Journal Issue

Russian Federation: Selected Issues

International Monetary Fund
Published Date:
September 2008
  • ShareShare
Show Summary Details

II. Financial Integration within the CIS24

A. Introduction

46. This chapter takes stock of the extent of Russia's financial integration with overseas markets, focusing in particular on the role of Russia within the CIS. As yet, preliminary analysis suggests that Russia has a somewhat limited impact on other CIS markets. Indeed, data on direct cross-border asset holdings indicate that Russia is a marginal source of funds within the region.

47. Noting the potential shortcomings of direct capital-flow data, however, the chapter pays particular attention to indirect evidence of financial integration between Russia and the rest of the CIS. Staff analysis of equity returns indicates that, although Russia is relatively well integrated into global capital markets, Russian developments have little effect on other regional financial markets. Further, in seeking to explain this apparent lack of correlation, the chapter finds that this result is largely to be expected, given the current level of financial development throughout the region—more specifically, the low degree of correlation reflects the small size of non-Russian CIS economies, along with their relatively illiquid, less-developed financial systems. In sum, the results suggest that the risks of spillovers from Russia to other CIS economies, through financial channels, are limited at this point in time.

48. Looking forward, staff will keep the process of financial integration within the CIS under close review. Financial development within the CIS is still at a very early stage. Moreover, given Russia's physical proximity and close cultural links with other CIS countries, and given that Moscow is already a key headquarters location for many transnational corporations operating within the post-Soviet area, Russia's potential as a regional financial center is substantial. In this context, staff analysis suggests that the degree of integration will likely increase over the medium term as the financial systems of CIS economies grow and develop further.

49. This chapter is organized as follows. Section B will outline the problems associated with measures of direct asset holdings, and will suggest the need for a more indirect gauge of intra-CIS financial integration. Section C will provide preliminary analysis of the correlation of CIS equity returns, both within the region and with other developed and emerging markets. Section D will estimate a gravity model to explore the possible reasons for the apparently low correlation of returns between Russia and other CIS countries. Section E will conclude.

B. Capital Flows

50. Measurement of intra-CIS capital flows is problematic, as useful data on direct financial linkages are difficult to obtain. The locational BIS International Banking Statistics covers bank flows between countries, but as Russia is not within the BIS reporting area, it is impossible to infer anything about the pattern of lending within the region. The IMF's Coordinated Portfolio Investment Survey (CPIS), on the other hand, covers cross-border portfolio flows and does have Russia as a reporting country. The most recent survey includes data for 2006, and its results for the CIS region are summarized in Table 1 below. The first feature of note is that Russia appears to be a marginal source of portfolio investment for other CIS countries; accounting for less than 0.4 percent of total inflows. However, closer inspection highlights a key issue. For many CIS countries, a sizable fraction of inflows originate in international financial centers. Taking Kazakhstan, for example, the data suggest that Guernsey is 50 times more important than Russia as a source of portfolio funding. But anecdotal evidence suggests that much of these inflows actually represent Russian funds, which have been channeled via international centers for privacy reasons. The actual importance to the region of Russian portfolio capital, therefore, is unknown.

Table 1.Total Portfolio Assets, 2006(US$ Million)
Investment from:USAJapanUKNetherlandsOffshore Financial Centerso/w Bahamas, Theo/w Bermudao/w Cayman Islandso/w Cypruso/w Guernseyo/w JerseyRussiaOther CountriesTotal
Investment in:
Kyrgyz Republic42-11--61270
Russian Federation48,4412,95730,9625,14411,01872583634,0444,1612,095-60,324158,847
Note: the data are derived from the creditor side for both assets and liabilities-- Indicates a zero value or a value less than US$ 500,000.…. Indicates an unavailable datumSource: Coordinated Portfolio Investment Survey
Note: the data are derived from the creditor side for both assets and liabilities-- Indicates a zero value or a value less than US$ 500,000.…. Indicates an unavailable datumSource: Coordinated Portfolio Investment Survey

51. This problem appears to be even more acute when considering direct investment flows. FDI is a vital source of foreign funding throughout the CIS. But for these countries, the diversion of funds through international financial centers is widespread. The attached diagram represents the flow of FDI in 2007 between Russia and two key CIS economies: Ukraine and Kazakhstan. As before, the direct flows between Russia and these countries are relatively small. Taking Ukraine, the bilateral flow represents only 3.3 percent of Russia's outward direct investment, and only 5.8 percent Ukraine's inward direct investment. By contrast, three key international centers account for almost 65 percent of total Russian outflows, and those same three centers represent over 50 percent of total Ukrainian inflows. It is impossible to know the exact amount, but it is a widely held view that much of this activity stems from unrecorded Russian investment within the CIS.

Source: National Central Banks, staff calculations.

52. Consequently, this chapter takes a more indirect approach to measure the role of Russia within the CIS. For real markets, full integration can be defined as a situation in which goods are able to move freely between two markets. Similarly, integrated financial markets imply that agents can trade financial assets freely within a specified area. However, when the volume of trade in goods—or assets—cannot be measured accurately, it is often preferable to look at more indirect measures. In the goods market, for example, full integration suggests that the law of one price applies, so that goods flow from one market to another until the price of same good is equalized. Analogously, in the case of financial integration, the return of assets with the same risk characteristics should be equalized across markets. Asset returns in integrated markets, therefore, should be closely correlated.

53. The close co movement of equity returns, in which developments in Russia are mirrored in other CIS markets, would suggest that Russia is an important source of capital within the region. Moreover, this should be the case regardless of whether assets are traded directly between CIS markets, or are instead traded indirectly via international intermediaries. If the markets are not well integrated, on the other hand, then asset prices and returns should be relatively independent. In this context, the next section will look at the co movement of equity returns between Russia and other financial markets, to gauge whether CIS markets are in fact well integrated with Russia.

C. Asset Returns: Preliminary Analysis


54. In measuring the co movement of asset returns, this chapter focuses on stock-market behavior across a range of markets. In order to ensure comparability and consistency, wherever possible we use the daily MSCI indices compiled by Morgan Stanley, as reported on Datastream International. In addition, as we are considering the viewpoint of a standard international investor, we study returns in U.S. dollars.25 For CIS markets, however, reliable equity indices are relatively rare, and so we are limited to data for Russia, Ukraine, Kazakhstan and Kyrgyzstan. In the case of Kazakhstan and Kyrgyzstan, equity index data is taken from the Kazakhstan Stock Exchange (KASE) and Kyrgyzstan Stock Exchange (KSE) directly. The behavior of the four included CIS stock market indices are illustrated in Figure 1 below.

Figure 1.CIS Stock Market Indices, 1999–2008

(USD Index, 1 Jan 2004 = 100)

Source: Datastream, KASE, KSE, staff calculations

55. Equity data is gathered for a range of developed, emerging, and CIS markets, and returns in each market are measured on a weekly basis. This helps eliminate any spurious correlation resulting from the fact that not all markets are within the same time zone. It also reduces the impact of the relative illiquidity of some of the smaller markets, where it may take a number of days to execute a particular trade. Returns are also defined in continuously compounded terms, so that the return over one period (a week) is defined as:

In this context, Pt is the value of the index at time t. The correlation matrix for key developed, European, and CIS markets is given below. As can be seen, it appears that most CIS markets are poorly correlated with other markets. The exception, however, is Russia, which seems relatively closely linked with global asset markets.

Correlation Matrix. Weekly Returns, 1999–2007
Source: Datastream, KASE, KSE, staff calculations
Source: Datastream, KASE, KSE, staff calculations

Principal Components Analysis

56. As a first step, this chapter employs Principal Components Analysis (PCA) to help build a rough picture of stock-market co movements: both within the CIS region, and between CIS and non-CIS countries. This is a widely used tool in multivariate statistics, and is specifically designed to help uncover underlying patterns within a given, often complex dataset—allowing us to extract the key driving factors that explain stock-return co movements across different markets. The methodology is outlined in more detail in the appendix.

57. As noted, apart from Russia, CIS markets appear relatively uncorrelated with other developed and emerging markets. It is still possible, however, that events in Russia may nonetheless be an important driver of market developments in other CIS countries, independent of developments elsewhere—i.e. CIS markets may be relatively integrated with Russia, but not with the global market. If this were the case, then we would expect our analysis to identify Russia and other CIS markets as a single, separately identifiable group.

58. PCA analysis suggests that, within our sample, there are three main independent factors that explain cross-country co movements in stock market returns. The attached table illustrates the correlation (the “loading coefficient”) of the factors with each individual market, focusing on instances where the correlation is greater than 0.5. As seen, the three extracted factors account for a large amount of variance for each of the different stock-return series, ranging from about 50 percent for Japan to almost 90 percent for Kyrgyzstan. 26

Loading Coeficients.
Factor 1Factor 2Factor 3Communality
United Kingdom0.8740.78
Czech Rep.0.8010.65
(blanks represent coefficient < 0.5)
(blanks represent coefficient < 0.5)

59. The first factor seems to represent a common global trend, which jointly influences both developed and emerging markets together. Russia falls into this group, and so appears to be driven by the same forces that drive most other active markets around the world. This would suggest that Russia is now effectively integrated within the international financial system.

60. The second factor might be characterized as a CIS-specific trend, which drives both Ukraine and Kazakhstan in tandem, but appears to exclude Russia. Consequently, while the financial markets of Ukraine and Kazakhstan may be closely related to each other, they appear to be isolated from global market trends, and are also relatively separate from developments within Russia. Kyrgyzstan, it seems, is somewhat secluded—from both Russia and other CIS countries—and is driven by a factor of its own.

61. In sum, although Russia is well integrated within international capital markets, global developments that impact Russia do not appear to flow on into other CIS markets. Moreover, there does not seem to be any significant pattern of regional comovements, in which idiosyncratic developments in Russian impact other CIS markets. Instead, non-Russian CIS markets are relatively separate from developments within Russia. On the one hand, this result is somewhat surprising, given the common language, legal traditions, and historical ties that are shared between CIS countries. On the other hand, however, this result may reflect the fact that, compared to Russia, the economies of other CIS countries are relatively small, with illiquid and underdeveloped financial systems. Financial markets in smaller CIS countries, therefore, may be influenced by a substantially different set of idiosyncratic trends. The remainder of this chapter will model the factors that typically determine the comovement between different countries' financial markets, to see whether they can explain the relatively low degree of correlation between Russia and other CIS countries

Factor loadings

D. Modeling Financial Linkages: A Gravity Approach


62. In seeking to explain the apparent lack of integration between Russia and other CIS markets, the chapter builds on existing empirical efforts to measure the determinants of cross-country financial integration. Studies have identified various key factors that determine the degree of integration, such as trade intensity (Chinn and Forbes, 2004), financial development (Dellas and Hess, 2005), and business-cycle synchronicity (Walti, 2004). Generally, these studies find some support for the explanatory power of such factors, but the results and conclusions often differ significantly. Martin and Rey (2004), on the other hand, have proposed a theory of capital flows from which a “gravity” equation emerges. The model's main result is that gross financial flows should depend inversely on transactions costs—such as the cost of gathering information across borders—and should depend proportionally on market size, as proxied by stock-market capitalization.

63. Empirical studies that have adopted this gravity-model framework have found that it is generally successful in explaining bilateral financial flows. The gravity equation has been the workhorse model for trade in goods since the 1960s. At core, the trade version seeks to explain the bilateral flow in goods between countries as a function of their two masses (GDPs) and their geographical distance. Unlike the goods trade, however, where distance is a useful proxy for transportation costs, the trade in financial assets is relatively weightless, so the role of geographical distance is less obvious. Nonetheless, Portes and Rey (2005) demonstrate that the gravity equation performs at least as well in explaining asset trade as it does in explaining goods trade. They further illustrate that geographical distance, in this case, serves as a valuable proxy for informational frictions that serve as a barrier to interaction between economic agents. Similar studies have shown that such informational considerations are also important within countries, so that domestic investment decisions are often biased in favor of projects that are relatively familiar, and that these tend to be those that are relatively close (Coval and Moskowitz, 1999). Other studies have confirmed that the gravity effect is also important for international portfolio flows (Berkel, 2006), and for FDI flows (Talamo, 2007).

64. Following this approach, the chapter adopts a gravity-model framework to explain co movements in stock prices across countries. As mentioned above, a lack of suitable data precludes us from modeling capital flows directly. However, for those countries with a viable stock exchange, we can model the impact of these flows indirectly by measuring the extent of correlation between stock-market returns. In this regard, we build on the work of other authors that have also modeled stock-market co movements within a gravity framework (Biene and Candelon, 2005; Flavin and others, 2002).

The Model

65. The model specification essentially augments the standard gravity equation. As with the trade model, the variables influencing the degree of stock-market correlation include the standard geographical and historical factors, such as distance, common borders, common language and colonial links, while country size is replaced with stock-market capitalization. Details on data sources and country coverage are provided in the appendix, which also includes a more thorough discussion of the specification and econometric methodology chosen. However, it should be noted that our sample includes indices from 30 stock markets—including from the four CIS countries discussed in the previous section—which implies a total of 435 pairs of countries for each time period. Again, we use standardized weekly returns to eliminate any spurious correlation owing to non-synchronous trading hours, and we then calculate 435 bivariate correlations for each year (1999–2007), generating a potential panel of 3,480 observations.

66. Given this chapter’s focus on the extent of integration between CIS markets, the model explicitly tests whether the degree of correlation within this subgroup is systematically different from that of other countries. For each bivariate observation, therefore, the model includes a dummy variable that indicates whether both countries are CIS members. If the coefficient on this dummy were positive, it would suggest that CIS markets tend to be more closely connected to each other, compared to a randomly-selected country pair. In this case, therefore, we might conclude that the impact of Russia on other CIS markets is larger than expected—even after controlling for these markets' relative size and close cultural links. On the other hand, if the coefficient were negative, it would suggest that the degree of correlation is unexpectedly low; perhaps indicating de facto restrictions on the free flow of capital. The model further allows for the possibility that this CIS effect may have changed over time, so it also includes an interaction term to indicate whether the observation is from the second half of the sample. In addition, given that correlations might be higher for economies that are similar, the model includes a variable that measures the relative disparity between the two countries' GDP (as represented by the relative share of the smallest partner). Finally, the model allows for possible dynamics by including a time trend and lagged dependent variable.

67. Thus, the bivariate correlation (σij, t) is modeled as:

As mentioned, details on the explanatory variables are provided in the appendix, but it should be noted here for clarification that this is a panel-data model, so we allow for possible non-modeled country-pair effects via a fixed-effect term µij.


68. Coefficient values all display the expected sign and are generally significant. The positive coefficient on the time trend confirms that, worldwide, the degree of integration between markets has increased markedly over time—a coefficient of 0.037 corresponds to trend increase in correlation of about 1½ percent per year. The coefficient on the lagged dependent variable, on the other hand, suggests that the persistence of a shock to crosscountry correlation is generally small, but is still statistically significant. As expected, greater distance between markets tends to reduce the degree of co movement, whereas markets with a large joint capitalization tend to move together more closely. Once we control for physical distance, however, it appears that other “informational distance” variables—such as the presence of a common language—have little extra explanatory power.27 Finally, economies that are closer in size exhibit a higher degree of market correlation.

Kiviet-correctedTwo-step systemCorrected two-step
Dependent Variable: Bivariate Correlation (transform)
Lagged correlation (transform)0.2730.2220.222
(Log) Distance----0.054-0.054
(Log) Joint Market Capitalization0.0490.1010.101
CIS Country---0.0460.046
CIS interaction term-0.0480.0040.004
Common Language---0.0110.011
Share of Smallest Partner-0.3040.6680.668
t-statistics in brackets* p<.1, ** p<.05,*** p<.01
t-statistics in brackets* p<.1, ** p<.05,*** p<.01

69. Controlling for the above factors, there is little evidence to suggest that CIS countries are systematically more integrated than other country pairs. The CIS dummy is not significantly different to zero, and does not seem to have changed throughout the sample period. This suggests that the poor correlation between Russia and other CIS countries is broadly what we might expect given their histories, location, and relative size. On the one hand, the fact that CIS countries are relatively close, often share a common border, and enjoy a common language, all suggest that Russian investors should have a relative advantage when investing in other CIS countries (and vice versa). From this point of view, therefore, we would expect that their financial markets would be relatively closely integrated. On the other hand, stock markets in Ukraine, Kazakhstan, and Krygyzstan are somewhat undercapitalized and illiquid, so we would expect that developments in the Russian market might not flow easily into other CIS markets. Similarly, compared to Russia, the economies of the other CIS countries are somewhat small, and this disparity in size will further tend to reduce the degree of financial integration.

70. Despite their cultural and geographical proximity, therefore, the low degree of integration between Russia and other CIS countries is not surprising; and reflects the relatively small and financially underdeveloped nature of non-Russian CIS economies. It might be argued that formal and informal restraints on capital inflows may also play a role in these CIS countries. However, if that were the case, we would expect that the degree of integration and correlation would be systematically lower than that exibited throughout the rest of the world—i.e. we would expect the CIS dummy to be significantly negative. With no evidence of this, we can tentatively conclude that there is little to suggest that defacto barriers to inflows are any greater in the CIS than those existing elsewhere.

E. Conclusion and Caveats

71. In light of Russia's clear potential as a regional financial center, this chapter has taken stock of the current level of financial integration between Russia and other countries within the CIS. At present, although Russia appears to be well integrated with global capital markets, the process of financial development throughout the rest of the region is still at a very early stage. As a result, developments in Russia still have a somewhat limited impact on other CIS markets.

72. It should be noted, however, that some of the above results may reflect weaknesses in the available data. Measuring integration in any context is a challenge, and this is especially so in the context of the CIS. As mentioned, data on direct asset holdings is scarce and potentially misleading, and this chapter has therefore turned to more indirect evidence of financial integration—focusing instead on the correlation between equity returns. But even here, reliable equity indices are available for only a few CIS markets. Moreover, cross-border equity flows are only a small part of capital flows within the region. On the latter point, in an ideal world with perfect markets and frictionless arbitrage, capital should be fungible, and so equity-market returns should also be influenced by non-equity capital movements. However, CIS financial markets are far from perfect, and are often fragmented. Thus, it is possible that Russian capital flows, though unmeasured and channeled through international centers, may have a greater role than implied by the minimal co movement of CIS equity returns.

73. The degree of regional integration, and the potential for financial spillovers between markets within the CIS, will be an ongoing focus of staff analysis. Currently, staff analysis suggests that the low degree of integration between Russia and other CIS markets can be explained by the small size and financial underdevelopment of non-Russian CIS economies. As these economies grow and develop further, however, the results suggest that geocultural proximity of CIS countries to one another will result in a greater degree of integration. Looking forward, therefore, we can expect that Russia will have a greater and greater influence on other markets in the region.


Principal Components Analysis

74 Principal-component models have been applied across a wide range of empirical investigations. PCA has been used to analyze global and regional business cycles, including the co movement of consumption, investment and output (Cerqueira, 2006), but an approach closer to this chapter uses the methodology to isolate closely-integrated groups of countries that are driven by a single common factor (Kozluk, 2008; Fernandez-Izquierdo and Lafuente, 2004). We are interested in the extent of financial integration between Russia and other CIS countries, and whether they form a single, separately-identifiable group

75. A brief outline is provided below. With i = 1..N stock-market indexes and t = 1..T observations, our dataset can be represented by the (N × T) matrix X. This matrix can then be decomposed into: r = 1..R orthogonal common factors, represented by the (R × T) matrix F; as well as a (N × R) matrix of coefficients P; and a matrix of idiosyncratic components ε.

Every ith column of P represents a series of “loading” coefficients, each one of which captures the impact of factor r on series i. The reduced number of factors R is not known a priori, so we determine this number statistically: the standard approach is to look at the largest eigenvalues of the covariance matrixCx=1T1X'X , where the number of factors R is determined by the number of eigenvalues greater than one.28 The loading matrix P, then, is comprised of the R eigenvectors associated with these largest eigenvalues. The attached chart illustrates the eigenvalue distribution for our dataset, indicating that there are three key factors that explain the behavior of returns within the sample.

The Gravity Model

76. The chosen specification augments the standard gravity equation. Explanatory variables include: (i) distance, measured by the direct distance between the two country's capitals; (ii) cultural and historical factors, such as common borders, common language and colonial links; (iii) market size, measured by the level of stock-market capitalization; and (iv) relative economic size, measured by the share of the smallest partner in the two countries' combined GDP. Geographical and cultural data is sourced from the Feenstra, Markusen, and Rose (FMR) dataset, available at Data on GDP and market capitalization are from the World Bank's WDI database. The sample covers indices from 30 stock markets29 sourced from Morgan Stanley's US$ MCSI indices. As with the principal component analysis, the model uses standardized weekly returns to eliminate any spurious correlation owing to non-synchronous trading hours. This dataset allows us to calculate 435 bivariate correlations for each year (1999–2007), generating a potential panel of 3,480 observations.

77. The specification also takes care to allow for possible dynamics. Noting the well-established finding that correlations between markets have tended to increase over the recent decade, the model includes a common time trend. And noting also that market volatility and correlation tend to cluster over time, the model allows for serial correlation in the dependent variable. The chapter focuses on whether the degree of correlation between CIS countries is systematically different from that of other countries, and therefore include a dummy variable that indicates whether both countries are CIS members. The model further allows for the possibility that this CIS effect may have changed over time, so also includes an interaction term to indicate whether the observation is from the second half of the sample. In addition, the model considers that correlations might be higher for economies that are similar, and so includes a variable that measures the relative disparity between the two countries' GDP (as represented by the relative share of the smallest partner). Finally, as correlations are bounded in the interval [-1,1], there is a potential for bias when observations take extreme values, so the dependent variable is reshaped according to a Fisher-Z transformation:

The transformed correlation between country i and country j, (σij, t), is thus contained in the new interval (-∞, ∞).

78. In sum, the (transformed) bivariate correlation (σij, t) is modeled as:

As mentioned, this is a panel-data model, so the specification allows for possible non-modeled country-pair effects via a fixed-effect term µij.

Econometric Discussion

79. The inclusion of a lagged dependent variable raises a number of econometric issues. It is well known that dynamic panel estimators tend to be biased in the presence of fixed effects, as the estimated effect is correlated with the lagged dependent variable by construction, rendering the latter variable endogenous. The extent of the bias is of order O(1/T), and so is not necessarily an issue for panel-data models with a long time-series. But our sample is moderately short, with T = 9, so this bias is a possible concern. Therefore, it is expected that, with a simple pooled-OLS estimator, the coefficient on the lagged dependent variable (ρ) will tend to be biased upward. Sweeping away the fixed effect via the least-squares dummy variable (LSDV) estimator, on the other hand, does not get rid of the problem, but instead tends to bias the coefficient downward. Comparing the two estimators, therefore, provides a rough guide as to the plausible range for the unbiased estimate. From the attached table, it appears that the dynamic coefficient lies somewhere between 0.07 and 0.46.

Dependent Variable: Bivariate Correlation (transform)
Lagged correlation (transform)0.4550.0734
Common Language0.0168---
(Log) Distance-0.0335---
Share of Smallest Partner0.468-0.233
(Log) Joint Market Capitalization0.05680.0543
Other variables not reported
t-statistics in brackets* p<.1, ** p<.05, *** p<.01
t-statistics in brackets* p<.1, ** p<.05, *** p<.01

80. One way of addressing this dynamic-panel bias is to calculate the potential bias directly, and then correct the LSDV estimate. This is the approach suggested by Kiviet (1995) and Bruno (2005)30, and should provide unbiased estimates for the coefficients for all time-varying variables. Unfortunately, the LSDV estimator also sweeps away any time-invariant variables, such as distance, and so is unable to provide coefficient values for many of our geographical variables of interest. As an alternative, however, rather than purging any fixed effects, the “system-GMM” approach instead estimates the effects by building a series of orthogonal instruments from the entire dataset.31 This estimator can thus provide coefficient values for time-invariant variables, and so is more useful for our purposes. Nonetheless, this approach often provides results that are implausible precise,32 especially for the more consistent two-step estimator, so we augment our analysis by applying the Windmeijer (2005) finite-sample correction to the system-GMM standard errors.

81. Our preferred specification, therefore, is the two-step system-GMM estimator, with corrected standard errors. Our results are outlined in main text, and include three separate estimators: Kiviet-corrected LSDV; uncorrected system-GMM; and system-GMM with corrected standard errors. The coefficients from all three approaches are generally comparable, with the Windmeijer correction resulting an a substantial upward revision in standard-error estimates.


    Barbara Berkel, 2006.“The EMU and German Cross-Border Portfolio Flows,” MEA Discussion Paper 06110,University of Mannheim.

    Beine, Michel A.R. andCandelon, Bertrand2007. “Liberalization and Stock Market Co-Movement between Emerging Economies”. CESifo Working Paper No. 2131

    • Search Google Scholar
    • Export Citation

    Blundell, Richard andBond, Stephen1998. “Initial conditions and moment restrictions in dynamic panel data models,”Journal of Econometrics,Vol. 87(1), pp. 115–143

    • Crossref
    • Search Google Scholar
    • Export Citation

    Bruno, G.S.F.2005. “Approximating the bias of the LSDV estimator for dynamic unbalanced panel data models”. Economics Letters,87, 361–366.

    • Search Google Scholar
    • Export Citation

    Cerqueira, P.2004. “How Pervasive is the World Business Cycle?,” CEUNEUROP Discussion Paper No.22,University of Coimbra.

    Coval, J. D. andMoskowitz, T. J.1999. “Home Bias at Home: Local Equity Preference in Domestic Portfolios,”Journal of Finance,Vol. 54, pp. 2045–2073.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Dellas, Harris &Hess, Martin2005. “Financial development and stock returns: A crosscountry analysis,”Journal of International Money and Finance,Elsevier,Vol. 24(6), pp. 891–912.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Fernandez-Izquierdo, Angeles andLafuente, Juan Angel2004. “International transmission of stock exchange volatility: Empirical evidence from the Asian crisis,”Global Finance Journal,Vol. 15(2), pp. 125–137

    • Search Google Scholar
    • Export Citation

    Flavin, Thomas,HurleyMargaret andRousseau, Fabrice2002. “Explaining Stock Market Correlation: A Gravity Model Approach”.The Manchester School,Vol. 70, pp. 87–106.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Forbes, Kristin andChinn, Menzie2004. “A Decomposition Of Global Linkages In Financial Markets Over Time,”Review of Economics and Statistics, No.86, pp. 705–722.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Kiviet, Jan F.,1995. “On bias, inconsistency, and efficiency of various estimators in dynamic panel data models,”Journal of Econometrics,Vol. 68(1), pp. 53–78

    • Crossref
    • Search Google Scholar
    • Export Citation

    Koźluk, Tomasz,2008. “Global and Regional Links Between Stock Markets - the case of Russia and China,”BOFIT Discussion Paper 4/2008.Bank of Finland.

    • Search Google Scholar
    • Export Citation

    Martin, Philippe &Rey, Helene2004. “Financial super-markets: size matters for asset trade,”Journal of International Economics,Vol. 64(2), pp. 335–361.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Portes, Richard &Rey, Helene2005. “The determinants of cross-border equity flows,”Journal of International Economics,Vol. 65(2), pp. 269–296.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Giuseppina Maria Chiara Talamo, 2007. “Institutions, FDI and the Gravity Model,”mimeo, Università di Palermo.

    Wälti, Sébastien,2004. “The Macroeconomic Determinants of Stock Market Synchronization”EFMA 2004 Basel Meetings Paper,January2004.

    • Search Google Scholar
    • Export Citation

    Windmeijer, Frank,2005. “A finite sample correction for the variance of linear efficient two-step GMM estimators,”Journal of Econometrics,Vol. 126(1), pp. 25–51

    • Crossref
    • Search Google Scholar
    • Export Citation

Prepared by Andrew Tiffin (EUR).


This is a standard approach. Large swings in the US$ exchange rate may yield larger observed correlations, but should have less of an impact on the relative strength of bivariate correlations.


For each country series, the total variance explained by all three factors, known as the “communality,” is given by the squared sum of the loading coefficients.


Other geocultural variables, such as a common-border dummy or an indicator of past colonial ties, are also insignificant, once we control for distance.


A factor with an associated eignenvalue less than one would have less overall explanatory power than a single stock-market index by itself.


Sample countries include: Germany; Japan; United Kingdom; United States; Argentina; Brazil; Chile; China; Columbia; Czech Republic; Egypt; Hungary; India; Indonesia; Israel; Jordan; Korea; Malaysia; Mexico; Morocco; Philippines; Poland; Russia; South Africa; Thailand; Turkey; Venezuela; Ukraine; Kazakhstan; and Kyrgyzstan.


This methodology calculates the extent of the bias using a consistent, instrumental variables estimator, and then corrects the coefficient values derived from the LSDV estimator. Standard errors are bootstrapped.


We use the system-GMM estimator outlined in Blundell and Bond (1998), as implemented by the xtabond2 procedure in Stata.


Efficient GMM tends to underweight variables with high second moments, and also deemphasizes outliers.

Other Resources Citing This Publication