Risk sharing from international factor income: explaining cross-country differences

Abstract

Access to world capital markets and net investment income flows between countries help protect national income from country-specific output shocks. I empirically study what factors explain cross-country differences in the extent of risk sharing from international factor income. An index of investor protection is the leading causal variable for the estimated amount of risk sharing over the 1985 to 2004 period. Improving investor protection in Russia to Denmark's level implies five times larger risk sharing compared to the sample average. These results indicate one possible way to reap large potential benefits from international risk sharing.

Keywords:

• risk sharing
• diversification
• international financial integration
• investor protection

JEL Classification::

• F36
• G15
• O17

Acknowledgements

I thank Jean Imbs, Sebnem Kalemli-Ozcan, Ayhan Kose, Gian Maria Milesi-Ferretti, David Papell, Eswar Prasad, Bent Sorensen, Andrey Ukhov, Kei-Mu Yi and an anonymous referee for their comments and suggestions.

Notes

¹In the literature smoothing of national income against output shocks is referred to as ‘income risk sharing’, ‘income smoothing’ or ‘income insurance’. I use the three terms interchangeably. I refer to the difference between country-level and the world-wide variables as ‘country-specific’ or ‘idiosyncratic’.

²Leibrecht and Scharler (2011) show that adjustment of consumption growth to output shocks is asymmetric and countries smooth the impact of adverse shocks on consumption faster than the impact of positive shocks. This result is consistent with the interpretation that borrowing constraints become binding relatively quickly after negative shocks and thereby reduce a country's ability to smooth consumption.

³In addition, Balli and Sorensen (2007) find a negligible extent of consumption smoothing through central fiscal institutions and private transfers (such as remittances) even for advanced countries. Canova and Ravn (1996) and Artis and Hoffmann (2008) argue that the observed lack of international risk sharing may be due to a lack of insurance against permanent shocks.

⁴The measure relies on a cross-sectional decomposition of GDP as in Asdrubali et al. (1996) and reflects the sensitivity of national income (GNI per capita) to idiosyncratic output (GDP per capita) fluctuations. It potentially varies from zero (no risk sharing) to 100% (full risk sharing) and represents a percentage of idiosyncratic risk to GDP insured through Net Factor Income (NFI) flows (compared to full risk sharing). We can obtain negative empirical estimates of the measure that would imply dis-smooting of income (see details in Section III).

⁵Fratzscher and Imbs concentrate on consumption risk sharing between pairs of countries and establish the positive interaction effect of institutions and trade openness for their definition of risk sharing. They also document that the effect of good institutions is larger in a sub-sample of more financially open countries.

⁶Crucini (1999), Sorensen and Yosha (1998), Kalemli-Ozcan et al. (2003), Becker and Hoffmann (2006) and Asdrubali and Kim (2008) document lower risk sharing between countries than within countries.

⁷The examples of ‘bonding’ are customized contracts and corporate charters with larger rights for investors than as required by the local law, listing companies in countries and on exchanges with better investment regime, or merger with the foreign companies.

⁸In international context, Robert Shiller proposed a scheme under which countries in a risk-sharing pool could issue claims on their output, Robert Merton advocates a network of GDP swaps executed bilaterally or by an intermediary (see Borensztein and Mauro, 2004 for a review). A multinational fiscal federation which smoothes income cross-sectionally is advocated by Forni and Reichlin (1999) for Europe.

⁹With insurance against asymmetric output shocks, people can specialize and safely engage in riskier but high-return investment projects and guarantee higher expected consumption growth and welfare, speed up capital accumulation, financial deepening and reduce uncertainty in the growth process (Helpman and Razin, 1978; Obstfeld, 1994a; Acemoglu and Zilibotti, 1997; Kalemli-Ozcan et al., 2003; Imbs, 2004).

¹⁰The reader is invited to refer to Asdrubali et al. (1996) for details.

¹¹The EU Structural Funds and the Cohesion Fund are examples of an international tax-transfer system that may contribute to risk sharing. The Funds are the financial instruments of the EU regional policy. They intend to support infrastructure and job-creating investment, promote employment conditions by funding training measures and finance environmental and transportation infrastructure projects.

¹²I highlight the major components of the national accounts important for my analysis. GNI was previously called Gross National Product. Gross National Saving in the above formula is loosely defined to include depreciation, international transfers and net saving (by governments, businesses and individuals).

¹³The SNA puts the concept of residence – not nationality or citizenship – as the basis of accounting. A resident within the economic territory engages in economic activities on a significant scale over a long period of time (usually 1 year or more). For more details, refer to the revised System of National Accounts 1993 (United Nations, 2005).

¹⁴The remittances are the current transfers in cash or in kind by resident migrant workers to their country of origin. Because of SNA's residency emphasis, remittances are excluded from NFI of the workers' home countries. For example, a Polish citizen earning income in the UK as a British resident would contribute to the UK's GNI; her remittances to Poland are not included in factor income flows to Poland. Instead, these transfers enter the secondary distribution of income account and do affect the level of Poland's Gross National Disposable Income.

¹⁵The literature often tends to discuss the costs and the benefits of risk sharing separately. Lewis (1999) postulates that the international risk sharing might be small because the costs to diversification exceed the benefits involved. This implies that both factors should be analysed simultaneously.

¹⁶I do not consider nonseparability between consumption and leisure and nontradable endowment because these frictions does not seem to explain low international risk sharing (Lewis, 1999).

¹⁷Heathcote and Perri (2004) show that under certain conditions – also fulfilled in my model – people who can reallocate the portfolio each period choose the same allocations as in one-period model with a single trade. Since people optimally choose not to reallocate their portfolios after the initial period, it is safe to assume that they cannot do this.

¹⁸Taking a short position in foreign shares in the beginning of the period means that the individual promises to deliver more than what she owns before trade, or 100% of their own endowment when χ = 1. Restricting χ to be no greater than one in the constraint (1) rules out such possibility. This, seemingly restrictive, assumption is not unreasonable if we remember that many developing countries cannot indeed borrow short or without collateral.

¹⁹I choose values of the free parameters that are consistent with international business cycles literature (see, e.g. Backus et al., 1992; Heathcote and Perri, 2004): A = 1, μ _y  = 2 and σ _z  = 0.02; these values generate a percentage SD of the common factor, σ _w , of 1%.

²⁰Information asymmetries could lead to the reversal of capital flows (as in the model of Gertler and Rogoff, 1990), explain differences in real interest rates across countries (Gordon and Bovenberg, 1996), financial asset returns, home bias and risk sharing (Martin and Rey, 2004). Portes and Rey (2005) empirically stress the importance of imperfect information and other credit market imperfections for bilateral equity flows.

²¹Tirole (2002) stresses that government actions can affect private willingness or ability to fulfill international contracts. When a government assumes private debt, it becomes sovereign debt. Reinhart et al. (2003) and Reinhart and Rogoff (2004) demonstrate that a country's default history explains its ability to withstand high risk of default and borrow internationally.

²²Melitz and Zumer (1999) introduce the structure where the extent of risk sharing β can vary by country. My approach closely follows Sorensen et al. (2007) and Demyanyk et al. (2007) who allow for interactions of risk sharing with multiple explanatory variables. Usually it is hard to obtain a balanced panel with a large number of explanatory variables without making the restrictive assumptions about missing data.

²³As discussed by Demyanyk et al. (2007), when the interaction terms are not included in the regression in Equation 7, β₀ approximately measures the amount of risk sharing at the average values of the variables RISK and BARRIERS. If the interaction terms are included but not demeaned, β₀ would represent the amount of risk sharing when RISK = 0 and BARRIERS = 0. I subtract the middle year of the sample period from time trend t for the same reason.

²⁴In context of risk sharing, one benchmark is the autarky equilibrium when countries consume the value of their GDP. A natural measure of output is then GDP deflated by the Consumer Price Index (CPI), not by a GDP deflator. I recalculate the nominal GDP and GNI data in local currency from the World Bank into real 1995 US dollars using the CPI and US dollar exchange rate for 1995. This translates GDP and GNI per capita to the amount of individual consumption that they can buy.

²⁵In empirical study it is impossible to calculate the true Whole World total GDP and GNI aggregates. The empirical notion of the world includes 23 high-income Organisation for Economic Cooperation and Development (OECD) countries by the World Bank classification that on average account for 80% of total GDP of my largest empirical sample.

²⁶They are obvious outliers Luxembourg and Switzerland with large net savings abroad (the estimates of for these two economies are 44% and 51%, respectively). The rest of outliers include several poor, hyperinflationary or war-ridden developing countries. They are Albania, Bangladesh (low ), Burkina Faso, Namibia, Nicaragua, Nigeria, Papua New Guinea, Sierra Leone and Suriname (high ). I also employ least absolute deviation regressions to check the role of outliers in driving the main findings.

²⁷Even though the illustrative example in Section II rules out these possibilities, I keep the countries with negative estimates of β in my main sample to exploit all available information in the empirical analysis.

²⁸The GDP weights capture the positive relation between trade volume and GDP. As explained in Alfaro et al. (2008, p. 353), this variable is a proxy not for geography (like ‘distance from equator’ and average distance between countries) but rather for information frictions. Because of the GDP weights, out of two equally remote countries based on average distance, the one which is ‘smaller’ economically would be in a greater disadvantage.

²⁹ICRG names this index ‘Investment Profile’. PRS Group claims that ICRG ratings are immune to the ‘perception bias’ because they do not rely on opinion surveys of any kind, in particular on opinion surveys of businesses operating in a particular country. I address these issues further in instrumental variables analysis.

³⁰I calculate a condition index as in Belsley (1991) and variance inflation factors (Kennedy, 1998) to evaluate the severity of multicollinearity. The condition index introduced by Belsley (1991) equals to the square root of the largest eigenvalue divided by the smallest eigenvalue of the data matrix. When there is no collinearity at all, the eigenvalues and the condition index will all be equal to 1. When the multicollinearity increases, the eigenvalues get close to zero the condition index rises. Belsley advises that when the condition index is larger than 30, the multicollinearity is a serious concern.

³¹Only in the regression with Output Risk and the measures of all four barriers the condition index is 46, which points to a severe multicollinearity.

³²I experiment with all indices from ICRG (e.g. law and order, quality of bureaucracy, corruption) and index of corruption from Transparency International. The coefficient of Investor Protection comes out most significant, stable and largest in absolute value in all specifications. I attempt to differentiate between different aspects of institutions in the robustness section.

³³It plots the residuals from the OLS regression of on Output Risk versus the residuals from the regression of Investor Protection on Output Risk in the main sample. By the Frisch–Waugh theorem, the coefficient in that regression is 2.18, exactly the same as the coefficient of Investor Protection in the multiple regression in Column 5, Table 3.

³⁴Then the OLS procedure would erroneously ‘assign’ the effect of the omitted variables to institutions. I addressed this problem, at least partially, by running multiple regressions with the investor protection index and other controls. In addition, a rating agency may observe an economy actively engaging into international risk sharing and assign better investment protection score to the country.

³⁵The 2SLS regressions including other barriers besides Investor Protection do not change the qualitative results in this section.

³⁶In Britain, the courts became fairly independent from the crown since as early as seventeenth century and, under the pressure of the parliament, protected private property and later investors. On the continent, the state dominated the courts or property owners. Today's commercial codes in France and Germany, adopted by Napoleon and Bismark in nineteenth century, preserved more limited investor rights. Since then, different variations of these three systems spread throughout the world.

³⁷I obtain similar results with a measure of the efficiency of the judicial (or administrative) system in the collection of overdue debt. The systems with complex and cumbersome court procedures hurt investors.

³⁸The estimate of risk sharing is expressed in percentage points. Given the 2SLS coefficient estimates for Investor Protection in Table 4, the average magnitude effect using the individual point estimates of 2SLS coefficients is 7%. From Column 3, the magnitude effect of moving from the lowest to the highest quartile in Investor Protection; is 8 percentage points ((6.8 − 5.0) × 4.43).

³⁹Investor protection is also conducive to reducing the volatility of income relative to that of output. In a regression, higher values of Investor Protection are associated with lower ratio of the SD of GNI growth to that of GDP growth. The results are available upon request. In a related study, Bekaert et al. (2006) find that countries with better investor protection experience larger decreases in consumption growth volatility after the opening of equity markets to foreign investors.

⁴⁰The significance level of Investor Protection is smaller in sub-samples than in the Whole World sample because there is somewhat less of cross-section variation in the investor protection index.

⁴¹This is the main variable of interest investigated by Kose et al. (2006). Trade openness can be complementary to financial integration as in Rajan and Zingales (2003). I also check for the role of volume-based measure of trade openness with similar results.

⁴²The institutional quality, broadly defined, is empirically important for the patterns of capital flows (Wei, 2000; Alfaro et al., 2008). The result is qualitatively similar with the other measures of institutions. In all these experiments the coefficient of institutional quality is positive but insignificant and the coefficient of Investor Protection is significant at least at 5%. Based on the diagnostic by Belsley (1991) (see footnote 30), multicollinearity is not a major concern in these regressions.

⁴³I got the similar result with the proxies for the prudence of macroeconomic policies, tax policy and other domestic fundamentals potentially important for financial integration (human capital, culture, climate, natural resources and the geographic location). The results are available.

⁴⁴The estimation involves two stages and assumes that the error variance differs across countries. In the first stage, I estimate the model by the OLS and obtain the SDs of the residuals for each country. In the second stage, I weigh each country with the inverse of its SD and re-estimate the model. I also correct for possible autocorrelation in residuals assuming that the error term in each country is Autoregression (AR(1)).

⁴⁵Since the ‘true’ date of the break is unknown, I try various years: the 1995, since after this year most of the countries in Eastern Europe were done with privatization programs, those with hyperinflation in early 1990s (mostly the former USSR and Eastern Europe, but also Peru and Nicaragua) finally stabilized the prices and we see an improvement in overall investor protection from about this time; the 1998, as the year past 1997 East Asian Crisis and of the Russian Crisis; the 2000, when most of the effects of the above crises must have been overcome and many of emerging markets turned to the positive economic growth; the 1991, as the very beginning of the period after the fall of Communism and also the end of the ‘lost decade’ (the 1980s) in Latin America. The coefficient of the dummy increases at the absolute value as we define the break later, which is consistent with the evidence of the upward trend in the degree of risk sharing documented in the original draft and by a recent literature on risk sharing. The time of the break is clearly the mid-1990s; when I define the date of break in 1991, it is not significant. The results for the other break dates are available upon request.

⁴⁶Under the new Polish law, any group with 5% of the company can demand an independent audit and each share in a public company may carry only one vote at a general shareholder meeting. In Hungary shareholders must command 10% of a company's votes to demand an audit or meeting and some shares can carry as high as 10 votes (Economist, 2001). The value of the index of minority investor protection I use in this article is 6.0 for Poland and just 4.3 for Hungary, indicating much better minority shareholder protection in Poland.

⁴⁷For a preliminary evidence for the positive effect of financial openness on other channels or risk sharing, see Sorensen et al. (2007), Fratzscher and Imbs (2009) and Kose et al. (2009) who find positive effects of openness on overall consumption risk sharing. Fratzscher and Imbs also document nonlinearity effects.

⁴⁸Carroll et al. (2008) discuss the estimation issues (such as measurement error and time aggregation) in aggregate consumption data in detail. Rogoff (2007) points on quite different aggregate volatility pattern depending on whether yearly or quarterly data are used.

⁴⁹Stockman and Tesar (1995) refer to the multiplicative disturbances to individual utility function as ‘taste shocks’. They interpret taste shocks very broadly to include consumer sentiment, policy shocks and so on.