The Size and Effects of Emigration and Remittances in the Western Balkans. A Forecasting Based on a Delphi Process

Introduction

The Western Balkans have always been recognized as facing high emigration rates, and hence as having large diasporas.^[1] At the end of 2013, 5.7 million persons originating from the Western Balkans were living abroad, setting the emigration rate at 31.2% (World Bank), ranging from 18.2% of the total population in Serbia to 45.3% in Montenegro. In the course of the dissolution of Yugoslavia, 3.5 million individuals left the region. Hence, the emigration rate has been at 19.4%, ranging from 9.1% in Serbia to 38.5% in Albania. Consequently, remittance flows have been sizeable: the entire region received 8.6 billion USD in 2015, ranging from 3.1% of GDP in Macedonia to 16.7% in Kosovo. These figures still considerably exceed the inflows from foreign direct investment and official development assistance, 5.0 and 2.4 billion USD, respectively, thereby reflecting the massive dependence of the region’s economies on money returned from the diaspora.

The micro- and macro-economic impacts of remittances have been widely documented. At the macro-level, remittances stabilize the balance of payments, hence contributing to closing the large and persistent trade gaps in many countries, and preserving macro-economic stability.^[2] At the micro-level, the development effects of remittances, with a certain degree of variety, have been documented for poverty alleviation, improving education and health outcomes, improving income distribution, steering entrepreneurial spirit, etc.^[3] The related research on the Western Balkan economies shows large social and developmental effects of remittances on recipient countries.^[4]

The great recession (2007-2011), followed by the European debt crisis (2012-2013), created concern that external resource flows to developing countries may undergo precipitous decline, which could then adversely affect the basic development indicators and potentially reverse some achievements over the preceding decade.^[5] After the crisis hit the region in 2008, the average annual decline in remittances received in the Western Balkans has been 1.8%, yet with distinctive patterns across countries. Hence, while remittance inflow in the Western Balkans maintained relative stability, it was shown not to be countercyclical, as the literature usually asserts.^[6] In addition, with the economic crisis and the fall of the Schengen wall in part of the Western Balkans, these countries faced accelerated emigration. Entire families, including more educated individuals and families, left for a better life. This raises the question whether the magnitude of remittances sent could be maintained for a prolonged time, hence potentially undermining their social and developmental effects in countries of origin. Therefore, policymakers in the region may benefit from a forwardlooking analysis of emigration rates and remittances.

This paper aims at forecasting the size and effect of remittances and emigration in four Western Balkan countries—Macedonia, Albania, Serbia, and Kosovo—through application of a qualitative method, the Delphi questionnaire. The paper contributes a couple of new findings to the current sparseness of knowledge in this area. First, to our knowledge, qualitative methods in migration and remittance forecasting have not previously been applied in the literature, and this aspect represents the main contribution this paper makes to the literature. Second, we apply the Delphi method to consensus building within and between two groups of respondents—experts and remittance receivers—which also has not been done before. Third, in addition to its analytical contribution, the paper also makes a policy contribution, by pointing out ways in which the current policy context may utilize these results and findings. This contribution may be particularly important, as remittances and emigration have not been mainstreamed in the respective government policies so far, despite some efforts to frame diaspora strategies currently underway.

The paper is organized as follows: Section Two briefly reviews the existing literature on forecasting migration and remittances flows; Section Three elaborates the underlying methodology; Section Four presents the results and offers a discussion; and Section Five concludes.

Background Literature

Remittances have rarely been forecast, even in the global literature. Forecasting has usually been done at the macro-level using relatively simple quantitative models.^[7] Factors such as the stock of migrants, income levels at destination and origin, costs of remitting, exchange rates, interest rate differentials, but also duration of emigration and frequency of family visits have been accounted for in the literature in attempting to predict future developments with remittances and migration. However, as Mohapatra and Ratha note: ‘While remittances are influenced by all of the above factors, their use in a forecasting exercise is constrained by the lack of reliable forecasts of the future evolution of these explanatory variables’.^[8] Mohapatra and Ratha model remittances in a gravity-like model, whereby the main determinants are incomes at origin and destination, assuming the stock of migrants remains unchanged. Their model correctly predicted the modest decline of remittances following the global crisis. Two other studies, on the other hand, used a set of time-series methods to forecast the trend of remittances in Albania and Nigeria, respectively.^[9] While the effort is worth considering, it may be too naive to not consider other factors that determine remittance flows, primarily the income of the emigrants and the needs of the household at origin.

Migration has been forecast more frequently, nevertheless. This has been done in a similar context—either through following past deterministic trends or through devising a form of gravity model. For example, Layard and colleagues forecast East-West migration in Europe after the EU enlargement, by extrapolating earlier findings on earlier South-North Europe and Mexico-United States migration.^[10] Similarly, another study, based on a gravity equation of net migration between Central and Eastern Europe and the then EU-15, forecast a 0.08 to 0.16% emigration caused by a 10% income gap in favour of EU-15.^[11] This study predicted a population outflow from Central and Eastern Europe to the EU-15 ranging between 590,000 and 1,180,000 persons a year, depending on the pace of income convergence. Other studies widely used econometric gravity-like regressions in forecasting migration with an overview of the economic model and econometric tools utilized in the migration forecasting.^[12]

Despite the general trend to forecast migration flows in quantitative frameworks, a few studies have attempted to predict migration size based on surveys. For example, Drbohlav used a two-round Delphi method with a sample of seventy experts in the first round and thirty-nine in the second, and obtained rough estimates of the magnitude and timing of the expected East-West migration flows in Europe.^[13] Two other studies have used large surveys of individuals who were asked about their intentions to leave the country and, based on their responses, the probability is derived of emigration patterns.^[14]

Data and Methodology

This study is based on the application of the Delphi method. It originated in a series of studies that the RAND Corporation conducted in the 1950s, with the objective of developing a technique to obtain the most reliable consensus of a group of experts.^[15] The method is structured in group communication, allowing a group of experts to deal with a complex problem by arriving at a consensus through several rounds of feedback and subsequent revision.^[16]Typically, the method is used in cases where judgemental information is indispensable, through a series of questionnaires interspersed with controlled opinion feedback.^[17]

The design of the Delphi questionnaire follows a similar pattern to the standard survey. It is usually composed of a few questions, asking either for a quantification of certain phenomena, or delineating several options in a multiple answer context. Alternatives in the latter case may be set by the researcher, or derived from a pre-survey among a larger group of field specialists. Then an appropriate group of respondents is selected; these are typically field experts, sufficiently qualified to answer the questions. After the first round of surveying is conducted, a second round questionnaire is designed, based on the first round one but this time including the aggregate responses (results) of the first round. A further questionnaire / survey is similarly produced in each subsequent round, offering the aggregated results of the previous round. In each subsequent round, respondents have the opportunity to revise their answers, based on the new set of data provided. The process is reiterated until a satisfactory degree of consensus is reached. The respondents are anonymous to each other (though not to the researcher) throughout the process. The final output is formed from the aggregation of all individual opinions. A more detailed technical execution of the Delphi method is presented in Okoli and Pawlowski, while an excellent literature review on the revived interest in the application of the Delphi method can be found in Landeta.^[18]

Devised in this way, the Delphi method has the measurement of consensus as its key component. However, group stability has to be achieved beforehand, i.e. consensus may be meaningless without group stability. Group stability is defined as ‘the consistency of responses between successive rounds of a study’.^[19]

For the purpose of this study, we conducted the Delphi method as follows. In each of the four countries, we devised two groups of respondents: i) about 10 field experts; and ii) about 20 remittance receivers, per country. Each expert was chosen by reference to their field of expertise, i.e. persons who research and/or are professionally investigating the dynamics of migration and remittances in each country. They included researchers at universities / institutes, professionals / researchers at the central banks and ministries of finance, professionals within agencies on migration / diaspora, and bankers of foreign payments divisions. This group is in the core of the Delphi design. Readers should note that the number of experts in the particular issue in the investigated countries is fairly small, so the authors made an effort to include all the experts dealing with the referent issues. The other group is composed of about twenty remittance receivers. While this group is not composed of experts per se, we believe each remittance receiver who, hence, has a migrant abroad, may also provide a reliable judgement on the direction, size, and effects of migration and remittances. The receivers were selected randomly based on datasets (registries) the authors had of remittance receivers in each country. Therefore, sufficient heterogeneity in the group should be maintained. The Delphi survey took place between August and October 2016.

The questionnaire used for this exercise is given in the Appendix. Each group participated in two rounds of surveying. There was also a third, bridging, round: Round 2 provided the aggregate result of the Round 1 conducted on the same group, and potential revision of the opinions was solicited. Round 3 provided the aggregate result of Round 2 on the other group; experts were provided with the aggregate results of the receivers’ group, and vice versa, and potential revision invited. Hence, we are tracking the extent of consensus not only within groups, but also between groups.

There is a wide variety of views in the literature on the methods used to measure consensus and group stability.^[20] In addition, standards for consensus measurement have never been rigorously set, supporting the uncontrolled proliferation of methods used.^[21] For the purposes of this study, we relied on the coefficient of variation to measure consensus level and change. A consistent decrease of this coefficient between rounds would signify an increase in consensus,^[22] while a coefficient of variation at or below 50% is assumed to signify that a reasonably good internal agreement has been reached.^[23] We set a slightly lower cut-off rate of 40%, suggesting that the consensus of 60% (calculated as 1—coeffident of variation) would be considered satisfactory for this study and a stop point after at least two rounds for the same group (hence, not counting the cross-round) have been conducted.

To measure group stability, we rely on three measures. The first is a standard t-test, which has been widely used in the literature.^[24] We use two variants of the t-test: for paired samples between Rounds 1 and 2, since we follow the evolution of answers from the same sample; and for independent samples between Rounds 2 and 3, since we compare experts with receivers. The second test is the Wilcoxon signed-rank test, whose null hypothesis is that the two samples follow the same distribution. Finally, we count the share of individual responses that were maintained between rounds in the total number of answers. For the questions with a continuous underlying variable, we consider an answer to be maintained when the subsequent round answer belongs to +/- 1 s.d. of the previous round, while for the categorical-variable questions we require that the answer is obtained again.

Results and Discussion

Consensus Forecasts

We begin analysing the results of the Delphi exercise by presenting some findings about the consensus achieved. In all cases, the consensus within groups exceeded 60%, even in the first round, which rose to over 70%, or even near or above 80%, in Round 2. These are considered satisfactory levels of consensus, adding robustness to the consensual forecasts as well providing grounds for limiting the number of within-group rounds.

As Round 3 was an inter-group round (experts were given the Round 2 answers from receivers, and vice versa), we were also interested in the intergroup consensus, which is given in Table 1. Not surprisingly, in most cases the consensus increased as compared with own-group consensus from the previous round, but also as compared with the other-group consensus of the previous round. However, in terms of the latter, there are some notable exceptions. First, the largest consensus drops can be observed in two cases: emigration-size forecasting, and the effect of remittances on the labour market. Second, it is evident that the experts’ consensus was predominantly reduced when they were given the results of receivers, i. e. they were considerably reluctant to change their opinion when it was different from that of the receivers’, and this imposed noise/ variability in the experts’ opinions. On the other hand, the receivers tended to more readily approximate/agree with the experts’ opinion. Serbia was a slight exception, where both groups showed noise with the answers of the other group, i.e. neither group changed their minds readily (see Table 1).

Table 1

Between-group consensus.

Country	Cross-rounds	Delphi questions
		Amount of remittances	Size of emigration	The emigration effects	The remittances’ effect on poverty	Remittances’ usage	Remittances’ effect on labor market
Macedonia	Round II, receivers	66.70%	90.80%	93.90%	82.10%	93.90%	86.50%
	Round III, experts	86.30%	64.60%	100.00%	89.50%	100.00%	68.40%
	Round II, experts	76.50%	65.00%	84.20%	84.20%	100.00%	84.20%
	Round III, receivers	86.20%	89.30%	93.90%	88.80%	100.00%	84.30%
Serbia	Round II, receivers	75.80%	67.90%	88.80%	75.30%	88.80%	81.70%
	Round III, experts	88.20%	68.90%	84.20%	79.30%	89.50%	61.30%
	Round II, experts	84.90%	78.20%	84.20%	79.30%	89.50%	68.40%
	Round III, receivers	80.90%	69.90%	88.80%	77.60%	90.60%	79.80%
Albania	Round II, receivers	56.52%	72.31%	77.64%	85.36%	79.77%	77.64%
	Round III, experts	81.24%	79.53%	79.30%	79.30%	74.18%	61.27%
	Round II, experts	76.31%	52.21%	79.30%	74.18%	74.18%	68.38%
	Round III, receivers	80.92%	67.45%	75.28%	79.77%	77.64%	75.28%
Kosovo	Round II, receivers	80.80%	57.60%	80.40%	84.40%	80.40%	78.20%
	Round III, experts	83.30%	60.60%	68.40%	61.30%	84.20%	68.40%
	Round II, experts	86.00%	41.10%	68.40%	84.20%	84.20%	68.40%
	Round III, receivers	86.00%	76.90%	75.80%	86.30%	84.40%	86.30%

1. Source: Authors’ calculations based on Delphi survey.

We now turn to observing the consensual forecasts. Figure 1 presents the consensual forecasts for the remittances’ flows in the next five years in the four investigated countries. The graph offers information on the size of remittances in 2016; the variation in the responses of both groups (presented through the minimum and maximum forecasts across rounds); and the achieved consensus after the final third cross-round. We first note that forecasts are positive for Macedonia and Serbia, and negative for Albania and Kosovo. In all cases except Albania, experts had lower forecast variation; and in the same cases, the final consensus has been geared toward the lower bounds of experts’ and remittances’ opinions. In Albania, receivers were more convinced in their forecasts than in the other three countries, and hence experts finally agreed close to receivers’ range of forecasts and close to experts’ upper bound forecasts. Overall, remittances are forecast in Macedonia and Serbia to rise over the next five-year period by 9% and 17.4%, respectively, implying respective annual increases of 1.8% and 3.3%. This may be driven by the fact that remittance flows in these two countries have been fairly stable and coming from stable economies such as Germany and Switzerland, as well reflecting the recent large waves of emigration whereby emigrants still have to maintain relatively strong ties with the family members left behind. On the other hand, the consensual forecast for Albania and Kosovo suggest annual remittances decline of 4.5% and 2.2%, respectively. These results may be driven by the fact that remittances in these two countries have been booming over previous years and this effect may fade out soon. In addition, in the cases of Albania and Kosovo, it is more likely that entire families have been emigrating more frequently than in the cases of Macedonia and Serbia, as well the recently increased number of irregular or seasonal labour emigrants working in low-skilled jobs who are hence able to generate less income (see Figure 1).

Figure 1

Remittance forecasts. Source: Authors’ calculations based on Delphi survey.

Similarly, Figure 2 presents the same information on the emigration forecasts. The consensual forecast is that emigration will subside in Macedonia, Albania and Kosovo, but will increase in Serbia, over the next five-year period. Noticeably, these results have been characterized by more variability, hence with a lower level of consensus among the respondents. For instance, in the case of Macedonia, the lack of consensus is visible from the non-overlapping ranges of experts and receivers (also reflected in consensus deterioration in Table 1), although by the end of the exercise, consensus improved and the final forecast is a slight reduction of emigration of 3% in 2021. With smaller variation but feeble consensus, the forecast for Albania suggests a decrease of emigration of 22% in the same year. The variation and consensus appeared better in Kosovo and Serbia, where respondents were finally in agreement that emigration will decelerate by 9% and accelerate by 13.5% in 2021, respectively. Emigration has been high recently, and the consensual forecast in Serbia suggests that this effect has not yet been exhausted, compared to the other three cases where large waves of emigration had already taken place. The expectation of subsiding emigration may be also driven by the readmission of irregular migrants, as now all the Western Balkan countries are considered ‘safe countries’, as compared to the crisis caused by the heavy emigration from the conflict areas in the Middle East (see Figure 2).

Table 2

Consensus forecasts on emigration and remittances effects.

		The emigration effects	Remittances’ effect on poverty	Remittances’ usage	Remittances’ effect on labor market
Macedonia	Experts	Negative, educated depart	As is	Current consumption	Inactivity
	Receivers	Negative, educated depart	As is	Current consumption	Employment
Serbia	Experts	Negative, educated depart	As is	Current consumption	Employment / Self-employment
	Receivers	Negative, educated depart	As is	Current consumption	Employment
Albania	Experts	Negative, educated depart	Weaker	Current consumption	Employment / Self-employment
	Receivers	Negative, educated depart	Weaker	Current consumption	Employment / Self-employment
Kosovo	Experts	Positive, because of remittances	Weaker	Current consumption	Employment
	Receivers	Negative, educated depart	As is	Current consumption	Employment

1. Source: Delphi survey.

Figure 2

Emigration forecasts. Source: Authors’ calculations based on Delphi survey.

Table 2 provides the consensual forecasts on the remaining questions of our Delphi survey, which refer to the effects of emigration and remittances. Both groups achieved consensus in most cases. For instance, emigration effects for the society have been forecast to be negative due to skilled labour emigration. However, there is no consensus on this aspect in Kosovo, where experts continue to consider the effect as positive due to the sheer size of remittances the economy receives (see Figure 1). This exclusion is probably driven by the comparatively larger remittances in Kosovo (16.7% of GDP) compared to the other three cases. Similarly, it is only in Kosovo where experts and receivers were not in agreement on the effect of remittances on poverty: the former agreed it would become weaker, while the latter it would remain as is, which largely corroborates the fact that remittances in Kosovo are relatively bigger. However, we need to note here that the agreed answer of Kosovar experts is not in agreement with their consensual result about emigration effect. While in the other three countries, experts and receivers ended up with consensus, it is notable that they agreed about maintaining remittances’ effect on poverty in Macedonia and Serbia, with reduction in Albania (see Table 2).

In all four countries, experts and receivers were in agreement that remittances will continue to support current consumption only; this is the question with the largest consensus achieved, even during the earlier rounds. On the other hand, Macedonia and Serbia see a disagreement on the remittances’ effect on the labour market, the lack of consensus being especially marked in Macedonia; namely, experts agreed that remittances will support inactivity, while receivers agree remittances support employment. The disagreement may be driven by a recent finding that remittances cause inactivity except for younger receivers, where they were found to promote self-employment.^[25] In Serbia, on the other hand, both groups agreed remittances will support activity, albeit experts favour employment, while receivers favour self-employment. There is no clear-cut evidence for what may be driving this divergence, but apparently receivers are more inclined to think that the money they receive is important for starting up micro-businesses. Finally, there has been a consensus achieved in Albania and Kosovo that remittances support activity. The last question exhibited the lowest consensus within the entire Delphi process. There are at least two sets of reasonings for this result: first, there is no clear-cut evidence about the effect of remittances on labour-market outcomes in the Western Balkans; therefore, an educated guess cannot be made, and hence respondents are mainly driven by their intuition on this question; second, experts may be driven by prevalent results from the literature from other countries (mainly Latin America), while respondents by particular cases (rather than systematic change) that remittances produce favourable labour market outcomes rather than inactivity.

Conclusions and Recommendations

The objective of this paper is to forecast the size and effects of remittances and emigration in Macedonia, Albania, Serbia and Kosovo, through application of a qualitative forecasting method, namely a Delphi questionnaire. We examined the consensus for two groups of respondents, experts and remittance receivers. By so doing, we could track consensus building and arrive at forecasts both within and between the two groups. In each of the four countries, ten experts and about twenty remittance receivers were consulted in three subsequent rounds, out of which the first two were on the same group, while the third round was a cross-round whereby average answers of receivers were given to experts, and vice versa.

In general, consensus on almost all investigated issues was evident even from Round 1, and this increased in Round 2, and even further in Round 3. However, it was notable that in the cross-round, Round 3, experts were on average more reluctant to adjust their average answer closer to that of the receivers, which gave rise to some noise in the results and reduced consensus. In general, this did not happen with receivers in Round 3.

The results suggest that remittances in the projected five-year period will increase in Macedonia (9%) and Serbia (17.4%), and will reduce in Albania (1.8%) and Kosovo (3.3%). With lower consensus, the results forecast that emigration will decelerate in Macedonia (3%), Albania (22%) and Kosovo (9%), and will accelerate in Serbia (13.5%) in 2021. Emigration effects have been forecast as negative due to skilled labour emigration, despite consensus on this issue lacking in Kosovo. Similarly, it is only in Kosovo where experts and receivers were not in agreement on the effect of remittances on poverty: the experts agreed it would become weaker, while the receivers agreed it would remain as it is; in the other three countries, experts and receivers consensually forecast maintaining the effect of remittances on poverty in Macedonia and Serbia, while weakening in Albania. In all four countries, experts and receivers were in agreement that remittances will continue to support current consumption only. On the other hand, the Macedonians lacked consensus on the effect of remittances on the labour market, whereby experts agreed that remittances will support inactivity while receivers favoured employment. On the other hand, consensus has been achieved in the other three countries that remittances will support labour market activity.

The findings of this paper may have two lines of implications. The first set is the implications for further research. As forecasting of remittances and migration has been sparse in the literature, the findings open a new strand of thinking whereby quantitative variables could be forecast through qualitative approaches. However, this should be in parallel with quantitative techniques, so that results are ultimately comparable. In particular, for the Western Balkan countries, a quantitative exercise to forecast remittances and migration (based on regressions) would aid our results, either by corroborating or rejecting them. The second set of implications is for policy. Remittances may not last forever and hence governments should have clear-cut evidence as to their possible flows and potentially plan actions on how to utilize them while they flow in. In particular, strengthening the longer-term effects of remittances, such as supporting their use for opening micro-businesses, may be possible. Similarly, projections about migration flows and their effects may be valuable for governments to undertake actions that will lessen the negative effects of such emigration, especially if the effect is projected to be negative due to the departure of highly educated workers.