Abstract
This paper attempts to assess the impact of the oil shock on South Asian economies, including Sri Lanka, as well as policy responses to deal with the shock, and the real income loss for low-income households if fuel subsidies were fully removed. The most important impact has been on the balance of payments followed by the results of poverty and social impact analysis (PSIA) conducted for Sri Lanka. Finally, the paper explores to what extent Sri Lanka’s large receipts of worker remittances serve as a hedge against shocks.
III. Are Workers’ Remittances A Hedge Against Shocks? The Case of Sri lanka12
A. Introduction
1. Despite Sri Lanka’s astonishing resilience, the country remains vulnerable to external shocks. With an average growth rate of 5 percent during the 1990s and 6–7 percent recently, Sri Lanka’s growth performance is remarkable. Nevertheless, the country’s export base is narrow, with garment and tea exports accounting for two thirds of merchandize exports, and the expiration of the Multifiber Agreement in 2005 has added to competitive pressures. Tourism, another major exchange earner, has recovered from the tsunami but faces new threats from a deteriorating security situation. Finally, the country’s heavy reliance on oil, particularly in energy generation, exposes it more than others to movements in world prices. Between 2003 and 2005 Sri Lanka’s oil balance deteriorated by 2.4 percentage points of GDP, compared to 1.7 percentage points of GDP for the average low-income country in Asia.
2. On the other hand, Sri Lanka has access to a large and relatively stable source of foreign exchange—workers’ remittances. Over the last two decades, workers’ remittances have increased by an annual average of 10 percent and since 1994 constitute the largest source of foreign financing for the island (8.3 percent of GDP in 2005). Some 4 percent of the Sri Lankan population work abroad, mostly in the oil rich Gulf states, making Sri Lanka one of the leading recipients of remittances as a share of GDP. Remittances are a particularly attractive source of foreign financing, because they are much more stable over time than private capital flows. In addition, they are unrequited transfers, which unlike other capital flows, do not create obligations in the future.
3. This paper explores to what extent, workers’ remittances have helped cushion Sri Lanka against economic shocks and are likely to do so in the future. It is widely believed that workers’ remittances are to a great extent motivated by altruism.13 Under this assumption, they should be negatively correlated with income in the home country and as such constitute an insurance against shocks. Similarly, they should be positively correlated with incomes and wages in the host countries—in this case the Gulf States—and, hence, provide a welcome hedge against rising oil prices. Alternatively, remittances could be motivated by investment considerations, in which case they should respond to interest rate differentials and, in general, be more aligned with the business cycle in the home country. Under either hypothesis, other macroeconomic variables are likely to have a bearing on the amount of money sent home, notably the exchange rate and the price level in the home country.
4. A better understanding of what determines remittances, and in particular their cyclical properties, could help assess the potential role of remittances as a shock absorber. The central question is whether remittance receipts respond to economic conditions in the home country. A few studies investigate the cyclicality of inward remittances. IMF (2005) reports the correlation between detrended global remittances and detrended world GDP and finds that remittances are procyclical, albeit to a lesser extent than official aid, exports, and portfolio investment. Using the same approach on a country-by-country basis, Giuliano and Ruiz-Arranz (2005) show that remitting patterns vary across countries with procyclical remittances observed in two thirds of developing countries, suggesting that investment considerations play an important role. More recently, in a related cross-country work, we estimate a gravity model of bilateral remittance flows for a limited number of developing countries and find that remittances are aligned with the business cycle in the home country. These results suggest that remittances can play some role, but perhaps not a major one, in limiting vulnerability to shocks. Being procyclical, remittances tend to falter when exports weaken and GDP growth slows. They also decline when the home investment and political climate worsens and do not seem to respond to adverse shocks at home.14
5. Most studies of the macroeconomic determinants of remittances have focused on one country only whose diaspora is concentrated in a known country, or small group of countries. Straubhaar (1986) finds that wages and employment in Germany have a positive impact on Turkish inward remittances. El Sakka and McNabb (1999) try to explain nominal remittances received by Egypt and find that remittances increase with Egyptian inflation and income abroad and decline with the black market premium. For India, Gupta (2005) regresses real inward remittances on oil prices and migrants’ real overseas earnings and finds the latter to enter significantly and positive. In a second class of regressions, she finds that changes in U.S. employment and an Indian drought dummy have a positive impact on the cyclical component of remittances. Bouhga-Hagbe (2004) uses a vector error correction (VEC) specification to model workers’ remittances received by Morocco. The cointegration vector suggests that, over the long run, inward remittances are positively correlated with French wages and negatively correlated with real GDP in Morocco.
6. These attempts to establish a relationship between workers’ remittances and a set of macroeconomic variables suffer from a number of shortcomings. Some studies fail to discuss and account for the time series properties of the variables under investigation, although regressions of nonstationary variables are known to be spurious. One study runs a regression in variations—supposedly removing any nonstationarity—but fails to test for co-integration, thus opening the door for omitted variable bias. Moreover, some of the macroeconomic variables, such as the exchange rate, the price level or GDP, could be affected by remittances on top of affecting remittances. However, most studies ignore issues of endogeneity and reverse causality. Bougha-Hagbe (2004) is the noteworthy exception in accounting for time series properties and endogeneity, but robustness of his findings are questionable given that a VEC model with 22 parameters is estimated using 35 observations.
7. Our analysis tries to overcome weaknesses in previous studies and shows that remittances to Sri Lanka may be less of a hedge against shocks than commonly believed. We estimate a Vector Error Correction (VEC) model for Sri Lanka remittance receipts using quarterly data from 1996 to 2004. Our main focus lies on the response of remittances to a number of macroeconomic variables, namely real GDP, CPI, exchange rate, interest rate, and oil price. We find that remittances are positively correlated with the oil price, but behave strongly procyclical, and decline when the Sri Lankan currency weakens.
B. Stylized Facts
8. Workers’ remittances in Sri Lanka increased at an average annual rate of 10 percent over the last 20 years and since the mid-1990s constitute the largest source of foreign financing. In 2005, workers’ remittances amounted to 8.3 percent of GDP, compared to 2.5 percent of GDP in official development assistance (ODA), 1 percent of GDP in FDI, and ½ percent of GDP in portfolio investment. Sri Lanka’s prime export, textile and garments, amounted to 12 percent of GDP (Figure III.1).
Sources of Foreign Financing, 1975–2004
(Percent of GDP)
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
Sources of Foreign Financing, 1975–2004
(Percent of GDP)
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
Sources of Foreign Financing, 1975–2004
(Percent of GDP)
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
9. The evolution of Sri Lanka’s inward remittances is broadly in line with the trend observed in global remittance flows. As a global aggregate, workers’ remittances have become the largest source of foreign financing after FDI, exceeding both official development assistance and portfolio investment by a wide margin. In 2005, remittances to developing countries amounted to $165 billion. Asia and the Pacific is the main destination region for remittances, accounting for 45 percent of the global total. Some of the surge in workers’ remittances may be attributable to better recording and a shift from informal to formal channels, particularly after September 11, 2001. However, underpinned by mounting demographic pressures in the developing world, remittance flows are unlikely to abate soon. In the case of Sri Lanka, persistent rural poverty, growing inequality, and ethnic tensions will continue to secure stable flows of remittances in the medium term.
10. Inward remittances are large relative to the size of the Sri Lankan economy. Among 13 developing countries of broadly equal size, Sri Lanka exhibits the fourth largest remittances-to-GDP ratio (Figure III.2). In Asia, Sri Lanka is surpassed only by the Philippines (13.7 percent of GDP), Mongolia (13.3 percent of GDP), and Nepal (12.9 percent of GDP).
11. Inward remittances are sensitive to swings in oil prices, as close to 85 percent of Sri Lankan migrants reside in countries which are net oil exporters. According to the Sri Lanka Bureau of Foreign Employment (2004), the number of overseas workers amounted to 744,100 in 2004, or 3.8 percent of the population. These workers are concentrated in a few countries with Saudi Arabia, Kuwait, United Arab Emirates, and Qatar hosting 80 percent.
12. Sri Lanka remittances are less volatile than private capital flows and ODA, confirming a pattern observed for global aggregates. The standard deviation of remittances amounts to 43 percent of the mean, compared with 51 percent for ODA, 68 percent of FDI and 164 percent for portfolio investment. Merchandise exports, on the other hand, are less volatile than remittances, deviating only 21 percent from the mean.
13. Remittances to Sri Lanka seem to be procyclical and, strikingly, more so than any other source of foreign exchange. Remittances and GDP, when detrended by the Hodrick-Prescott filter, show a correlation of almost 70 percent over the period 1975–2004; slightly higher than the correlation of exports and GDP. Private capital flows and GDP are positively correlated at only 20 percent, while ODA is counter-cyclical. Figure III.3 plots remittance receipts against a couple of macroeconomic aggregates. The procyclicality of remittances is borne out by the first figure, which plots the log-differences of Sri Lanka remittance receipts and GDP over 1985–2005. Since the mid-1990s remittances and GDP seem to be moving in log-step. Noteworthy also the year 2001, when Sri Lanka was hit by a number of severe shocks, including a military attack on the Colombo airport, disruptions of the power supply, and severe weather. GDP contracted for the first time in 50 years and remittances recorded the lowest growth in more than 10 years. Similar responses of remittances to dramatic changes in the home country’s economic conditions have been observed for other countries (World Bank, 2003). In the Philippines, remittances rose steadily throughout the early 1990s, but became more volatile with the financial crisis in the late 1990s. In Turkey, remittance receipts increased for most of the 1990s, but started to decline when the crisis hit in 1999 and 2000.
Sri Lanka: Stylized Facts About Remittances
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
Sri Lanka: Stylized Facts About Remittances
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
Sri Lanka: Stylized Facts About Remittances
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
Sri Lanka: Correlations of Remittances and Macroeconomic Variables
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
Sri Lanka: Correlations of Remittances and Macroeconomic Variables
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
Sri Lanka: Correlations of Remittances and Macroeconomic Variables
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
14. Since the mid-1990s, one also observes a strong positive correlation between remittance receipts and GDP of the workers’ host countries. But, although Sri Lankans seek employment mostly in oil exporting countries, the correlation with oil prices is less clear-cut. Moreover, remittances appear to be negatively correlated with the exchange rate (implying less remittances when the currency weakens) and the interest rate (if at all), but not correlated with the price level.
C. Econometric Analysis
15. We estimate a VEC model for Sri Lanka to determine the response of remittance receipts to shocks in macroeconomic variables. The choice of a VEC model was based on the following considerations:
Most of the macroeconomic variables are endogenous, suggesting a multi-equation estimation. Many of the macroeconomic variables are likely to affect remittances but, given the magnitude of remittances, are also likely to be affected by remittances. Giuliano and Ruiz-Arranz (2005) provide evidence that remittances boost GDP by easing liquidity constraints. Large remittance flows may also lead to Dutch disease phenomena, that is nominal currency appreciation or inflation (or both). In fact, anecdotal evidence points to a link between large remittance inflows and house price inflation in some Latin American countries. Moreover, some of the macroeconomic variables, such as interest rates, exchange rates, and CPI are likely to be correlated among themselves. To account for these interactions without imposing too many restrictions we opt for a Vector Auto Regression (VAR).
Many of the variables are nonstationary, suggesting an estimation in first (or higher) differences. If nonstationary time series are regressed upon each other, the resulting t-statistics are biased upward, implying a correlation between variables, when in fact there is none. Differencing the time series until they become stationary is the usual way out. We, therefore, estimate a VAR in first differences.
The variables may be cointegrated, suggesting the inclusion of the cointegration relationship as an additional regressor. Estimating a simple VAR in differences, when the variables are cointegrated—that is a linear combination of the variables in levels is stationary—introduces omitted variable bias. In effect, one ignores important information about the variables’ long-term relationship. We therefore estimate a VEC model, a VAR in first differences that includes the long-term relationship of the endogenous variables as an additional regressor.
16. Our dataset covers the period 1996–2004 on a quarterly basis.15 While annual remittance data is available back to 1975, such a series would contain fewer observations and is more likely to contain structural breaks. As potential shock variables the dataset includes what the literature usually refers to as macroeconomic determinants of remittances, namely real GDP and CPI in the receiving country, the exchange rate, and a relative rate of return. Another common determinant, real GDP in the host country, is not available on a quarterly basis for the Gulf States. It is proxied by the world oil price, which is of more immediate interest for this study.
17. The data used in the analysis are drawn from the IMF databases and the country’s national statistics. Remittance (REM) data, in millions of U.S. dollar, is taken from the IMF Balance of Payments Statistics Yearbook and comprises the line items workers’ remittances, compensation of employees, and migrant capital transfers. Real GDP (GDP), in billions of rupees and at 1996 prices, is taken from the WEO database. The relative rate of return (I), in percentage points, is calculated as the difference between Sri Lanka’s interest rate on 3-months fixed deposits and the LIBOR on 3-months dollar deposits. Data on interest rates, the Colombo consumer price index (CPI), the rupee/dollar exchange rate (E) and the oil price (OIL)—a simple average of UK Brent, Dubai, and West Texas crude prices—all stem from the International Financial Statistics Yearbook. Data are in levels and not seasonally adjusted.
18. We first test for the presence of unit roots in the macroeconomic time series using the augmented Dickey Fuller test and find that all series are integrated of order one. To determine the appropriate lag length we start with a large number of lags and subsequently eliminate lags with insignificant coefficients. The choice of model, that is whether to include an intercept or time trend, is based on the approach of Doldado and others (1990). Under this approach, one starts with the least restrictive of plausible models and then introduces restrictions until the null hypothesis of a unit root is rejected (if at all). As shown in Table III.1, the data series are found to be nonstationary in levels (have unit roots) and stationary in first differences. Hence, all series are integrated of order one.
Sri Lanka: Augmented Dickey Fuller Test for Non-Stationarity 1/
(Sample: 1995Q1–2004Q4)
*** denote rejection at the 1 percent level.
Model 1 includes trend and intercept; Model 2 includes intercept, but no trend; and Model 3 includes neither.
Sri Lanka: Augmented Dickey Fuller Test for Non-Stationarity 1/
(Sample: 1995Q1–2004Q4)
| Level | First Difference | |||||
|---|---|---|---|---|---|---|
| t-ADF | Lags | Model 2/ | t-ADF | Lags | Model 2/ | |
| Remittances | 1.28 | 3 | 2 | -5.34 *** | 2 | 3 |
| Real GDP | 1.16 | 3 | 2 | -13.28 *** | 1 | 3 |
| CPI | 0.84 | 0 | 2 | -2.91 *** | 0 | 3 |
| Exchange rate | -0.25 | 1 | 2 | -4.43 *** | 0 | 2 |
| Interest differential | -1.13 | 1 | 3 | -2.66 *** | 0 | 3 |
| Oil price | 1.57 | 0 | 3 | -4.68 *** | 0 | 3 |
*** denote rejection at the 1 percent level.
Model 1 includes trend and intercept; Model 2 includes intercept, but no trend; and Model 3 includes neither.
Sri Lanka: Augmented Dickey Fuller Test for Non-Stationarity 1/
(Sample: 1995Q1–2004Q4)
| Level | First Difference | |||||
|---|---|---|---|---|---|---|
| t-ADF | Lags | Model 2/ | t-ADF | Lags | Model 2/ | |
| Remittances | 1.28 | 3 | 2 | -5.34 *** | 2 | 3 |
| Real GDP | 1.16 | 3 | 2 | -13.28 *** | 1 | 3 |
| CPI | 0.84 | 0 | 2 | -2.91 *** | 0 | 3 |
| Exchange rate | -0.25 | 1 | 2 | -4.43 *** | 0 | 2 |
| Interest differential | -1.13 | 1 | 3 | -2.66 *** | 0 | 3 |
| Oil price | 1.57 | 0 | 3 | -4.68 *** | 0 | 3 |
*** denote rejection at the 1 percent level.
Model 1 includes trend and intercept; Model 2 includes intercept, but no trend; and Model 3 includes neither.
19. Next, we test for the existence of a cointegration vector following Johansen (1991) and find one cointegration relationship. The vector auto regression includes two lags as suggested by the LR test statistic and most other selection criteria. As shown in Table III.2, the trace statistic indicates the existence of at most 3 cointegration vectors, while the eigenvalue statistic confirms the existence of at most 1 cointegration vector. Over the long run, remittances move with the other macroeconomic variables based on the following cointegrating relationship (t-statistic in parenthesis):
Sri Lanka: Cointegration Test for REM, GDP, CPI, E, I, OIL 1/
Rejection at the 1, 5, and 10 percent level denoted by ***, **, and *, respectively.
Sri Lanka: Cointegration Test for REM, GDP, CPI, E, I, OIL 1/
| Hypothesized Number ofCointegration Vector(s) | Eigenvalue | Trace Statistic | Maximum Eigenvalue Statistic |
|---|---|---|---|
| None | 0.81 | 143.23 *** | 58.87 *** |
| At most 1 | 0.61 | 84.36 *** | 33.45 |
| At most 2 | 0.49 | 50.90 ** | 24.27 |
| At most 3 | 0.43 | 26.63 | 20.20 |
Rejection at the 1, 5, and 10 percent level denoted by ***, **, and *, respectively.
Sri Lanka: Cointegration Test for REM, GDP, CPI, E, I, OIL 1/
| Hypothesized Number ofCointegration Vector(s) | Eigenvalue | Trace Statistic | Maximum Eigenvalue Statistic |
|---|---|---|---|
| None | 0.81 | 143.23 *** | 58.87 *** |
| At most 1 | 0.61 | 84.36 *** | 33.45 |
| At most 2 | 0.49 | 50.90 ** | 24.27 |
| At most 3 | 0.43 | 26.63 | 20.20 |
Rejection at the 1, 5, and 10 percent level denoted by ***, **, and *, respectively.
20. The estimation of the VEC model and the impulse response functions confirm the evidence presented in Section B, namely that remittances are procyclical and increase with oil prices. The estimates of the VEC model are presented in Appendix III.1. The impulse response functions illustrate how one standard deviation shocks in the oil price, the exchange rate, GDP, the interest rate, and CPI affect remittances (Figure III.4).
Remittances are procyclical: remittances increase when economic activity in the home country accelerates and they decrease when economic conditions deteriorate, an indication that investment considerations are at play. In particular, an increase in real GDP by 5.5 billion of 1996 rupees (2 percent) leads to an increase in remittances by $30 million (2 percent). This suggests that remittances respond to investment opportunities and the business and political climates in the home country as much as to altruistic and insurance considerations. It also implies that remittance flows may not be as important to smooth fluctuations or shocks in the economy as commonly believed.
Remittances fall when the exchange rate weakens: a one percent depreciation of the rupee against the dollar leads to a $15 million reduction in remittances (1 percent) in the first year and another $5 million reduction thereafter. Depreciation of the rupee reduces remittances as migrants may need to send less dollars in order to buy the same goods basket at home as before the depreciation.
Remittances increase with oil prices: an oil price increase of $2.8 per barrel is associated with a $15 million increase in remittances (1 percent). In the case of Sri Lanka, oil prices may be a good proxy for the economic activity in its migrants’ host countries. This result suggests that greater economic activity in the host country increases the chances of employment and wages, allowing migrants to send more remittances.
Remittances fall with higher interest rates: an increase in interest rates in Sri Lanka (vis-à-vis LIBOR) by ½ percentage point results in an $18 million reduction inremittances.16 The result that interest rate differentials are not important in attracting remittances is in line with earlier research and could explain the relative stability of remittances when compared with other types of capital flows. This finding together with that regarding the positive association of remittances with GDP growth suggests that remittances do not seem to respond to investment in interest bearing assets (i.e., financial investment) but they respond to productive investment opportunities.
Inflation does not seem to have any significant effect on remittances: CPI shocks do not trigger any change in remittance flows. This result is in contrast with our related cross-country work that finds that higher inflation in the home country encourage remittances to compensate for the loss of purchasing power at home.
Sri Lanka: Impulse Response Functions
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
Sri Lanka: Impulse Response Functions
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
Sri Lanka: Impulse Response Functions
Citation: IMF Staff Country Reports 2006, 447; 10.5089/9781451823578.002.A003
21. Our main findings are robust to other specifications. We reduce the lag length from 2 to 1, to decrease the number of estimated parameters from 14 to 8. The cointegration relationship stays broadly the same. The impulse response functions are qualitatively the same and quantitatively very similar, except for I which has less of an impact. Estimating bivariate correlations, we find that REM are cointegrated with GDP, E, and I. The impulse response functions are qualitatively the same, but the impact of the exchange rate is larger and the impact of GDP and I is smaller in the bivariate setting. Finally, we estimate a VEC model in REM, GDP, E, and I, the variables that are found to be correlated with REM on a bivariate basis. Cointegration equation and impulse response functions are qualitatively the same as in the baseline specification, and quantitatively very similar.
D. Conclusion
22. Remittance receipts seem to be procyclical in Sri Lanka, undermining their usefulness as a shock absorber upon deterioration in economic fundamentals. This paper explores to what extent Sri Lanka’s large receipts of workers’ remittances serve as a hedge against macroeconomic shocks. Both descriptive evidence and econometric analysis show that workers’ remittances are positively correlated with real GDP but negatively correlated with the strength of the rupee exchange rate and interest rate differentials, limiting their potential as shock absorber.
23. However, remittances are positively correlated with oil prices, offering a hedge against oil shocks. This is particularly important in Sri Lanka with a large percent of its migrant population working in the oil rich Gulf States and with oil imports accounting for more than 20 percent of total imports. During the most recent oil shock, robust growth in remittance flows has contributed to finance the current account, strengthen the balance of payments and accumulate reserves.
24. The evidence about what motivates remittances is mixed, but altruism may be less of a factor in Sri Lanka than commonly believed. The procyclicality of remittances calls into question the notion that remittances are largely motivated by altruism. At the same time we fail to confirm portfolio considerations as a prime motive, since no positive link is established between remittances and relative rates of return. This result is in line with our cross-country research paper that applies a gravity model to explain bilateral remittance flows. We find a positive association between remittance receipts and the dependency ratio in the home country, suggesting that helping those at home is an important motive. However, remittances appear to be procyclical, suggesting an investment motive. Remittances are also sensitive to the investment and political climate in the home and host countries, again suggesting that (non-financial) investment considerations play an important role.
25. The results suggest that while remittances should be encouraged they should not be seen as a panacea. Remittances can yield important economic benefits to Sri Lanka, providing financing and supporting consumption and investment. They can also play an important role in the regional development of the country and in reducing vulnerability to oil shocks. On the other hand, they may be of limited value in absorbing shocks to macroeconomic fundamentals (GDP and exchange rate). While it is important to continue facilitating remittance inflows with policies directed at reducing transaction costs, promoting financial sector development, and improving the business climate, remittances should not be seen as a substitute for government policy and structural reform.
Appendix III.1. Vector Error Correction Estimates 1/
Sample: 1996Q1–2004Q4; 36 observations; standard errors in parantheses.
Appendix III.1. Vector Error Correction Estimates 1/
| Error Correction | D(REM) | D(OIL) | D(I) | D(GDP) | D(E) | D(CPI) |
|---|---|---|---|---|---|---|
| CointEq1 | −0.730 | −0.012 | −0.002 | 0.446 | 0.015 | −0.011 |
| (0.358) | (0.043) | (0.008) | (0.088) | (0.017) | (0.032) | |
| D(REM(-1)) | -0.317 | -0.006 | -0.003 | -0.369 | -0.028 | 0.046 |
| (0.304) | (0.037) | (0.007) | (0.075) | (0.014) | (0.027) | |
| D(REM(-2)) | -0.101 | 0.014 | -0.003 | -0.212 | -0.018 | 0.025 |
| (0.219) | (0.027) | (0.005) | (0.054) | (0.010) | (0.020) | |
| D(OIL(-1)) | 0.736 | 0.236 | -0.014 | -0.365 | -0.047 | -0.077 |
| (1.784) | (0.216) | (0.042) | (0.438) | (0.085) | (0.160) | |
| D(OIL(-2)) | 1.649 | 0.138 | 0.024 | 0.232 | 0.102 | 0.058 |
| (1.709) | (0.207) | (0.040) | (0.419) | (0.081) | (0.154) | |
| D(I(-1)) | -8.007 | -0.796 | 0.555 | -3.419 | 0.479 | -1.030 |
| (11.132) | (1.347) | (0.261) | (2.731) | (0.528) | (1.001) | |
| D(I(-2)) | 3.586 | 1.122 | -0.160 | -1.498 | -0.433 | 1.143 |
| (9.913) | (1.199) | (0.233) | (2.432) | (0.470) | (0.891) | |
| D(GDP(-1)) | -0.750 | -0.046 | -0.005 | 0.813 | 0.018 | 0.046 |
| (0.825) | (0.100) | (0.019) | (0.203) | (0.039) | (0.074) | |
| D(GDP(-2)) | -1.014 | -0.062 | 0.015 | 0.065 | 0.067 | -0.036 |
| (0.811) | (0.098) | (0.019) | (0.199) | (0.038) | (0.073) | |
| D(E(-1)) | -7.858 | 0.737 | 0.048 | 1.395 | 0.280 | 0.715 |
| (4.559) | (0.551) | (0.107) | (1.119) | (0.216) | (0.410) | |
| D(E(-2)) | 2.860 | -1.200 | 0.229 | 2.871 | 0.418 | -0.134 |
| (5.117) | (0.619) | (0.120) | (1.255) | (0.243) | (0.460) | |
| D(CPI(-1)) | 3.186 | -0.104 | -0.038 | -0.427 | 0.043 | 0.399 |
| (2.698) | (0.326) | (0.063) | (0.662) | (0.128) | (0.243) | |
| D(CPI(-2)) | 1.127 | -0.341 | 0.017 | -1.189 | -0.167 | -0.370 |
| (2.889) | (0.349) | (0.068) | (0.709) | (0.137) | (0.260) | |
| C | 6.138 | 2.143 | -0.379 | 1.214 | 0.835 | 0.992 |
| (10.771) | (1.303) | (0.253) | (2.643) | (0.511) | (0.968) | |
| R-squared | 0.70 | 0.35 | 0.62 | 0.92 | 0.53 | 0.48 |
| Adj. R-squared | 0.53 | -0.03 | 0.40 | 0.87 | 0.25 | 0.17 |
| Sum sq. resids | 11831.09 | 173.10 | 6.52 | 712.23 | 26.60 | 95.62 |
| S.E. equation | 23.19 | 2.80 | 0.54 | 5.69 | 1.10 | 2.08 |
| F-statistic | 3.99 | 0.92 | 2.78 | 19.59 | 1.89 | 1.57 |
| Log likelihood | -155.39 | -79.35 | -20.32 | -104.81 | -45.64 | -68.67 |
| Akaike AIC | 9.41 | 5.19 | 1.91 | 6.60 | 3.31 | 4.59 |
| Schwarz SC | 10.03 | 5.80 | 2.52 | 7.22 | 3.93 | 5.21 |
| Mean dependent | 5.46 | 0.72 | -0.10 | 2.35 | 1.42 | 2.33 |
| S.D. dependent | 33.68 | 2.76 | 0.70 | 16.00 | 1.27 | 2.29 |
| Determinant resid covariance | 2,088.23 | |||||
| Log likelihood | -444.08 | |||||
| Akaike information criterion | 29.67 | |||||
| Schwarz criterion | 33.63 | |||||
Sample: 1996Q1–2004Q4; 36 observations; standard errors in parantheses.
Appendix III.1. Vector Error Correction Estimates 1/
| Error Correction | D(REM) | D(OIL) | D(I) | D(GDP) | D(E) | D(CPI) |
|---|---|---|---|---|---|---|
| CointEq1 | −0.730 | −0.012 | −0.002 | 0.446 | 0.015 | −0.011 |
| (0.358) | (0.043) | (0.008) | (0.088) | (0.017) | (0.032) | |
| D(REM(-1)) | -0.317 | -0.006 | -0.003 | -0.369 | -0.028 | 0.046 |
| (0.304) | (0.037) | (0.007) | (0.075) | (0.014) | (0.027) | |
| D(REM(-2)) | -0.101 | 0.014 | -0.003 | -0.212 | -0.018 | 0.025 |
| (0.219) | (0.027) | (0.005) | (0.054) | (0.010) | (0.020) | |
| D(OIL(-1)) | 0.736 | 0.236 | -0.014 | -0.365 | -0.047 | -0.077 |
| (1.784) | (0.216) | (0.042) | (0.438) | (0.085) | (0.160) | |
| D(OIL(-2)) | 1.649 | 0.138 | 0.024 | 0.232 | 0.102 | 0.058 |
| (1.709) | (0.207) | (0.040) | (0.419) | (0.081) | (0.154) | |
| D(I(-1)) | -8.007 | -0.796 | 0.555 | -3.419 | 0.479 | -1.030 |
| (11.132) | (1.347) | (0.261) | (2.731) | (0.528) | (1.001) | |
| D(I(-2)) | 3.586 | 1.122 | -0.160 | -1.498 | -0.433 | 1.143 |
| (9.913) | (1.199) | (0.233) | (2.432) | (0.470) | (0.891) | |
| D(GDP(-1)) | -0.750 | -0.046 | -0.005 | 0.813 | 0.018 | 0.046 |
| (0.825) | (0.100) | (0.019) | (0.203) | (0.039) | (0.074) | |
| D(GDP(-2)) | -1.014 | -0.062 | 0.015 | 0.065 | 0.067 | -0.036 |
| (0.811) | (0.098) | (0.019) | (0.199) | (0.038) | (0.073) | |
| D(E(-1)) | -7.858 | 0.737 | 0.048 | 1.395 | 0.280 | 0.715 |
| (4.559) | (0.551) | (0.107) | (1.119) | (0.216) | (0.410) | |
| D(E(-2)) | 2.860 | -1.200 | 0.229 | 2.871 | 0.418 | -0.134 |
| (5.117) | (0.619) | (0.120) | (1.255) | (0.243) | (0.460) | |
| D(CPI(-1)) | 3.186 | -0.104 | -0.038 | -0.427 | 0.043 | 0.399 |
| (2.698) | (0.326) | (0.063) | (0.662) | (0.128) | (0.243) | |
| D(CPI(-2)) | 1.127 | -0.341 | 0.017 | -1.189 | -0.167 | -0.370 |
| (2.889) | (0.349) | (0.068) | (0.709) | (0.137) | (0.260) | |
| C | 6.138 | 2.143 | -0.379 | 1.214 | 0.835 | 0.992 |
| (10.771) | (1.303) | (0.253) | (2.643) | (0.511) | (0.968) | |
| R-squared | 0.70 | 0.35 | 0.62 | 0.92 | 0.53 | 0.48 |
| Adj. R-squared | 0.53 | -0.03 | 0.40 | 0.87 | 0.25 | 0.17 |
| Sum sq. resids | 11831.09 | 173.10 | 6.52 | 712.23 | 26.60 | 95.62 |
| S.E. equation | 23.19 | 2.80 | 0.54 | 5.69 | 1.10 | 2.08 |
| F-statistic | 3.99 | 0.92 | 2.78 | 19.59 | 1.89 | 1.57 |
| Log likelihood | -155.39 | -79.35 | -20.32 | -104.81 | -45.64 | -68.67 |
| Akaike AIC | 9.41 | 5.19 | 1.91 | 6.60 | 3.31 | 4.59 |
| Schwarz SC | 10.03 | 5.80 | 2.52 | 7.22 | 3.93 | 5.21 |
| Mean dependent | 5.46 | 0.72 | -0.10 | 2.35 | 1.42 | 2.33 |
| S.D. dependent | 33.68 | 2.76 | 0.70 | 16.00 | 1.27 | 2.29 |
| Determinant resid covariance | 2,088.23 | |||||
| Log likelihood | -444.08 | |||||
| Akaike information criterion | 29.67 | |||||
| Schwarz criterion | 33.63 | |||||
Sample: 1996Q1–2004Q4; 36 observations; standard errors in parantheses.
References
Bouhga-Hagbe, Jacques, 2004, “A Theory of Workers’ Remittances with an Application to Morocco,” IMF Working Paper 04/194 (Washington: International Monetary Fund).
El-Sakka, Mohammed I., and Robert McNabb, 1999, “The Macroeconomic Determinants of Emigrant Remittances,” World Development, Vol. 27, No. 8, pp. 1493 –1502.
Giuliano, Paola, and Marta Ruiz-Arranz, 2005, “Remittances, Financial Development, and Growth,” IMF Working Paper 05/234 (Washington: International Monetary Fund).
Gupta, Poonam, 2005, “Macroeconomic Determinants of Remittances: Evidence from India,” IMF Working Paper 05/224 (Washington: International Monetary Fund).
International Monetary Fund, 2005, World Economic Outlook, April 2005 (Washington).
Johansen, Søren, 1991, “Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models,” Econometrica, Vol. 59, No. 6, pp. 1551 –1580.
Lucas, Robert E. B., and Oded Stark, 1985, “Motivations to Remit: Evidence from Botswana,” Journal of Political Economy, Vol. 93, pp. 901 –918.
Lueth, Erik, and Marta Ruiz-Arranz, 2006, “A Gravity Model of Workers’ Remittances,” IMF Working Paper (forthcoming) (Washington: International Monetary Fund).
Rajan, Raghuram; and Arvind Subramanian, 2005, “What Undermines Aid’s Impact on Growth?,” IMF Working Paper 05/126 (Washington: International Monetary Fund).
Rapoport, Hillel, and Frédéric Docquier, 2005, “The Economics of Migrants’ Remittances,” IZA Discussion Paper 1531 (Bonn, Germany: Institute for the Study of Labor).
Ratha, Dilip, 2003, “Workers’ Remittances: An Important and Stable Source of External Development Finance,” Global Development Finance 2003, pp. 157 –175 (Washington: World Bank).
Sayan, Serdar, 2006, “Business Cycles and Workers’ Remittances: How Do Migrant Workers Respond to Cyclical Movements of GDP at Home?,” IMF Working Paper 06/52 (Washington: International Monetary Fund).
Sri Lanka Bureau of Foreign Employment, 2004, Annual Statistical Report of Foreign Employment (Colombo).
Straubhaar, Thomas, 1986, “International Migrant Workers’ Remittances: The Case of Turkey,” Weltwirtschaftliches Archiv, Vol. 122, pp. 728 –740.
Prepared by Erik Lueth and Marta Ruiz-Arranz.
Given the data coverage, the analysis does not capture the impact of the tsunami on remittance flows.
While remittances increase in interest rates according to the cointegration relationship, this only holds under the ceteris paribus assumption. The VEC model implies that all variables are interdependent, and so a positive shock to the interest rate depresses GDP, which in turn depresses remittances—an effect that dominates the direct impact of interest rate on remittances.
