Appendix I: Multivariate Hierarchical Cluster Analysis
Abiad, A., D. Leigh, and A. Mody, 2009, “Financial Integration, Capital Mobility, and Income Convergence,” Economic Policy, Vol. 24, No. 58, pp. 241–305.
Abramson C., R. Andrews, I. Currim, and M. Jones, 2000, “Parameter Bias from Unobserved Effects in the Multinomial Logit Model of Consumer Choice,” Journal of Marketing Research, Vol. XXXVII, 410–426.
Bakker B. B. and A. M. Gulde, 2010, “The Credit Boom-Bust in Emerging Europe: Bad Luck or Bad Policies?” IMF Working Paper (forthcoming).
Ostry J., A. Ghosh, K. Habermeier, M. Chamon, M. Qureshi, and D. Reinhardt, 2010, “Capital Inflows: The Role of Controls,” IMF Staff Position Note SPN/10/04.
Schadler, Susan, Ashoka Mody, Abdul Abiad, and Daniel Leigh, 2007, “Growth in the Central and Eastern European Countries of the European Union,” IMF Occasional Paper No. 252, January 03, 2007
The author thanks Thanos Arvanitis for extensive discussions and guidance, as well as Bas Bakker, Holger Floerkemeier, Albert Jaeger, Yuko Kinoshita, Zuzana Murgasova, and Jesmin Rahman for useful comments and suggestions. The author is also grateful to Dustin Smith for his excellent technical assistance.
Abiad, Leigh, and Mody (2009) provide a useful analysis of the role of the “downhill” flow of capital in facilitating income convergence with Western Europe.
As some of countries in the sample were already in crisis during 2008, the year of 2007 represents a good proxy for the end of the boom cycle. Results are reasonably robust to the choice of the benchmark years.
While a number of alternative metrics (e.g., the extent of currency mismatches, the composition of capital inflows, and deviation of the current account balance from the norm) of external vulnerability could be studied, the two-variable grouping used here has important advantages of tractability and ease of interpretation.
It is important to stress heterogeneity of countries in the low vulnerability cluster as it covers countries of very different income level, ranging from Albania (10 percent of Euro Area level) to Czech Republic (over 50 percent of Euro Area level), which was recently recognized as an advanced economy. To highlight these critical differences, the low external vulnerability cluster is further clustered into two sub-groups based on the level of per capita income in 2003.
The two input variables enter cluster analysis in non-standardized way and thus differences in levels and variances influence variables’ relative importance in cluster determination. As a result, the stock vulnerabilities (external debt) dominate determination of dissimilarity between two individual countries, particularly for high vulnerability cluster. The role of the flow vulnerabilities (current account balance) is to separate the low and the medium vulnerability clusters.
As suggested by poor EBRD Transition Indicator and ICRG Institutional Quality ratings.
While economy’s openness is a very broad concept, for the purpose of this paper, trade and financial openness are defined as sums of exports and imports of goods and services and overall external assets and liabilities expressed as a ratio to GDP.
Bakker and Gulde (2010) show that countries where nominal exchange rate appreciated showed less signs of overheating and lower nominal wage increases. As a result, external competitiveness in these countries was better preserved.
It is important to recognize that, despite appearance, the six growth-vulnerability outcomes do not have a well-defined ordered structure. While it can be argued that the high growth/low vulnerability cluster is clearly “superior” to the low growth/high vulnerability cluster, the choice between, for example, low growth/low vulnerability and high growth/high vulnerability clusters is less obvious.
Structural policy measures are also likely to be important determinants of a growth model given their influence on the business environment and investment climate. These measures are omitted from the empirical analysis, however, on account of the difficulty quantifying their impact.
The reference cluster is chosen to facilitate interpretation of policy implications and has no bearing for the model estimation implemented by the generalized linear latent and mixed model procedure.
As in any choice model in a panel setting, the model may be subject to the presence of unobserved effects, including the presence of state dependence. However, this type of “country-branding” is unlikely to be excessively strong in the sample as countries frequently transited across clusters. To account for the unobserved factors, the model assumes presence of a random effect at the country level. Whether this approach is fully successful to account for the unobserved effects—completely eliminating potential for the presence of the parameter bias—is an empirical question that lies outside of the scope of this paper (see Abramson et al, 2000).
Jones and Olken (2005) show that changes in both exports-to-GDP and trade openness (exports plus imports-to-GDP) are positively associated with up-breaks in economic growth. Increasing importance of exports may signal the efficiency gains arising from cross-sector reallocation of factors toward the country’s comparative advantage. Increasing trade openness may signal increase productivity through increased scale economies, enhanced technology spillovers, and efficiency improvements. But it may also reflect growth-subtracting spillovers from buoyant domestic consumption.
Strictly speaking, the level and the composition of capital flows to a country is a joint outcome of external factors (as it reflects global liquidity conditions), macroeconomic policies (as it reflects investors risk perceptions), and structural characteristics of the economy (as it reflects availability of business opportunities and ease of doing business).
Relative risk ratio measures the risk of a country being in the current cluster relative to the exposure (one unit increase in the underlining variable): RR=P(GV=ij)/P(GV=23). The relative risk ratio of less (greater) than one suggests that the current growth-vulnerability cluster is less (more) likely than the reference cluster.
Slovakia exemplified a successful transition: the country, widely considered a difficult case in the 1990s, undertook sweeping structural reforms, ran high current account deficits, mostly financed by FDI, then saw a surge in exports, and current account deficits normalizing back to sustainable levels.
Bakker and Gulde (2010) argue that fiscal policy in some countries in Eastern Europe was too loose from a demand management prospective as spending was particularly high in overheating countries. Similarly, Rahman (2010) finds evidence of significant pro-cyclicality of the government expenditures in the region.
IMF staff estimates that the ratio of spending on public wages and social transfers to public investment averaged 3.5 in Slovakia during 1995-2002, compared to about 6.5 in Croatia during 2002-07.
With overly 90 percent of banking system being foreign owned, certain prudential measures (e.g., introduction of marginal reserve requirement rate) encouraged parent banks to fund their Croatian subsidiaries through beefing up their equity rather than by debt financing. This raised banking system buffers and, to some extent, moderated the pace of external debt accumulation. On the other hand, bank credit ceiling were only partially effective in limiting private sector credit growth as best corporate clients shifted to direct cross-border financing.
Development of the appropriate policy mix for individual countries lies outside of the scope of this paper. It will depend critically on the specific circumstances of each country, including potential constraints that may arise from the memberships in the EU and the WTO.
The use of targeted reductions of tariffs on intermediate inputs—an option for reducing input costs for tradables—in some countries may be limited by a need to harmonize the tariff structure with the European Union.
In countries where the social partnership with labor unions in the private sector is difficult to institute, wage cuts in the public sector would have to lead the way, envisaging demonstration effects for the private sector.
EBRD (2008) argues that increased government interventions are unavoidable in the aftermath of the crisis, but, in doing so, the focus should be on preserving market incentives and transparency.
See Everitt and Dunn (1991) for a detailed discussion on the use of hierarchical clustering in applied multivariate data analysis.
Alternative linkage methods (e.g., complete, centroid, and group-average) were also tested and generally produced similar groupings of the countries.