Does remittance to Africa really reduce labor force participation and increase reservation wage? A Quasiexperimental

Transcription

1 Does remittance to Africa really reduce labor force participation and increase reservation wage? A Quasiexperimental evidence from Ghana Edward Asiedu University of Ghana Business School (UGBS) University of Passau, Germany Nurokinan Chimbar University of Ghana Business School (UGBS) 1

2 Does remittance to Africa really reduce labor force participation and increase reservation wage? A Quasi-experimental evidence from Ghana 1 Edward Asiedu University of Ghana Business School and Chair of Development Economics, University of Passau, Germany Nurokinan Chimbar University of Ghana Business School Abstract In 2015, remittance receipts into Africa was about USD 67 billion, an equivalent of 12 percent of the total global remittances. Over 50 percent of this figure (USD 39 billion) was sent to countries in Sub Saharan Africa (SSA) with Nigeria, Ghana, Zimbabwe, Senegal, Kenya and Uganda receiving the largest proportions. Ghana ranked second after Nigeria, with total remittance of USD $4.9billion, being 7 percent of overall remittances into Africa, 13.6 percent of Gross Domestic Product (GDP) and the second largest external flows behind Foreign Direct Investment (FDI). Since remittances remain a crucial element in the migration ecosystem, we set out to examine how remittances, being an outcome of labour mobility affects labour supply and reservation wage given that migration takes away members of the household labour force. We used micro level data from Ghana Living Standards Survey (GLSS) to estimate impact at the individual level. A sample of remittance receiving individuals were compared with non-recipients on common pretreatment covariates using propensity score matching technique. With different estimators for average treatment effect on the treated, we found a strong negative association between remittances and labour force participation (LFP) but no evidence to support the a priori hypothesis that the former increases reservation wage and reduces hours of work. Finally, using a sub-sample of the employed, we examined whether reservation wage of recipients responds differently. Results from this estimation was statistically insignificant. Based on this, we conclude that remittances reduce labour force participation of recipients in Ghana. At the household level, we found that about 90 percent of the remittance-receiving households channeled remittance income into consumption expenditure rather than investing in the productive sector. As labour markets remain non-competitive and people are able to meet their basic needs with remittance income, the incentive to participate in the labour market declines. High unemployment rates in Africa contributes to this phenomenon when remittance income more than compensate the discouragement people face due to lack of jobs. Based on this, we suggest that African governments should implement wage policies that make the labour market competitive, so that even with alternative sources of income such as remittances, people will be encouraged to participate in the labour market. There is also the need for policy environment that will create the incentives to attract remittance income into productive sectors. This will encourage self-business and serve as an economic tradeoff for skilled labour loss from migration. Keywords: Remittances, Labour Force Participation, Reservation wage, Quasi experiment, Africa, Ghana 1 s: Asiedu, edasiedu@ug.edu.gh(corresponding author); Chimbar,nchimbar@gmail.com. We are grateful for valuable comments from Michael Grimm on an earlier draft. 2

3 1 Introduction Because of instability in the global financial environment, the amount of external financial flows to developing countries in general has been on the decline in the last decade. However, personal remittances in the midst of this global crisis continued to rise, an indication that the latter is less volatile relative to other external flows (Ratha, 2005; Lubambu, 2014). In the case of Ghana, personal remittance flows have seen dramatic increases since the 1970s. The average amount of remittance from 1979 to 1990 was a little over US $2.8million (International Monetary Fund, 2017). Two decades later, the share of remittances increased to about US $135million as of 2010 and US $2.1billion in Over these years, personal remittances have been the second largest external flow to Ghana behind Foreign Direct Investment (FDI) and almost twice Official Flows. In 2015, remittance receipts in Ghana stood at US $4.9 billion being 13.6 percent of Gross Domestic Product (GDP) and 7 percent of the total remittance flows to Africa. This exceeded the total value of FDI, which recorded a little over US $3.1billion (a decline of about 5 percent) and makes Ghana the largest recipient of remittances in Africa after Nigeria. While these are registered remittances, Mazzucato, van den Boom, and Nsowah- Nuamah (2005) indicate that based on data from Ghana Transnational Networks (Ghana TransNet) 2, unregistered remittances to Ghana accounted for about 65 percent of total remittances. Given that about US $4.9 billion was received in 2015, what this estimation 2 Transnational Networks is a research program that investigates how transnational networks affect local economies in Ghana and in The Netherlands. Transnational networks are formed when migrants create and maintain linkages with people in their home countries and through the networks, goods, information, money and ideas flow. 2

4 suggests is that, total remittances to Ghana could be around US $9 billion. Apart from being a source of foreign exchange earnings to government, at the micro level, remittances serve as an important source of household finance and can trigger a number of welfare benefits. Figure 1 shows top three (3) external capital flows to Ghana in the last decade and half. $6, $5, $4, $3, $2, $1, $- IDA grants Personal remittances FDI Official Flows Figure 1: Personal Remittances and Other Inflows to Ghana ( ) Source: IMF balance of payment data Generally, remittances are viewed as transfers sent by migrants to their households or relatives at home. This is in line with how the IMF balance of payment data classifies personal remittances. The sixth edition of the IMF s Balance of Payment Manual explains personal remittances as transfers either in kind or cash made by workers employed in an economy where they are not resident to their relatives at home. This makes the benefits of labour mobility (remittances) an important topic to discuss because migration is viewed as brain drain in the country of origin and may have implications for labour supply. However, at the same time, Ratha (2003) believes that output losses resulting from skilled labour mobility could be offset by remittances. For instance, the incomes of recipient households improve as a result of remittance and triggers multiplier effects on both consumption and 3

5 investment leading to output growth (ibid.). But, there are mixed empirical findings on the actual impact of remittances at both the micro and macro levels. Elsewhere, studies on the effect of remittances on labour participation or supply abound. However, in Africa, studies have tended to focus on the impact of remittances at the macro level: macro-volatility, socioeconomic development and general welfare of the poor. For instance, Quartey and Blankson (2005) looked at the impact of remittances on macrovolatility on poor households; Quartey (2006) focused on the general welfare of the poor while Mintah and Nikoi (2015) concentrated on the impact of remittances on socioeconomic development in Ghana. The effects of personal remittances on labour force participation remain largely an unstudied area and particularly so using data from the micro level. Therefore, this research intends to investigate whether remittances to Africa lead to reduction in labour force participation and increases reservation wage with evidence from Ghana. Labour Force Participation (LFP) measures the proportion of a country s working-age population that engages actively in the labour market, either by working or looking for work. It provides an indication of the relative size of the supply of labour available to engage in the production of goods and services (Ghana Statistical Service, 2014). The labour market stimulates economic growth because of the output contribution to GDP. Thus, factors that have the potential to influence labour supply patterns are worth studying. In Ghana, the structure of the LFP rate has not changed significantly since From 1984 to 2013, about 75 percent rise in the LFP rate has been recorded compared to almost 4000 percentage rise in personal remittances over the same period. To have a 4

6 comprehensive coverage of labour market issues, data from round six (6) of the Ghana Living Standards Survey (GLSS 6) was used. The GLSS 6 is a nationwide household survey that collects data on labour market activities, migration and remittances every five (5) years. Using the large representative data from Ghana Living Standard Survey, we find that remittances impacted negatively on labour force participation i.e., those who are actively engaging in the labour market. This result is consistent with findings by Cox-Edwards and Rodriguez-Oreggia (2009) for Mexico. Obviously, the effect of remittances on labour force participation is not due to the proxity of the developing country to the closest developed country. Ghana, which happens to b relatively farther away from Europe (the closest developed region) compared to the distance between Mexico and the United States, we still find a negative impact of remittances on labour force participation. Thus, the negative effect of remittances on labour participation is not merely due to aspiration or easiness to migrateto a developed country, which in turn reduces labour force participation. Further, we estimated the impact of remittances by looking at hours of work and reservation wage as outcomes. We found no evidence to support the relationship between remittances, hour of work and reservation wage. This is incongruent with the hypothesis suggested by the neo-classical labour-leisure model that remittance-receiving households will have high wage expectations compared to non-recipients. Finally, receiving remittance did not have any effect on labour hours for household members who were in active employment in the last one year. About 90 percent of remittance income was spent on consumption rather than production where the multiplier effect can be significant. We note that consumption will 5

7 continue to dominate in terms of the application of remittance income because the labour market in Africa remains non-competitive. The effect of discouragement from unemployment eclipses the incentive to remain active in the labour market especially when remittance income sufficiently provides the basic needs of household members. The rest of the paper is structured as follows. Section 2 presents the background and context of the study. Section 3 presents related literature, particularly literature on the impact of remittance on labour force participation in other regions. Section 4 touches on the theoretical connection between remittances and labour force participation. Section 5 explains the data and the empirical strategy adopted for our empirical estimation. Section 6 discusses the empirical results. Finally, section 7 provides a summary of our findings, conclusion and policy recommendations. 2 Background and Context Given that remittance income is an important outcome of labour mobility and the latter affects labour supply due to an absent member of the household, there is the need to study how remittances, being the expected economic tradeoff for skill labour loss, affects the labour supply of migrant s household members. The sensitivity of the overall labour supply situation in Sub Saharan Africa relative to the stability and increasing flows of remittances to the region is important because the labour market serves as the principal source of income and livelihoods for many household members. Establishing the nexus between remittance and LFP is also relevant because taking Ghana for instance, the structure of labour force participation rate has not changed significantly relative to the 6

8 consistent growth in remittance volumes to the country. While national LFP rates hover around percent since 1989 (25% increase), remittances have increased by over 4000 percent in the last 3 decades. Estimates of long-term global LFP rate show that a further decline to below 63 percent is expected of the global workforce by 2030 and the falling participation rates could be due to changing demographics and discouragement arising from lack of jobs (ILO, 2015). However, remittances also have the potential to increase recipients non-labour income which can affect labour reservation wages. Within the context of the prevailing world economic order where real wages continue to lag behind productivity as a result of the world economic crisis, the aggregate effect of lower wages and employment can consequently lower participation rates. While this is a major development concern because of its impact on growth, studies in Ghana and for that matter Africa, have paid little attention to how remittance flows affect labour supply. Available studies have only focused on the impact of remittances on poverty and growth with the latter being at the macro level. The story of direct recipients of remittance as far as their participation in the labour market in Ghana and Africa as a whole is concerned, has not been told in detail and this explains why this study empirically examined whether remittances influenced labour force participation of recipients, labour hours and reservation wage. The study is significant in two principal ways. The findings will augment the literature available by situating the impact of remittances in an African context. While this will be 7

9 useful for people in academia, stakeholders such as government can leverage on the findings to design policy frameworks to maximize the economic tradeoff of skilled labour migration. The scant empirical evidence on Ghana and for that matter in Africa makes it open for vague assumptions to be made based on experiences from other jurisdictions where studies have been conducted. But, because of geopolitical, economic, cultural and social differences across countries, findings may not be uniform and this gives relevance to this study. It is also expected that the findings will help players in the remittance chain with specific information to support their ongoing initiatives. Remittance flows and Trend Globally, remittance flows have been increasing in volume over the last two and half decades. Between 1990 and 2000, the share of remittances across the world increased from US $67 billion US dollars to US $121 billion, an equivalent of 79 percent (World Bank Indicators, 2017). A decade and half later, international remittances increased by about 320 percent from US $131 billion dollars in 2001 to US $552 billion dollars as of Over the same period, FDI and Official Development Assistance (ODA) and Aid increased by 173 percent and 120 percent respectively. While there are regional and country level variations in terms of receipt of remittances, the latter has established its role as the second largest source of financial flows next to FDI across the world. Figure 2 illustrates the trend of global remittance flows. 8

10 FDI ODA & Official Aid Remittances Figure 2: Global Trends in External Flows Source: IMF balance of payment estimates In Africa, total remittance receipts as at 2015 stood at 11.2 percent of the total global remittance flows. Since 2010, remittance receipts in Africa have been increasing except in 2013 when there was about 1.37 percent decline over the inflows in 2012 but resumed an upward trend in 2014 and 2015 (Table 1). The continental picture is reflected in countries south of the Sahara and Ghana is no exception. What is significant is that between 2014 and 2015, Ghana recorded about 148 percent rise in total remittance receipts (ibid.). The increase in volume and stability in remittances has catapulted into visibility its relevance in development finance as compared to other external flows. Table 1: Remittance Flows to Africa relative to the Global Trend 2010 (000,000) 2011 (000,000) 2012 (000,000) 2013 (000,000) 2014 (000,000) 2015 (000,000) Ghana 135 2,134 2,155 1,863 2,007 4,982 SSA* 31,265 35,761 36,246 36,366 36,944 39,713 Africa 52,433 59,581 64,461 63,580 67,256 67,377 World 417, , , , , ,317 Source: IMF balance of payment data SSA* means Sub-Saharan Africa 9

11 Between 1990 to early 2000, ODA and official aid were more dominant in Africa than FDI and remittances (Figure 3). The dominance of ODA and official aid relative to FDI is understood because the political climate in Africa at the time was unstable and many economies in the continent before the 1990 showed a significant decline in real per capita Gross Domestic Product (GDP). Culagouski cited in Husain and Underwood (1991) indicated that almost 20 percent decline in real GDP was recorded between 1974 and Out of 40 countries in Sub Saharan Africa (SSA), only 10 recorded an average growth rate exceeding 5 percent a year (Husain and Underwood, 1991). Furthermore, in half of the 40 countries, Husain and Underwood noted that economic growth fell short of population growth. Around 2010, remittance inflows into Africa surpassed both ODA and FDI. The trend continued through to In Ghana, ODA and official aid showed resilience relative to FDI and remittances from 1990 until around 2006 when it fell below FDI and subsequently below remittances around 2012 (Figure 4). The decline in official aid and ODA can be explained by the fact that Ghana attained a middle-income status and coupled with the discovery and drilling of crude oil, donors began to redirect official aid to other countries since Ghana is no more classified as a country in need of aid. 10

12 USD $(Millions) Millions USD $ Millions USD FDI ODA & Official Aid Remittances Years Figure 3: External Flows to Africa Source: IM balance of payment estimates FDI ODA & Official Aid Remittances Years Figure 4: External Flows to Ghana 50,000 40,000 30,000 20,000 10,000 - SSA Years Figure 5: Remittance Flows to Sub Saharan Africa Source: IM balance of payment estimates 3 Literature Review 3.1 Impact of Remittances on LFP Different country-level impact studies on remittances and labour force participation have generated mixed results. But generally, empirical findings have drawn attention to three key outcomes: inverse relationship, positive relationship and no relationship. The findings have largely been drawn from country level data on living standards, which are collected nationally. Although, 11

13 the equivalent of this data in Ghana is the Ghana Living Standards Survey, studies on remittances in Ghana have tended to focus on the macro level impact compared to the micro impact. Between 1990 and early 2000, Jamaica experienced a phenomenon of high unemployment rate and a declining LFP rate (Kim, 2007). With this high unemployment rate, real wages increased from around J$900 to J$1, 200 between 1990 and While it was expected that in a functioning labour market, the increase in real wages will be transient in view of the persistent unemployment rate, increasing trend in real wages continued. To understand why labour supply declined even when real wages were increasing, Kim (2007) examined whether remittances played any role in the reservation wage and individual labour supply decisions in Jamaica. According to Kim, both high unemployment and increasing wages point to two things: an inefficient labour market that is not able to absorb excess labour or too high reservation wage among the workforce given the level of productivity. Where the labour force is significantly aging, Kim believed that the average reservation wage would increase given the fact that the aged workforce will have reduced interest in working. To compensate the aged workforce for the loss of interest in work, Kim argued that employers will pay a premium for experienced workers because the population of young workers is shrinking. This does not only imply that increased reservation wages can be accounted for only if the workforce is significantly aging, there are other potential factors that can contribute to increased reservation wage and one of them is remittance income. Jamaica is one of the countries where remittances constituted about 2 percent of total household income in 1990 and about 6 percent by the year 2000 (ibid.). Apart from being one of the countries with the highest household recipients of remittances in the Latin America and the Caribbean (LAC) 12

14 region, Jim noted that at least about 25 percent of Jamaican citizens lived outside Jamaica and the growth in Jamaican population overseas is estimated to be about twice the national population growth rate of Jamaica. Using data from the Survey of Living Conditions (SLCs) and Labour Force Survey (LFS) in Jamaica which considered a sample of 2,000 households and 7,000 individuals, Kim (2007) provided evidence that remittance income negatively impacted on labour market participation decisions. The findings showed that if an individual received a remittance, he or she was 3.6 percent less likely to participate in the labour market. However, in terms of the hours worked per week, the study did not find any significant correlation with remittances. The findings were based on a constructed pseudo-panel and cross sectional data from both the SLCs and LFS. The two surveys maintained the same clusters in the surveys over time. From the cluster level, household indicators were traced because participating households did not change. Despite this, one methodological limitation is the likelihood of having different individuals within a household responding to the surveys over time. While this may be corrected when constructing the pseudo-panel data, the results will be weakened by the reduced sample. Nevertheless, Kim concluded that increased remittance flows have a negative impact on labour force participation because it increases the reservation wage. In Rodriguez and Tiongson (2001), migration was seen to play a role in labour supply decisions of both migrants and non-migrants much the same way remittance also impacted on labour force participation. According to them, labour supply decisions will depend on the type of task a migrant 13

15 performed in the household production process before migrating. Where migrants were considered substitutes to non-migrants in household production, there was the likelihood that migration will lead to decrease in labour supply decisions. On the other hand, labour supply was likely to increase if the migrant played a complementary productive role in the household prior to embarking on the migration. They further argued that remittances were a part of migration and once non-migrants received remittances, it was seen to be an additional non-labour income and an increase in the latter reduces the participation of recipients in the labour market. Both Rodriguez and Tiongson empirically tested these assumptions using household survey data for overseas contract workers (migrants and non-migrants) in the Philippines in 1991 alongside data from the Labour Force Survey (1991). Estimating the impact of migration, a control group made of non-migrants in Manila with an equally high incidence of migration was used. Evidence drawn from their study suggested that migration affected labour participation and hours worked by non-migrants. In other words, non-migrants tend to consume more leisure than work and this was observed to be higher among males than females. The gender dimension was not only observed when the impact of migration on LFP was measured but also on remittances. Remittances to non-migrant members of the family reduced their participation in the labour market, but was observed to be stronger for men than women. This gender difference is explained by the high involvement of men in paid work as compared to women. Because women already are less involved in paid work, their labour supply is not significantly determined by changes in non-labour income. In the case of men, due to their high 14

16 participation in paid labour, their labour supply is less responsive to wage changes as compared to changes in non-labour income such as remittances (Becker, 1985; Rodriguez & Tiongson, 2001). Using 2002 data from the Mexican National Employment Survey (Encuesta Nacional de Empleo Trimestral, ENET), Cox-Edwards and Rodriguez-Oreggia (2009) also investigated how persistent remittances impacted LFP. Their analysis was based on a propensity score matching with a comparison group in recipient households. Although Mexico offers a rich context to investigate the effects of remittances because of its migration pattern and the flow of the former, the study found that despite the 1.6 million people who sent remittances to Mexico between 1997 and 2002, no evidence of an impact was observed on labour force participation. However, their study found a positive change in the participation within a subsample for women in urban areas. This observation suggested that women who received remittances in urban areas increased their economic opportunities through non-farm activities. In Nicaragua, Funkhouser (1992) reported both positive and negative effects of remittances. Noting that remittances constituted the second largest external flow to Managua aside coffee, the study found that remittances tended to have a positive effect on self-employment for non-migrants but showed a negative income effect on labour force participation. The implication of this finding is that, as non-migrants receive remittances, their participation in wage labour reduces because the remittance increases their non-labour income and raises the reservation wage at which a person will be willing to work. On the other hand, for the positive effect, remittances tend to serve as capital for non-migrants to engage in self-employment activities. Empirical investigation into the establishment of self-business have highlighted capital constraint as a major bottleneck. Testing 15

17 finance-constraint hypothesis, Blanchflower and Oswald (1998) used data on inheritance and gifts found that individuals who have received inheritances or gifts are more likely to run their own businesses. They observed that the effect was not due to individuals inheriting family businesses but using the gifts and inheritance as capital to start self-business. The use of gifts can be linked to the role remittances play in capital formation. Analysing the general economic effects of remittances, Ratha (2005) made a number of observations across different economies. He posited that remittances boost the foreign exchange reserves of recipient countries and contribute to output growth if invested or create multiplier effects in the case of consumption. These effects, according to Ratha offset some of the output losses in developing countries arising from skilled labour loss through migration. Citing Orozco (2000) and Lucas (1985), Ratha mentioned that in five Sub-Saharan African countries, emigration contributed to labour supply losses and reduced crop production in the short run. But this saw a reversed trend in the long run where investment from remittances improved crop production and cattle accumulation. Focusing on macroeconomic determinants of remittances and implications for household expenditure and labour supply in Kosovo, Havolli (2015) noted that results for individuals (per capita) were insignificant in so far as the involvement of the active working age group in the labour market was concerned. The results were uniform for both men and women. However, with the same data, Hovilli realized that the impact of aggregate remittance for households was positive with an increased probability of males becoming economically active. 16

18 In Sub-Saharan Africa (SSA), Ssozi and Asongu (2014) used a production function to estimate the medium through which remittances affected worker output. With a sample of 31 SSA countries, Ssozi and Asongu found that remittances directly increased output per worker if the worker was educated. They observed this outcome by interacting education and remittances because the effects of remittances alone showed a negative relationship. In their view, remittances that are invested in education directly increase workers income. This analysis was done using macro level data, controlling for FDI and ODA. In Mali, Gubert (2002) interviewed households in Kayes area western Mali to test the hypothesis that remittance is a motivation derived from the need to insure households left at home from risk and shocks. Evidence from his study supported the hypothesis that indeed, remittances were a form of insurance against different categories of risk and shocks for household members left at home. However, Gubert was cautious that issues of moral hazards could emerge in the remittance chain just as any form of insurance whereby recipient households could reduce their labour participation efforts to be eligible for financial assistance in the form of remittance. In any ways, Gubert s observation supported the idea that remittances negatively affect labour force participation of receiving households. In Ghana, studies on remittances have not focused on LFP although the GLSS provides national representative data on labour force participation, which can be used for micro level analysis. Studies that have been sighted have focused on other outcomes than Labour Force Participation. Quartey and Blankson (2005) looked at the impact of remittances on the macro - volatility on poor households, while Quartey (2006) investigated the impact of remittances on welfare of the poor. 17

19 The study found that remittances helped to improve household welfare by reducing the effects of economic shocks. Mintah and Nikoi (2015) concentrated on the impact of remittances on socioeconomic development in Ghana. 3.2 Non-Labour Income, Reservation Wage and LFP In examining the sensitivity of LFP to remittances, reservation wage has surfaced as an important element to consider. Reservation wage is how much extra earnings the individual would require to be induced to give up a unit of leisure when he or she is not working at all (Edwards and Rodriguez-Oreggia, 2006). In other words, it is the wage at which an individual is willing to work. In Jamaica, Kim (2007) found that individuals and households with higher remittances were more likely to have a higher reservation wage than families that did not receive remittances. Increased reservation wage implied that people who were employed were more likely to move out of the labour market and those who were unemployed became less incentivized to look for a job. Kim s findings are akin to the observation made by Rodriguez and Tiongson (2001) to the effect that wealth or non-labour income plays a role by way of reducing labour supply because recipient families will demand more leisure. Despite the argument by Rodriguez and Tiongson does not suggest that reducing labour supply were a result of increased reservation wage, higher remittances produce a strong income effect which reduces labour supply. Rodriguez and Tiongson refer to this as the dependence of international migration whose strength is amplified by family dynamics between the migrant and recipients of remittances. 18

20 Cox-Edwards and Rodriguez-Oreggia (2009) also noted the effect of reservation wage on LFP in Mexico. Using persistent and recent remittances to households, they observed what they termed discouraged participation 3 in labour market activities among remittance receiving households. This discouragement according to Cox-Edwards and Rodriguez-Oreggia was because the amount remitted to households more than offset losses associated with emigrant members of the household and this resulted in increased reservation wage of household members who were left at home. 4 Theoretical Framework Labour Leisure Choice Model This study is grounded in the neoclassical economic model of labour leisure choice. The model is used to explain labour supply decisions by looking at the variables that determine whether an individual will choose to work or enjoy leisure. The assumption underlying the model is that individuals will enjoy life or derive utility from consuming goods and leisure, but the former has to be bought in the market using labour or non-labour income. Where individuals depend on labour income to purchase their desired goods, it means people must work to earn income that will be used in purchasing the goods. If people decide to work, they consume less of leisure, but will be financially capable of consuming more goods. Alternatively, if people decided to enjoy more leisure without working, they will be unable to consume more goods, but can afford a cedi value of goods based on their non-labour income 3 The reduced rate of labor force participation as a result of persistent remittances to households left behind (Cox- Edwards and Rodriguez-Oreggia, 2006) 19

21 amount. The model therefore isolates wage and non-wage income as key economic variables that determine the time to be allocated to labour market activities or leisure. This means the enjoyment of life or maximisation of utility by an individual is subject to three constraints: wage income, nonwage income and leisure time. Assuming that the utility is given by u whereas consumption and leisure are denoted by c and l, respectively, then utility derived from consuming both goods and leisure is represented mathematically as follows: u = f(c, l).. (1) Given that consumption is a function of a person s income (labour and non-labour income) and labour wage rate is given by w and the number of hours a person allocates to labour market activity is represented by h, a mathematical notation for consumption in monetary value (Ghana cedis) can be derived as follows if non-labour income is given as r which is independent of the number of hours worked: c = wh + r.. ( 2) Again, if both the number of hours worked plus the leisure a person enjoys is equal to the total time available to an individual and given as T, then total time will be T= h + l and this can be substituted in equation 2 and rewritten as follows because h will be equal to T- l. c = w(t l) + r.. (3) Alternatively, this can be written as: c = (wt wl) + r.. (4) Or c = (wt + r) wl.. (5) 20

22 The above equations are budget constraint equations given on the assumption that wage rate remains constant. With equation 3, in so far as a person decides not to work, it will still be possible to consume r value of goods because r represents the cedi equivalent of non-labour income and this can be income from investment or remittances. To contextualize this model for this study, r will represent remittances received by an individual. To represent equation 3 on a budget constraint curve, wt which is the wage income will be represented as w. In figure 6, a person will be able to purchase goods and service at c0 while enjoying leisure at t0 given that he has a wage income of w. With c0 and t0, the person derives utility at point A where the indifference curve u1 is tangential to the budge line w, t. As the income increases through remittances (non-labour income) denoted by r, total income becomes w+r thus pushing the budget line upwards to w+r, t. At this point, if the wage income effect is strong, the individual will reduce his leisure time to t1 while enjoying a higher utility (u2) at point B. Similarly, if the wage income is weak and the substitution effect is strong, the individual will prefer to stay out of the labour market to enjoy more leisure at t3 but will be able to purchase goods and services at c3 with a utility of u3 because of the effect of remittances. 21

23 Figure 6: Labour Leisure Budget Constraint Curve The decision to continue being in the labour market will depend on whether the effect of remittances received is strong enough to trade off working hours or not. This is what Cox-Edwards and Rodriguez-Oreggia (2009) referred to as discouraged participation. This arises because the income effect increases the reservation wage of recipients and this leads to a reduction in their willingness to participate in the labour market. At any given point in time, where the budget constraint line moves upwards, there is an opportunity to purchase more goods, but hours of work will reduce if the wage income effect is weak. And whenever non-wage income (remittances) is more than the wage income, there is the tendency for a person to substitute work for leisure. 5 Data and empirical strategy The study relied on a national micro level data collected through the Ghana Living Standards Survey (GLSS). The survey is conducted every five (5) years by the Ghana Statistical Service 22

24 (GSS) to assess the living standards of households and individuals on core indicators of welfare, health, economic, social and labour market activities. The first GLSS was conducted in 1987 followed by the second in 1988 and the third in 1991/1992. The fourth, fifth and sixth rounds were conducted in 1998/1999, 2005/2006 and 2012/2013. The GLSS 6, being the most recent survey, sampled 1,200 Primary Sampling Units (PSUs) and 18,000 households but 16,772 households being 93 percent was eventually enumerated involving 72,372 individual observations. The data was collected over a one-year period starting October 18, 2012 to October 17, Empirical Models and Strategies This study is observational by design. In estimating treatment effect or causal inference, estimates can be biased, especially because randomization of treatment is not possible in observational studies (Nicholas, 2007). To overcome this conundrum of possible bias, Khandker, Koolwal, & Samad (2010) proposed the mimicking of randomisation by constructing an observational analogue of the randomised experiment to avoid biased estimation of the impact of treatment arising from the non-randomisation of assignment to treatment group. This often occurs because of wrong model specification where both observed and unobserved variables might not equal zero to pass for exogeneity. To correct for endogeneity, this study used Propensity Score Matching (PSM) to estimate the Average Treatment effect on the Treated. Propensity score matching estimates each individual s propensity to receive a binary treatment using a probit or logit model (Nicholas, 2007). According to Caliendo and Kopeinig (2005), estimating the probability of participation against nonparticipation using a probit or a logit model does not really matter because they return similar outcomes. Unlike direct matching, PSM corrects weaknesses associated with the former by 23

25 addressing issues of multidimensionality and the counterfactual. Cox-Edwards and Rodriguez- Oreggia (2009) used a similar method in estimating the effects of remittances on LFP in Mexico. The problem of counterfactual will not arise in experimental or longitudinal studies because of the opportunity to examine the direct effect of treatment by observing the behaviour of subjects before intervention and post intervention. But, the GLSS does not collect longitudinal data where the same sample of remittance receiving households can be followed over a certain time period. To measure impact, therefore, a comparison of recipient households with non-recipient households has to be done but using PSM. Matching involves comparing treatment and control samples with similar observable characteristics. Where there are no matches, samples are dropped because a common point for comparison does not exist. According to Dehejia and Wahba (2002:151), when; The relevant differences between the two samples are captured in the observable (pretreatment) characteristics or covariates, which occur when outcomes are independent of assignment to treatment conditional on pretreatment covariates, matching methods can yield an unbiased estimate. Assumptions in Propensity Score Matching In PSM, matching is done for both treated and control groups on the pretreatment covariates (x) independent of the outcome variables (Yi1, Yi0). This is called the conditional independence assumption mimicking the randomized assignment (Yi1, Yi0 ), X T). The conditional independence assumption requires that given a set of observable covariates Xi that are not affected by treatment Ti, potential outcomes Yi1, Yi0 are independent of treatment assignment Ti (Khandker, Koolwal, & Samad, 2010). 24

26 This is expressed as: Yi1, Yi0 (Ti (Xi (1) Also, matching assumes a common support where individuals with the same characteristics (x) should have a positive but not perfect probability (pr) of being either participants or nonparticipants. In Heckman, LaLonde, and Smith (1999), the common support condition ensures that treatment observations have comparison observations nearby in the propensity score distribution. The statistical expression of the common support region is given as: 0<Pr(Ti=1 Xi) <1 (2) Unlike direct matching which is done using the covariates, PSM is done using propensity scores (also referred as pscore) which is the probability that an individual will be in the treatment group because of his/her observed covariates. The pscore is estimated using either probit or logit model. This reduces the information about Xi into a single dimension and based on this matching is done. Pr(Ti=1 Xi) =Pr(Xi) (3) Within the common support region, the distributions of the treated group and the non-treated group must be the same to satisfy the balancing property. According to Khandker, Koolwal and Samad (2010), the pscores for the treated and non-treated groups can be the same, but if misspecification exists in the equation, the distributions can be observationally dissimilar and therefore the need to ensure the balancing property is achieved. Pr(Xi Ti = 1) = Pr(Xi Ti = 0) (4) 25

27 According to Rosenbaum and Rubin (1983), for the independence assumption to be satisfied, if potential outcomes are independent of treatment conditional on covariates Xi as shown in equation (1), then they are also independent of treatment conditional on propensity scores, Pr(xi). This is expressed as: Yi1, Yi0 Ti Xi Yi1, Yi0 Ti Pr(Xi) (5) Estimating the ATT with PSM Estimating ATT is conditional on satisfying the assumptions of conditional independence and common support. ATT is estimated as the mean difference in the outcome variables Yi1, Yi0 over the common support and weighted on non-recipients of remittances by the propensity score distribution of remittance receiving individuals. The ATT estimator is expressed as: ATTPSM = Ep(xi) Ti=1 {E[Yi1 Ti=1, p(xi)] E[Yi0 Ti=0, p(xi)]} (5) The ATT estimator is also written as follows: ATT PSM = 1 N T1 [ Y i1 i T 1 ω(i, j)y j0 ] j T 0 Source: Smith and Todd, 2005 Where NT1 is the number of remittance recipients and ω(i, j ) is the weight used to aggregate LFP outcomes for the matched non-recipients of remittance j. Determination of Impact Given that the assumptions of conditional independence (unconfoundedness) and common support (overlap condition) are satisfied and the distribution of the treated (remittance households) and control (non-remittance households) groups are balanced, differences in outcomes (LFP) between 26

28 the two can be considered as the impact of remittance. Different ATT estimators can be used to determine mean differences or impact of remittance. We discuss four of the ATT estimators used in this study (nearest neighbor, radius matching, stratification matching and kernel matching). According to Cox-Edwards & Rodriguez-Oreggia (2009:1006) the various ATT estimators try to overcome the problem that the probability of treatment conditional on the value of covariates is a continuous variable, with the probability of observing two units with exactly the same value of the Score being in principle zero. In addition to this, direct nearest neighbour test was run as a second level robustness check on the outcome variable (LFP).) Types of ATT estimators Nearest-neighbour matching ensures that each treatment unit is matched to the comparison unit with the closest propensity score and computes the difference in outcome (Yi1-Yi0) for each pair and then the mean difference across pairs. This is done by sorting all records by the propensity score and searches both forward and backward for the closest control unit(s) (Nicholas, 2007; Cox- Edwards & Rodriguez-Oreggia, 2009; Khandker et al., 2010). Matching with nearest neighbour can be done in two ways: matching with replacement or without replacement (Nicholas, 2007; Khandker et al., 2010). This study used matching with replacement to allow for the same nonrecipient of remittance to be used as a match for other remittance recipients. The choice of matching with replacement over matching without replacement is because it reduces bias and increases the average quality of matching (Caliendo & Kopeinig, 2005). Kernel matching relies on the weighted average of all non-treated individuals to construct the counterfactual match for each treated individual. An advantage of this approach is that low variance is achieved because more information is used. According to Khandker, Koolwal, and 27

29 Samad (2010), kernel matching uses a weighted average of all non-treated units to construct the counterfactual match for the treated units and this is written in the notational format below. Supposing Pi is the propensity score for a treated unit i and Pj is the propensity score for a nontreated unit j, the weight for Kernel matching is given as: ω(i, j) KM = K ( P j P i a n ) k c ( P k P i a n ) Stratification matching ensures that the common support of the propensity scores is subdivided into strata where the treatment effect within each strata is calculated using the mean difference in the outcome variable. Rosenbaum and Rubin (1983) refer to this as the sub-classification or interval matching. Literature provides that five subclasses are mostly enough to remove about 95 percent of bias associated with one single covariate (Cochrane & Chambers, 1965). Radius matching uses the one-to-many caliper matching algorithm to match as many observations within a predefined distance around the propensity scores. In other words, each treatment unit is matched with a corresponding control unit that is within a predefined interval of the treatment unit s propensity score. Robust estimation using Direct Nearest Neighbour Matching In addition to the propensity score estimation. We also computed impact by comparing outcomes between treated and a control group using direct nearest neighbour matching for all variables. According to Abadie, Herr, Imbens and Drukker (2004), direct nearest neighbor allows for multidimensional matching on a set of variable and adjusts for bias of treatment effect without assuming 28

30 homoscedasticity (constant treatment effect); in other words, using heteroskedastic robust variance estimators. Matching with this estimator allows for the estimation of the counterfactual by comparing observations using a distant metric on a subset of variables. This was used as a form of robustness check to the normal propensity score estimators as the outcome improves the model specification. Model specifications The models to test the hypothesis that remittances reduce labour force participation were based on the working force observations aged 18 years and above. First, four models were used involving the total sample. Secondly, nine (9) subsamples were constructed to test impact across all the subsamples. Outcome variables for all these models were labour force participation (dummy), reservation wage (continuous variable), hours worked in last seven (7) days and twelve (12) months. The last two outcome variables were used as proxies for labour force participation. However, hours worked in the last seven (7) days was excluded in the subsample estimation. The sample was split into urban and rural samples, sample of men and women, sample of urban men, and urban women, sample of rural men and rural women and lastly, sample of employed and unemployed. The final set of covariates, which were included in the propensity score estimation, are as follows: Household size (continuous variable), sex of household members (dummy: 1 if hh member is male, 0 otherwise), location of hh members (dummy: 1 if urban & 2 if rural), existence of migrant hh member (dummy: 1 = yes & 2 = no), educational expenditure of hh members (continuous variable), relationship of migrant to hh member (dummies: 1 if yes & 0 if no for spouse & son respectively), 29

31 record of morbidity in hh (dummy: 1 if yes & 0 if no), district (local government area) and hh gross income (continuous variable). In the probit regression, six (6) out of the ten (10) covariates specified were significant meaning that they predicted the probability of an individual receiving remittance. The coefficient of hh size was negative with a probability of 3.9%; sex (p-value =0%); location (p-value =0%); existence of migrant member (p-value = 0.8%); morbidity (p-value =0%) and district (p-value =0.2%). The rest of the covariates were not strong in predicting the probability of receiving remittance (see Table 2). Initially, sex of household head was included in the model specification, but was dropped due to collinearity. Age of hh member, marital status and whether emigrant accessed loan before migrating were also included as covariates but the balancing property was not achieved hence the need to drop them from the model. 30