled column also show considerable marginal effects, suggesting that the probability of migrating, in general, changed according to the characteristics of these periods, possibly due to the financial crisis.

In relation to the interest variable, Table 4 shows that if unemployment duration of the representative person increases by one day, the probability that they will migrate increases by approximately 0.0001 in the case of the 2007-I semester, yet only by 0.00008 if it concerns the 2008-III semester, or 0.00003 if it is the 2010-I one, that is, the effect decreases after the crisis by almost 30%, and then by 70% compared to 2007-I. Considering that the marginal effects and the expected probability depend on the characteristics of the representative individual, it is interesting to analyze how the effect changes for different lengths of unemployment duration. To do this, from the coefficients obtained in the model in Table 3, the expected probabilities of migrating for the representative individual defined above are calculated for a range of unemployment duration between 0 and 10,000 days, keeping the values of the other covariates constant.

Note: States were selected at random, making sure there was at least one from each region of the country, including the capital.

Sources: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

Graph 7 Estimated Effect of Unemployment Duration on the Probability of Migrating by Period and Selected States 

From the sample of states shown in Graph 7, it can be noticed that the estimated probability of migrating increases with unemployment duration, although the largest marginal effects occur at very high unemployment levels, above the maximum level stated in the samples examined (see Table 1). As argued above, it is clear that the effect of unemployment duration decreases considerably after the financial crisis, primarily for the period 2010-I. Considering the sample of states selected, the largest drop in the effect of unemployment duration occurred in Chiapas, and the lesser occurred in Quintana Roo.

Note: States were selected at random, making sure there was at least one from each region of the country, including the capital.

Source: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

Graph 8 Comparison of the Estimated Effect of Unemployment Duration on the Probability of Migrating for Selected States (2007-I) 

On the other hand, Graph 8 shows that the estimated effect of unemployment duration on the probability of migrating can vary between states. Considering the selection of states shown in the graph, Jalisco and Mexico City show the least effect on the probability of migrating for the same level of unemployment duration; this result was expected, as these are two of the three most economically important states in the country, which makes them poles of labor attraction. On the contrary, the states of Quintana Roo and Chiapas showed the most noticeable effects.

Sensitivity Test

Returning to what was stated in the methodological section, the model estimates are made taking into account unemployment duration as the time that the unemployed individual spends looking for work, until the moment that they report the end date (reported duration), instead of the time of the interview (see Table 4). As shown in Table 5, under this specification of the RD duration explanatory variable, the estimates show some slight differences when it comes to the ID duration specification. For example, the probabilities of migrating are relatively higher for the representative individual, but the marginal effects are proportionally the same, i.e. the effects of temporal dichotomies as well as of sex and marital status continue to be greater.

Table 5 Marginal Effects of the Reported Duration of Unemployment 

Var Indep	Pooled I-D	Pooled I-T	Valores
duration DR	0.000133	0.00013	2000
P2*	-0.027201	0.12257	0
P3*	-0.045937	-0.15412	0
duration.P2	-0.000052	-0.00004	0
duration.P3	-0.000098	-0.00009	0
age	0.002284	0.00108	35
age2	-0.000099	-0.00008	1225
Educ	0.000784	-0.00163	9
educ2	-0.000033	0.00008	81
woman*	-0.033241	-0.04652	0
married*	-0.068866	-0.06639	1
ocupinc	-0.0000005	-0.000003	0
salary	0.000124	0.00010	1405
border*	0.000889	-0.01601	0
south*	0.004421	-0.02136	0
State dichotomies	Sí	Sí	0
Interactions with other covariates	No	Sí	0
Pr(migrates=1)	0.1585	0.1585

Note: The I-D Pooled column corresponds to the logit estimate of the pooled model (Pooled column of Table 2) including only the interaction of the temporal dichotomies with unemployment duration variable; the I-T Pooled column corresponds to the logit estimate of the pooled model that includes the interaction of the temporal dichotomies with all the regressors (estimate from Table 3). The Values column shows the values of the variables for which the marginal effects were calculated. * marks dichotomous variables, so the marginal effect shows the change in probability when the variable changes from 0 to 1. P2 is a dichotomy equal to 1 if it is the 2008-III quarter, and P3 if it is quarter 2010-I. Effects in bold are not significant at 95%.

Source: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

In relation to unemployment duration, the reported specification (RD duration) shows a slightly higher probability of migrating, compared to the specification of the interview (ID duration), for a certain level of unemployment duration. Graph 9 exemplifies the differences in the effect of unemployment duration for both specifications and the three periods for a representative individual living in Mexico City. Based on this graph, it is evident that for the three periods the specification of the duration of reported unemployment (RD duration) has a greater effect on the probability of migrating, especially on the mean values of the distribution (between 3,000 and 8,000 days). However, given that 99% of individuals who report a positive duration of unemployment do not last more than 500 days unemployed, the real differences are not significant.

Note: RD: Reported duration of unemployment (end-date stated by individual). ID: Interview duration of unemployment (end date is that of the interview).

Source: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

Graph 9 Estimated Probability of Migrating by Different Specifications of Unemployment Duration and Periods - Mexico City 

Multinomial Logit

To test whether there are differences in the effect of unemployment duration on the probability of migrating abroad or within the country, a multinomial logit model is estimated in which the dependent variable represents one of three mutually exclusive alternatives: does not migrate (base alternative), migrates internally (within Mexico), or migrates internationally; the explanatory variables of the model are the same as in the model shown in Table 3, i.e. equation 1 is complemented with interactions of the temporal dichotomies with the other covariates, including state dummies. It is worth noting that this estimate assumes the alternatives to be independent, i.e. there is no correlation among them.

One concern pertains to the assumption that individuals perceive the three alternatives as different. A more relaxed assumption is that individuals view the migration alternatives as similar, in which case it would be more appropriate to estimate a nested logit. In this sense, the justification of the assumption can relate to the difference in the costs of migrating internationally (within Mexico) or internally, that is, to the extent that the costs of migrating abroad are much higher than the costs of migrating internally, the alternatives become more independent and therefore the results of the multinomial logit are valid.

In correspondence with the previous results, Table 6 shows that a longer unemployment duration increases the probability of migrating, regardless of the destination (within Mexico or abroad). Additionally, the results show that for the first period (2007-I) a longer duration of unemployment increased the probability of migrating abroad to a relatively greater extent than within the country, both alternatives contrasted against the option of not migrating.

Table 6 Multinomial Logit 

Ind. Var.	Migrating internally	Migrating abroad
duration	0.00062	0.00134
	(0.00003)**	(0.00004)**
Period2	2.96931	-2.33212
	(0.09337)**	(0.28722)**
Period3	-4.49196	11.71601
	(0.10203)**	(0.31823)**
P2.duration	0.00013	-0.00537
	(0.00004)**	(0.00034)**
P3.duration	-0.00045	-0.00070
	(0.00004)**	(0.00005)**
Age	-0.00275	0.05838
	(0.00050)**	(0.00109)**
age2	-0.00051	-0.00108
	(0.00001)**	(0.00001)**
Educ	-0.02605	0.13093
	(0.00086)**	(0.00212)**
educ2	0.00181	-0.01147
	(0.00004)**	(0.00013)**
Woman	-0.18862	-2.03788
	(0.00198)**	(0.00639)**
Married	-0.48443	-0.04680
	(0.00255)**	(0.00576)**
Ocupinc	0.00000	-0.00013
	(0.00000)**	(0.00000)**
Salary	0.00132	-0.00231
	(0.00004)**	(0.00012)**
Border	0.05032	-0.48796
	(0.01043)**	(0.01776)**
South	-0.01077	-0.55247
	(0.01013)	(0.01540)**
N		164 911 922
Ll		-17 389 304
Chi2		1 946 120**
Pseudo_R2		0.053

Note: The coefficients of the state dichotomies and the interactions of the temporal dichotomies with the rest of the covariates are omitted. ** marks significance at 99%.

Source: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

Additionally, our results show that the onset of the crisis had different effects on unemployment duration for the two alternatives since in the second period the duration had a greater effect on the probability of migrating internally, but such effect was considerably lesser on the probability of migrating abroad; later, in the third period, the effect decreased (compared to the previous period) on the probability of migrating internally, but increased on the probability of migrating abroad. Graph 10 shows these effects for a representative individual from Mexico City.

Note: The graph corresponds to probabilities for a representative individual in Mexico City. MA: The probability of migrating abroad. MI: The probability of migrating internally.

Source: Own elaboration based on the National Occupation and Employment Survey (^{INEGI, 2021}).

Graph 10 Estimated Probability of Migrating and Duration of Unemployment by Period and Type of Migration - Mexico City 

Graph 10 confirms the general results presented above: the probability of migrating increases with unemployment duration, regardless of whether it is abroad or within the country. In addition to this, taking into account the representative individual living in Mexico City, it becomes evident that the effects are different depending on whether they migrate abroad or within Mexico, and that such differences vary depending on whether the effects are analyzed before or after the crisis. For example, before the crisis, at a given duration of unemployment, the probability of migrating abroad was greater than within Mexico; likewise, the effect of unemployment duration (represented by the slope of the curve) was greater on the probability of migrating abroad. However, for the second period, the probability of migrating abroad was considerably reduced (compared to the probability of migrating internally) and the effect of unemployment duration on this probability became practically insignificant. The roles later shifted places, since the probability of migrating abroad increased considerably in relation to the probability of migrating internally, the latter practically ceasing to be influenced by unemployment duration.

Discussion and closing remarks

This paper analyzed structural breaks caused by the 2008 financial crisis in the effect of unemployment duration on the probability of migrating. Microdata of the Mexican population was used for three periods between 2007 and 2010, and by estimating binomial and multinomial logit models, the results show that unemployment duration had a positive effect on the probability of migrating, regardless of whether such migration was within Mexico or abroad. In addition, the results of the binomial model show that after the crisis this effect decreased in magnitude, but when differentiated by the type of migration (multinomial model), the effect first increased on the probability of migrating internally and decreased on the probability of migrating abroad, yet later the opposite happened. Likewise, the results show that the probability of migrating upon reaching a certain duration of unemployment varies by state. However and generally speaking, the probability of migrating is explained primarily by factors not accounted for in the models, and the differences found are narrow at short durations of unemployment.

These results suggest that the structural breaks caused by the crisis in the relationship between migration and unemployment duration may be detrimental in the sense that, on the one hand, the crisis reduces the mobility of the productive factor within the country, thus limiting positive socioeconomic effects such as reducing inequality and increasing the efficiency of the labor factor, and on the other hand, it causes a greater probability of migrating abroad. In this sense, given that the probability of migrating of people with higher education increased after the crisis, this may justify the argument that the crisis caused brain drain, although confirming such argument is outside of the scope of this study. Finally, these results gain relevance given the expectations of a world recession caused by recent events (the Great Confinement crisis) and the questions about the expected effects. Although the present crisis is anticipated to have particular characteristics and much more profound socio-economic effects, the experience of the 2008 financial crisis can provide useful information on expected changes in the relationship between unemployment duration and migration in the coming months.

Translation: Fernando Llanas.