Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
European Integration and the Feldstein-Horioka Puzzle Margarita Katsimi and Gylfi Zoega European Integration and the Feldstein-Horioka Puzzle Margarita Katsimia and Gylfi Zoegab December 10, 2015 Abstract We apply the differences-in-differences method to study the effect of the European single market in 1993 and the euro in 1999 on the Feldstein-Horioka equation where countries outside the single market serve as a control group and those within as a treatment group. We find structural breaks that coincide with both events, in addition to the financial crisis in 2008. The results suggest that the correlation between investment and savings depends on institutions, exchange rate risk and credit risk. Furthermore, the pattern of capital flows within the single market leaves a significant part of the flows unexplained by fundamentals. 9971 words. Keywords: Feldstein-Horioka puzzle, European integration. JEL Classification: E2 a Department of International and Economic Studies, Athens University of Economics and Business, 76 Patision Avenue, 10434 Athens. CESifo, Munich, Germany, [email protected] b Corresponding author. Department of Economics, University of Iceland, 2 Saemundargata, 101 Reykjavik, Iceland. Department of Economics, Mathematics and Statistics, Birkbeck, University of London, Malet St, London WC1E 7HX, [email protected]. CESifo, Munich, Germany. Tel: +354 525 5239. Fax: +354 552 1331. Email: [email protected]. We are grateful to two anonymous referees for their comments on an earlier draft. 0 1. Introduction A large and growing literature has attempted to explain the findings of Feldstein and Horioka (1980) that savings and investment are correlated across countries. In this paper we take advantage of a natural experiment to test whether the beginning of the European single market in 1993 and the introduction of the euro in 1999 coincided with a structural break in the savings-investment relationship in the European countries but not in the rest of our sample of 30 countries. We use the differences-indifferences method by using countries that are outside the single market as a control group and assuming that the saving-investment relationship would have evolved in the same manner in countries currently within and outside the single market if these steps towards European integration had not been taken. We can then show how the effect of the single market and the euro affected the relationship for various groups of European countries, such as those within and outside the euro zone, the northern part of the euro zone and its southern periphery. The FH equation is the following (1) where I denotes gross capital formation, S is savings, Y denotes GDP, u is an error term and j is a country index. If capital was perfectly mobile across countries we would find the coefficient b to be close to zero because a fall in savings in one country would not affect domestic interest rates or investment. However, Feldstein and Horioka found that the estimated coefficient of the saving rate was 0.887 in a cross section of industrialized countries for the period 1960 and 1974. In a world of capital mobility the finding appears puzzling since a fall in savings in one country should not affect domestic interest rates or investment. The authors attributed their finding to barriers to capital mobility.1 1 Several authors have provided explanations for the FH puzzle. Coakley et al. (1996) use the fact that the difference between saving and investment equals the current account surplus in the national accounts to focus their explanation on current account imbalances. According to these authors, excessive current account deficits may raise the interest rate faced by a country in international capital markets making further borrowing difficult. Similarly, as argued by Tobin (1983) and Summers (1988), governments may dislike deficits for financial stability reasons and surpluses since they indicate room for expansionary policies. Katsimi and Moutos (2009) find that the FH result is impervious to the use of a broader view (i.e., including human capital) of investment and saving. Bai and Zhang (2010) show that financial frictions can explain the FH puzzle. 1 This paper makes two contributions in the Feldstein-Horioka literature: First, it shows that capital mobility as measured by the coefficient of savings in the FeldsteinHorioka equation is both time-varying and not unidirectional and then uses the Frankel (1992) theoretical framework for interpreting changes in the coefficient taking place in Europe in the past two decades. Frankel points out that a zero saving retention coefficient (the FH perfect capital null) requires both real parity to hold and national saving to be uncorrelated with the error term in the FH regression. The second contribution is that the paper attempts to explain factors affecting capital flows in Europe by exploring the pattern of capital mobility before the financial crisis of 2008. The results suggest that a large part of the pattern is unexplained by fundamentals, suggesting that one of the unintended consequences of European integration may have been a flow of capital from core to peripheral economies that led to asset price bubbles and financial instability in the periphery. We will estimate the FH equation 1960-2014 for a sample of 19 EU Single market and 11 non-single-market countries in order to test for structural breaks in the equation coinciding with steps towards European integration and the recent financial crisis. We first use a rolling 10-year panel regression, which is novel in the FH literature in showing the variability of capital mobility over time, especially in Europe. This stands in contrast to the conventional treatment of using time-averaged cross-sectional regressions. Next, we perform panel estimation for the whole sample period and test for structural breaks in the FH coefficient. By not confining our sample to the EU countries we can compare and contrast structural breaks between the euro zone, member countries of the single market that do not use the euro and countries that do not belong to the single market. Using the differences-in-differences method allows us to find the effect of the single market and the euro on the FH coefficient for European countries. We will then investigate the determinants of current account imbalances, which by definition equal the difference between saving and investment, and argue that one unintended consequence of increased capital mobility within the EU may have been asset price bubbles and financial instability in the periphery economies. 2 2. Theoretical framework Frankel (1992) explains how the FH puzzle may arise if other variables affect both savings and investment. Thus countries with low tax rates may have both high saving and investment rates. There is also the possibility that real interest rates are not equalized across countries due to a country premium, exchange rate premium or the expected depreciation of the currency. In this case a fall in savings in one country may make real interest rates rise and investment fall. For the saving-investment correlation to be zero the real interest parity must hold. The real interest rate differential can be written as (2) where and denote domestic and foreign nominal interest rates, and are expected domestic and foreign inflation, fd is the forward discount on the domestic currency and is the expected depreciation of the domestic currency. The first term on the RHS of equation (2) is the covered interest differential and is defined by Frankel (1992) as the 'country premium' since it captures country specific factors that may not allow real interest rates to be equal such as the possibility of capital controls or default risk. The second and third terms represent the exchange risk premium and the expected real depreciation and together they form the 'currency premium'. The elimination of both the 'country' and the 'currency' premium implies that the real interest parity holds. Two major institutional reforms increasing capital mobility have taken place in recent decades. The first was the single European market in 1993 and the second the introduction of the euro in 1999 (EMU). These may have affected the country and the currency premium in equation (2). The introduction of the single market involved the removal of capital controls with obvious implications for the country premium in equation (2) above. The introduction of EMU may also have been important for the currency premium. Regions within countries do not face currency risk when it comes to intra-country capital movements. Thus regional authorities can be expected to be less concerned about current account imbalances than national authorities. Real rates of interest should also be the same across regions with given intra-country purchasing power parity. Since the correlation between savings and investment should for this reason be much higher between countries than between regions within a country, the 3 establishment of the euro zone in 1999 may have coincided with a structural break in the coefficient of savings in the Feldstein-Horioka equation.2 Blanchard and Giavazzi (2002) studied current account balances, savings and investment and found that the FH puzzle had disappeared in the euro zone. Telatar et al. (2007) looked at nine EU countries for the 1970-2002 period and found that six of these countries moved from a low to a high capital mobility regime with EMU. Supportive evidence for a negative EMU effect on the saving retention coefficient has also been found by Kumar and Rao (2009) in a sample of 13 OECD countries and Choudhry and Kling (2014) in a sample of 252 countries. Smitz and von Hagen (2011) find that current account imbalances widened significantly in the euro area since the beginning of EMU while the current account balance of the euro zone as a whole remained almost balanced. They find that capital flows follow differences in capital endowments. These authors interpret their results as indicative of deepening in financial market integration in the euro area and go so far as to claim that the observed current account imbalances were a sign of the proper functioning of the euro area. These studies do not look at differences between various groups of countries within and outside the single market nor do they include the period of the recent financial crisis. Choudhry and Kling find that capital mobility increased until the beginning of the financial crisis and declined thereafter. Johnson and Lamdin (2014) also find a positive impact of the financial crisis on the coefficient of the saving ratio for the euro zone and non-euro countries of the EU. Specifically, they estimate the FH equation for a sample of 40 OECD countries for the 1980-2012 period and find that b increases (by 0.06) in the 2006-2012 period for the EU and euro zone countries while for the rest of the countries the savings ratio coefficient remains unaffected. However, somewhat surprisingly, they find that this effect comes from the 2006-2008 period and disappears in the subsequent years. Although the Feldstein-Horioka (FH) equation was initially estimated as a crosssection relationship, most recent studies use panel estimation techniques that allow both for country heterogeneity and time-series variation. Many of these studies involve testing for cointegration between savings and investment using a variety of 2 These results are supported by Helliwell (1998), Bayoumi and Rose (1993) for British regions, Helliwell and McKitrick (1998) for Canadian regions and Sinn (1992) for U.S. regions. 4 standard or more sophisticated econometric techniques,3 while other studies attempt to study the FH relationship using estimators that are consistent even in the absence of cointegration.4 We depart from the earlier studies by classifying the 30 countries in our sample into groups and then use the differences-in-differences method to find the effect of institutional changes in Europe (single market and the euro) on the FH coefficient using the countries outside the single market as a “control group”. One group of countries includes the single market countries, another group has eurozone countries in northern Europe, the fourth has eurozone countries in southern Europe and fifth and final group has countries in the single market who are not a part of the eurozone. Moreover, we extend the sample of observations to 2014 giving us six observations for each country after the onset of the financial crisis in 2008. In the following section we estimate a rolling ten-year window panel regression for sub-groups of countries in order to describe the time series variation in the coefficient of the saving rate in contrast with the static coefficient estimation from long-run averaged cross-sectional regressions. This then provides a rationale for investigating the time-series variation in the coefficient of savings, such as testing for structural breaks due to institutional changes. We explore whether the FH coefficient is sensitive to institutional changes, in particular the removal of capital controls in Europe in the 1990s and EMU in 1999, and the financial crisis starting in 2008. One could argue that the creation of the single market should decrease the 'country' premium for participating countries while the fiscal crisis after 2008 must have worked towards the opposite direction due to a rise in default risks. Also one would expect the adoption of the common currency to reduce the 'currency premium' for euro area countries. 3. Statistical results We observe savings and investment for 30 OECD countries going back to 1960 and ending in 2014. Unit-root tests and panel tests for cointegration between savings and 3 Miller (1988) is the first study using standard cointegration techniques while later studies such as Caporale et al. (2005) employ various asymptotically efficient cointegration estimators. For a survey see Apergis and Tsoumas (2009) and Kumar and Bhaskara (2009). 4 De Vita and Abbott (2002), Kollias et al (2008) and Coakley et al (2004). 5 investment show that over the period 1960-2014 many of the series have a unit root and it is difficult to reject cointegration for the whole period, easier for the sub period 1993-2014 (see Table A1). The value of the LM test of Hansen (1992) rejects the null of parameter stability for all countries but Belgium, Finland and Israel. The figure below highlights the evolution of the FH coefficients from equation (1) over time by plotting estimates for b using a "rolling window" of 10 observations.5, 6 Figure 1. The evolution of FH coefficients 1.2 1.0 euro zone 0.8 outside single market 0.6 0.4 0.2 0.0 single market - non-euro -0.2 -0.4 1980 1985 1990 1995 2000 2005 2010 The coefficients are estimated using panel estimation with fixed effects, a moving ten-year sample and White cross-section standard errors & covariance (d.f. corrected). The figure shows that capital mobility increased more in the single-market countries after 1993 than in the non-single market ones – the FH coefficient fell by more – and that the subsequent fall in capital mobility after the financial crisis is strong in the single market countries. It is interesting that capital mobility increased even more in the non-euro single-market countries than in the euro zone after 1999. Also, note that the FH coefficient has fallen in the past couple of years in the single market countries – both within and outside the euro zone – indicating greater capital mobility. Tables 1 and 2 show the results of (unbalanced) panel estimations of the FH equation for all countries as well as groups of countries defined in the Appendix. In 5 6 For example the observation in 1980 is estimated using 1971-1980 data. The list of countries belonging to each group is given in Appendix II. . 6 Table 1 we estimate equation (1) with OLS while in Table 2 we perform IV estimations using the lagged savings ratio as an instrument in order to address a possible endogeneity problem pointed out in the literature (see Feldstein and Horioka; 1980, Frankel; 1992). 7 In all estimations we use country fixed effects in order to account for the remaining countries' characteristics. We allow for three time dummies to test for structural breaks; one for the years 1993-98, another for 1999-2007 and the third one for 2008-2014. The time dummies are also interacted with the saving rate to test for structural breaks in the coefficient b in the equation above. A Wald test rejects the hypothesis of a constant intercept term as well as a constant coefficient of the saving rate across time. Comparing the OLS coefficients in Table 1 to the IV coefficients in Table 2 one can see that they are qualitatively quite similar. The values of the coefficients on the saving ratio are somewhat higher in the IV regressions while the coefficients of the interaction of savings with the time dummies have very similar values across estimation methods in most cases. For the whole sample of 30 countries we get a coefficient of 0.537 for the saving rate in Table 1 (and of 0.638 in Table 2). The southern euro-zone countries seem to have the lowest capital mobility over the whole sample period reflected in a coefficient of 0.81 in OLS estimations and 0.86 in IV regressions, which is the highest among the group of countries we examine. Countries outside the single market have a coefficient lower than countries inside the single market (0.48 compared to 0.65 in Table 1 and 0.56 compared to 0.75 in Table 2). To the extent that this difference reflects lower capital mobility in the single market countries, it may also explain the need for promoting free movement of capital through the creation of the single market. Within the single market, the results from IV regressions indicate a similar saving coefficient for the euro-zone and the non-euro countries whereas OLS estimates point to a higher saving coefficient for countries using the euro. Turning to the impact of the single market, the adoption of the euro and the economic crisis, one can see in Table 1 that the coefficient of the saving rate for the whole sample of countries shifts downwards in 1999-2007 to 0.076 and then back in 2008-2014 to 0.279. Similarly, in Table 2 the coefficient shifts downwards in 19997 We have also obtained similar results using alternative instruments such as the ratio of the population aged 65+ and the benefit replacement rate. However, in order not to restrict our sample substantially due to the availability of these instruments across countries and over time, we decided to perform the IV regressions using the lagged savings ratio as an instrument. See, amongst others, Kasuga (2004). 7 2007 to 0.197 and then back in 2008-2014 to 0.424. This indicates greater capital mobility in the 1999-2007 period, which is partly reversed during the crisis. In contrast, both OLS and IV estimates indicate that the creation of the single market does not seem to affect b when looking at the whole sample of countries. In column (2) of Table 1 we see that, as expected, the creation of the single market does affect b for the single-market countries. The coefficient of the saving rate falls significantly from 0.648 to 0.443 in 1993-98 and further to -0.026 in 1999-2007. About half of the fall in b is due to euro area creation is offset by the crisis after 2008 when it rises to 0.277. Interestingly, in Table 2 the effect of the single market was more significant to the single-market countries that did not later adopt the euro. One can see in column (4) for the non-euro single-market countries that the coefficient of the saving rate falls significantly from 0.682 to 0.487 in 1993-98 and further to 0.039 in 1999-07. Again about half of the fall in b due to euro area creation is offset by the crisis. Although in countries outside the European single market capital mobility appears not to be affected by the single market, euro zone creation seems to have coincided with a decrease in the value of b in the whole sample. However, this effect is much weaker for the non-single market countries (becomes statistically insignificant in the OLS regressions) and much stronger for the euro zone. Column (3) of Table 2 shows that the adoption of the euro coincides with a decrease of the coefficient of the savings ratio for countries outside the single market to 0.376, which is, however, five times higher than the same coefficient for single market countries (0.078). The effect is greater within the euro zone as shown in column (5) of Tables 1 and 2 where the creation of the euro area decreases the value of the savings ratio coefficient from 0.71 to -0.02 according to the OLS estimates and from 0.769 to 0.093 in IV regressions. The financial crisis leads to a rise in the saving ratio coefficient throughout the EU but it does not affect non-single market countries. The southern periphery of the euro zone is affected more than its northern part and the single-market countries that do not use the euro. In lines 9-14 in Tables 1 and 2 we perform a differences-in-differences analysis to show the results more clearly. This can be explained in terms of the equation below where superscript E denotes a group of European countries (the treatment group), 8 NSM the group of countries outside the single market (the control group) and t measures the time period; 60-92, 93-98, 99-07 or 08-14. (3) The equation calculates the difference in the change of the FH coefficient b between time periods between a group of European countries, on the one hand, and the group of non-single-market countries, on the other hand, which is equivalent to the difference between time periods of the within-group difference. Thus we find whether the coefficient of savings in equation (1) decreased more in the euro zone – or the northern part of the euro zone, the southern part or the single-market countries that do not have the euro – than in the control group, which has the non-single-market countries. As shown in lines 11-14 in Tables 1 and 2 the effect of the single market caused the largest fall in b in the southern part of the euro zone and in the non-euro part of the single market. The introduction of the euro has, as expected, the biggest effect in the euro zone, especially in the northern countries. Finally, the financial crisis starting in 2008 raised the value of b similarly in the euro zone and non-euro zone single market countries. Frankel's (1992) theoretical framework presented in equation (2) allows us to understand better the rise in the saving ratio after 2008. The fiscal crisis after 2008 was a clear case of a rise in the 'country risk' reflected by the significant rise in default risks. There is also the possibility of a euro exit creating currency risk in some of the countries. 4. Factors affecting capital flows Increased capital mobility in the European single market starting in 1993 and in the euro zone starting in 1999 may have increased economic efficiency; capital flowing to equalize marginal rates of return. Differences in the age structure of the population across countries, in output per capita and in growth prospects can thus be expected to have played a role in affecting capital flows. There is also the possibility that the capital flows have been destabilizing by generating unsustainable credit expansions and asset price bubbles. Thus, capital flows have been blamed for the recent financial crises in countries such as Greece, Iceland, Ireland and Spain. 9 In this section we explore to which extent observed current account imbalances can be explained by differences in the population age structure of different countries and by differences in growth rates and output per capita, on the one hand, and to which extent they are unexplained hence possibly causing asset price bubbles and financial instability, on the other hand. The life-cycle hypothesis (see Modigliani, 1975) and the national account identity imply that the age structure of the population should affect current-account balances so that a country with proportionately more young and old people should have deficits and a country with proportionately more middle-aged people should have current account surpluses. Studies of the relationship between the age structure and the current account go back to the work of Leff (1969) who regarded dependency rates as the determinant of saving rates.8 In addition, we include variables measuring economic growth defined as the rate of growth of output per capita, the level of output per capita and budget balances for the government as a ratio to GDP. Economic growth is included because in the medium to long term higher growth rates may reflect differences in the rate of technological progress which may coincide with an inflow of capital. A country having high rates of technological progress could be expected to have a greater demand for capital that would be expected to generate a capital inflow and current account deficits. The level of GDP per capita is included because a lower level of output per capita may reflect a higher marginal product of capital in a given country if its steady-state output per capita is the same as elsewhere. Thus Schmitz and von Hagen (2011) find that the elasticity of net capital inflows with respect to output per capita increases significantly after the adoption of the common currency for euro zone countries. Finally, government budget balances may be related to the current account since the sum of private and public savings net of investment has to equal the current account surplus. While the presence of such “twin deficits” does not prove prima facie evidence for causality, the link between the two may be stronger within a single currency area. 8 A relationship between age structure variables and domestic savings has been found by many authors, such as Fry and Mason (1982), Mason (1987, 1988), Higgins (1998), Lindh (1999), Herbertsson and Zoega (1999), Bloom et al., (2010) and Gudmundsson and Zoega (2014). In an interesting paper, Taylor and Williamson (1994) explain the capital flow from Britain to Australia, Canada, the U.S. and Argentina in the late 19th century, early 20th century by high youth dependency ratios in the colonies. 10 Table 3 reports the results for the sample of 30 countries for the period 1981 to 2007.9 We end the sample in 2007 so as not to allow the financial turbulence of the years 2008-2011 to affect our estimates. The dependent variable is the current account surplus as a share of GDP (in per cent). Among regressors we include two age groups, 0-24 and 65+ (in percentage of total population) in addition to the annual rate of growth of output per capita (in percentage), the level of output per capita (US thousands of dollars per capita at 2005 prices) and the budget balance as a share of GDP (in percentage). We start by introducing time dummies; the first takes the value one for each year since the start of the single market in 1993 and zero for all earlier years and another that takes the value one for each year since the start of the euro zone in 1999 and zero for all other years. We report the results of the pooled estimation in column (1) and the fixed effect estimation in column (2) using annual data. In columns (3) and (4) we then remove the time dummies and replace them with a dummy for the single market countries and a dummy for the euro zone countries. 10 The effect of each of the variables can then be found by looking at three coefficient estimates when the relevant variable is interacted with each dummy variable. Thus if one were interested in the effect of having a higher budget surplus on the current account surplus in column (4) for a euro zone country, one would add up the three coefficients of that variable that are statistically significant – the coefficient of the budget surplus variable, the coefficient of the variable interacted with the dummy variable for membership of the single market, and the coefficient of the variable interacted with membership of the euro zone. In this case the only significant coefficient is that of the variable interacted with the dummy variable for the single market, which is 0.29, implying a positive effect on the current account – a 1% budget surplus goes with 0.29% current account surplus. The results show that current account surpluses tend to be associated with government budget surpluses in the single market and high level of GDP per capita, the latter effect especially strong in the euro zone. High growth rates have a negative coefficient in the last time period (1999-2007) when time dummies are used but in the 9 Korea excluded because of missing data on government budget balances. When a single-market country enters the euro zone the value of the single-market dummy remains equal to one while the value of the euro zone dummy changes from zero to one. If a country remains in the single market and does not enter the euro zone (Sweden for example) then the single market dummy also remains equal to one. 10 11 euro zone high growth and current account surpluses appear to go together. Thus it is the output per capita and the budget surplus variables that most clearly affect current account surpluses in the expected way. The relationship between the current account and the demographic variables is such that the higher the proportion of the old, the smaller the current account surplus or the larger the deficit in the period 1999-2007 when fixed effects are used. However, this relationship does not appear in the fixed effect estimation for the euro zone in column (4). As shown by Gudmundsson and Zoega (2014) this should not come as a surprise because based on the age structure Germany, Austria, Belgium and Denmark should have smaller surpluses while the age structure cannot fully explain the current account deficits of Greece, Portugal and Spain. Somewhat surprisingly, the higher the proportion of the young, the larger the current account surplus. Again, this shows that the age structure of the population cannot explain the pattern of current account surpluses in this sample of countries. Looking closer at the results using time dummies in columns (1) and (2), the growth variable has a negative and statistically significant coefficient during the period 1999-2007 implying that a 5% increase in the rate of growth (as from 1% to 6%) reduces the current account by 1.15% of GDP (using the coefficient estimate 0.23 = -0.33 + 0.10 from the fixed effect estimation). Note that the sign of the coefficient is positive before 1999. Output per capita has a positive and statistically significant coefficient in all periods using the fixed effect estimation. Finally, the budget surplus has a positive and significant coefficient after 1993 under the fixed effects estimation and for the period after 1999 using the pooled estimator, implying that a 5% increase in the budget surplus (as a fraction of GDP) will make the current account improve by 1.15% of GDP using the fixed effect estimator. The variable measuring the share of the population over 65 (old) is statistically significant and negative in the post 1999 period. During this period the current account surplus is smaller the greater the proportion of old people in the population. However, it is positive in the period 1993-1998. The coefficient for the young is, in contrast to what was expected, positive after 1999 (-0.39+0.64=0.25) when using the fixed effects estimator, albeit smaller than the coefficient of the old. Now turning to columns (3) and (4) where the time dummies are replaced by dummy variables for the single-market countries and the euro zone countries we find that within the euro zone high output per capita has a positive effect on the current 12 account surplus. This implies that within the euro zone current account deficits – that is capital inflows – coincide with low GDP per capita. Budget surpluses are also associated with current account surpluses within the single market. What is surprising, however, is the current account surpluses within the euro zone are not explained by the population under 25 and over 65 years of age. There is even a positive relationship between the share of the population in these age groups and the current account surplus within the single market.11 The R-square of the pooled estimate in column (1) of Table 3 is 0.43 while the fixed effect estimate has an R-square of 0.76. This implies that a significant share of the capital flows is unexplained by the factors considered here. The residuals from the four regressions show that Germany, Austria, Canada, Norway, Sweden and Switzerland developed unexplained current account surpluses in the 2000s while Belgium, Estonia, Greece, Ireland, Iceland, Italy, New Zealand, Portugal, Spain and the U.K. had unexplained deficits during this period. This opens the possibility that speculative capital flows from northern European countries to the southern periphery of the euro zone, Estonia, Ireland, and Iceland may have caused the financial crises that hit these countries in the late 2000s. Investors may thus have wrongly assumed that the country premium (credit risk) in equation (2) was permanently lowered in all EU economies whereas it returned with a vengeance during the financial crisis. In Figure 2 we take the average value of the residuals for 2001-2007 from the fixed effect estimation of our equation in column (4) in Table 3 and plot them against the actual current account surpluses. 12 Note that the countries with the largest unexplained deficits suffered financial crises in 2008-2009 while the ones with the unexplained surpluses did not have a crisis. The current account deficits reflect a capital inflow that preceded the crises, which went through the banks in the case of Iceland, Ireland and Spain and the sovereign in the case of Greece and Portugal.13 11 Using a 55 country sample used by Gudmundsson and Zoega (2014) we find a stronger effect of the age structure in the predicted direction. 12 Australia (aut), Austria (aus), Belgium (bel), Canada (can), Chech Republic (cze), Denmark (den), Estonia (est), Finland (fin), France (fra), Germany (ger), Greece (gre), Iceland (ice), Ireland (ire), Italy (ita), Japan (jap), Netherlands (net), Norway (nor), New Zealand (nze), Poland (pol), Portugal (por), Poland (pol), Slovenia (slove), Slovakia (slova), Spain (spa), Sweden (swe), Switzerland (swi), UK (uk) and the US (us). 13 A major financial crisis hit Belgium in 2008 when two large banks, Fortis and Dexia, encountered liquidity as well as solvency problems. At the same time British banks had problems requiring the government to offer a rescue package intended to restore confidence in the banking system. The bursting of a house price bubble made the Irish banks insolvent requiring the government to stabilise them by taking on large liabilities. A housing bubble also burst in Spain in. The Icelandic banks failed 13 Figure 2. The current account and its unexplained component 15 current account (%) 10 nor swi net 5 ger den aus can isr chi slove fra uk ita ire pol cze nz aut us slova spa gre est por -5 ice -10 swe jap bel 0 fin -15 -6 -4 -2 0 2 4 6 residual from Table 3 We next reduce the sample of countries to the six surplus countries (Austria, Canada, Germany, Norway, Sweden and Switzerland) and ten deficit countries (Belgium, Estonia, Greece, Ireland, Iceland, Italy, New Zealand, Portugal, Spain and the U.K.). In Table 4 we estimate the FH equation presented in Tables 1 and 2 and discussed in the previous section for these six surplus and the ten deficit countries. As in Tables 1 and 2, the IV estimations produce in general higher coefficients than OLS. One can see that the creation of the single market reduced the coefficient for the saving rate more in the deficit countries than in the surplus countries. As a result of the single market creation deficit countries seem to have experienced a higher rise in capital mobility than the surplus countries reflected in a savings rate coefficient of 0.26 in OLS estimations and of 0.44 in IV estimations instead of 0.38 and 0.52 respectively. In contrast, the creation of the euro area seems to have led to a slightly stronger decrease in the savings coefficient for the surplus countries. The financial crisis led to a decrease in capital mobility in both groups however, although the impact was stronger for the surplus countries. The rise in the saving rate coefficient for the surplus countries was roughly as high as its initial fall due to the creation of the in October 2008. Greece and Portugal had a sovereign debt crisis in 2010 and. The Baltics also had a crisis when domestic credit expansion came to an end. 14 single market while it was much smaller for the deficit countries. Specifically the financial crisis increased the coefficient to 0.33 in surplus countries and to 0.16 in deficit countries according to the OLS estimates.14 Finally, we explore further whether the explanatory variables can account for the pattern of current account surpluses and deficits by repeating the estimation in Table 3 for this narrower set of countries. The results are shown in Table A2 in Appendix I. The R-square of both the pooled as well as the panel regressions in columns (1) to (4) tend to be lower than before in Table 3. Moreover, the residuals show the same pattern of rising surpluses in the first group of countries as before, most importantly in Germany, while the residuals in the second group, such as in the crisis countries of Iceland, Ireland, Portugal, Spain and Greece, show rising deficits in the 2000s. The model thus does not help explain the pattern of surpluses and deficits in the 2000s based on fundamentals leading up to the financial crises in the deficit countries. 5. Conclusions Frankel (1992) pointed out that for investment and savings to be perfectly uncorrelated requires real interest rates to be equalised across countries – thus a zero country and currency premium. The single market could be expected to have reduced the country premium by making its country commit to the free flow of capital and the euro would by definition have abolished the currency premium. Finally, the financial crisis has increased the country premium and even raised the currency premium through the threat of euro exit by some members of the euro zone. Our results confirm changes to the coefficient of savings in the FH equation that coincide with the single market, the introduction of the euro and the financial crisis beyond the changes observed in the control group of countries outside the single market. We find that the increase in the FH measure of capital mobility coincides with the start of the European single market in 1993 and the introduction of the euro in 1999. We find that the former effect occurs in single-market countries that remain outside the euro zone whereas the latter is greater in the euro zone countries. The increased capital mobility within the euro zone consists of capital coming from nations with a higher output per capita and government budget surpluses. Also, the financial crisis that started in 2008 has led to a fall in the FH measure of capital 14 The difference is smaller when the IV estimator is used. 15 mobility in the single market. This recent development has occurred in both the member states of the single market that use the euro as well as in those countries that use their own currencies. However, capital mobility has increased in the past couple of years in the single market, both within and outside the euro zone. In contrast, the financial crisis has not affected the magnitude of the savings ratio coefficient in nonsingle market countries significantly. We conclude that the institutional changes that followed the Maastricht treaty that created the single market in capital and eventually the introduction of the euro increased capital mobility in the member countries beyond what was observed in countries outside the euro zone. The financial crisis also reduced capital mobility more in these countries. Lastly, we cannot account fully for the pattern of current account surpluses, which suggests that the capital flows may have had destabilising effects within the single market. Acknowledgements We gratefully acknowledge comments by Ron Smith and Jerry Coakley and seminar participants at Athens University of Economics and Business and the Central Bank of Turkey on an earlier draft. This work was supported by the Icelandic Research Fund (IRF – grant number 130551-052) and the University of Iceland Research Fund. 16 Table 1. Panel estimation : OLS with F.E. All countries Single market Non-single market Euro zone Euro zone - north Euro zone - south (3) 16.577*** (5.87) Single market noneuro (4) 12.968*** (9.30) (1) 15.342*** (9.00) (2) 10.982*** (11.11) (5) 9.437*** (6.54) (6) 10.211*** (6.82) (7) 7.155** (3.53) c93-98 -0.835 (-0.61) 1.583 (1.17) -0.790 (-0.68) 1.379 (0.76) -0.128 (-0.05) -1.952 (-0.63) 4.696*** (2.85) c99-07 8.538*** (3.44) 13.210*** (5.02) 2.604 (0.80) 10.728** (2.42) 14.628*** (4.67) 7.980* (1.81) 12.672*** (2.81) c08-14 1.941 (1.15) 4.153*** (3.31) -0.639 (-0.24) 2.375** (2.05) 5.315*** (3.23) 5.450 (1.37) 6.923*** (4.67) S/Y 0.537*** (8.21) 0.648*** (15.80) 0.476*** (4.60) 0.505*** (7.71) 0.710*** (12.03) 0.688*** (9.89) 0.810*** (10.13) d93-98*s -0.057 (-0.86) -0.205*** (-3.12) 0.020 (0.50) -0.220** (-2.36) -0.119 (-0.91) -0.049 (-0.33) -0.317*** (-3.61) d99-07*s -0.461*** (-4.38) -0.674*** (-6.14) -0.215* (-1.81) -0.593** (-3.54) -0.726*** (-4.93) -0.487** (-2.41) -0.517** (-2.03) d08-14*s -0.258*** (-3.73) -0.057 -0.404 0.203 -0.077 -0.169 0.067 30 1127 0.658 -0.371*** (-7.40) -0.205 -0.469 0.303 -0.225 -0.234 0.167 18 669 0.624 -0.079 (-0.87) 0.020 -0.235 0.136 8 349 0.744 -0.288*** (-5.46) -0.220 -0.373 0.305 -0.240 -0.138 0.169 5 201 0.655 -0.423*** (-7.18) -0.119 -0.607 0.303 -0.139 -0.372 0.167 11 428 0.601 -0.441** (-3.03) -0.049 -0.438 0.046 -0.069 -0.203 -0.090 7 265 0.584 -0.450*** (-7.68) -0.317 -0.2 0.067 -0.337 0.035 -0.069 4 163 0.641 6.22(0.00) 7.49(0.00) 8.44(0.00) 18.89(0.00) 1.02(0.44) 2.00(0.20) 2.15(0.24) 11.03(0.02) 13.03(0.00) 31.98(0.00) 2.65(0.14)) 5.80(0.03) 15.49(0.02) 824(0.00) Constant Difference: 60-92 and 93-98 Difference: 93-98 and 99-07 Difference: 99-07 and 08-14 Differences in differences 93-98/60-92 Differences in differences 99-07/93-98 Differences in differences 08-14/99-07 Countries (clusters) Observations R squared Wald test F: c93-98= c99-07= c08-12=0 F: d93-98= d99-07= d08-12=0 Standard errors clustered by country d.f. corrected. t-ratios in parentheses. Significance denoted by stars: *** at the 1% level, ** at the 5% and * at the 10% level of significance. 17 Table 2. Panel estimation, IV with F.E. All countries Single market Non-single market Single market noneuro Euro zone Euro zone - north Euro zone - south (1) 8.505*** (12.12) (2) 5.549*** (5.53) (3) 10.094*** (10.66) (4) 6.900*** (4.33) (5) 8.652*** (4.97) (6) 2.038 (0.92) (7) 5.574 (1.63) c93-98 -2.271* (-1.79) 0.063 (0.04) -1.013 (-0.64) 1.325 (0.80) -4.313 (-1.54) -5.649* (-1.67) 6.318 (0.72) c99-07 8.127*** (7.08) 13.313*** (8.75) 1.893 (1.30) 11.954*** (5.66) 13.681*** (6.51) 5.006 (1.62) 12.710*** (3.03) c08-14 1.133 (1.07) 3.690*** (2.62) -0.291 (-0.21) 3.164 (1.61) 3.214 (1.33) -1.235 (-0.22) 5.422 (1.30) S/Y 0.638*** (17.95) 0.753*** (15.16) 0.559*** (11.12) 0.682*** (7.89) 0.769*** (10.47) 0.830*** (9.48) 0.860*** (5.66) d93-98*s 0.014 (0.24) -0.125* (-1.84) 0.030 (0.39) -0.195** (-2.55) 0.081 (0.64) 0.124 (0.83) -0.388 (-0.93) d99-07*s -0.441*** (-9.05) -0.675*** (-10.42) -0.183*** (-3.03) -0.643*** (-7.86) -0.676*** (-7.29) -0.371*** (-2.84) -0.504** (-2.56) d08-14*s -0.214*** (-4.72) 0.014 -0.455 0.227 -0.016 -0.242 0.139 30 1086 0.650 -0.335*** (-5.57) -0.125* -0.55 0.34 -0.155 -0.337 0.252 18 649 0.593 -0.095 (-1.62) 0.030 -0.213 0.088 8 334 0.739 -0.317*** (-3.97) -0.195 -0.448 0.326 -0.235 -0.235 0.238 5 194 0.619 -0.299*** (-2.82) 0.081 -0.757 0.377 0.051 -0.544 0.289 11 416 0.567 -0.134 (-0.55) 0.124 -0.495 0.237 0.094 -0.282 0.149 7 258 0.495 -0.315 (-1.48) -0.388 -0.116 0.189 -0.418 0.097 0.101 4 158 0.610 69.38(0.00) 97.72(0.00) 89.50(0.00) 17.353(0.00) 3.38(0.34) 10.83(0.01) 34.33(0.00) 65.43(0.00) 72.23(0.00) 68.61(0.00) 7.48(0.06) 10.91(0.01) 1.39(0.01) 6.63(0.08) Constant Difference: 60-92 and 93-98 Difference: 93-98 and 99-07 Difference: 99-07 and 08-14 Differences in differences 93-98/60-92 Differences in differences 99-07/93-98 Differences in differences 08-14/99-07 Countries (clusters) Observations R squared Wald test F: F: Standard errors clustered by country d.f. corrected. t-ratios in parentheses. Significance denoted by stars: *** at the 1% level, ** at the 5% and * at the 10% level of significance. 18 Table 3. The current account and fundamentals (1981-2007) Periods Constant Young Old Growth Output per capita Budget surplus as share of GDP (1) Pooled 0.20 (0.05) -0.09 (1.51) 0.12 (1.22) (2) F.E. -28.52* (3.02) 0.64* (3.07) 0.16 (0.85) 0.20* (2.11) 0.001* (6.15) 0.10* (2.30) 0.002* (3.40) Growth 0.05 (0.88) 0.001 (0.01) Budget surplus as share of GDP 1993-1998 Constant (93-09) Young Old Growth Output per capita Budget surplus as share of GDP Young Old Growth Output per capita Budget surplus as share of GDP Cross sections Observations R-squared Young Old Output per capita 0.09 (1.06) 0.001* (12.45) 0.14* (2.91) 0.002* (3.29) 0.06** (1.70) 0.02 (0.43) -27.24* (6.82) 0.31* (3.58) 1.42* (12.26) -0.65* (4.51) -0.003* (2.25) 0.39* (6.95) -17.73* (3.53) 0.34* (3.06) 0.73* (4.55) -0.27* (2.46) -0.002** (1.82) 0.29* (4.90) -7.79 (0.80) 0.04 (0.24) -0.72* (2.06) 0.82* (5.06) 0.54* (18.69) 0.01 (0.08) 29 586 0.49 -22.81* (2.62) 0.20 (1.29) 0.41 (1.38) 0.34* (2.48) 0.04* (6.91) 0.15 (0.35) 29 586 0.76 Single market countries -12.62* (2.91) 0.11 (1.41) 0.74* (5.60) -0.19 (1.06) -0.001* (2.64) 0.11 (1.17) -4.05 (0.63) 0.04 (0.29) 0.47* (2.99) 0.05 (0.54) -0.001* (2.97) 0.13** (1.89) 1999-2007 Constant (99-09) Constant Participation in European integration (3) (4) Pooled F.E. 2.64 -12.44** (1.24) (1.92) -0.16* 0.22** (3.79) (1.70) 0.10* 0.16 (1.98) (1.09) Constant (93-09) Young Old Growth Output per capita Budget surplus as share of GDP Euro zone countries 7.26 (1.49) -0.01 (0.17) -0.38* (2.63) -0.75* (4.51) 0.0003* (2.41) 0.48* (5.64) 29 586 0.43 24.87* (3.68) -0.39* (2.96) -0.76* (3.92) -0.33* (2.55) 0.001 (4.87) 0.23* (2.71) 29 586 0.76 Constant (99-09) Young Old Growth Output per capita Budget surplus as share of GDP Cross sections Observations R-squared White cross-section standard errors & covariance (d.f. corrected p). t-ratios in parentheses. Significance denoted by stars: * at the 5% and ** at the 10% level of significance. 19 Table 4. Panel estimation, OLS and IV with F.E.: Surplus and Deficit Countries Surplus countries (OLS) 11.422*** (6.60) Surplus countries (IV) 5.361*** (2.67) Deficit countries (OLS) 9.213*** (6.35) Deficit countries (IV) 4.577** (2.49) c93-98 3.206 (1.71) 2.737 (1.62) 5.943*** (3.29) 4.810** (2.02) c99-07 9.672** (3.43) 9.742*** (5.17) 11.788* (1.95) 11.040*** (3.39) c08-14 2.604 (0.89) 2.962 (1.26) 6.077** (2.79) 3.946* (1.67) S/Y 0.631*** (9.53) 0.733*** (10.63) 0.681*** (9.53) 0.791*** (8.37) d93-98*s -0.248** (-3.19) -0.214*** (-2.74) -0.416*** (-4.43) -0.345*** (-2.91) d99-07*s -0.562*** (-5.07) -0.565*** (-7.11) -0.551 (-1.76) -0.482*** (-2.94) d08-14*s -0.302** (-3.51) -0.248 -0.314 0.26 6 294 0.648 -0.321*** (-3.47) -0.214 -0.351 0.244 6 285 0.636 -0.517*** (-3.66) -0.416 -0.135 0.034 10 357 0.620 -0.338*** (-2.67) -0.345 -0.137 0.144 10 342 0.586 3.95(0.07) 10.43(0.01) 30.80(0.00) 53.16(0.00) 4.99(0.03) 13.46(0.00) 12.00(0.00) 12.14(0.01) Constant Difference: 60-92 and 93-98 Difference: 93-98 and 99-07 Difference: 99-07 and 08-14 Countries (clusters) Observations R squared Wald test F: F: Standard errors clustered by country (d.f. corrected). t-ratios in parentheses. Significance denoted by stars: *** at the 1% level, ** at the 5% and * at the 10% level of significance. Appendix I Table A1. Panel cointegration test (Pedroni) Null Hypothesis: No cointegration 1993-2014 Panel v-Statistic Panel rho-Statistic Panel PP-Statistic Panel ADF-Statistic Observations: Cross-sections: Statistic -0.91 2.30 1.41 -0.86 Prob. 0.972 0.989 0.921 0.195 Weig.St. -1.71 1.21 -0.27 0.14 682 30 (1 dropped) 1960-2014 Prob. 0.96 0.89 0.39 0.56 Statistic 0.34 -2.66 -2.33 -5.61 Prob. 0.366 0.004 0.010 0.000 Weig.St. 0.241 -7.08 -6.44 -5.93 1705 30 (1 dropped) Trend automatic lag length selection based on AIC with lags from 1 to 9: Deterministic intercept and trend. Newey-West automatic bandwidth selection and Bartlett kernel. 20 Prob. 0.405 0.000 0.000 0.000 Table A2. The current account and fundamentals (1981-2007) – Narrow sample Periods Constant Young Old Growth Output per capita Budget surplus as share of GDP 1993-2007 Constant (93-09) Young Old Growth Output per capita Budget surplus as share of GDP (1) Pooled 3.73 (1.03) -0.16 (2.73)* 0.04 (0.41) 0.21 (2.02)* -0.0001 (0.23) 0.12 (1.66)** (2) F.E. -3.89 (0.29) 0.39 (1.41) -0.67 (1.99)* 0.18 (2.28)* -0.01 (1.75)** 0.19 (2.71)* -24.05 (5.63)* 0.39 (4.94)* 0.91 (6.96)* -0.44 (2.15)* 0.0005 (0.66) 0.06 (0.61) -34.00 (4.44)* 0.67 (4.19)* 1.10 (5.51)* -0.16 (1.17) 0.0002 (0.26) 0.00 (0.05) 1999-2007 Constant (99-09) Young Old Growth Output per capita Budget surplus as share of GDP Cross sections Observations R-squared Constant Young Old Growth Output per capita Budget surplus as share of GDP Single market countries Constant (93-09) Young Old Growth Output per capita Budget surplus as share of GDP Participation in European integration (3) (4) Pooled F.E. 8.34* 57.24* (2.84) (5.57) -0.22* -0.69* (3.33) (3.78) -0.11 -2.19* (1.36) (7.40) 0.14 0.17* (1.34) (2.33) -0.001 -0.01* (1.00) (3.78) 0.16* 0.24* (3.60) (4.71) -48.40* (7.92) 0.58* (5.31) 2.23* (8.78) -0.98* (3.28) -0.0002 (0.16) 0.41* (4.62) -65.87* (5.64) 1.01* (4.84) 2.40* (6.46) -0.55* (3.14) 0.001* (2.15) 0.14** (1.72) -14.23 (1.52) 0.10 (0.51) -0.63* (2.06) 0.95* (3.27) 0.61* (21.69) -0.39* (2.54) 16 383 0.54 -43.87* (3.33) 0.40 (1.63) 1.21* (2.94) 0.51* (2.67) 0.35* (7.19) -0.30** (1.72) 16 383 0.73 Euro zone countries 20.90 (1.99)* -0.32 (1.45) -0.66 (2.49)* -0.66 (2.74)* -0.003 (2.49)* 0.78 (8.27)* 16 383 0.39 22.38 (2.55)* -0.42 (2.22)* -0.49 (2.20)* -0.37 (2.86)* 0.0004 (0.28) 0.44 (5.43)* 16 383 0.71 Constant (99-09) Young Old Growth Output per capita Budget surplus as share of GDP Cross sections Observations R-squared White cross-section standard errors & covariance (d.f. corrected). t-ratios in parentheses. Significance denoted by stars: * at the 5% and ** at the 10% level of significance. 21 Appendix II EU- single market: Austria, Belgium, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy, Netherlands, Norway, Portugal, Spain, Sweden Switzerland, U.K. Non single market: Australia, Canada, Chile, Israel, Japan, Korea, New Zealand, U.S. EU- euro: Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal, Spain. EU-euro-north: Austria, Belgium, Finland, France, Germany, Ireland and the Netherlands EU-euro-south: Greece, Italy, Portugal, Spain Single market – non euro: Iceland, Norway, Sweden, Switzerland, the U.K. Data Gross capital formation as a share of GDP. Source: World Bank. Gross national saving as a share of GDP. Source: World Bank General government net lending/borrowing. Source: IMF. Real GDP per capita in US collars. Calculated as the ratio of output-side real GDP at chained PPPs (in mil. 2005US$) and population in millions. Source: Penn-World Tables. Share of population under age 25 and 65 and older. Source: World Bank. 22 References Apergis, N. and Tsoumas, C. (2009). “A survey of the Feldstein-Horioka puzzle: What has been done and where we stand,” Research in Economics, Vol. 63, pp. 6476. Bai, Y. and Zhang, J (2010). "Solving the Feldstein-Horioka puzzle with financial frictions„, Econometrica , Vol. 78(2), pp. 603-632. Bayoumi, T. and Rose, A. (1993). “Domestic saving and intra-national capital flows,” European Economics Review, Vol. 37 (6), pp. 1197–1202. Blanchard, O. and Giavazzi, G. (2001). “Current account deficits in the Euro area: The end of the Feldstein-Horioka puzzle?” Brookings Papers on Economic Activity, 2, pp. 147-209. Bloom, D., Canning, D., and Fink, G. (2010). “Implications of population ageing for economic growth,” Oxford Review of Economic Policy, Vol. 26 (4), pp. 583-612. Caporale, G.M., Panopoulou, E., and Pittis, N. (2005). "The Feldstein-Horioka puzzle revisited: A Monte-Carlo study" Journal of International Money and Finance, Vol. 24, pp. 1143-1149. Choudhry, T., Jayasekera, R. and Kling, G. (2014). “The global financial crisis and the European single market: The end of integration?” Journal of International Money and Finance, Vol. 49, pp. 191-196. Coakley, J., Kulasi, F. and Smith, R. (1996). “Current account solvency and the Feldstein-Horioka Puzzle,” The Economic Journal, Vol. 106, pp. 620-627. Coakley, J., Fuertes, A.M. and Spagnolo, F. (2004). “Is the Feldstein–Horioka Puzzle History?”, The Manchester School, Vol. 72, pp. 569-590. De Vita, G. and Abbot, A. (2002). “Are saving and investment cointegrated? An ARL bounds testing approach,” Economics Letters, Vol. 77 (2), pp. 293-299. Feldstein, M. and Horioka, C. (1980). “Domestic saving and international capital flows,” Economic Journal, Vol. 90 (2), pp. 314-329. Frankel, J. A. (1992). “International capital mobility: A review,” The American Economic Review Papers and Proceedings, Vol. 82 (2), pp. 197-202. Fry, M. and Mason, A. (1982). “The variable rate-of-growth effect in the life-cycle model,” Economic Inquiry, Vol. 20, pp. 232-233. Gudmundsson, G.S. and Zoega, G. (2014). “Age structure and the current account,” Economics Letters, Vol. 123 (2), pp. 183-186. Hansen, B.E. (1992). “Tests for parameter instability in regressions with I(1) Processes,” Journal of Business and Economic Statistics, Vol. 10, pp. 321-336. 23 Helliwell, J. F. and McKitrick, R. (1998). “Comparing capital mobility across provincial and national borders,” NBER Working Paper No. 6624. Herbertsson, T. T., Zoega, G. (1999). Trade surpluses and life-cycle saving behavior. Economics Letters, Vol. 65, pp. 227-237. Higgins, M. (1998). “Demography, national savings, and international capital flows,” International Economic Review Vol. 39 (2), pp. 343-369. Johnson, M.A. and Lamdin, D.J. (2014). “Investment and saving and the euro crisis: A new look at Feldstein-Horioka,” Journal of Economics and Business, Vol. 76, pp. 101-114. Kasuga, H., 2004. "Saving-investment correlations in developing countries". Economics Letters, Vol. 83, pp. 371-376. Katsimi, M. and Moutos, T. (2009). A note on human capital and the FeldsteinHorioka puzzle", The Manchester School, 7 Vol. 7, pp. 398-409. Kollias, C., Mylonidis, N. and Paleologou, S.M. (2008). “The Feldstein–Horioka puzzle across EU members: Evidence from the ARDL bounds approach and panel data,” International Review of Economics & Finance, Vol. 17, pp. 380-87. Kumar, S. and Bhaskara Rao, B. (2009). "A time series approach to the FeldsteinHorioka puzzle with panel data from the OECD countries". MPRA Paper 18464, University Library of Munich, Germany. Leff, N.H. (1969). Dependency rates and saving rates. American Economic Review, Vol. 59 (5), pp. 886–896. Lindh, T. (1999). “Age structure and economic policy: The case of saving and growth,” Population Research and Policy Review, Vol. 18, pp. 261-277. Mason, A. (1987). “National Savings Rates and Population Growth: A New Model and New Evidence,” in D. Johnson and R. Lee, eds., Population Growth and Economic Development: Issues and Evidence, Wisconsin University Press, pp. 523560. Mason, A (1988). “Saving, Economic growth and Demographic Change,” Population and Development Review, Vol. 14, pp. 113-144. Miller, S.M., (1988). "Are saving and investment co-integrated?", Economics Letters 27, pp. 31-34. Modigliani, F., (1975). “The life-cycle hypothesis of saving twenty years later.” In: Parkin, M. (Ed.), Contemporary Issues in Economics, Manchester University Press. 24 Schmitz, B. and Von Hagen, J. (2011). "Current account imbalances and financial integration in the euro area“, Journal of International Money and Finance, Vol. 30, pp. 1676-1695. Sinn, S. (1992). “Saving-investment correlations and capital mobility: On the evidence from annual data,” The Economic Journal, Vol. 102 (414), pp. 1162-1170. Summers, L. (1988). “Tax policy and international competitiveness,” in Jacob Frenkel, ed., International Aspects of Fiscal Policies, Chicago: University of Chicago Press, pp. 349-75. Taylor, A. and Williamson, J. (1994). “Capital flows to the new world as an intergenerational transfer,” Journal of Political Economy, Vol. 102, pp. 348-369. Tobin, J. (1983). “Comment,” European Economic Review, Vol. 21, pp. 153-6. World Data Bank, “Health, Nutrition and Population Statistics”. http://databank.worldbank. 25