Household Social Expenditure in Ghana: Examining the Ex-post Effects and Vulnerability to Poverty Authors: 1

We estimate the effect of household social expenditure on vulnerability to poverty using the four latest cross-sectional waves of Ghana Living Standard Survey (GLSS) from 1999 to 2017. Using a 3-Stage Least Square and Quantile Regression, our results show a widening consumption ex-post welfare gap between the poorest households and the non-poor households in a per-cedi social expenditure. Also, we estimate the probability of an ex-ante poverty using vulnerability to expected poverty. The results, however, indicate that regardless of poverty status, household vulnerability to poverty increased consistently between 1999 and 2017, and the very poor households showing the severest vulnerability. Hence, it is concluded that social expenditure increases the chances of a poor household falling into chronic poverty a non-poor household into transient poverty in the future.


Introduction
Mankind interact, communicate, and share parts of their lives with others from their immediate to the farthest relations (Sen 1985, Shefrin andThaler 1988). Such communal sharing often come in the form of social assemblies referred to as social functions, gatherings, events, or simply social ceremonies. Social events, in most circumstances, have brought people together for ceremonies that would rarely happen in one's lifetime and are usually ceremonies for the passage of time. Therefore, social events are activities that an individual participates and derives satisfaction on the account of so many others participating as well. The interest is a shared or a common one. It could either be for shared happiness, commemoration, community development, solidarity, and so on. They include funerals, weddings, festivals, parties, and other ceremonies that often bring two or more people together for a shared interest. However, the elaborate nature of consumption in some of these social events cannot be overemphasised.
At the global stage, the Goal 1 of the Sustainable Development Goals (SDGs) of the United Nations London 2015). By 2017, funeral costs had risen by 70.6 percent whilst wages had increased by only 20 percent over the previous decade (Royal London 2017). Also, the Royal London National Funeral Cost Index, 2017 showed an increase in the cost of funerals ahead of inflation. Unfortunately, it is reported that the UK's most vulnerable citizens are those taking on these increased levels of funeral debt (Quarker Social Action 2017). Gradually, "public health funerals" or "paupers' funerals" which are organised by local authorities for deceased persons who neither have relatives nor friends are rising because there is evidence that the cost of funerals now prevents some families from having funeral services for their deceased (Quarker Social Action 2015, 2017. Likewise, in the US, the average funeral cost is between $7,000 -$10,000 ("How Much Does the Average Funeral Cost?" 2018), and a lot of poorer households are being put under financial distress ("Funeral Poverty in the 21st Century" 2014). Also, in Eastern Europe and many parts of Russia, the average cost of weddings ranged from $1,000 in Slovakia to $15,000 in Russia ("Survey Compares Countries Wedding Spending Habits" 2013).
The situation is widespread in developing countries. In China, Chen (2014), Chen and Zhang (2012) have found that social spending on funerals and festivals militate against early child development in rural China. Likewise, Bloch et al. (2004), Rao (2001) have shown that elaborate social expenditure perpetuates rural poverty in India. Yet, the practice is largely indispensable in the lives of the poor. Banerjee and Duflo (2007) indicated that festival expenses took a significant share of the budget for the majority of poor households in developing countries.
So according to the study by Banerjee and Duflo, in Udaipur, about 99 percent of the very poor families expended on weddings, funerals, or religious festivals, and the average household expended about 10 percent of its yearly budget on festivals. In South Africa, 90 percent of families who live on less than $1 a day spent money on festivals. In Pakistan, Indonesia, and Cote d'Ivoire, more than 50 percent did likewise (Banerjee & Duflo 2007). According to Aker and Sawyer (2016), households in Niger who are often unable to save for education, health, or agriculture purposes did spend to celebrate the Tabaski holiday. Similarly, South African households could also spend about a year's income to bury a departed member of the family (Shimeles and Woldemichael 2013).
In Ghana, according to Ghana Statistical Service's report on the poverty profile in 2008, about 32 percent of Ghanaians were poor living below $2 a day. Yet, the average funeral in Ghana then costs between US$2000and US$3500 (Butu 2013, Ghana Statistical Service 2008; costing between 1000 and calories per adult equivalent of food per day, even if they were to spend all their incomes on food (Ghana Statistical Service 2014).
Funerals have become an avenue for the display of wealth (Butu 2013, Jufare 2008. It is, therefore, expected that the elaborate consumption of these social events is likely to overstretch the budget of some households within a cohort that has wide income disparities. Interestingly, such high expenditure cut across all types of households including the poor ones. Moreover, these social events are a source of merry and relaxation as well as prestige and esteem which form part of the needs of individuals and householdsspecifically, their social needs. These social needs, according to Maslow (1943Maslow ( , 1954 theory of human needs, form part of the pyramidal needs of an individual. Social ceremonies on the pyramid constitute a higher need apart from the basic human needs such as food, shelter, clothing, sex, and housing. Martin (2016) argued that "We mistakenly assume that there's no way a person can or should possibly worry about self-esteem if they're hungry". According to Martin, it is not surprising to find the poor in deprived regions who are active on social media even in times of unmet basic needs. It is therefore not surprising to find lavish expenditure on funerals, weddings, and festivals in poor homes and regions.
According to Mazzucato et al. (2006), money and death are inextricably interwoven. Every death triggers a flow of money and the funeral business flourishes. The elaborate funeral celebrations during which no trouble or expense is spared contrast sharply with the daily struggle for the primary necessities of life. They have become great public events, where families compete for prestige and respect by showing off wealth, and by publicly conforming to norms of solidarity and respect for the dead. Families would spend whatever assets possible just to bury the dead regardless of the lasting consequences for posterity.
Similarly, marriage ceremonies in Ghana have become westernised to the very extent that the couples-tobe usually hold two separate marriage ceremonies before they are socially accepted to be properly married.
To some extent, some religious organisations do not recognise a traditional marriage as legitimate until one performs a 'western-style' wedding ceremony.
The implications of these social expenditure are not far-fetched, especially in the context of mass poverty and poor standard of living. Non-productive expenditures like these would likely aggravate the disease of poverty and misery among the people. In situations where one could sell off productive lands and plantations just to organise lofty weddings and funerals (Case et al. 2008, De Witte 2003, there is nothing to expect from unnecessary hardships for the household. Newly wedded couples would have to necessarily restart their whole lives as a bountiful amount of life savings would have been expended on their wedding parties as Aker and Sawyer (2016) have found.
In the United Kingdom, for example, in 2014, funeral cost rose exponentially far more than the rise in the cost of living (Royal London 2015). By 2017, the average cost of a funeral had risen by almost 71 percent whilst wages had increased by only 20 percent over the previous decade (Royal London 2017).
Unfortunately, it is reported that it is the country's most vulnerable citizens who are taking on these increased levels of funeral debt (Quarker Social Action 2017). Also, in other parts of Eastern Europe, the average wedding cost ranged from $1,000 in Slovakia to $15,000 in Russia ("Survey Compares Countries' Wedding Spending Habits" 2013).
Likewise in the US, the cost of an average funeral was between $7,000 -$10,000 ("How Much Does the Average Funeral Cost?" 2018), and a lot more poor households are being put under financial distress ("Funeral Poverty in the 21st Century" 2014). In China, Chen and Zhang (2012) found that social expenditure on funerals and festivals militate against early child development in rural China. Likewise, Rao (2001) and Bloch, Rao, and Desai (2004) have shown that elaborate social spending perpetuates rural poverty in India. Furthermore, Banerjee and Duflo (2007) indicated that, more surprisingly, expenditure on festivals forms an important part of the budget for many extremely poor households in developing countries. Households in Niger spend to celebrate the holiday of Tabaski but are often unable to meet savings goals for education, health, or agriculture expenses (Aker and Sawyer 2016). Similarly, South African households could also spend about a year's income to bury a departed member of the family (Case, Garrib, Menendez, and Olgiati 2008).
The first of the Sustainable Development Goals (SDGs) is about ending poverty in all of its forms with the targets of eradicating extreme poverty and reducing absolute poverty for all people by the year 2030. However, according to Chaudhuri (2003), any anti-poverty intervention must, of first importance, be "necessarily going beyond the catalogue of who is currently poor, how poor they are, and why they are poor to an assessment of households' vulnerability to poverty -who is likely to be poor, how likely are they to be poor, how poor are they likely to be, and why are they likely to be poor". In this regard, the assessment of future poverty aligned with household social expenditure becomes essential to policy much more than focusing on current poverty levels. Hence, the SDGs will be well achieved if policies are targeted at issues that threaten the vulnerability of both the poor and the non-poor in society, especially in the light of social expenditure which commands high shares of household expenditure and savings.

Determinants of Household Welfare -Quantile Regression
Household welfare measured by consumption expenditure per adult equivalent scale (Annim, Mariwah, and Sebu 2012;Asenso-Okyere et al. 2000;Donkoh, Alhassan and Nkegbe 2014;Ghana Statistical Service 2014) is influenced by household idiosyncratic characteristics and, sometimes, external variables (Deacon 1992;Diallo and Wodon 2007). Following the works of Browning and Lusardi (1996), Chaudhuri (2003), Coulombe and Wodon (2007), Diallo and Wodon (2007) and Shimeles and Woldemichael (2013), variables such as age, sex, marital status, education, working status of household head as well as durable assets which serve as a store of wealth were included in the determination of household welfare. One of the most profound ways of analysing the effect of social expenditure is to estimate arbitrarily to represent categories of households from the bottom percentile to the highest, as shown in Table 1, at different quantiles (10th, 25th, 50th, 75th and 90th percentiles) of welfare. These quantiles are chosen first, report coefficients of exogenous variables in the currency unit (₵, cedi), and elasticity effect for social expenditure only. Elasticity coefficients for control variables have been omitted for the sake of parsimony, which could be made available to any interested party upon request.
To begin with the analysis of results, Table 1 shows a positive relationship between household welfare and social expenditure. Similar results for 2012/2013, 2005/2006 and 1998/1999 and presented in Table S1, Table S2, and Table S3, respectively. A household in the 10 th percentile that observed an expenditure on social events averagely increases the household welfare by about ₵892 more than other households in the same percentile that did not incur social expenditure. This is no surprise since social expenditure is also captured as part of the total welfare estimate for each household.
Hence, the welfare of households that make social expenditure is, on average, higher than those that did not spend on such social events like funerals, weddings, parties, etc. Also, households in the 25 th percentile add ₵987 to their consumption welfare more than households that did not make expenditure on social events, an increase over the average for the 10 th percentile households.
Likewise, households in the 50 th , 75 th and 90 th percentiles increasingly have ₵1161.7, ₵1572.5, and ₵1685.2, respectively, more than households that did not expend on social events. The increasing amount added to household welfare as over the percentile indicates that as households get richer greater sums of cedis are spent on funerals, weddings, and the like.  An important aspect of this analysis is to consider the trend of the elasticity effect of social expenditure on household welfare. According to Table 2, the elasticity effect on welfare increases horizontally as the welfare also increases from the 10 th to the 90 th percentiles. That is, addition to welfare is greater among poorer households than richer households. This is because as poor households spend on funerals, weddings, or religious festivals every year, the amounts spent represent a significant proportion of their budget. Therefore, for richer households, social spending would rather represent a smaller proportion of their entire household budget compared with poor ones. For instance, in 2016 the percentage (elasticity) of welfare due to social spending decreased from about 66 percent for the bottom 10% to 38 percent for the top 10%. Similarly, the elasticity effect of social expenditure on welfare for all preceding years to 2016/2017 also increased as welfare percentiles increased. Most importantly, each percentile consistently increased their social expenditure as a proportion of household welfare. Therefore, Table 2 shows that, for example, the 10 th percentile of households between 1999 to 2017, increased their social expenditure as a proportion of welfare from about 16 percent to about 66 percent, whereas that of the 25 th percentile households increased from 12 percent to 63 percent. The 50 th percentile households also increased from 13 percent to 59 percent while social expenditure as a proportion of welfare jumped from 10 percent to 48 percent for households in the 75 th percentile; and lastly, that of households in the 90 th percentile rose from less than one percent to 38 percent.

Testing of Effect sizes between Quantiles
This section now discusses the results of the hypothesis tests conducted on the coefficients of social expenditure across different quantiles of the simultaneous quantile regression estimates above using linear combination (lincom) tests. First, 2016/2017 is considered followed by 2012/2013, 2005/2006, and then 1998/1999, in that order. Table 3 presents a composite hypothesis testing results of the effect of social expenditure on household welfare between lower and upper welfare quantiles. Only significant tests are shown in Table 3. Table 3 shows that for all years, that is, 2016/2017, 2012/2013, 2005/2006, and 1998/1999, there were differences in the effect of social expenditure on household welfare between the 10 th and the 25 th , 50 th , 75 th, and 90th percentile; and between the 25 th and the 50 th , 75 th and 90th percentile. These statistical differences imply that the extremely poor households (10 th percentile) have more addition to welfare for a cedi spent on social events than the very rich households (75th and 90th percentile). This could be explained in the sense that, in the ex-post, poor households have smaller consumption expenditure such that an additional cedi would have high effect on household welfare than rich households. Hence, poor households in the 10 th percentile of welfare in 2016/2017, for example, have approximately 25 percent and 32 percent more than the 75th and the 90th percentiles respectively. Also, the tests for the upper percentiles (that is, 25th, 50th, 75th, and 90th percentiles) indicate that even between the median and the rich households, the poorer ones have greater addition to welfare than the richer households in the ex-post analysis. The reasons are just as is in the case between the 10th, 75th, and 90th percentiles. The implication of the above hypotheses is straightforward. That is, the effect of social expenditure on welfare between the poorest households and others increases towards higher quantiles of welfare. Thus, in the neighbourhood of the 10th percentile of welfare, there is no evidence of differences in the effect of social expenditure. However, extending the neighbourhood towards higher quantiles like 50th, 75th, and 90th quantiles reveal such differential effect. This suggests that in the expost, poor households benefit from social expenditure than rich households. The latter statement would lead one to argue social expenditure as a form of social investment which yields welfare returns.

Results from
As this argument may be partly true, in the presence of functional, resourceful social networks, it may not be true if such expenditure does not yield commensurate benefits. According to Case et al. (2008), Haq et al. (2009), Mazzucato et al. (2006, and Chen and Zhang (2012), household social expenditure tend to be an unproductive venture with the possibility of squeezing out essential components of household consumption like food, health, education and so forth at a different period. In this regard, the next section discusses the future vulnerability to poverty as a result of the household's engagement in social expenditure as a step further to examine the ex-ante effect apart from the ex-post gains.

Vulnerability to Expected Poverty
This section discusses the vulnerability to poverty due to social expenditure. The process followed the works of Chaudhuri (2003) and Shimeles and Woldemichael (2013) Table 4 where 62 percent of all households were vulnerable to poverty. This comprised 80 percent of very poor households, 96 percent very poor households, 91 percent poor households and 51 percent of non-poor households. In this case, there are a lot of households that are prone to poverty even the non-poor is not fully spared. Comparatively, estimates captured in Table 5 indicate a slight reduction in vulnerability for all categories of households as social expenditure dummy is added to the consumption model. However, the decline in vulnerability is mainly influenced by a decrease in the percentage of vulnerable nonpoor households, that is, from 51 percent in Table 4 to 50 percent in Table 5 whereas the very poor and the poor households saw an increase in their vulnerabilities to 97 percent and 92 percent, respectively. The implication could be that the poverty levels of poorer households are likely to be negatively affected in the future as they incur social expenditure compared to those who did not make such social expenditure. Next, Table 6 shows the fact that for the total sample of households, 29 percent became vulnerable after observing social expenditure as an exogenous (shock) variable for 2016/2017. This represents a rise in the sample average from 27 percent to 29 percent. Again, there was about a 1 percent increase in the number of vulnerable households for all poverty status or categories. It suggests, therefore, that social expenditure increases slightly the vulnerability to poverty of all, especially the very poor households. By this, one could argue that, although in the ex-post analysis, poor households may have had a more positive effect than the rich households, social expenditure does not seem good for the very poor in future analysis.
This argument is true especially when consumption of social events are often tied to societal norms and relative consumption, poor households that stretch their budgets to meet present expenditure on social events may be counted to have improved their welfare by increasing consumption expenditure but would have to face the dire consequences in the future. So, this finding points out the negative future effect that social expenditure brings on households that are poor but would still want to follow the herd. In the wake of extravagant funerals, festivals and weddings in developing countries, the ex-ante analysis points to future permanent or transitional poverty for poor households that would venture what is the preserve of the rich.
From Table 6 which contains estimates of vulnerability without social expenditure, a greater number of households (that is, 64.14%; 9939 households), including the non-poor, were vulnerable to poverty compared with others who were not. It could be seen that the very poor and poor households form the majority, as usually known. On the other hand, Table 7 shows the fact that for the total sample of households, 63.81 percent are vulnerable after social expenditure was introduced as an exogenous (shock) variable. This represents a drop in the sample average from 64.14 percent to 63.81 percent which looks good.
However, considering the constituents of the sample average, it could be seen that the decline in vulnerability for the entire sample was as a result of a fall in the vulnerability of the Non-poor only (that is, 87.04% in Table 6 to 85.90% in Table 7). This means that vulnerability rather increased from 92.37 percent in Table 6 to 92.51 percent in Table 7 for the Very poor while for the Poor, it rose from 91.73 percent in Table 6 to 91.80 in Table 7. It suggests, therefore, that social expenditure increases slightly the vulnerability to poverty of the Very poor by 0.14 percent and the Poor by 0.07 percent.      Table 10 and Table 11 respectively. It is evident here that vulnerability for the non-poor has increased to 28.84 percent while the probabilities for the very poor and poor declined slightly to 66.24 percent 59.41 percent respectively. However, it could be seen that despite the drop in vulnerability for the very poor compared with Table 10, about 27 more households in that category became vulnerable to poverty. The results of the hypothesis tests also showed that the effect of social expenditure on welfare between the poorest households and others increases towards higher quantiles of welfare. This suggests that in the current poverty analysis, poor households benefit from social expenditure than rich households. Thus, in the neighborhood of the 10th percentile of welfare, there is no evidence of differences in the effect of social expenditure. However, extending the neighbourhood towards higher quantiles like 50th, 75th, and 90th quantiles reveal such differential effect. Hence, it could be said that social expenditure contributes significantly to the welfare of very poor than non-poor households and that this difference in effect widens between the poorest and other households of higher levels of welfare.
On the contrary, vulnerability estimates show the negative repercussions of household social expenditure behaviour to future poverty for all years of the survey. By comparison across the years, it is seen that baseline vulnerability to poverty without social expenditure increased from 38.43 percent to 64.14 percent of the sampled population. As a result, vulnerability to poverty including social expenditure also increased from 38.43 percent in 1999 to 63.81 percent in 2017. This is an indication that vulnerability among the populace rose despite Ghana's middle-income status achieved over the same period.
By this, it could be argued that even though in the current poverty analysis, poor households may be seen to have increased their welfare through social expenditure, the practice is not good for the poor in the future poverty analysis. As a result, the notion of social investment through social expenditure may not be entirely true for the poor in these instances, especially without a compensating reciprocation. This argument is true especially when consumption of social events is tied to the societal norms and relative consumption, poor households that outstretch their budget in order to meet present expenditure on social events may be counted to have improved their welfare but would have to face the dire consequences in the future. By this finding, we point out the negative future effect that social expenditure brings on poor households. In the wake of extravagant funerals, festivals, and weddings in developing countries, the ex-ante analysis points to future permanent or transitional poverty for poor and non-poor households respectively.

Conclusion
Directly from the hypotheses tested in this study, it is concluded that very poor households benefit more in terms of welfare than non-poor households and that the difference in the effect of social expenditure widens between the poorest and other households, moving towards higher levels

Methods
The effect of social expenditure on household welfare is presented into two main parts. First, is the usual poverty analysis which computes the effect of an exogenous factor (social expenditure) on an outcome variable (household welfare)? In this study, this was achieved using a three-stage least square (3SLS) estimator and simultaneous quantile regression (Sqreg) approaches. This is explained in a moment. The second part of the study analyses the effect of social expenditure on future poverty levels (vulnerability to expected poverty) of households, adopting the vulnerability approach proposed by Chaudhuri (2003). This approach, too, is explained in detail in succeeding paragraphs after brief discussions on 3SLS and Sqreg.
To begin with, the 3SLS estimator is used to estimating the effect of social expenditure on household welfare due to the potential endogeneity, specifically, bi-causality, between the outcome variable and social expenditure. That is the likelihood of households with higher welfare incurring higher social expenditure, and higher social expenditure influencing higher household welfare. The stochastic process generating the consumption welfare of a household h is given as: Where lnWh is the log of household consumption welfare, Xh represents a set of observable However, εh in most instances are correlated with the outcome variable, leading to endogeneity challenges. Literature exists in the treatment of such a relationship of the model using a two-stage least square estimator (2SLS). However, unlike a 2-stage least squares approach, a 3-stage least square is more efficient, according to Cameron and Trivedi (2005). In a system of equations where an explanatory variable becomes an outcome variable in other reduced equation(s), the error terms among the equations are expected to be correlated. 3SLS uses an instrumental-variables approach to produce consistent estimates and generalized least squares (GLS) to account for the correlation structure in the disturbances across the equations (Cameron andTrivedi 2005, Zellner andTheil 1962).
According to Zellner and Theil (1962), three-stage least squares estimates are obtained by estimating a set of nonlinear (or linear) equations with cross-equation constraints imposed, but with a diagonal covariance matrix of the disturbances across equations. This is the constrained two-stage least squares estimator. They further explained that the parameter estimates thus obtained are used to form a consistent estimate of the covariance matrix of the disturbances, which is then used as a weighting matrix when the model is re-estimated to obtain new values of the parameters.
Subsequently, simultaneous quantile regression (Sqreg) was used to estimate the effect of social spending (using predicted values from the 3SLS estimation above) on household welfare at various welfare quantiles. Simultaneous quantile regression is a quantile regression technique that estimates different quantiles concurrently (Cameron andTrivedi 2005, Zellner andTheil 1962 On the other hand, vulnerability to expected poverty is estimated following the methods used in the works of Chaudhuri (2003), Dercon (2002), Hoddinott and Quisumbing (2010). Vulnerability is considered as the probability of consuming below an established welfare threshold Z. Thus, the probability that a household at time t would consume below the absolute poverty line. Vulnerability, V, is given as: Where lnCh is the logarithm of the household's consumption expenditure per equivalent scale at time t and Z is the absolute poverty line. The stochastic process generating the consumption of a household h is assumed as: Where lnCh is the logarithm of consumption expenditure per equivalent scale, Xh represents a set of observable household and household head's characteristics; β is a vector of parameters; and eh is a meanzero, constant disturbance term that captures idiosyncratic factors (shocks) that contribute to different per capita consumption levels for households that are otherwise observationally equivalent (Chaudhuri 2003). Further, it is also assumed that the variance of eh is given by: We estimate β and θ using a three-step feasible generalized least squares (FGLS) procedure as in Chaudhuri (2003) and Shimeles and Woldemichael (2013). First, equation (3) is estimated using an ordinary least squares (OLS) procedure. Then, the estimated residuals from equation (2) to estimate the following equation using OLS.
The predictions from equation (5) OLS estimation of equation (7) yields a consistent and asymptotically efficient estimate of β. Using the estimates β and θ to directly estimate expected log consumption: and the variance of the log consumption is also given as: By assuming that consumption is log-normally distributed, it becomes possible to form an estimate of the probability that a household with the characteristics, , will be poor. Letting Φ(.) denote the cumulative density of the standard normal, we obtain the probability values given by equation (2). As according to literature (Chaudhuri 2003, Hill and Porter 2017, Hoddinott and Quisumbing 2010Shimeles and Woldemichael 2013), this study considers a household as vulnerable to poverty if is greater than a probability threshold P: = 1, > 0, ≤ Also, the study adopts the commonly used threshold of 0.5 for P such that a vulnerable household is one whose probability exceeds 0.5.