Immigrants’ employment and the business cycle in Spain: taking account of gender and origin

The purpose of this paper is to provide an empirical overview of the relationship between the business cycle and the immigrant employment cycle in Spain, differentiating by gender and origin, during the period 1999–2017. The study focuses on three main goals. First, we identify and compare the phases and the main features of the female and male immigrant employment cycles and the business cycle by means of their turning points. This target also includes the analysis of the differences in the cyclical pattern between immigrants’ and natives’ employment. Second, we wonder to what extent the global financial crisis initiated in the middle of 2007 modified the cyclical fluctuations of immigrants’ employment and whether female and male employment presents significant differences. Finally, we study whether immigrants’ origin plays some role in employment fluctuations during the pre- and crisis periods. We find that in the Spanish economic upturns, the labour market is more prone to create immigrant males’ employment than females’, while in the downturns females’ employment shows more resilience. With regard to the natives, immigrants remain unemployed longer in downturns and are laid off before the economy reaches its peak. However, a stronger gain of immigrants’ employment appears to be present in periods of expansion and a weaker loss in periods of contraction. In the contraction periods, immigrant females present the highest chances of keeping their employment.


Introduction
There is no doubt that since the end of the 1990s, immigration has become an important issue in the Spanish society and economy. From 2000 to 2005, the number of immigrants quadrupled, reaching in 2010 roughly 12% of the total population. In this year and due to the global economic crisis, a trend towards change was discernible, with the total number of immigrants reduced in both absolute and relative terms. At the beginning of 2016, immigrants comprised about 9.9% of the total population-two points below the pre-crisis levels. The highest proportion of immigrants comes from eastern European countries, South America, and Africa. At the same time, over the years, Spanish immigration has become more gender balanced among foreign-born females and males.
A large percentage of the immigrant population are between 20 and 45 years old and play an active role in the labour market. They usually have activity and unemployment rates above those of natives. Data from the Spanish National Statistics Institute (INE) indicate that the immigrant activity rate was around 75% and that of natives around 60%. With respect to the unemployment rate, before the 2007 crisis, the rate for immigrants was around 12%, and that for natives was around 8%. During the core of the crisis, the figures increased to 30% and 21%, respectively. Moreover, more than 30% of immigrants are overqualified for the job they carry out (García 2009;Godenau et al. 2014). 1 This framework provides the basis for an interesting analysis about how immigrants fare in the labour market and what is their linkage with the business cycle. Research findings on the impact of the business cycle on the immigrants' labour force demonstrate that the issue is not straightforward. After World War II, the "buffer theory" became popular which argued that during expansion phases, immigrants were helpful to the growth and they had to leave or be sent back during economic downturns (Dobson et al. 2009). However, this special relationship between migratory flows and business cycles was left out because reality showed that economic contractions did not involve a significant decrease in migration rate. There are many factors that explain this circumstance. Among others, we highlight the globalisation of the immigration, the lack of legal status that makes it difficult to return, the economic conditions in the country of origin, the difficulty to move between the origin and destination and the fact that immigrants are not a homogeneous group and their reactions are different during an economic downturn (Bradatan and Kolloju 2015).
Since the early 1970s, the international economy underwent important changes fostered by globalisation, the increasing privatisation processes and new technologies, among other key factors. These changes entailed replacing mass production with flexible specialisation. Post-Fordism became the dominant system and was accompanied by the decline of regulation and the rise of global markets. The new capitalism system brought about the deregulation of the labour market, but not people's mobility. Portes and De Wind (2004) indicate that economic reasons demand that the borders be opened, whereas political and legal logic demand closing them. Globalisation has increased the complexity of migration flows and has raised new motivations related to personal self-realisation (King 2006). However, Massey (2007) points out that the immigrants are still vulnerable due to their class, race and gender. They are vulnerable, which is required by the labour market of the developed countries (Sassen 1988) and, in addition, most of them are concentrated in lower occupational categories in destination countries. This stage traps immigrants in the lower tiers of society, but at the same time, entails a special relationship between their employment and the business cycle fluctuations.
Currently, there is a broad consensus in recognising that immigrants relative to natives show larger unemployment sensitivity to business cycles (Aydemir 2003;Zadodny 2009, 2010a, b;Dustmann et al. 2010;Mandelman and Zlate 2012;Prean and Mayr 2016;Clemens and Hart 2017). However, the core of the matter is far from being as simple as it might seem at first sight. This necessitates broadening the scope of research. For instance, it is necessary to analyse whether the response of the immigrants' labour market supply to movements in economic activity depends on the gender and the origin of the immigrant labour force. We seek to explore this issue.
The purpose of this paper is to provide an empirical overview of the relationship between the business cycle and the immigrant employment cycle in Spain, differentiating by gender and origin, during the period 1999-2017. The study focuses on three main goals. First, we identify and compare the phases of the female and male immigrant employment cycles and the phases of the business cycle by means of their turning points. This target also includes the analysis of the differences in the cyclical pattern between immigrants' and natives' employment. A well-established methodology is used to examine the intensity of the cycles' synchronisation. Second, we wonder to what extent the crisis initiated in the middle of 2007 modified the cyclical fluctuations of immigrants' employment and whether there are significant differences between female and male employment. Finally, we study whether immigrants' origin plays a role in their employment fluctuations both pre-and during the crisis. We carry out a shift-share analysis to find out whether origin matters.
Analyses of the differences between immigrants and natives in employment over the business cycle in Spain have already been conducted (Lacuesta and Puente 2010;De la Rica and Polonyankina 2013;Carrasco and García-Pérez 2015). However, no work has looked at distinctions between gender and place of origin. This paper fills this gap. Such distinctions become noteworthy points in our empirical analysis. In contrast to other studies in this field, what is lost with a holistic perspective is gained through more accurate results. Therefore, our work becomes more significant. In this sense, understanding how immigrants from different origins and different genders face business cycle fluctuations is a key issue, not only from a demographic point of view, but also for any policy recommendation. As this paper offers this information, we will be able to discuss whether the measures adopted by the Spanish government were shaped correctly to optimise the effect of economic growth and to minimise the effects of the downturns in immigrants' employment. On the other hand, if we find that gender and origin matter, any well-defined model trying to estimate the interactions between immigrants' employment and economic factors in Spain, should include variables to proxy both gender and origin.
The overall conclusions are the following. First, the Spanish economy has more capacity for growth than creating immigrant employment, being more evident in the case of females. However, in contraction phases, females show higher resilience to losing their employment than do males. Second, immigrants are more sensitive to the GDP upturns than natives. Finally, the study also confirms that the origin of immigrants matters. We find that origin influences the immigrant employment fluctuations, in both the pre-and crisis periods.
The paper is organised as follows. Section 2 describes the data used in this study. Section 3 identifies and compares the cycle phases by means of their turning points. The section closes by exploring the cycles' synchronisation. Section 4 discusses to what extent the 2007 crisis modified the cyclical fluctuations of immigrants' employment and whether there are significant differences between females and males. In Sect. 5 we study whether the origin of immigrants matters. The conclusions are drawn in Sect. 6.

Sources and data
The dataset used in this paper covers 17 years with a quarterly frequency. We analyse the interactions between the fluctuations in immigrant employment, native employment and the Spanish business cycle during the period 1999-4Q/2017-2Q. The source of the data is the Spanish National Statistics Institute (INE). The business cycle indicator is GDP in real terms from Quarterly Spanish National Accounts. The employment data comes from the Spanish Labour Force Survey (LFS; EPA in Spanish). The survey provides quarterly information based on the immigrants' birth place, irrespective of their legal situation. The employed people are defined as all those aged 16 or over who, during a given reference period (week prior to the week of the interview), declared having a job as an employee, wage-earner, or engaged in an activity on their own behalf (self-employed). Foreigners are included in the LFS if they are going to live in Spain for over a year (except persons in the military or foreign diplomats). To perform our analysis, we distinguish between immigrant males' employment (IM), immigrant females' employment (IF), native males' employment (NM) and native females' employment (NF). 2 It must be stressed there are some aspects related to the LFS that affect our analysis. First, in 2002 and in 2005 the database incorporated new population projections, which, among other questions, aimed to better capture the increase of the immigrant population in Spain. These changes imply that a structural break may exist, which would not make the series comparable. However, we carried out the Cusum test (5% significance) and the results show that there is a structural permanence of the parameters. Second, the LFS only incorporates individuals who live in registered households, therefore, the survey does not include information about collective establishments, which are the place of residence of a significant number of immigrants, especially the newcomers. This circumstance gives rise to a slight underestimation of the immigrant population and must be taken into account in order to interpret the results and the conclusions. Nevertheless, in spite of this limitation, the LFS remains one of the most significant databases about the Spanish labour market.
Based on the pioneer paper of Burns and Mitchell (1946), the fluctuations of the business cycle can be identified following different approaches. Our study is based on the 'growth cycle', which defines the cycle as fluctuations in periods of fast growth followed by periods of slow growth. Concretely, we use the growth cycle named 'deviation cycle' due to the fact that the analysis is based on fluctuations of the series with respect to the trend (Lucas 1977). 3 Thus, the cyclical component is calculated as the deviation of the original series with respect to the trend using the Hodrick-Prescott filter (HP). 4 The cyclical component fluctuations determine the cycle and its turning points: peaks (P) and troughs (T). We identify two phases in the cycle: expansion and contraction. The cyclical component increases in the expansion phase, trough to peak, and decreases in the contraction phase, peak to trough.
Therefore, the variables are at quarterly frequency, seasonally adjusted, and in log-levels; thus, IF, IM, NF, NM and GDP are the log-levels of the cyclical components.

Turning points and main features of cycles
To identify the turning points, our study works with the algorithm of Bry and Boschan (1971) adapted to quarterly data by Harding and Pagan (2002). The procedure entails certain censoring rules. First, a local maximum/minimum is defined as the highest/lowest point between the two preceding and following quarters to its position. That is, y t is a peak at time t if y t is the maximum (y t−2 …,y t+2 ), and it is a trough if y t is the minimum (y t−2 …,y t+2 ). Second, a peak must be followed by a trough and vice versa to ensure that only completed cycles may be used. Finally, the minimum phase length is three quarters, and the minimum complete cycle length is five quarters. 5 Once we identify the turning points and define the expansion and contraction phases, we will present the features of the cycle in terms of duration, amplitude, delay, and synchronisation. The duration (D) of a contraction (expansion) is the number of quarters that elapse between a peak and the next trough (trough and peak). The duration of the cycle is the number of quarters that elapse between peak and peak (trough and trough). The amplitude of a contraction (expansion) is calculated as the percentage change between the cyclical component value in the trough (peak) and the value in the previous peak (trough). The amplitude estimates the gains in terms of production (employment) in expansion phases and the losses in contraction phases. The delay is the number of quarters that a series lags (leads) in reaching the turning point with respect to a reference series; in our study the reference series is GDP. Throughout the paper, we focus on the average of the duration, amplitude, and delay.
By applying the turning-point algorithm, we obtain a set of peaks and troughs for the variables analysed. Table 1 presents these and the main features related to the duration, amplitude, and delay. 6 All variables present two complete trough-to-trough cycles.
Analysing firstly the immigrants' data, we can see that the average duration of the GDP contractive phases is shorter than the duration of the IF and IM, 10.7 quarters in front of 12.6 and 19.0, respectively. Instead, during expansion phases, the average duration of immigrant employment, females and males, is quite similar to the GDP (9, 9.5, and 9.5). Regarding the average lag, the males' employment lags in the troughs and in the peaks, while the females' coincides in the troughs and leads in the peaks. In addition, the greatest amplitude, both in expansion and in contraction (absolute terms), is shown by GDP, followed by males.
If the natives' data are taken into account, some differences should be noted. In contraction phases, the natives' average duration is shorter than that of the immigrants' and closer to the GDP, however, the losses are higher than those of the immigrants. That said, altogether, the female immigrants' employment is the most resilient to the contraction periods. In expansion phases, the average duration of the natives' employment is greater and farther from the GDP average duration and the females' (males') average amplitude show better (worse) results than that of the immigrants. 5 Another well-known approach to identifying the turning points is the NBER methodology, which is provided freely by the Spanish Economic Association. It could be thought that calculating turning points under different methodologies could be a way to improve the analysis and to validate the results. However, analysts do not recommend adopting a multidimensional perspective which implies different dating methodologies and different censoring rules. There is no sense in using different methodologies because it is probable that turning points do not coincide, thus, the results of one methodology would not be useful to increase the robustness of the results of the other so they suggest choosing a specific method instead of several (see for instance Abad et al. 2000). Some findings can be deducted. First, immigrant employment displays a quite similar possibility of continued expansion over time than does economic activity, and greater difficulty in breaking down contractions. Second, during upward episodes, the capacity of the economy to generate immigrant employment is lower than its own capacity for growth. Third, in the upward periods, the Spanish economy is more prone to create male immigrant employment than female, but, in the downward periods, females have higher resilience to losing their employment. Compared with the natives, immigrants break later (sooner) their contraction (expansion) phases, that is, they are more sensitive to the GDP upturns. Overall, the Spanish labour market destroys (creates) less (more) immigrants' jobs than natives'.
Since the immigrants' employment is sectorally specialised in specifics jobs, their relationship with GDP is largely due to their relationship with such sectors. We have disaggregated the GDP into four economic sectors: agricultural, industry, construction and services. Table 2 shows the sectoral turning points. The male immigrants' employment is quite concentrated in the construction sector; during the period of study, construction averaged about 20%. 7 By comparing the male immigrants' turning points (Table 1) and the turning points of the four sectors (Table 2), we can confirm that the male immigrants' employment is driven by the fluctuations of the construction sector. The framework of the female immigrants' employment is more specific than that of the males. Throughout the period, on average more than 90% are employed in the services sector, which means that their relationship with the GDP is largely determined by their relationship with the services sector. This can be seen in Tables 1 and 2. The features of the cyclical phases affect the synchronisation of cycles. Synchronisation is defined as the amount of time two series (i, j) are in the same phase. The analysis of cycle synchronisation is performed in two ways. First, as expounded by Harding and Pagan (2002), we use the concordance index (I), which for two cycles (i, j) is defined as: where S it (S jt ) is a binary variable that takes the value 1 when i (j) is in expansion and 0 when it is in contraction, and T is the number of observations. The index varies between 1 and 0, where 1 denotes perfect concordance and 0 denotes perfect absence of concordance. The concordance index is not able to determine whether the co-movements are statistically significant. To overcome this difficulty, the correlation coefficient (ρ) between S it and S jt is estimated using the generalised method of moments (GMM) proposed by Harding and Pagan (2006). The moment condition is: where μ s and σ s are, respectively, the mean and the standard deviation of the series S it and S jt . This yields the following estimator: We use the heteroskedastic and autocorrelation consistent (HAC) estimation procedure of Newey and West (1987) with Bartlett weights. The statistical significance can therefore be contrasted using the t-ratio.
The concordance index (I) achieves the values shown in Table 3. The immigrant male index is 0.73 and the female one is 0.68. The second indicator, the t-student of the correlation coefficient estimation between S ti and S tj (ρ), shows that only that of the male is statistically significant. This means that, overall, the immigrant employment fluctuations are pro-cyclical, although male employment and the business cycle present stronger pro-cyclical co-movements than female employment and the business cycle. In general, these results are in line with that of the natives, where the main difference is that the pro-cyclical native female employment is significant, this is, it is more coupled with GDP than immigrant female employment.
The indexes above provide information on the synchronisation over the entire period. In order to conduct a disaggregated analysis of synchronisation, we calculate an asymmetry index (AI) that quantifies the trend of synchronisation over time (Larsson et al. 2009, citing Hassler 2003. The asymmetry index between two variables (i, j) in the period t is defined as: where c* is the cyclical component of the series GDP, IF, IM, NF and NM, standardised using the standard deviation (σ): In the period t, lower values of AI imply higher levels of synchronisation, with 0 as the minimum value, which implies that both variables are fully synchronised. Figures 1 and 2 plot the AI and the adjusted linear trend (immigrants and natives, respectively) and below we can find the linear trend estimated equation. Overall, on the one hand, the immigrant female employment's AI is higher than that of the male. On the other hand, the negative slope of the immigrant male lineal trend contrasts with the stability of that of the females, both immigrants and natives, and with the positive slope of that of the native males. The outcomes reaffirm and widen those related to the synchronisation: across the whole period, the male immigrant employment is the most pro-cyclical and, over time, draws closer to the business cycle, while the female immigrant employment remains asymmetrical.

Business cycle and 2007 crisis
We wonder to what extent the crisis initiated in the middle of 2007 modified the cyclical fluctuations of immigrants' employment and whether there are significant differences between immigrants and natives. We separate the AI into two periods: pre-crisis and crisis. The pre-crisis period lasts until 2008-1Q because in this quarter  (Table 1). Figure 3 plots the pre-crisis period. The immigrant employment of females and males presents different behaviour. Both have positive slopes, which means that in these quarters employment held up certain degree of decoupling with respect the GDP; however, that of the females presents greater levels and a steeper slope than that of the males. The female immigrant employment displays more independence with respect to the business cycle than does the male employment. Figure 4 indicates that the natives also have positive slopes and that the native females' employment also reaches higher levels than that of the males, but in this case, the natives' employment goes away from GDP more than the immigrants' employment, this is, it is more asymmetric with respect to the business cycle.

3
The crisis period implies changes in all the groups. As we can see in Figs. 5 and 6, the slopes of males, both natives and immigrants, and native females become negative; hence, the crisis period implies an increase of their pro-cyclical behaviour. Instead, the crisis made the fluctuations of the female immigrants' employment more asymmetric. 8 These results might mean that an 'added worker effect' is present in immigrant employment behaviour, but it is not evident in native employment. It is common in the literature covering economic models of family utility maximisation to conclude that, on the one hand, during downturn periods, a temporary increase of the labour market supply can be found because individuals try to compensate for family income losses, due to their partners' job loss, by entering the labour market or by increasing their working hours. This is known as the 'added worker effect' (Humphery 1940;Woytinsky 1940;Long 1958;Mincer 1962Mincer , 1966. 9 In general, it seems to be a female phenomenon, 10 which could explain why females are more counter-cyclical in response to the business cycle (Hotchkiss and Robertson 2012; Belaire-Franch and Peiró 2015). Fromentin et al. (2016) state that the employment of immigrant males has been severely hit by the crisis, while it affected women only slightly. The crisis empowered working women because for some immigrant families their wages became the main or the only family income (Domínguez-Mújica et al. 2014). This is what our findings show. Immigrant females' employment is less pro-cyclical, both in expansion and in contraction phases, and tends to move away from the business cycle. This behaviour becomes more evident during the 2007 crisis. Hence, an 'added worker effect' might have been the answer for Spanish immigrant families whose employed members experienced layoffs or restrictions in work hours.

Does origin matter?
We have found different behaviour between immigrant females' and males' employment, which has become more acute after the 2007 crisis. We introduce in this section a new variable in the analysis: the origin of the immigrants. We study to what extent origin plays some role in immigrants' employment fluctuations both pre-and crisis. We carry out a shift-share analysis to find out whether origin matters. We follow the steps below.
1. We divide immigrants into five categories (i), 11 based on their birthplace: Europe, Africa, North America, Central and South America, and Asia. 12 2. We define the pre-crisis amplitude (A) of IF and IM as the percentage of change of these variables between the last GDP peak before the crisis (2008-1Q), and the first GDP trough of the period of study (2004-4Q). In the same way, the crisis amplitude (A) of IF and IM is the percentage of change of these variables between the last GDP trough of the crisis period (2013-3Q) and the last GDP peak before the crisis (2008-1Q).
In both cases, we define A as: where, c is the the cyclical component of the series IF and IM; t is the initial period; t + m is the final period; i is the origin of the immigrants; and j is the gender.
3. We apply a shift-share analysis to identify if origin determines the evolution of the immigrant employment cyclical components. The expression (1) can be rewritten as: where, A n × c ij (t) is the growth that the cyclical component of the immigrant employment of origin i and gender j would have observed if it had grown as the immigrant employment as a whole. (A in − A n )× c ij (t) indicates which part of the discrepancy between the growth of immigrant employment as a whole and that of c ij , is due to the fact that immigrant employment of origin i has grown differently from that of the whole (we named it the Origin-Mix effect: OM nj ).
(A ij − A in )× c ij (t) indicates which part of the discrepancy between the growth of immigrant employment as a whole and that of c ij is due to the different growth of the gender j in the origin i (Gender-Share effect: GS ij ). 4. We plot an XY-graph where in the abscissa is represented as GS ij and in the ordinate OM ij . In order to show a quick overview of the divergences, we define each quadrant with respect to the shift-share analysis as follows: Upper-right quadrant: advantages both in origin and in gender. Bottom-right quadrant: advantages in gender and disadvantages in origin. Upper-left quadrant: advantages in origin and disadvantages in gender. Bottom-left quadrant: disadvantages both in origin and in gender. 13 Figure 7 indicates that 60% of the categories obtain advantages in origin, with North American immigrants standing out. Among this percentage, gender is significant for European and African immigrants. In the first case, females have disadvantages while in the second case are males who have them. Asian and South American immigrants show disadvantages in origin and gender is not significant for Asian immigrants but is the determining variable for South American. In this sense, South American males present advantages and, instead, females present disadvantages. It should be noted that the worst position is being a South American female. Figure 8 denotes that over time, all cohorts have undergone changes. The categories 13 In the crisis period, we work in absolute terms to be able to express the advantages and disadvantages in gender and origin in the same way as shown for the pre-crisis period.  In order to quantify the losses (gains) associated with the crisis, we calculate DOM nj as the difference between the OM nj in the crisis period and in the pre-crisis, and DGS ij as the difference between the GS ij in the crisis period and in the pre-crisis: Table 4 shows the results. According to the DOM nj , the origin that has lost the most advantages with the crisis is North America followed by Africa and Europe. In contrast, South America has improved the most, followed far behind by Asia. If we take into account the DGS ij , the difference between males and females is evident. All female values are positive, which means that the crisis has improved their relative situation. It is worth noting that the South American females have gone from the worst situation to the best. Table 4 also contains the Total-Shift effect (TS ij ), calculated as the sum of DOM nj and DGS ij . The figures stress that, compared to the precrisis period, 60% of categories have worsened, including all the male categories (except Asians) and females from North America and Africa.
So far we have carried out the analysis of the origin based on the immigrants' birthplace, but given that the economic reason is the main cause of the Spanish migratory flows, it is necessary to explore categories of immigrants from an economic perspective. Thus, we carry out a shift-share analysis to find out whether economic origin matters. To define the economic categories, we follow the World Bank methodology and we divide the countries into five income groupings (i), according to the GDP per capita, PPP, constant 2011 international $, (GDP pc ), of their country of birth: Very high (VH) ≥ 51,000 GDP pc High (H) ≥ 38,000 GDP pc Mid-high (MH) ≥ 26,000 GDP pc DOM nj = OM nj(crisis) −OM nj(pre -crisis) DGS ij = Gs ij(crisis) −GS ij(pre -crisis) . where, c is the cyclical component of the series IF and IM; t is the initial period; t + m is the final period; i is the economic categories (VH, H, MH, ML, L); j is the gender; A is the amplitude; Economic origin-mix effect: EOM nj ; Economic gendershare effect: EGS ij . Figure 9 allows us to stress two main facts. First, in the pre-crisis period, the economic origin advantages are concentrated with the employees from countries with Pre-crisis economic shift-share effects 14 The five categories have been established dividing the difference between the highest GDP pc and the lowest between five and applying the result to define each quartile. Since the Norway GDP pc reached an extremely high value, it has been left out of the calculations. Table 8 in the "Appendix" shows the countries in each category. mid-high GDP pc and higher categories. Second, regardless of category, females have disadvantages and, in general, males have advantages. The crisis changes this situation. Figure 10 plots a relative fall of the advantages in the three highest ratings of the GDP pc and in the males. The female outcomes reveal that economic origin plays an important role. In the previous shift-share analysis, where the birthplace was the classification criterion, the crisis period implied that all female categories had advantages; however, the results change when economic criterion is used in the analysis. On the one hand, the advantages appear in the mid-high GDP pc and lower categories; on the other hand, the two highest GDP pc categories show disadvantages.  As in the previous analysis, in order to quantify the losses (gains) associated with the crisis, we calculate DEOM nj as the difference between the EOM nj in the crisis period and in the pre-crisis, and DEGS ij as the difference between the EGS ij in the crisis period and in the pre-crisis. Table 5 shows the results. According to the DEOM nj , the economic origin categories that have lost the most advantages with the crisis are those with the highest ratings of GDP pc . In contrast, the lowest has achieved a slight improvement. If we take into account the DEGS ij values, the difference between males and females becomes evident. All female values are positive, especially mid-low, mid-high and low categories. Hence, as in the previous shiftshare analysis, we confirm again the greater resilience of females.
Different reasons explain both shift-share results. First mid-high and lower GDP pc categories include, in general, immigrant employment concentrated in sectors highly sensitive to the economic cycle, such as construction or industry, where most workers are men (see, for example, Carrasco and García-Pérez 2015;Motellón and López-Bazo 2015;Murillo-Huertas and Simón 2017). By contrast, female employment is more concentrated in activities such as elder care, health care, and domestic work, which are less cyclical and act as 'shelter sectors' (Aysa-Lastra and Cachón 2012). These facts are especially significant in the Spanish economy, where the construction sector gathered 20% of the male immigrant employment, and where roughly 60% of the female immigrant communities are domestic and care workers, cleaners, and waitresses. Furthermore, most of the females from South America and the Philippines have the additional benefit of speaking the same language and are socially much closer to Spanish culture than other immigrants, which is very important in those kinds of jobs. 15 Another question that should be taken into account is the process of return produced as a result of the depth of the crisis. Since the crisis emerged, the Spanish Government promoted a process of return by financing voluntary return programmes. According to the data of the Ministry of Employment and Social Security, from 2009 until 2016, 31,660 immigrants took part in the programmes, where 2009 had the highest percent at 27%. Parella and Petroff (2014) analyse these voluntary return programmes and indicate that they did not have the expected success. The main reasons stressed by the authors are the lack of incentives, the rights lost by the immigrants if they decided to return, and the perception that the origin country situation is worse than that of Spain.
However, added to this official exit, there was a "spontaneous return", that is, returns of immigrants with their own resources without being involved in any programme. 16 The Residential Variation Statistics (RVS), provided by the Spanish National Institute of Statistics, is a good indicator of any way of returning by reporting on the immigrant cancellations from the Municipal Census Registry (Padrón Municipal, in Spanish terminology). 17 From 2008 to 2016 the figure reached 2.9 million, which implies that during these years, annually, between 6 and 7% of all immigrants left Spain. In any case, returns reduced the impact of the crisis on the immigrant population.
In our study, returns allow us to reaffirm the greater effect of the crisis on the males since the profile of the immigrant who returns is a working-age male aged 25-35 years. On the other hand, the main destination countries are the leading contributors of the past migratory flows: Romania, Colombia, Ecuador and Morocco (Parella and Petroff 2014); these countries worsened the males' situation during the crisis with regard to the pre-crisis period (Tables 4, 5, and 8 of the "Appendix").
The worse results of the immigrants coming from the highest GDP pc categories are explained by the fact that they generally are skilled and have higher levels of education. Although some studies, such as those by Medina et al. (2010) and Miguélez et al. (2014), state that higher levels of education act as protection for immigrants' employment, our findings do not seem to agree. In our opinion, there are two main factors that might explain why the worst change observed corresponds to European and North American immigrants. On the one hand, the nature of the 2007 crisis also implied the destruction of many high-skill jobs. The literature suggests that immigrants coming from developed countries perform better in the destination country than the immigrants from developing countries (see for instance Simón et al. 2008 or Díaz-Serrano 2010). However, the 2008 economic crisis changed this generally accepted view significantly. On the other hand, higher education levels allow people to search longer for alternative employment after becoming unemployed (Xuanren and Sakamoto 2016). Both facts may be behind the worse situation of those cohorts during the crisis.

Conclusions
Our paper contributes to the literature that seeks to analyse the immigrant employment cycle. The study provides evidence on the interaction between Spanish, immigrant and native business cycles using data over the period 1999-2017. The paper had three main goals. First, we identify and compare the phases of the cycles, pointing out their main features and their synchronisation level. Second, we point out the effects of the 2007 crisis. Finally, we examine the role played by the origin of immigrants in our results. The entire inquiry is carried out by distinguishing between female and male immigrants.
We highlight the following results. First, in the Spanish economic upturns, the labour market is more prone to create immigrant males' employment than females', while in the downturns females' employment shows more resilience. Second, with regard to the natives, immigrants remain unemployed longer in downturns and are laid off before the economy reaches its peak. However, a stronger gain of immigrants' employment appears to be present in periods of expansion and a weaker loss in periods of contraction. Third, in the contraction periods, immigrant females present the highest chances of keeping their employment.
The synchronisation analysis has demonstrated that the male immigrants' employment is more synchronised with the business cycle than that of the females. We do not find significant differences between this trend and that of the natives. Also, by including in the analysis the effects of the 2007 crisis, the outcomes indicate that this behaviour continues. Female immigrant employment still shows more decoupling with respect to the business cycle than male employment, which becomes even closer to the business cycle than in the pre-crisis period.
These results are mainly explained by three factors highlighted in our analysis. The first factor is a significant concentration of the male immigrant workers in sectors such as construction or industry, which are highly sensitive to the business cycle. A second factor is the increased presence over time the female immigrant workers in certain services related to health care, people care, and domestic work, which have demonstrated less sensitivity to the business cycle. The third factor is the presence of an 'added worker effect' which pushes immigrant females to apply for jobs when the employed members of their family are fired or experienced restrictions in their work hours.
The evidence from the birth place shift-share analysis confirms that the origin of immigrants matters. We find that, in the course of the business cycle, different immigrant groups are affected in different ways by the economic up-and downturns. Furthermore, all cohorts have undergone changes over time with respect to their relation with the business cycle. The categories that have experienced the worst changes between the pre-and crisis periods are North American and European male immigrants. On the other hand, South American and Asian females are those who have improved the most.
The economic shift-share analysis underlines the worse results of the immigrants coming from the highest GDP pc categories. During the crisis years, the Spanish labour market destroyed many high-skill jobs. Hence, although it is true that the crisis hit the immigrants from developing countries who were hired in low-skilled jobs, it is also true that the situation for immigrants coming from high GDP pc countries, generally high-skilled, worsened with the crisis.
Both shift-share analyses corroborate that the effect of the crisis had less influence on the lay-off rate for females than males. Thus, the birth place and the economic origin reaffirm the turning points and the synchronization analysis which state the greater resilience of females in front of the crisis.
If the origin of the immigrants matters and business cycle fluctuations affect their employment differently than that of the natives, it seems necessary to ask about the effectiveness of the different migration policies followed by the Spanish government before and during the economic crisis. We argue that if one subgroup is significantly different from the others, specific policies for them should be defined. However, this was not the reality.
This paper does not address the migration policies. The literature in this field can guide us further. In this sense, Sánchez (2011);Conejero (2012) and Torres et al. (2017) point out that the immigration measures were excessively focused on the control and admission processes, being neither diverse nor proactive. In expansion phases, they did not try to take advantage of immigration as a driving factor on the labour market and, in contraction phases, they did not mitigate their weakness. Sometimes they have been defined as a reactive and chaotic process of decision making. Also, while it is true that policies of immigration took into account the dimension of integration, Laparra and Martínez (2008) state that the measures were shared by three administration levels: central, regional and local, and so often their main properties were the lack of coordination and funds.
In addition, the measures adopted by the Spanish government in reaction to the crisis worked in three main ways: reducing the work permits quota, hardening of the Foreign Residents Law and voluntary returns programmes. Studies by Domínguez-Mújica et al. (2014) or Carrasco (2014) point out that the figures show that these restrictive measures failed. In times of contraction, restriction measures are a bad way to fight unemployment. It is a mistake to believe the "buffer theory" is going to work. The measures could have been replaced by workers' requalification policies, considering immigrants' specificities according to origin and according to gender instead of focusing on restrictions or voluntary return programmes.
We conclude indicating that, we demonstrate that the cycle fluctuations did not equally affect the groups and that the origin and gender matter, in our opinion, such divergence should have been recognised instead of treating all groups homogenously and as if temporary workers were only needed in times of expansion. As the literature in this field indicates, the measures adopted by the Spanish government were not appropriate to optimise the effect of the economic growth and to minimise the effects of the downturns in the immigrant labour force. Policy makers should take note of the immigrants' divergences in order to implement more efficient migration policy measures to address future shocks.