This Selected Issues paper evaluates Hungary’s growth and current account performance by using a simple empirical model that provides benchmarks to measure GDP growth rates and current account deficits. The cross-country analysis suggests that in general, larger current account deficits are associated with faster income convergence. The model’s benchmark for Hungary suggests that its current account deficit has been larger than would be expected based on the income convergence process. The paper describes the motivation for, and specifics of, the modeling strategy, and the data used in the analysis.

Abstract

This Selected Issues paper evaluates Hungary’s growth and current account performance by using a simple empirical model that provides benchmarks to measure GDP growth rates and current account deficits. The cross-country analysis suggests that in general, larger current account deficits are associated with faster income convergence. The model’s benchmark for Hungary suggests that its current account deficit has been larger than would be expected based on the income convergence process. The paper describes the motivation for, and specifics of, the modeling strategy, and the data used in the analysis.

I. Growth and Current Account Performance: Results from a Cross-Country Model1

A. Introduction

1. The European Union’s (EU’s) new member states have experienced relatively high rates of real income growth, alongside large external deficits. Hungary, for example, has experienced growth of about 4 percent and current account deficits of about 8 percent of GDP over the past five years, while the EU as a whole grew by 2 percent over the same period, running a balanced external account. A relatively high rate of real GDP growth is to be expected in the new member states as they catch up to the real income levels of western Europe. Similarly, increased investment to support this higher growth, as well as strong consumption based on expectations of higher future income, should generate large current account deficits.

2. What is needed are benchmarks against which to measure these GDP growth rates and current account deficits. Standard models of growth theory suggest that countries that are further behind should be expected to grow faster than other countries, all other things equal. Lower-income countries would also be expected to run larger current account deficits, as the investment opportunities are greater and the expected increase in future income is larger. But is there a quantitative benchmark one can establish to assess whether GDP growth and the current account deficits in these countries are where they should be, based on the predictions of a model of income catch-up and intertemporal optimization? For example, is Lithuania’s current account deficit of 8.5 percent of GDP justifiable because it is enabling higher growth? Does convergence suggest a higher growth rate for Hungary than its current growth rate of 4 percent?

3. This chapter describes a simple empirical model that provides such benchmarks, and uses them to evaluate Hungary’s growth and current account performance. The model, described in more detail in a forthcoming working paper, was developed to explore whether larger current account deficits allow lower-income countries of the EU to converge more quickly to the income levels of advanced EU countries. It also recognizes that, while growth can be influenced by the running of external deficits, the latter can also be influenced by growth. It does this by combining a modified neoclassical growth regression with a current account regression similar to the one used by Blanchard and Giavazzi (2002). Predictions of the model are used as benchmarks for countries’ current performance.

4. The chapter has two main findings. First, the cross-country analysis suggests that, in general, larger current account deficits (capital inflows) are associated with faster income convergence. Second, the model’s benchmark for Hungary suggests that its current account deficit has been larger than would be expected based on the income convergence process. Hungary’s current account deficit of 9 percent of GDP was about 2 percentage points of GDP higher than the central prediction of the model. And growth, though just within the 95 percent confidence band, was about 1 percentage point less than the central prediction.

5. The rest of the chapter proceeds as follows. Section B describes the motivation for, and specifics of, the modeling strategy. The data used in the analysis are described in Section C, while Section D discusses the estimation results and analyses Hungary’s growth and current account performance. Section E concludes.

B. The Model

Stylized facts and theoretical motivation

6. There is a clear link between growth and the current account. Scatterplots of GDP growth against the current account show that higher growth is associated with larger current account deficits, both over short one-year horizons (Figure 1) and over longer periods (Figure 2). Simple as they are, these scatterplots are already informative, as they allow us to roughly approximate the growth rates that are associated with a given level of the current account. The best-fit line suggests that an increase in the current account deficit of 2 percentage points of GDP is associated with a 1–1.4 percentage point increase in GDP growth. Deviations from the best-fit line are also informative, as they show that some countries are growing rapidly at present without incurring significant external liabilities, while others could be expected to grow faster given the level of the current account (or conversely, that they should have a smaller current account deficit given their present growth rates).

Figure 1 and 2.
Figure 1 and 2.

Current Account Balances and GDP Growth in the EU

Citation: IMF Staff Country Reports 2005, 215; 10.5089/9781451818024.002.A001

Source: IMF World Economic Outlook database.

7. Theoretical models predict that external borrowing will be driven by future growth prospects. To the extent that low per capita incomes reflect relative scarcity of capital, we should find lower levels of per capita income associated with greater external borrowing to finance high-return investments. In addition, Blanchard and Giavazzi (2002) note that the growth rate of income can also affect the current account, as it not only is an indicator of future growth prospects, but also captures cyclical effects of output movements on the current account.

8. But current account deficits can also influence growth, in two ways. Most obviously, external borrowing removes constraints on investment and consumption. Investment is no longer constrained by domestic savings, and domestic expenditure (on both investment and consumption) is no longer constrained by domestic income.2 This suggests that current account deficits have a direct, positive effect on growth.

9. A second effect is suggested by open-economy versions of the neoclassical growth model, as elaborated, for example, by Barro and Sala-i-Martin (2000). If international capital were perfectly mobile and borrowers were unconstrained, convergence would be achieved instantly. A more realistic variant of the model incorporates constraints on international credit. Specifically, if some forms of capital (e.g., human capital) provide unacceptable security for loans, while other forms of physical capital are acceptable because creditors can take possession in the event of default, then foreign debt can be positive but cannot exceed the quantity of physical capital. In such a model, Barro and Sala-i-Martin write, “the opportunity to borrow on the world credit market … will turn out to affect the speed of convergence” (p. 105). Empirically, this suggests that the coefficient on per capita income in standard growth regressions may itself be influenced by the current account, as explained below.

Empirical specification

10. Two equations, one for the current account and one for growth, are estimated simultaneously. The first equation is a modified growth regression, which allows for variation in the speed of convergence:

Δyit=x+(α1t+α2cait1)(yit1yt1*)+α3cait1+α4Z1,it(1)

In equation (1), growth in per capita income in country i in year t, Δyit, depends on lagged income relative to the steady state income level, (yit1yt1*). The steady state income level, yt*, is allowed to change over time but is assumed to be the same for all countries in the sample. If poor countries grow faster as they converge to income levels of their richer neighbors, then the coefficient on lagged relative income should be negative. Here, this “speed of convergence” coefficient consists of two parts: a part that is influenced by the current account, α2cait-1, and an independent part, α1t. If current account deficits (cait< 0) accelerate income convergence, then coefficient α2 should be positive. The specification also allows for the possibility that the current account influences actual growth directly, and this effect is captured by the terms α3cait-1. In addition, growth is allowed to be influenced by standard neoclassical growth controls, such as the proportion of the population with secondary schooling, population growth, and the investment-to-GDP ratio, which are included in matrix Z1,it. Finally, growth is also influenced by the rate of technological progress, x. However, time variation in this and in other global cyclical factors suggest augmenting equation (1) by a year dummy, Dt which equals one in year t and zero otherwise. The equation can be rewritten as:

Δyit=α0t+(α1t+α2cait1)yit1+α3cait1+α4z1,it+(u1t+v1i+ε1it)(2)

where the term α0txα1tyt1*+α5Dt, and (u1t + ν1i + ε1it) represents a mean-zero composite error term.

11. Equation (3)—the second simultaneous equation of the model—describes the dynamics of the current account. The current account-to-GDP ratio in country i in year t, cait, depends on the current level of income, yit, on current growth, Δyit, and on the dependency ratio, Z2:

cait=β1t(yityt*)+β2tΔyit+β3z2,it(3)

The specification, which is identical to the one used by Blanchard and Giavazzi (2002), allows the effect of income per capita on the current account to vary over time. If the process of increasing financial integration in Europe has enabled poor countries to borrow more and rich countries to lend more, then one would expect the coefficient on relative income, β1t to increase over time. Current growth also enters the equation, both as a predictor of future income and in order to capture cyclical effects of output movements on the current account. The effect of growth on the current account is also allowed to vary over time. The dependency ratio captures intertemporal effects of demographic changes. Other things equal, a country with a relatively high dependency ratio is expected to save less.3 Finally, as in the growth equation, the equation has a common time effect, captured by the year dummy, Dt The equation can thus be rewritten as:

cait=β0t+β1tyit+β2tΔyit+β3z2,it+(u2t+v2i+ε2it)(4)

where the term β0tβ1tyt*+β4Dt, and (u2t + ν2i + ε2it) represents a mean-zero composite error term.

12. The estimation method used is three-stage least squares (3SLS), a standard technique for simultaneous estimation of simultaneous equations in the panel data context. This method, first proposed by Zellner and Theil (1962), permits the estimation a system of equations, in which some of the explanatory variables are endogenous. Here, both the current account and growth are explanatory variables and are endogenous. The three-stage least squares procedure uses an instrumental variable approach to produce consistent estimates and generalized least squares (GLS) to account for the correlation structure in the disturbances across the equations. For further discussion of the 3SLS approach to estimation, see, for instance, Greene (2003, pp. 405–407).

13. Once the parameters have been estimated, the model can be used to generate predicted values of the current account and growth, along with 95 percent confidence intervals. These benchmark values can be compared with actual outcomes to assess the performance of growth and the current account. The (in-sample) predicted values are obtained using the equations:

Δyit^=α0t^+(α1t^+α4^cait1)yit1+α2^cait1+α3^Z1,it(5)

and

cait^=β0t^+β1t^yit+β2t^Δyit+β3^Z2,it(6)

where the “^” superscripts denote estimates. For each period t, the matrix of prediction standard errors is denoted by st. The standard errors are computed using the following formula:

St=xtVxt(7)

where xt is the matrix of right-hand-side variables up to and including period t, and V is the estimated variance covariance matrix of the parameter estimates. Standard error bands around the predicted values can then be computed using a band of ±1.96 times the prediction standard errors.

C. Data

14. The sample covers the period 1975–2004 and includes the 25 countries of the EU (EU-25). For the new member states, the sample starts in 1995 to avoid the structural breaks associated with the shift to a market economy. Following Blanchard and Giavazzi (2002), income per capita is constructed using real GDP per capita at purchasing power parity (PPP) from the Penn World Tables, Version 6.1 (CICUP, 2002), extended to 2004 using per capita real GDP growth rates from the IMF’s World Economic Outlook database.4 The current account is measured as a ratio to GDP and is taken from the IMF’s International Financial Statistics.5 The dependency ratio is constructed using population and labor force data from the Penn World Tables.

15. Data on the neoclassical growth controls (schooling, investment shares, and population growth) come from the growth data set compiled by Bosworth and Collins (2003). The Bosworth-Collins data set takes several standard data sets and extends them to cover 84 countries over the 40-year period to 2000. Data on education attainment are from Barro and Lee (1993 and 2000). Capital stock and investment data are originally from a World Bank study by Nehru and Dhareshwar (1993), modified and extended to 2000 by Bosworth and Collins. And data on population growth are from the Penn World Tables. The Bosworth-Collins country sample, however, does not cover any of the countries from central and eastern Europe, so data for these countries was taken from Doyle, Kuijs and Jiang (2001) and cover the period 1996–2004. Data for many of the control variables are, unfortunately, available only up to 2000; schooling data are available only up to 1999. So the model with the full set of controls is estimated over a restricted sample (1975–99). To establish benchmarks up to 2004, the model is estimated without the additional controls for the full sample (1975–2004).

D. Results

Evidence of convergence and of increasing financial integration

16. A plot of income per capita against subsequent growth over 1975–2004 suggests absolute convergence in Europe. Relatively poor countries with low income per capita are growing relatively quickly, as they catch up with the higher income levels of their richer neighbors (Figure 3). The best-fit line is significant, with a p-value of less than 1 percent.

Figure 3.
Figure 3.

Absolute Convergence in Europe, 1975-2004

Citation: IMF Staff Country Reports 2005, 215; 10.5089/9781451818024.002.A001

17. This finding is supported by the regression results. We begin by plotting the set of estimated coefficients α1t and β1t against time. As these are time-varying, it is easiest to view them in a chart rather than in a table. Results are reported for both the EU-25 countries and a subset, the EU-15, namely, the EU-25 minus the 2004 accession countries. Figure 4 shows estimates of the speed of convergence parameter, α1t. For the EU-25, the coefficient is negative most of the time, in line with the hypothesis of absolute income convergence. The estimate, however, only becomes significant at the 5 percent level in the five years following 1999, as well as in 1990 and 1997. For the EU-15, α1t is again negative most of the time but insignificant at the five percent level in all but three periods. Although this suggests that convergence is faster in the new accession countries, it also reflects the fact that much of the convergence process has been completed for the EU-15, so that the convergence parameter for this subset of countries is less precisely estimated.

Figure 4.
Figure 4.

Estimates of the Speed-of-Convergence Parameter, 1975-2004 1/

Citation: IMF Staff Country Reports 2005, 215; 10.5089/9781451818024.002.A001

1/ The dashed lines show the upper and lower bounds of the 95 percent confidence interval.

18. The regressions also find that relative income significantly affects the current account. Figure 5 shows estimates of β1t, the effect of relative income on the current account. For both sets of countries, the coefficient is nearly always positive, in line with the intuition that relatively poor countries run larger current account deficits. For both sets of countries, the effect also becomes significant (at the five percent level) only in the late 1990s. The increased dependence of the current account on income is in line with the Blanchard and Giavazzi (2002) finding that greater financial integration in Europe in the 1990s allowed relatively poor countries to borrow more from their richer neighbors. However, there is a more obvious increasing trend in the parameter for the EU-15, and the estimated coefficients reach a higher value by the sample’s end. For example, the estimated coefficient of 0.2 in 2000 for the EU-15 sample implies, that, other things equal, for a country with an income per capita that is 40 percent below the EU steady state level (roughly the case of Hungary, if one proxies the steady state with average EU income), the current account-to-GDP ratio should be about 8 percentage points lower than the EU steady state. However, the estimated coefficient of about 0.1 for the EU-25 implies, that, other things equal, a country with an income per capita about 40 percent below the EU steady state should have a current account-to-GDP ratio only about 4 percentage points lower than the EU steady state. A plausible explanation for this greater EU-15 effect is that the EU-15 has been more integrated financially than the EU-25 over the sample period.

Figure 5.
Figure 5.

Impact of Income Per Capita on Current Account, 1975-2004 1/

Citation: IMF Staff Country Reports 2005, 215; 10.5089/9781451818024.002.A001

1/ The dashed lines show the upper and lower bounds of the 95 percent confidence interval.

19. Estimates of the remaining model parameters indicate that current account deficits raise the speed of convergence. Table 1 displays the estimates of a selection of the remaining model parameters. The estimate of parameter α2—which measures the impact of current accounts on the speed of convergence—is positive and significant at the 1 percent level, in line with the hypothesis discussed in paragraph 10 that current account deficits increase the speed of convergence. A 1 percentage point increase in the current account deficit raises the speed of convergence by 0.26 percentage point per year. In addition, the estimate of α3 is negative and significant, suggesting that the current account also affects growth directly. Overall, the model’s fit is good, with an R2 of 51 and 58 percent for the growth and current account equations, respectively.

Table 1.

Estimates of Current Account Effect on Growth

Estimation of equations:

Δyit = α0t + (α1t + α2cait-1)yit-1 + α3cait-1 + α4Z1,it + (u1t + ν1i + ε1it) cait = β0t + β1tyit + β2tΔyit + β3Z2,it + (u2t + ν2i + ε2it)

article image

20. These results are robust to changes in model specification. First, the robustness of the results is tested to exclude the new member states, and the results appear in column 2. For this subset of countries, the EU-15, the signs of the estimates of α2 and α3 remain unchanged, although the magnitude and significance of these coefficients increase slightly. Second, the robustness to including country dummies is tested and the results appear in column 3. The inclusion of country dummies strengthens both the direct effect of the current account on growth, and its indirect effect via the speed of convergence. Finally, we find that the results are robust to including all the standard neoclassical controls. As noted above, the sample using all the controls is shorter, as schooling data is available only to 1999, and the other neoclassical controls are available only to 2000.

21. The baseline estimates of the convergence speeds have implications for the number of years needed to converge to the steady state. A useful measure of this convergence time is the half-life, which measures the number of years needed to halve the gap between the current income level and the steady state. To estimate the half-life, the estimates of parameters α1t and α2 are used in Equation (8); the last available value of α1t, that for 2004, is used. To illustrate how current account deficits can accelerate the process of convergence, the half-lives are computed for a number of possible current account deficits. The half-life estimates appear in Table 2, along with estimates of the three-quarter lives, computed using Equation 9. As Table 2 suggests, raising the current account deficit from zero to 5 percent of GDP reduces the three-quarter-life from 31 to 24 years.

Table 2.

Estimated Convergence Half and Three-Quarter Lives

(In years)

article image
halflife=|In(1/2)In(α1t+α2cait1)|(8)
threequarterlife=|In(1/4)In(α1t+α2cait1)|(9)

Model predictions for current account and growth

22. The model’s predicted values can be used as benchmarks for evaluating how much of growth and the current account can be explained using catch-up considerations. Figures 6 through 8 show the (in-sample) predictions of the current account and growth, along with 95 percent confidence intervals. Because the model is based on the premise of intertemporal optimization, subject to the intertemporal budget constraint, the current account predictions are consistent with sustainability.

Figure 6.
Figure 6.

Hungary: Growth and Current Account Predictions, 1997-2004 1/

Citation: IMF Staff Country Reports 2005, 215; 10.5089/9781451818024.002.A001

1/ The dashed lines show the upper and lower bounds of the 95 percent confidence interval.

23. The model predicts higher growth and a smaller current account deficit for Hungary than is presently the case (Figure 6). As a result of the crisis in 1994–95, growth remained below its catch-up potential growth rate in 1996 and 1997. But from 1998 to 2000, Hungary grew at or above the model’s predicted growth rate, at rates of 4–5 percent. Since 2001, growth has slowed, both in absolute terms and relative to the model prediction. By 2004, actual growth of 4 percent was approximately 1 percentage point below the central prediction of the model. The current account deficit (Figure 6, right panel) was smaller than predicted in 1996–97, following the macroeconomic adjustment to the crisis. But since then it has widened substantially. The 2004 current account deficit of about 9 percent of GDP is about 2 percentage points higher than what the model predicts.

24. Lithuania is an example of a country where current performance seems to be more closely in line with the model’s predictions (Figure 7).6 Lithuania’s recession in 1999 is associated with the Russian financial crisis. But, apart from this contraction, Lithuania’s growth performance has roughly been in line with the model’s predicted growth rate; its current growth rate of about 7 percent coincides with the model’s central prediction. The current account deficit has also been in line with the model’s prediction, and in recent years has been smaller than what the model suggests.

Figure 7.
Figure 7.

Lithuania: Growth and Current Account Predictions, 1996-2004 1/

Citation: IMF Staff Country Reports 2005, 215; 10.5089/9781451818024.002.A001

1/ The dashed lines show the upper and lower bounds of the 95 percent confidence interval.

25. As Hungary’s income per capita increases, it is expected to run smaller current account deficits and to experience lower growth. Taking as an example a relatively rich country, such as France, the model predicts a current account position near to balance or in surplus, with growth in the 1–2 percent range. As Figure 8 suggests, this prediction fits the experience of France well.

Figure 8.
Figure 8.

Growth and Current Account Predictions for France, 1985-2004 1/

Citation: IMF Staff Country Reports 2005, 215; 10.5089/9781451818024.002.A001

1/ The dashed lines show the upper and lower bounds of the 95 percent confidence interval.

E. Conclusions

26. This chapter describes a simple empirical model that provides benchmarks to evaluate Hungary’s current growth and current account performance. The model, described in more detail in a forthcoming working paper, was developed to explore whether larger current account deficits allow lower-income countries of the EU to converge more quickly to the income levels of advanced EU countries. It also recognizes that, while growth can be influenced by the running of external deficits, the latter can also be influenced by growth. The model does this by combining a modified neoclassical growth regression with a current account regression similar to the one used by Blanchard and Giavazzi (2002). Predictions of the model are used as benchmarks for countries’ current performance.

27. The chapter has two main findings. First, the cross-country analysis suggests that, in general, larger current account deficits (capital inflows) are associated with faster income convergence. Second, the model’s benchmark for Hungary suggests that its current account deficit has been larger than would be expected based on the income convergence process. Hungary’s current account deficit of 9 percent of GDP was about 2½ percent of GDP higher than the central prediction of the model. Though just within the 95 percent confidence band, growth was about 1 percentage point less than the central prediction.

References

  • Barro, Robert J., and Jong-Wha Lee, 1993, “International Comparisons of Educational Attainment,Journal of Monetary Economics Vol. 32 (December), pp. 363 –94.

    • Search Google Scholar
    • Export Citation
  • ———, 2000, “International Data on Educational Attainment: Updates and Implications.NBER Working Paper No. 7911 (Cambridge, Massachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation
  • ———, and Xavier Sala-i-Martin, 2000, Economic Growth (Cambridge, Massachusetts: MIT Press).

  • Blanchard, Olivier, and Francesco Giavazzi, 2002, “Current Account Deficits in the Euro Area: The End of the Feldstein-Horioka Puzzle?Brookings Papers on Economic Activity: 2, Brookings Institution, pp. 147 –209.

    • Search Google Scholar
    • Export Citation
  • Bosworth, Barry, and Susan M. Collins, 2003, “The Empirics of Growth: An Update,Brookings Papers on Economic Activity: 2, Brookings Institution, pp. 113 –206.

    • Search Google Scholar
    • Export Citation
  • CICUP, 2002, Penn World Tables, Version 6.1 (Philadelphia, Pennsylvania: Center for International Comparisons at the University of Pennsylvania).

    • Search Google Scholar
    • Export Citation
  • Doyle, Peter, Louis Kuijs, and Guorong Jiang, 2001, “Real Convergence to EU Income Levels: Central Europe from 1990 to the Long Term”, IMF Working Paper 01/146 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Greene, William H., 2003, Econometric Analysis, 5th ed. (Upper Saddle River, New Jersey: Prentice-Hall).

  • Nehru, Vikram, and Ashok Dhareshwar, 1993, “A New Database on Physical Capital Stock: Sources, Methodology and Results,” Revista de Análisis EconÓmico, Vol. 8 (June), pp. 37 –59.

    • Search Google Scholar
    • Export Citation
  • Zellner, Arnold, and Henri Theil, 1962, “Three-Stage Least Squares: Simultaneous Estimation of Simultaneous Equations,Econometrica, Vol. 30 (January), pp. 54 –78.

    • Search Google Scholar
    • Export Citation
1

Prepared by Abdul Abiad and Daniel Leigh.

2

A second mechanism through which openness can affect growth is technological spillover effects. As Barro and Sala-i-Martin (2000) note, trade, and especially investment linkages, may promote greater technological diffusion from leader countries to follower countries.

3

Following Blanchard and Giavazzi (2002), the dependency ratio is constructed as the ratio of population to the labor force.

4

For certain countries, such as Ireland, gross national income (GNI) per capita may be a better proxy for income per capita. However, due to the unavailability of real PPP GNI data, GDP data are used instead.

5

Data for 2004 are based on the IMF’s World Economic Outlook database. Luxembourg is excluded from the sample due to its highly idiosyncratic behavior, with reported current account surpluses that are consistently in the range of 10–15 percent of GDP. The other idiosyncratic cases are (i) Greece in 1975, with a very volatile current account position in the aftermath of the collapse of the military regime—the current account deficit reached 30 percent of GDP; (ii) Ireland in 1978–80, with large unsustainable current account deficits following the second oil price shock and large fiscal deficits in excess of 12 percent of GDP; (iii) Portugal in 1975–82, during which the current account and growth were very volatile in the aftermath of revolution, the loss of colonies, the second oil price shock, and the loss of control of fiscal policy. These three cases are dealt with by introducing country dummies interacted with the specific years in question.

6

This finding for Lithuania is consistent with the results obtained using a simplified version of the model in a selected issues paper for the Lithuania 2004 Article IV Consultations.

Hungary: Selected Issues
Author: International Monetary Fund