3 Data and Data Analysis The yearly data for the period 1970 - 2008 used in this study are the real GDP per capita, consumption expenditures, exports, and imports, all at constant prices in US Dollars 1990 extracted from National Accounts Main Aggregates of the United Nations Statistical Database online. All the data were then reconverted to prices of 2008, the last year in our sample for which data are available. We use the real GDP per capita to investigate business cycles synchronicity whereas exports and imports along with the real GDP were used to compute the degree of openness for each country. Although financial openness is also important, it is extremely difficult to find complete data on financial flows for the 217 countries included in our study. Chinn and Ito’s (2008) index of financial openness (Kaopen_2007.xls) contains too many missing information in between the years for so many countries to make it usable to our end. The monthly spot oil price data (West Texas Intermediate) were downloaded from the Dow Jones Industrial Average website and were then expressed in yearly average prior to their conversion in real terms. Data on consumption per capita growth, trade openness, and changes in oil prices were used to search for an explanation of the business cycle synchronicity.
The data on real GDP per capita were used to classify the countries as per the World Development Indicators (WDI) published by the World Bank. The World Bank classifies countries as High Income OECD (HIC_OECD), High Income non-OECD (HIC_other), Upper Middle-Income (UMC), Low Middle-Income (LMIC) and Low Income (LIC) as a proxy for the relative degree of economic development of countries. Accordingly, countries are grouped as LICs if their income per capita is $825 or less; LMCs, $826–3,255; UMCs, $3,256–10,065; and HICs, $10,066 or more. The World Bank views low-income and middle-income economies as developing economies.4
After classifying the data, we end up with an unbalanced panel of 57, 60, 38, 23, and 30 countries in the respective categories. We decomposed each series into a trend and a cycle using a penalty parameter of 6.25 for the Hodrick-Prescott filter as suggested by Ravn and Uhlig (2002) for annual data. The output gap was computed as the ratio of cycle over trend. A negative (positive) value indicates actual output is below (above) trend. As a prelude to the empirical analysis, we computed the correlation coefficient for each pair of countries within each income category. The results are presented in Table 2, indicating that, with the exception of LMCs, there are more positive than negative correlation of the business cycles within each group. For example, we found positive correlations as a share of total correlation within LICs, LMCs, UMCs, HICs-OECD, and HICs-other to be 60, 48, 57, 83, and 61 percent, respectively. The minimax and the maximax of the positive correlations vary widely in range [0.06, 0.88], [0.06, 0.95], [0.07, 0.87], [0.25, 0.84], and [0.004, 0.77], respectively for each income group. In his classical business cycles in Latin America, Mejía-Reyes (1999) considers a Pearson’s corrected contingency coefficient (CCcorr) less than 40 percent as a clear sign of “low” association, between 40 and 60 percent as “mild”, and greater than 60 percent as an indication of strong association of the cycles. Along the same line, we attest that countries of similar levels of income are on average characterized by more positive than negative association of their cycles and the tendency towards synchronicity of the group irrespective of the correlation magnitudes of the individual pairs range from mild to strong. The HICs-OECD is the group with the largest share of positive correlation. Thus far, the data seem to suggest that within groups there is more tendency towards synchronization of the cycles than not, but nothing can be said across groups at this point.
3.1 Empirical Results
3.1.1 Business Cycles Synchronicity within Groups
Do countries of similar income levels on average follow similar business cycles? We answer this question by examining the synchronicity measure proposed by Mink et al. (2007) depicted in Equations (5) and (6). We used different criteria to choose the reference cycle. We select a country of reference for each group on the basis of the minimum real income per capita gap (Min_Gap). Our contention here is that the closer a country’s output is to its potential level, the more it tends to trade with the rest of the world, as a result a number of countries tend to see an increase in their own levels of output for that same reason. We also computed the average and the median growth rate for each country to select within each income category the country with the maximum of the averages (Max_Average), the median of the averages (Median_Average), and the maximum of the medians as reference cycles. The median of all observed output gaps was also considered following Mink et. al.’s strategy. Since this time series could not be associated to any country in particular, we refer to it as a fictitious country (Fictitious). As can be seen from Table 3, despite the comprehensive nature of our selection criteria for the reference cycle, none of the criteria selected the US as reference. Hence, we analyze the linkage between US cycles and cycles of other OECD countries on an ad hoc basis. As could be expected, Table 4 shows that the US is, behind the fictitious country, the country with tighter co-movement of cycles with the remaining countries of the group than any other countries selected through the battery of computations used.
We proceeded in four steps to arrive at the SR for each group of countries in the pairing of the reference cycle and each individual cycle.5 First, we counted the number of synchronized cycles (+1) and divergent cycles (-1) and obtained a total equal to the number of cycles over the years for each pair of countries. Second, we computed the share of +1s of the total. Third, we produced a count of the shares of +1s greater than 0.50. Fourth, we calculated the SR as the count of shares of +1s greater than 0.50 over the total (+1s and -1s).
For the multivariate formula, we computed the horizontal average of the +1s and -1s stemming from the matching of the reference cycle with the individual cycle at every point in time. This calculation produced a column series of 39 observations between -1 and +1. Positive values indicate tendency towards synchronization whereas negative values indicate just the opposite. The multivariate SR is computed as the count of the positive averages over the total of all averages (positive and negative).
We summarize in Table 4 the SR for each income group for both the bivariate and the multivariate framework.6 The results are quite similar. We find that the fictitious country represented by the median of all observed output gaps as the reference cycle is strongly synchronized with the individual countries of each group irrespective of the income levels taken into consideration.On average LICs (0.51, 0.57), LMCs (0.51, 0.62), and UMCs (0.54, 0.57) are characterized each by mild association of the cycles, whereas HICs-OECD (0.78, 0.65) and HICs-other (0.63, 0.63) exhibit each stronger association of the cycles with some variations within pools depending on the criterion used for the reference cycle. The averages for HIC-OECD do not change much when the US is excluded from the calculation (0.75, 0.65). These results indeed suggest that countries of similar income levels do share similar business cycles.
3.1.2 Business Cycles Synchronicity across Groups
To determine whether the synchronization of cycles that we observe within groups for countries of relatively similar degree of economic development also extends across groups, we pooled all the countries with ratio of synchronized cycle with the reference country greater than 0.50 irrespective of the income group they belong to. We ended up with 99 countries for the minimum gaps, 169 for the medians, 96 for the maximum averages and the maximum medians (MAMMs) with Bhutan, 101 for the MAMMs with Vietnam, and 112 for the median averages. For each of these pools of synchronized cycles, we calculated the bivariate and the multivariate SRs using the country dictated by the minimum gap criterion as the reference cycle. We also used the US as an ad hoc global reference for each pool, and whenever the minimum gap criterion produced a country that we believed was not likely to make sense as a reference, we experimented with the countries that served as references in the original bivariate analysis by selecting the one with the minimum average output gap (e.g., the minimum of the minima criterion). For example, the pool of median averages selected France as the reference, but the previous reference cycles with United Kingdom (UK), Liechtenstein, Poland, Paraguay, and Equatorial Guinea pointed to Equatorial Guinea as the reference when the minimum output gap is identified for these six possible candidates. In cases like this one, we chose UK as another ad hoc country, since it is the largest economy of choices available. Similar treatments were given to other pools. Table 5 supplies details about the selection process of the reference cycle for each large pool of countries, which is representative of the dynamics governing the world economy. We chose the reference cycles for 5 possible groupings of the world.7 Australia, France, and the US were selected for the Minimum Gap pool; Ireland and the US for the MAMM with Vietnam pool; Ireland, Monaco, and the US for the MAMM with Bhutan pool; UK, France, and the US for the Median Average pool; and France and the US for the Median pool.
Table 6 presents the results pertaining to the synchronization of the business cycle at the world level for various representations of the world. For the bivariate and multivariate formula respectively, we find the SR to lie between 0.60 and 0.65 for the Minimum Gap pool on average, 0.61 and 0.59 for the MAMM with Bhutan pool, 0.61 and 0.62 for the MAMM with Vietnam pool, 0.62 and 0.72 for the Median Average pool, and 0.65 and 0.70 for the Median pool. Adhering to the criteria of determination, low (SR < 0.40), mild (0.40 ≤SR≤ 0.60), and strong (SR > 0.60) association of the cycles, there is ‘strong’ evidence that a common world business cycle exists despite income per capita differences across countries. This finding is quite interesting, but posits quite an intriguing question: which of the countries is the driver of the world business cycle? Well, we know for sure that it is not any of the less-developed or developing countries. Since we had chosen the US as an ad hoc reference cycle, it bears asking how this selection influenced the results. As can be seen from Table 6, the average SR across the five pools when the US is left out lies between 0.62 and 0.65, which is still a strong association of the cycles as per the benchmark. Therefore, our choice did not taint the results. It is difficult to compare the countries that were selected as references to determine which one emerges as the principal driver of the world business cycle, because the countries are not the same across pools. It is worth noting, however, that in each pool where both France and the US were present, the SR with France was on average greater than that of the US, thereby suggesting a greater role of France at the world stage, which is contrary to what one would expect when compared to the US, the largest economy of the world.
We brought further robustness to the finding of the common world business cycle by focusing solely on the countries originally determined as references in Table 3. We therefore searched for the reference among the references. The minimum average gap criterion of the 17 countries had selected Vietnam as the global reference cycle, but this choice was not used since Vietnam is a LIC. As a result, we use the US, UK, and China (for its growing importance) which were part of this pool as ad hoc references. Since France was never selected by the many criteria used in the original pairing of the countries, it was this time incorporated as an extra ad hoc country for comparison with the US, upon the finding that France might be the driver of the world business cycle. We find the bivariate and multivariate SRs to be respectively (0.63, 0.61) for the US, (0.56, 0.47) for the UK, (0.53, 0.53) for China, and (0.59, 0.63) for France with the rest of the 16 reference countries. With 10 of these countries the US and France equally share a common cycle more than 50 percent of the time, whereas it is 9 for both UK and China. These results suggest that France is equally important as the US in leading the world business cycle.
3.1.3 Did The 1990s Onward Make a Difference to Business Cycle Synchronicity Within and Across Income Groups? The 1990s have always been regarded as a turning point in international trade and finance. The Uruguay Round completed in 1994 was the last leg of trade negotiations and administrative reforms under the General Agreement on Tariffs and Trade, or GATT that cut tariff rates around the world, a whopping 40 percent from developed countries. It was anticipated that such cuts would produce substantial increase in world trade, but only a small increase was observed since average tariff rate had only fallen from 6.3 to 3.9 percent (Krugman et al., 2012, p.238; Schott, 1994). The creation of the World Trade Organization (WTO) in 1995 to replace the GATT aimed at fostering further trade among nations in goods and in services by implementing dispute settlement procedures to resolve trade disputes in a timely manner. According to the Organization of Economic Cooperation and Development (OECD), the world economy was expected to gain more than $200 billion annually once the Uruguay Round agreement was fully implemented, an estimate, of course, that is not devoid of controversy.
Added to the creation of the WTO, progresses in communications and transportation systems that started since the late 1970s have facilitated capital market integration as more countries embraced deregulation to foster investment at home and to compete with others in attracting more capital. Not only did capital flow to many developing countries, technology transfers also took place as multinational corporations established subsidiaries in host countries. The resulting effect was higher level of consumption and income. The 1990s onward was considered as the rebirth of globalization, which was interrupted by the First World War and other subsequent events until the end of the second oil shock of 1979. It is often reckoned that the world has become more integrated than before as a result of the globalization process. The dissenting view, however, is that this level of integration has made countries more vulnerable to financial crises, and the poorest countries are the most affected since they are least likely to be attractive to foreign investors due to existing political strife and lack of proper infrastructure at home. The antiglobalization movement also contends that redistribution of income from the rich to the poor is not at all possible in a globalized world since governments in less-developed countries do not have the means necessary to make the rich pay taxes since they can relocate their capital to other low-tax countries at little or no cost.
Regardless of the merits of the ongoing debate as to whether the costs of globalization exceed its benefits, one issue is certain: the world has become more vulnerable to shocks than before. The recent housing crisis that originated in the United States in 2007 that triggered financial crisis in many countries and a worldwide recession is a consequence of the tight linkages of modern economies. It also appeared that even countries with sound macroeconomic fundamentals could not escape since investors had to retire their capital from these countries to mitigate losses they had suffered in large international financial markets. Our contention in this paper is that if it is true that globalization is welfare improving, growth in countries with, say, high income per capita shall spillover countries with low income per capita through international trade and foreign direct investment (FDI) effects. We shall therefore observe a more synchronized response of income per capita to shocks across countries for the 1990s onwards than for the two previous decades, which would dictate that globalization has indeed benefited countries. To assess this claim, we split the full sample into two sub-samples covering the period 1970-1989 and 1990-2008 and investigate the business cycle synchronicity across and within income groups according to the five selection criteria of the reference cycle presented in Table 8. The maximum average and the maximum median growth rate criteria select the same countries as reference cycles, though these countries differ across sub-periods.
We present in Table 9 the synchronicity of cycles within income groups based on the various reference cycles for each subsample. The results are presented side by side to allow for comparison. We find that, on average, there is a mild association of the cycles for LICs, LMCs, and HIC-OTHERs on the basis of the bivariate synchronicity measure regardless of the sample period under consideration. UMCs lie at the border of mild and strong associations whereas HIC-OECDs are characterized by a strong association of the cycles for both sub-samples. The multivariate measure of synchronicity in most cases shows greater association than the bivariate measure. The results do not differ too much when the synchronicity measure based on the median output gap, which is the largest in each income category, is discarded from the computation of the global average. The most important result portrayed in Table 9 is that the synchronicity measure for the period 1970-1989 is superior to that of 1990-2008 (the globalization wave) for all income groups but HIC-OECDs. This is quite surprising since one would expect tighter linkage of the cycles within income group due to trade liberalization, FDI, and globalization overall.
What can possibility explain the lesser synchronicity of the cycles from the 1990s onward? One feature of the business cycle linkages uncovered in Table 9 is that only cycles of countries with similar income levels are being matched. It is quite possible that these countries do not trade much with each other since they are all trying to reach the larger markets of Europe and North America, for example. The 1970-1989 period which precedes the Uruguay Round could perhaps represent a period of more trade for countries of similar income since the concessions on tariff an non-tariff barriers from the developed world were not yet made. Less-developed and developing countries had no other choices than to foment trade within their own groupings. Well, this argument would be a hard sell because when we peruse Tables 10 and Table 11 where countries with synchronous cycles are grouped as a possible picture of the world economy irrespective of their level of income, the outcome is pretty much similar.
Table 10 shows the selection of the reference cycles when we pool countries with synchronous cycles based on the minimum output gap, the maximum average, maximum median, and median average growth rate criteria. We used the Minimin criterion to choose the global reference cycle in each large pool. However, two countries kept reappearing as world reference cycles; Luxembourg for the period 1970-1989 and Sweden for the period 1990-2008. Since these countries were too small to be drivers of the world business cycle, a global reference other than these two countries was then selected according to the same criterion. If in that round we did not find a high income country, we applied the Minimin criterion to OECD countries but Luxembourg or Sweden. In addition, we selected France and the USA as ad hoc reference cycles. This information is fed into Table 11, which unequivocally shows that there was more synchronicity of the cycles in the 1970s and 1980s combined than in the 1990s and 2000s combined, despite the globalization wave of the latter period. Further analysis of the data using the reference cycles of each income group to investigate synchronicity with major international players such as the United States, United Kingdom, China, and France reveals on average the association of the cycles range from mild to strong for period 1970-1989 and from low to borderline strong for the period 1990-2008, as per Table 12. These results do not support either the view that globalization has led countries to move along the same wavelength. Table 12 is also quite intuitive, for example, it shows that countries with cycles linked to Qatar (a reference cycle for HIC-OTHER) will likely react similarly to economic disturbances affecting the US, UK, China, and France whereas countries with cycles linked to Democratic Republic of Congo or Haiti will not share such similarity.
We believe Todaro and Smith’s (2003) introduction to globalization contains a more plausible explanation of the weaker synchronicity of the cycles supported by the data for most income groups or makeup of the world with the exception of the OECD countries. Todaro and Smith observe that while FDI was flowing in promising developing areas such as Asia and part of Latin America, foreign aid has been declining substantially over the years for the majority of less-developed and developing countries, including those living in abject poverty. In their view, although it is true that developed countries have become more open due to globalization; widespread protectionist policies are still being practiced by the most advanced OECD countries in agriculture and textiles where the less developed countries could enjoy a competitive advantage. It bears acknowledging that for Todaro and Smith openness to globalization itself does not inevitably forestall growth, at least among more developed countries. In their view,globalization has been a key to rapid growth in countries such as South Korea, China, India, among others. But globalization does carry the seed for inequality to accentuate across and within countries as some people and countries may not receive their fair share of economic windfall. Well-known examples are the growing disparities between coastal and inland China, between countries in Africa and countries in Asia or Latin America, Dominican Republic and Haiti, among others. Another factor that we think is relevant in understanding the cross period synchronicity is immigration policies erected by developed countries that started in mid 1980s but further intensified in the 1990s onward. Impediments to labor mobility across countries (not between developed countries)) have been intensified in contrast with the 1970s and the early 1980s. During these times it was relatively easier for workers from less developed countries to seek economic refuge in abundant countries so that they could provide for their relatives back home and even bring them along after certain time. All these factors, in our view, may be contributors to the underlying relative weak association of the cycles for the period 1990-2008.8 Further perusal of Tables 9, 11, and 12 also provides further convincing evidence that France is indeed an important driver of the world business cycle. The synchronicity measure with France as an ad hoc global reference is greater than that with the United States, whether we use the bivariate or the multivariate formula or whether we use full sample or subsamples. This finding is quite puzzling since France is not even the second largest economy of the world after the United States. Is there an explanation for such surprising results? Well, it is quite possible that as the largest economy, the US cycle may move concurrently with cycles of other OECD countries that enjoy similar income levels and perhaps similar habit formation, whereas France as a former colonist might enjoy concurrent cycles with most less-developed and developing countries, and for being a European country, proximity to other European Countries and Africa might place France in position to trade with the rest of the world at a lower cost than the US. One way to capture this spatial explanation of the tighter business cycle linkage is by introducing leads and lags into the analysis.9 We formally test whether it makes a difference in the results if we assume that disturbances affecting the rest of the world take some time before they can have some effects on the US or whether the pulse of the US economy is an indication of trouble or opportunities to come for the rest of the world. We calculate both measures of synchronicity for the full and subsample periods and present the results in Table 13. Unequivocally, the results show that US synchronicity measures are far superior to those of France when lagged cycles are matched with the contemporaneous cycles of the rest of the world. These findings are by and large in accordance with the chronology of modern economic crises, safe for those originated in the US including the most recent financial crisis.
3.2 Is There an Explanation for the Synchronicity of Business Cycles across Income Groups? Thus far, we have shown that despite differences in income levels which would normally dictate asynchronous business cycle, there is a common world business cycle and economies of the US and France play a pivotal role as drivers. This finding raises a fundamental question as to what underlies the commonality in the world business cycle. We took two approaches to investigate the role that globalization might have in increasing economic interdependence among countries. We use dynamic panel data model with synchronized output gap as the dependent variable and panel logit regression with the synchronicity measure as the left-hand-side variable to gauge the influence that growth in consumption per capita, real oil prices, and increased trade openness might exert on these two variables. The theory is clear on the choice of the explanatory variables. Consumption is the largest component of aggregate demand, changes in tastes and preferences, real incomes, real interest rates, expectations about future incomes and future prices do impact real output.10 It is also the case that the more open an economy is, the more vulnerable it is to international shocks. Gains from trade accrue to consumers in terms of lower prices for their consumption products and availability of wider range of differentiated products. Therefore, there is interaction between these two variables in explaining synchronicity. Oil remains one of the most important sources of energy in the production of output. Shocks to real oil prices do have severe repercussions on real income regardless of whether a country is a net exporter or a net importer. Our contention is that ups and downs in the level of economic activities across countries of different income levels can be synchronized if they revolve around these global variables (See Baxter and Stockman 1989, and Baxter 1991, 1995). However, there is contention in the literature whether standard multicountry business- cycle models can capture the relationship between trade and business cycle comovement (see Kose and Yi, 2001, 2006; and Kouparitsas, 1997a, 1997b).
We attempt to model the explanation using three main panel regression models: Panel Logit, Panel Linear estimation and Dynamic Panel estimation. For each model we used both the full set of observations (1970-2007) as well as subsamples (1970-1989 and 1990-2007)
For the linear and dynamic panel data models, we used the output gap of the countries with SR above 0.50 as per the criteria listed in Tables 3 and 4.
3.2.1 Panel Logit Estimation
For the panel logit regression, we convert the synchronicity measures of Equations 5 and 6 into a binary variable such as:
The logistic probability function is given by:
We rewrite this equation for estimation purpose as:
Where, OPi,tis real oil prices, Ci,t is consumption and TRi,t is the trade openness of country i at time t. All variables are expressed in percentage change using natural log differences for the first two variables. TRi,tis measured as the sum of exports and imports as a share of real GDP for each country. The growth in TRi,tcaptures the increase in trade openness as opposed to just the openness of a country over time. Hence, we ask to what extent further openness of domestic economies to the world is likely conducive to synchronization of business cycles across income groups.
For the lagged values model:
For both models, the Hausman test is employed to determine whether a model that allows for the possibility of a correlation between unobserved country characteristics and the predictor variables (fixed effects) or one that assumes that the variation across countries is random and uncorrelated with the predictor variables (random effects) is more appropriate. Results of the tests showed that the differences across countries have no influence on the dependent variable, hence fixed effects estimation was applied to all the regressions
The results from the robust fixed effects estimation presented in Tables 14 and 15 indicate that only real oil prices is consistently significant, whether we include lagged variables or not as explanatory variables. The lagged values of real oil prices appear to have the most significant explanatory power in determining synchronicity of output gap across countries and methods. Also, in all models estimated, the sign was mostly positive suggesting that real oil prices play an important role in shaping the dynamics of output gaps.11