External Debt Evolution when Global Financial Markets are Incomplete

External Debt Evolution when Global Financial Markets are Incomplete Alexis Anagnostopoulos1 and Gregorios Siourounis2 THIS VERSION: MARCH 2015 Abstract The aim of this paper is to show how a country’s output growth, employment, consumption, investment, interest rates and external debt behave in the presence of international asset market incompleteness. We show that when a country issues a full set of contingent claims, external debt is a stationary, mean reverting process, leading to the conclusion that adverse shocks do not have permanent long run effects. Examining a panel of 60 developed and developing countries from 1970 until 2008 and assessing stationarity with panel unit root tests reveals that external debt is a nonstationary process whereas the growth rate of output is a stationary process regardless of income or region stratum. We then show that this behavior is best accounted by a model of incomplete markets, where agents issue only one-period risk free assets. This is of paramount importance since it shows that countries that experience transitory adverse shocks can accumulate large amounts of external debt over long periods of time. JEL CLASSIFICATION: F41, F32, F21 KEY WORDS: complete vs. incomplete international financial markets, net foreign assets accumulation, external debt, output growth, panel unit root tests, and impulse response functions. Acknowledgements: We would like to thank Andrew Scott, Albert Marcet, Demetrios Christopoulos and Demetrios Thomakos. All erros remain ours.

1

Department of Economics, University of Stony Brook, New York - USA. Email: [email protected] 2

Corresponding Author: Department of Economic and Regional Studies, Panteion University of Social and Political Sciences, Athens- Greece. E-mail: [email protected].

1

1 Introduction The present paper studies the evidence that international financial markets incompleteness is an important feature that can explain the evolution of a country’s external debt and output growth in the presence of productivity shocks. Recent financial turmoil in the heart of Europe, the Eurozone, shows a great need to understand how financial linkages between countries behave in the presence of adverse shocks as well as the extent of policy intervention needed. In order to address that, we use a two country general equilibrium model that allows interaction between the financial markets of the two countries. We then characterize the response process of net foreign assets accumulation or external debt3 and output growth under both complete and incomplete markets and use this analysis as a road map to test empirically the behavior of these variables in a panel of 60 countries. We show that the behavior of external debt and output growth can be rationalized by the incomplete markets model. Given the recent debt crisis in three Eurozone countries, namely Greece, Ireland and Portugal and the attention that has been given in characterizing the theoretical properties of external debt 4 and the vast recent empirical evidence on the behavior of a country’s external asset position,5 understanding the influencing factors and implications of international financial markets is of particular interest to policy makers. Accumulation of foreign assets and/or liabilities represent an important global linkage. Identifying the sources of the fluctuations in world asset trade under complete and incomplete markets can contribute to understanding its sustainability and likely future trends as well as its effects on local economies in the presence of adverse productivity shocks. Market incompleteness does not lead to immediate answers as regards the behavior of foreign assets accumulation and output growth since the Euler equation system for complete and incomplete markets is very similar. Recent literature that introduces incomplete markets into asset pricing models suggest that this does not make a considerable difference in the behavior of other variables in the model like employment, consumption, investment etc. 6 Also a number of authors have suggested that the choice of the maturity of assets issued within a country, the choice of issuance of foreign currency denominated debt and a wide variety of options, futures and indexed debt can serve as a ground to replicate a complete markets model  to a great extent.7 3

In this paper we use external debt as equivalent to net foreign assets accumulation. See the work of Bekaert and Harvey (2000), Hau and Rey (2002). 5 In terms of empirical work on international financial integration, see for example, Henry (2000), Levine et al. (2000), Edison et al (2002), Edison and Warnock (2002) and Gourinchas and Jeanne (2003), that amongst others, have looked at the impact of international financial integration on various indicators. Obstfeld and Taylor (2002) provide a wide-ranging historical overview, including analysis of the longrun changes in gross asset trade. Lane and Milesi-Ferretti (2001a, 2003) give a thorough examination of the international asset position for a large set of developed countries. Portes and Rey (2002) document the effects of information asymmetries on the determinants of cross-border equity flows. 6 See, for example, Backus, Kehoe and Kydland (1992) and Marcet and Singleton (1999). See Athanasoulis and van Wincoop (2000) for a discussion of the literature and an attempt to account for these differences. 7 See, for example, Lloyd-Ellis and Zhu (2000) and Buera and Nicolini (2000). 4

2

Our work is most closely related to that of Marcet and Scott (2009), who show that incompleteness in the domestic bond markets provides a consistent explanation of the behavior of government debt. We apply a similar idea to external debt as opposed to government debt and we provide some additional empirical tests aimed at examining the stationarity of external debt. Our analysis, of the responsiveness of external debt and output growth in the presence of productivity shocks under complete and incomplete markets in an open economy general equilibrium model, enables us to construct a straightforward test of international financial markets completeness: under complete markets stationary shocks to productivity translate to external debt and output growth processes that are also stationary. In contrast, under incomplete markets stationary productivity shocks lead to a non-stationary response of external debt while the response of output growth is still stationary. We first calibrate the two models and provide impulse responses of all variables in the system and then simulate synthetic data of the same dimensions as our sample. To empirically evaluate the theoretical conjectures we use a panel of 60 developed and developing countries around the globe and test for stationarity of external debt and output growth. We also explore the properties of the impulse responses of employment, domestic consumption and investment. We find that regardless of income and region stratum, according to World Bank Classifications, external debt’s response to a productivity shock is a non-stationary process whereas output growth response to the same shock is a stationary process. Thus, the data seems to be consistent with the notion that global financial markets are incomplete. We believe that our findings have important implications for the current research agenda on financial globalization as well as the recent reaction of policy makers in the Eurozone, the ECB and the IMF regarding the rescue packages for Ireland, Portugal and Greece. In one sense, the intervention of these institutions and provision of funding at below-market interest rates represents an attempt to replicate the complete markets insurance paradigm: in response to an adverse shock, an insurance contract would stipulate payments to the country experiencing the shock. In turn, this can have the desirable effect of limiting the explosion of debt, even in our stylized model which abstracts from the possibility of default. Our findings have also implications for the recent debate on reforming the “international financial architecture”. A commonly held view is that capital flows to less developed economies are excessively low, and that the international financial architecture should be designed so as to increase the access of less developed countries to international financial markets. 8 This also relates to the recent experience of peripheral Eurozone countries which are in dire need to capital inflows. Our paper suggests that international financial markets incompleteness is an important inefficiency that is likely to go along with foreign capital scarcity. Countries have much more to gain from improving their domestic financial structures so as to attract foreign

8

However, Gourinchas and Jeanne (2003) document that the welfare gain from switching from financial autarky to perfect capital mobility are relatively small even for countries that receive a lot of capital inflows.

3

capital and boost growth given that domestic savings are not enough to propel new investment, consumption and employment. The rest of the paper is organized as follows. Section 2 presents the benchmark model that assumes perfect international financial integration and discusses its implications. Section 3 describes the data set. Section 4 presents the econometric methods used and the empirical behavior of GDP growth, external debt, employment, consumption and investment. Section 5 attempts to reconcile theory and data by considering a modified version of the benchmark model where financial integration is less than complete. Finally, Section 6 concludes and suggests possible future directions.

2 2.1

Complete markets Model

We use a two country general equilibrium model, as in Backus, Kehoe and Kydland (1992), as our main theoretical guide. This is a simple simultaneous choice dynamic model where representative agents in each country i choose consumption, labor, investment in physical capital and financial assets, as an optimal response to exogenous shocks to productivity. To be more precise, production in each country is achieved through the combination of two inputs, capital and labor. Capital has to be invested one period before it is used in production whereas labor can be decided contemporaneously with the resolution of uncertainty. The combination of inputs (the technology) takes place in the usual Cobb-Douglas fashion (1) F(ki, t − 1, lit) = kαi, t − 1l1 − αit where we have assumed constant returns to scale in production and focus on a representative firm. Actual output produced is subject to uncertainty relating to the level of productivity, zit of the inputs at any given period. The effect of productivity is multiplicative so that production at any given period t in country i is given by (2) yit = zitF(ki, t − 1, lit) Total production is exactly equal to total income for the representative agent (3) total incomet = ritki, t − 1 + witlit and profit maximization gives the usual competitive factor prices (4) wit = zitFl(ki, t − 1, lit) rit = zitFk(ki, t − 1, lit) On the demand side, the representative consumer can use her income for consumption cit, investment in the domestic firm iit or savings/borrowing in the form of foreign assets. Note that this is a single good framework meaning that the same good is produced, consumed or invested. Investment is transformed to productive capital on a one-to-one basis and a portion δ of capital depreciates every period so that (5) kit = (1 − δ)ki, t − 1 + iit We assume initially that international financial markets are complete so that the two countries can hedge against any idiosyncratic risk.  This is equivalent to saying that the representative agent in Greece, Ireland, Portugal or any other country in the Eurozone, has access to an array of contingent claims that can completely insure him against any idiosyncratic shock by providing instant capital inflow to compensate for any income 4

decrease caused by an adverse productivity shock.9  Thus, the budget constraint for each country reads (6) cit + iit + ⌠s ∈ Spt(s)b1t(s)ds = yit + b1, t − 1(st) for all t, st ∈ S Here we specify complete markets by assuming the existence of a full set of contingent claims b1t(s). At any state st, agents buy a portfolio of securities indexed by next period’s state s. Let b1t(s) denote the quantity of each of those contingent claims bought at price pt(s). Each contingent claim pays 1 if st + 1 = s and 0 otherwise, as usual. Finally, we have two market clearing conditions, for goods (7) c1t + c2t + i1t + i2t = y1t + y2t and for assets b1t(s) + b2t(s) = 0 Because of the lack of frictions and assumption of complete markets, allocations in this environment will be first-best. As a result, we can obtain these allocations by solving the corresponding social planner’s problem, that is max{{cit, lit, kit, bit}2i = 1}∞t = 0E0∑∞ βt{Ω1u(c1t, l1t) + Ω2u(c2t, l2t)} 𝑡=0 subject to c1t + c2t + i1t + i2t = y1t + y2t , kit = (1 − δ)ki, t − 1 + iit , yit = zitF(ki, t − 1, lit) We specify welfare weights Ω1 = Ω2 = 1 since the two countries are ex ante identical and only differ in the ex post realization of their exogenous shocks. The algorithm used is a Parameterized Expectations Algorithm where expectations are assumed to be loglinear in the state variables st = (k1, t − 1, k2, t − 1, z1t, z2t).10

2.2 External Debt and Net Exports In a closed economy framework we would assert that total domestic production yit must equal total domestic absorption cit + iit. But we have assumed that there is at least some level of international trading in financial assets and fully unimpeded international trade in goods. From the point of view of the good(s) market, the difference between domestic production and absorption is simply the country’s net exports. Alternatively, one could think of this in terms of the usual accounting definitions of current and capital accounts. The current account is simply a country’s net transactions in goods and services. This would include the balance of trade (in goods and services) plus net income on foreign assets as well as net cash transfers. This model follows the vast majority of international models in assuming away the presence of transfers such as humanitarian aid. In this single good framework, the balance of trade is summarized by our net exports variable (8) nxit = yit − cit − iit The last part of the current account is investment income. This is straightforward to understand if we assume that the only financial transaction that can be carried out

9

Within countries, financial markets are complete by assumption, which allows us to focus on a representative agent. 10 Technical appendix available upon request.

5

between the two countries is borrowing and lending in one period discount bonds. When a country has negative nxit, i.e. when it is a net importer of goods and services, it needs to find a way to finance these additional products that it is receiving. This can be done by borrowing the equivalent amount from the other country promising to repay next period the principal plus interest. So it would borrow bit now and have to repay (1 + rt)bit at period t + 1. As a result, net exports at any period t have to equal the net amount of borrowing at that period, that is the new bonds sold at t, bit, less the repayment of debt from the previous period together with any interest (1 + rt − 1)bi, t − 1. So the last part of the current account is simply rt − 1bi, t − 1, the investment income which in the above case would be negative. Now bit does not have to be narrowly defined as borrowing, it is any type of claim resident(s) in one country have on assets in the other (with the only simplification here being that those claims have to be redeemed and, if needed, bought again at every period). Therefore, the change in bit is simply the net transactions in assets between the two countries at t, the capital account. The simple identity which ensures the current and capital accounts sum to 0 is thus nxit + rtbit − (bit − bi, t − 1) = 0 and this is also the representative consumer’s budget constraint (9) cit + iit + bit = yit + (1 + rt − 1)bi, t − 1 Given that the exogenous productivity shock will be specified to have a continuous support, the above case where there is essentially one financial asset with fixed return is one of extreme international financial market segmentation. The complete markets counterpart of this model is one where there exist a full set of Arrow-Debreu securities b1t(s), priced at pt(s) and paying off 1 in case st + 1 = s and 0 otherwise. The budget constraint would be in that case (10) cit + iit + ⌠s ∈ Spt(s)b1t(s)ds = yit + b1, t − 1(st) for all t, st ∈ S Using the social planner solution, equilibrium allocations are obtained abstracting from prices and asset holdings. Prices are then recovered from the Euler equations of the competitive equilibrium problem (11) pt =

𝛽𝐸𝑡𝑢𝑐(𝑐𝑖, 𝑡 + 1, 𝑙𝑖, 𝑡 + 1) 𝑢𝑐(𝑐𝑖, 𝑡, 𝑙𝑖, 𝑡)

The level of international borrowing, evaluated at market prices is then recovered using the intertemporal budget constraint 𝑗 (12) b1, t − 1(st) = Et∑∞ 𝑗=0 𝛽

𝑢𝑐(𝑐𝑖, 𝑡 + 𝑗, 𝑙𝑖, 𝑡 + 𝑗) 𝑢𝑐(𝑐𝑖, 𝑡, 𝑙𝑖, 𝑡)

( − nxi, t + j)

which is derived by substituting the Euler equation in the period-by-period budget constraint and rolling forward. Imposition of the transversality condition ensures that the limit term is 0. We can rewrite this as 𝑗 (13) b1, t − 1(st) = − nxi, t + Et∑∞ 𝑗=1 𝛽

𝑢𝑐(𝑐𝑖, 𝑡 + 𝑗, 𝑙𝑖, 𝑡 + 𝑗) 𝑢𝑐(𝑐𝑖, 𝑡, 𝑙𝑖, 𝑡)

( − nxi, t + j)

Comparing with the budget constraint above, it is straightforward to see that external debt ∫s ∈ Spt(s)b1t(s)ds evaluated at market prices is equal to 𝑗 (14) Et∑∞ 𝑗=1 𝛽

𝑢𝑐(𝑐𝑖, 𝑡 + 𝑗, 𝑙𝑖, 𝑡 + 𝑗) 𝑢𝑐(𝑐𝑖, 𝑡, 𝑙𝑖, 𝑡)

6

( − nxi, t + j)

This means that external debt at any time t must equal the expectation of future discounted sums of net imports. This expression is used to obtain net foreign assets (NFA) given a simulation of allocations as a function of the state variables at t. Before solving for equilibrium laws of motion of the variables of interest, we can make one point about the role of international financial markets by just looking at the budget constraint (10). Consider a state of the world s where an unexpected negative productivity shock occurs, reducing current production (income). Under complete markets, complete insurance has been achieved by holding the optimal portfolio of securities and thus the payoff bS(st − 1, st = s) received from the contingent claim that pays at state s should exactly offset the fall in production .11 In particular, there is no reason why we should expect the value of the debt portfolio (net foreign liabilities, represented by the integral on the left side of (2)) to change. An immediate implication is that, when a full set of contingent claims is available for trading, the net asset position of the country is not affected by shocks. Instead, shocks are absorbed by compensating payments from contingent claims which provide the required insurance. More generally, external debt might respond to shocks but in a stationary manner, as shown below.

2.3

Calibration and Simulation Results

To address the question of the persistence of external debt relative to that of GDP, we simulate the theoretical economy and use panel data unit root tests (explained in section 3) on the two simulated series. The use of the whole panel of our dataset in our empirical estimates presents some difficulties with respect to consistency of simulated and actual data. Given our non-linear numerical procedure for the solution of the theoretical model, it is not feasible to specify the model as one comprising of 60 countries. As a first pass, we use a two country model calibrated to match the long run properties of the US and an average of the rest of the OECD countries, where we have imposed symmetry between the two parties. 12

3

Data

The data we are using for net foreign asset accumulation (defined here as external debt) is taken from Lane and Milesi-Ferretti (2001) extended to include data for later dates. The data spans from 1970 until 2008 for 60 countries around the globe. The measure we are using is an estimate of the net external asset position based on adjusted cumulative current account divided by real per capita GDP. Per capita growth rate is calculated as the log difference of real per capita GDP taken from the World Bank. 11

This holds exactly in the absence of aggregate uncertainty. In the more general case considered subsequently, asset holdings will only insure against idiosyncratic risk. 12 We have also used time series of a single country and compare it with simulated time series of the same size. Results are the same but the power of unit root tests is much smaller and we find it more difficult to reject the null of a unit root in NFA. By using panel data tests we are making it harder for our model to pass the test.

7

Table 1 gives the summary statistics for the two variables as well as a frequency graph and the breakdown of the data according to region and income strata.13

4

Econometrics

4.1 Testing the model The theoretical implication of the complete markets model is that growth rate of output and external debt is stationary processes. This means that any external productivity shock is fully compensated so that any disruptions noticed in the domestic consumption and investment paths are reversed quickly followed by full income level recovery. To empirically evaluate the above theoretical conjecture we need to test for stationarity for each country in our sample. However, it is well known that single equation unit root tests, when applied in small macroeconomic data sets, have very low power in rejecting the null hypothesis of a unit root. In addition, Taylor (2001) shows that in order to achieve a power of an ADF test of 0.5, you need at least 228 observations, when most available yearly macroeconomic data sets do not exceed 40! Therefore, exploiting cross-sectional information may increase the power of unit-root tests. A variety of procedures for the analysis of unit roots in a panel context have been developed. The emphasis in this development is the attempt to combine information from the time series dimension with that obtained from the cross-sectional dimension, in the hope that inference about the existence of unit roots and cointegration can be made more straightforward and precise by taking account of the latter.14 Given that many interesting relations involve relatively short time--series dimensions, and the well-known low power of conventional unit root tests when applied to a single time series,15 there may be considerable potential for tests that can be employed in an environment where the time series may be of limited length, but very similar data may be available across a cross--section of countries. However, a variety of issues arise when panel data are employed in testing for unit roots. Some of the tests proposed require a balanced panel (no missing data for any i nor t) whereas others allow for an unbalanced panel setting. In a panel context, that is, with a set of time series one may form the null hypothesis as a generalization of the standard Dickey-Fuller test, in that all series in the panel are assumed to exhibit nonstationary behavior. This null might be rejected if a fraction of the series in the panel appear to be stationary. Each of the different test procedures proposed in the literature have their advantages and disadvantages. We thus, employ three different test procedures that are briefly described below to try to circumvent some of them and increase the robustness of our results. Sarno and Taylor developed a multivariate analogue to the ADF test (MADF hereafter, 1998, 1998) as an extension of a test developed by Abuaf and Jorion several

13

We use the World Bank income classification. See http://www.worldbank.org. See also Baltagi (2001), pp.235. 15 See also Taylor(2001) for a relevant discussion. 14

8

years earlier. In this test, a single autoregressive parameter is estimated over a panel, by applying Zellner’s SUR estimator in the N equations corresponding to the units of the panel. Since SUR can only be employed where T>>N the test may only be used where this condition is satisfied. Thus, it is not a suitable test for small-T large-N panels, such as those often employed in a cross--country context. Each equation is specified as a kth-order autoregression, and the test involves testing the hypothesis that the sum of the coefficients on the autoregressive polynomial is unity. The null hypothesis states that this condition is satisfied over the N equations. Thus, this null will be violated if even one of the series in the panel is stationary. A rejection should thus not be taken to indicate that each of the series in the panel is stationary, but rather an indication that the condition that all series are I(1) does not receive empirical support. Critical values are nonstandard, and have been generated by simulation of a response surface. Table 2 reports the statistics of this test only for the sub-samples based on regional and income stratums since only then, the condition of T>N is met. It can be seen that for growth rates the test rejects the null hypothesis of a unit root in all sub-samples, except for Europe and Central Asia. The same test fails to reject the null hypothesis of a unit root in external debt in Latin American and Caribean, South East Asian and low income countries. However, as noted above this only indicates that, in at least one of the countries in each of these sub-samples, external debt is a stationary process. Since the test is not conclusive and given the fact that it does not allow for any individual and/or time effects for each of the individual series we extend our analysis by employing two additional tests. One of the first unit root tests to be developed for panel data is that of Levin and Lin as published in Levin, Lin and Chu (LLC hereafter, 2002). Their test is based on analysis of the equation: (14) Δyi, t = ai + δi, t + θt + ρiyi, t − 1 + ξi, t, fori = 1, 2, ...N and t = 1, 2, ...T This model allows for two-way fixed effects (a and δ) as well as unit-specific time trends. The unit-specific fixed effects are an important source of heterogeneity, since the coefficient of the lagged dependent variable is restricted to be homogeneous across all units of the panel. The test involves the null hypothesis H0:ρi = 0 for all i against the alternative H1:ρi < 0 for all i with auxiliary assumptions, under the null, also being required about the coefficients relating to the deterministic components. Like most of the unit root tests in the literature, LLC assumes that the individual processes are cross-sectional independent. Given this assumption, they derive conditions under which the pooled OLS estimate of ρ will have a standard normal distribution under the null hypothesis. Their work focuses on the asymptotic distributions of this pooled panel estimate of ρ under different assumptions on the existence of fixed effects and homogeneous time trends. The LLC test may be viewed as a pooled Dickey--Fuller (or ADF) test, potentially with differing lag lengths across the units of the panel. Unlike the MADF test, it is applicable to small-T large-N panels. This allows us to test the stationarity of growth rates and external debt for the full panel as well as the different sub-samples. Table 3 reports the test statistic of the LLC test. Regardless of the stratum we consider, the test is conclusive: growth rates are stationary whereas external debt is 9

non-stationary. We next use a test that allows for different individual mean reverting dynamics. The Im--Pesaran--Shin test (IPS hereafter, 1997) extends the LLC framework to allow for heterogeneity in the value of ρi under the alternative hypothesis. Given the same equation as before: (15) Δyi, t = ai + δi, t + θt + ρiyi, t − 1 + ξi, t, fori = 1, 2, ...Nandt = 1, 2, ...T The null and alternative hypotheses are defined as: H0 : ρi = 0 for every i H1 : ρi < 0 for i = 1, 2, ...N1 and ρi = 0 for i = N1 + 1, N1 + 2, ...N Thus under the null hypothesis, all series in the panel are non-stationary processes; under the alternative, a fraction of the series in the panel are assumed to be stationary. This is in contrast to the LLC test, which presumes that all series are stationary under the alternative hypothesis. The errors are assumed to be serially autocorrelated, with different serial correlation properties and differing variances across units. IPS propose the use of a group-mean Lagrange multiplier statistic to test the null hypothesis. The ADF regressions (perhaps of differing lag lengths) are computed for each unit, and a standardized statistic computed as the average of the LM tests for each equation. Adjustment factors (available in their paper) are used to derive a test statistic that is distributed standard Normal under the null hypothesis. IPS also propose the use of a group--mean t-bar statistic, where the t statistics from each ADF test are averaged across the panel; again, adjustment factors are needed to translate the distribution of tbar into a standard Normal variate under the null hypothesis. IPS demonstrate that their test has better finite sample performance than that of LLC. Table 4 reports the t-bar statistic of the IPS test. As in the LLC case the test is conclusive: growth rates are stationary whereas external debt is non-stationary regardless of region or income stratum. Acknowledging the limitations of panel unit root tests we do find strong empirical evidence that the behavior implied by conventional open economy models with complete international asset markets does not match the real data generating process. This means that in reality even developed economies with rich and deep financial markets cannot secure against adverse productivity shocks. If this is the case in free global capital markets, a call for intervention might be inevitable. We next develop an open economy model with incomplete international asset markets to show that it does imply a behavior of output and external debt consistent with the real data.

5 Incomplete Markets 5.1

Model

The budget constraint of a household when only a one period risk free asset is available is given in equation (9). First order conditions across asset market structures are strikingly similar.16 In fact, the only difference lies in the risk sharing condition. Under complete markets, the marginal utility of consumption is equated (up to a 16

Technical appendix available upon request.

10

constant of proportionality) across countries. The corresponding condition under market incompleteness, indicates that the ratio of marginal utilities of consumption is no longer constant. It fluctuates in response to changing expectations (16)

𝑢𝑐(𝑐1𝑡, 𝑙1𝑡) 𝑢𝑐(𝑐2𝑡, 𝑙2𝑡)

𝐸𝑡𝑢𝑐(𝑐1, 𝑡 + 1, 𝑙1, 𝑡 + 1)

= 𝐸𝑡𝑢𝑐(𝑐2, 𝑡 + 1, 𝑙2, 𝑡 + 1)

Seminal papers in the field have assessed this model’s capability to reproduce the stylized facts of international business cycles.17 The consensus from those studies is that, even though this asset market structure assumption brings theoretical second moments of consumption and inputs closer to the data, the effect is very small. The argument put forward is that, in this setup, it is too easy for a country to borrow during bad times and repay in good times, thus keeping allocations very close to the complete markets ones. The model assumes no default and, hence, can support large increases in debt positions as equilibrium outcomes. In practice when external debt becomes extreme, the possibility of default creates difficulties in additional borrowing. The most recent illustration of this has been in the Eurozone with countries like Greece, Portugal and Ireland having difficulties raising funds from international markets. We stress exactly this point by showing that international market incompleteness does not allow a country to hedge against idiosyncratic risks, thus giving space for active international policy like the one undertaken by the Eurozone, the ECB and the IMF to bail out Greece, Ireland and Portugal. Along this dimension, we demonstrate a striking improvement in the predictions of this model caused by the assumption of segregated international financial markets. As shown in section 4, external debt exhibits a substantial amount of persistence. This feature was not present in the model of section 2. External debt was shown to exhibit a level of persistence that was less than or equal to the persistence of GDP due to the presence of contingency claims against bad states of the world. Full insurance against idiosyncratic risk was bought one period ahead, so that an exogenous fall in income was compensated fully by a pay check from the asset portfolio. In particular, the new choice of insurance portfolio for next period did not need to change significantly, meaning that external borrowing did not change significantly. Under incomplete markets, a fall in income has to be offset by a rise in borrowing, which in turn implies higher interest payments next period and generates persistence in net foreign liabilities (or external debt) that is much stronger than persistence in GDP. That is exactly the experience of Greece, Ireland and Portugal during the recent financial crisis. In practice this translates into a stationary response process for output growth and a process for external debt that is so close to nonstationary that it is impossible to reject even the panel data unit root tests we use here. Thus the model rationalizes the empirical evidence presented in the last section.

5.2 Economic Interpretation

17

Baxter and Crucini (1995) and Kollmann (1995) are two examples.

11

We attempt to explain the different behavior of external debt, employment, consumption and investments under the two assumptions on asset market structure by analyzing fundamental impulse response functions. That is, the system is simulated for a number of periods with innovations set at 0 until all variables converge to their steady state values. Subsequently, for period t say, a positive innovation in country 1 is set equal to one standard deviation of the shocks. Innovations are set to 0 again from then on. So we simulate the responses to a one period positive productivity shock. Impulse responses produced in this way are plotted in figure (1). We have argued that we live in a world with global incomplete financial markets. In this case, efficient allocation of resources explains why the country with high relative productivity shows a positive response in investment regardless of market structure. Consider, for example, the MOUs for Greece, Portugal and Ireland in 2010, and the relevant bailout programs that increased debt in all countries de facto (since all countries were locked-out from global financial markets) but up to a certain limit, as exactly the incomplete markets model predicts. A positive productivity shock due to immediate implementation of internal reforms explains the different reactions of output, investment and debt accumulation between Ireland and Greece. Ireland experienced a positive productivity shock by implementing swiftly all necessary reforms and at the same time received the necessary funds to support it, which increased investment immediately creating output growth, employment, consumption, and a sustainable debt repayment path. Greece on the other hand, implementing very little if any major reform, spend all bailout funds to just repay previous loans and fund government deficits with no impact on domestic investments and all other macroeconomic variables. The impulse response for debt is a consequence of the impulse response for net exports (after all, debt is simply the expected future discounted sum of net imports). Initially net exports fall so debt increases, the relative magnitude of the response mirroring the one for net exports. Since access to external borrowing is limited by the bailout programs, countries with higher initial exports need to borrow much since they can fund domestic investments by net exports income. In our example, Ireland in 2010 had positive net exports thus it needed much less bailout funds to support its postreforms positive productivity shock. Greece on the other hand was not only running a huge net exports deficit but in the complete absence of swift reforms had no productivity shock to take advance from! Bailout funds were spend to smooth out government spending and pay previous debt obligations. Under incomplete markets, debt is fully repaid much faster, owing mainly to the fact that it is much smaller from the beginning. Given that massive investment has taken place in the beginning of the process, output is well above average and the country is in a position to start lending. Lending increases until almost balanced trade is achieved. From then on, the country runs a very small trade deficit and thus a small debt burden to be repaid eventually in the distant future.

6 Implications for future research 12

We provide evidence that international financial markets incompleteness is an important feature that can explain the evolution of a country’s external debt and output growth. We characterize the response of external debt and output growth under both complete and incomplete markets and we use this analysis as a road map to test empirically the behavior of these variables in a panel of 60 countries. Our results are suggestive of the fact that the issuance of a variety of different assets that can be traded internationally do not correspond to the behavior predicted by a model of complete international financial markets. Under incomplete markets, any productivity shock will generate a very persistent, close to non-stationary response process for external debt regardless of income or region stratum, which indicates that a country is nearly insolvent if not unlucky! Our analysis suggests that there exist productivity shocks that are not hedged by the current international financial markets and that countries have much more to gain from improving their domestic financial structures so as to upgrade international financial markets’ completeness to boost domestic productivity and international capital mobility. If most of the inequality in world income is explained by differences in productivity or domestic distortions, then the question of how the capital account structure interacts with domestic allocative efficiency seems quite relevant. Understanding the determinants of countries’ productivity is a central question in development economics, which considers a very rich array of explanations: technological innovation and diffusion, the legal regime and property rights, policies and institutions, and the financial and goods market structure. The neoclassical approach makes the strong assumption that these determinants are not affected at all by the capital account structure. Our results are consistent with the recent developments of the literature on growth and convergence in international perspective. In contrast with early papers that stressed factor accumulation as a source of growth (Mankiw, Romer and Weil (1992); Barro, Mankiw and Sala-i-Martin (1995)), the literature has moved towards the view that total factor productivity rather than factor accumulation accounts for most of income differences across countries (Hall and Jones (1999); Easterly and Levine (2001)). Moreover, the results are consistent with recent developments on the importance of financial structure in economic growth (Shleifer et al. (1998)). The present analysis is done within a two country symmetric general equilibrium model with perfectly competitive markets. We expect that our prediction of the behavior of external debt and output growth will prevail in a richer model (although verifiable through future research) as well, as long as foreign assets are used as a buffer stock to smooth productivity shocks. Some very relevant questions that we do not address here, are: what is the contribution of interest payments? How important is the size of the country (not significant in the data as documented in Lane and Millesi-Ferreti (2003))? What is the effect of exchange rates (2-good model)? How do economies respond to net exports shocks and do they increase insurance or exacerbate debt problems? Cole and Obstfeld (1991) show that terms of trade’s movement act as insurance under financial autarky. We leave all these for future research. 13

References [1] Abuaf, N. and P. Jorion. Purchasing power parity in the long run. Journal of Finance 45, 1990, 157-174. [2] Athanasoulis, Stefano and Eric vanWincoop, “Growth Uncertainty and Risksharing,” Journal of Monetary Economics, 2000, 45, 477---505. [3] Levine, Ross et al. (2000), “Finance and Sources of Growth,” Journal of Financial Economics 58, 261-300. [4] Edison, Hali and Frank Warnock (2002), “A Simple Measure of the Intensity of Capital Controls,” Journal of Empirical Finance, forthcoming. [5] Edison, Hali, Ross Levine, Luca A. Ricci and Torsten M. Sløk (2002), “International financial integration and economic growth,” Journal of International Money and Finance 21 no. 6 (November), 749-776. [6] Cole, Harold L. and Maurice Obstfeld (1991): “Commodity Trade and International Risk Sharing: How Much Do Financial Markets Matter?” Journal of Monetary Economics, 28, 3-24. [7] Obstfeld, Maurice and Alan M. Taylor (2002), “Globalization and Capital Markets,” NBER Working Paper 8846, March. [8] O’Donnell, Barry (2002), International Financial Integration and Economic Performance, PhD Dissertation, Trinity College Dublin. [9] Taylor, A. 2001. "Potential Pitfalls for the Purchasing-Power-Parity Puzzle? Sampling and Specification Biases in Mean-Reversion Tests of the Law of One Price", Econometrica 69, 2, 473-498. [10] Portes, Richard and Hélène Rey (2002), “The Determinants of Cross-Border Equity Flows: The Geography of Information,” mimeo, Princeton University. [11] Lane, Philip R. and Gian Maria Milesi-Ferretti (2001a), “The External Wealth of Nations: Measures of Foreign Assets and Liabilities for Industrial and Developing Nations,”\ Journal of International Economics 55, 263-294. [12] Lane, P. R. and G. M. Milesi-Ferretti, International Financial Integration. CEPR Discussion Paper No. 3769. International Monetary Fund Staff Papers, forthcoming. Revised January 2003 [13] Levin, Andrew and Lin, Chien-Fu. Unit Root Tests in Panel Data: New Results", University of California Discussion Paper No. 93-56, 1993. 14

[14] Sarno, Lucio and Mark P. Taylor. Real exchange rates under the current float: unequivocal evidence of mean reversion. Economics Letters, 60, 1998, 131-137. [15] Taylor, M. and L. Sarno, 1998. The behavior of real exchange rates during the post--Bretton Woods period. Journal of International Economics, 46, 281--312. [16] Banerjee, Anindya. Panel Data Unit Roots and Cointegration: An Overview. Oxford Bulletin of Economics and Statistics, Special Issue, 607-629, 1999. [17] Levin, Andrew, Lin, Chien-Fu and Chia-Shang James Chu. Unit Root Tests in Panel Data: Asymptotic and Finite Sample Properties. Journal of Econometrics, 108, 1-24, 2002. [18] Maddala, G.S. and In-Moo Kim. Unit Roots, Cointegration, and Structural Change, Cambridge: Cambridge University Press, 1998. Taylor, Mark P. and Lucio Sarno. The behavior of real exchange rates during the post-Bretton Woods period. Journal of International Economics, 46, 1998, 281-312. [19] Baxter, M. and Crucini, M. "Business Cycles and the Asset Structure of Foreign Trade". International Economic Review, November 1995, 36(4) [20] Kollmann, R. "Incomplete Asset Markets and the Cross-Country Correlation Puzzle". Journal of Economic Dynamics and Control, 1996, Vol. 20, pp. 945-961. [21] Kyung So Im, M. Hashem Pesaran, Yongcheol Shin, Testing for Unit Roots in Heterogeneous Panels. Unpublished Working Paper, Dept. of Applied Economics, University of Cambridge, Dec. 1997 (available at \ http://www.econ.cam.ac.uk/faculty/pesaran/lm.pdf) [22] Levin, Andrew and Lin, Chien-Fu. Unit Root Tests in Panel Data: New Results, University of California at San Diego Discussion Paper No. 93-56, 1993. [23] Marcet, A. and Marimon, R. (1999), “Recursive Contracts” (Manuscript, Universitat Pompeu Fabra). [24] Stockan, A. C. and Tesar, L. (1995),“Tastes and Technology in a Two-Country Model of the Business Cycle: Explaining International Comovements”, American Economic Review, 85, 168---185. [25] Hau, H. and H. Rey, 2002. "Exchange rates, equity prices and capital Flows", manuscript [26] Shleifer, A., R. La Porta, F. Lopez-de-Silanes, and R. Vishny. Law and Finance, Journal of Political Economy, December 1998.

15

[27] Mankiw, N. Gregory, David Romer, and David Weil, “A Contribution to the Empirics of Economic growth,” Quarterly Journal of Economics, 1992, 107 (2), 407--38. [28] Marcet, A. and Andrew Scott, “Debt and Deficit Fluctuations and the Structure of Bond Markets”, Journal of Economic Theory, 2009, 144 (2), 473-501. [29] Hall, Robert and Charles Jones, “Why Do Some Countries Produce So Much More Output Per Worker than Others?,” Quarterly Journal of Economics, 1999, 114 (1), 83--116. [30] Barro, Robert, Gregory Mankiw, and Xavier Sala-i-Martin, “Capital Mobility in Neoclassical Models of Growth,” American Economic Review, 1995, 85 (1), 103--115. [31] Easterly, William and Ross Levine, “It’s Not Factor Accumulation: Stylized Facts and Growth Models,”\ The World Bank Economic Review, 2001, 15 (2), 177---219. [32] Den Haan, W.J. and A. Marcet, 1990, Solving the stochastic growth model by parameterizing expectations, Journal of Business and Economic Statistics vol 8 no 1, 31-34. [33] Backus, D., Kehoe, P. and F. Kydland, 1992, “International Real Business Cycles”, The Journal of Political Economy, 100 (4), 745-775.

16

Table 1. Summary statistics Variable Observations Mean S.D. Min Growth

Max

1726

0.022 0.042 -0.285 0.243

Net Assets 1726

-1.969 6.46 -72.50 60.99

Table 2. The MADF test. Δyi, t = ρiyi, t − 1 + ξi, t H0:ρi = 1vs.H1:ρi < 1foralli All

Growth

Net Assets

T>N

T>N

By Region East Asia & Pacific

152.778∗∗ 9.497

Europe & Central Asia

19.278

4.799

Latin America & Caribbean 799.434∗∗ 121.28∗∗ North America

380.120∗∗ 29.349

South Asia

391.096** 694.150∗∗

Sub-Saharan Africa

169.9∗∗

2.729

Middle East & North Africa 48.362** 1.30 By Income High

2222.4∗∗

486.334∗∗

Low

53.166** 2.920

Middle-Low

350.128** 162.251∗∗

Middle-High

730.681∗∗ 79.177**

Note: **, indicates significance in the 5% level. The 5% critical values for the test are 27.491 (when augmented with 1 lag) and 28.150 (when augmented with 2 lags). Regional and Income classifications are taken from the World Bank. The estimates are based on a panel of 60 countries over the period 1970 - 2000. The MADF test requires that we estimate Zellner’s SUR estimator for each unit of the panel and test the null hypothesis.

17

Table 3. The Levin-Lin-Chu test Δyi, t = ai + δi, t + θt + ρiyi, t − 1 + ξi, t H0:ρi = 0vs.H1:ρi < 0foralli Growth All

Net Assets

-0.743 (0.000) 0.0451 (0.999)

By Region East Asia & Pacific

-10.963(0.000) 0.046 (0.999)

Europe & Central Asia

-12.99(0.000) NA

Latin America & Caribbean -18.4(0.000)

0.044 (0.999)

North America

-14.012(0.000) -0.071 (0.6303)

7South Asia

-19.17(0.000) 0.099 (0.999)

Sub-Saharan Africa

-14.219(0.000) 0.002 (0.934)

Middle East & North Africa -31.5(0.000)

0.074 (0.999)

By Income High

-19.03(0.000) 9.076(0.999)

Low

-29.40(0.000) -2.603(0.6028)

Middle-Low

-24.97(0.000) -4.860(0.8613)

Middle-High

-16.99(0.000) -1.771(0.9985)

Note: ***, indicates significance in the 1% level. Regional and Income classifications are taken from the World Bank. The estimates are based on a panel of 60 countries over the period 1970 - 2000.

Table 4. The Im-Pesaran-Shin test Δyi, t = ai + δi, t + θt + ρiyi, t − 1 + ξi, t H0:ρi = 0vs.H1:ρi < 0notforalli Growth All

Net Assets

-4.850 (0.000) 1.756(0.999)

By Region East Asia & Pacific

-3.856(0.000) 0.314(0.999)

Europe & Central Asia

-2.998(0.000) -1.448(0.946) 18

Latin America & Caribbean -3.739(0.000) -1.321(0.999) North America

-5.847(0.000) -0.502(0.919)

7South Asia

-3.854(0.000) -0.198(0.999)

Sub-Saharan Africa

-6.435(0.000) -1.122(0.980)

Middle East & North Africa -5.209(0.000) -1.256(0.963) By Income High

-3.879(0.000) -1.418(0.969)

Low

-4.543(0.000) -0.371 (0.994)

Middle-Low

-4.047(0.000) -0.882(0.999)

Middle-High

-3.864(0.000) 0.524 (0.999)

Note: ***, indicates significance in the 1% level. Critical values are obtained through a monte carlo simulation. Regional and Income classifications are taken from the World Bank. The estimates are based on a panel of 60 countries over the period 1970 - 2000.

19