اندازه گیری یکپارچگی بازار مالی در دراز مدت : آیا U شکل وجود دارد؟

Using long time series for sovereign bond markets of fifteen industrialized economies from 1875 to 2009, I find that financial market integration by the end of the 20th century was higher than in earlier periods and exhibited a J-shaped trend with a trough in the 1920s. The main reason for the higher financial integration seen today is the recent extensive globalization. Around the turn of the 20th century, countries frequently drifted apart. Conversely, in recent years, the bond markets of most countries have moved together. Both policy variables and the global market environment play a role in explaining the time variation in integration, while “unexplained” changes in the overall level of country risk are also empirically important. My methodology, based on principal components analysis, is immune to outliers and accounts for global and country-specific shocks and, hence, can capture trends in financial integration more accurately than standard techniques such as simple correlations.

The extent of international financial integration has important implications for economic theory and policy debates. The relative degree of financial integration during the two capital market booms, before World War I and after the collapse of the Bretton Woods system, remains subject to disagreement. Typical measures of integration include proxies for intensity of legal restrictions on cross-border capital flows, price-based criteria, and quantity-based criteria.1Quinn (2003) argues that financial markets were more integrated during the pre-WWI era, whereas Mauro et al. (2002) find that they are more integrated post-Bretton Woods. Others, including Obstfeld and Taylor, 2003 and Obstfeld and Taylor, 2004 and Goetzmann et al. (2005), argue that financial markets demonstrate a U-shape and hence an equal amount of integration before 1914 and after 1971. It is important to know which period has been associated with a higher degree of capital market integration because these periods differ drastically in terms of the economic environment and policies. I argue that mixed results in the literature are a result of studies using different methodologies and the failure to differentiate between global or country shocks. To address these concerns, I propose a systematic methodology based on the method of principal components that has several advantages. First, it accounts for several dimensions of integration including market co-movement and segmentation, within a straightforward statistical methodology that is widely used in microeconomic research. Second, it is robust to the presence of outliers or heavy-tailed distributions. Third, the current method is robust to the choice of a reference country (such as the United States or Great Britain). Fourth, the methodology has a clear theory-based interpretation. Finally, using this method I was able to account for global shocks while several other methodologies spuriously interpret large global shocks as integration because common global shocks make financial variables move together. The focus of this paper is on financial markets integration from the prospective of investors in financial assets or financial arbitrageurs, as opposed to integration of commodity markets or markets for real assets. Standard no-arbitrage theory predicts that, when investors in financial markets are neutral to exchange (or currency) risk and market frictions are negligible, free international capital flows (financial arbitrage) result in the Uncovered Interest Parity (UIP) condition. This result implies that similar assets in different locations have the same expected nominal rate of return regardless of exposure to the exchange risk. However, literature has accumulated abundant evidence of non-negligible exchange, default, and political risk across countries and over time. Further, these risks may result in persistent and volatile risk premia and hinder the ability of countries to tap into international capital markets. If these risks depend on, or are correlated with, legal restrictions to capital flows or the underdevelopment of financial markets, international arbitrage opportunities may also be limited. These combined factors reduce financial integration. I do not expect to find perfect capital mobility anywhere in history given all the evidence from the literature. Rather, I intend to concentrate on a weaker notion of integration characterized by smaller and more stabile risk premia that would result in a higher comovement (but not necessarily equalization) of a country’s financial returns. 2 Even if a greater comovement is driven by common global shocks, the fact that such shocks propagate across countries and these shocks are frequent might also be a sign of greater interconnection between individual economies ( Bordo et al., 2001). I also verify how comovement of returns has changed over time conditional on time-varying determinants of the risk premia. My empirical methodology is based on principal component analysis (PCA). The PCA is a non-parametric empirical methodology used to reduce the dimensionality of data and describe common features of a set of economic variables. This method transforms the observed data vectors into new variables referred to as components, which are linear combinations of the original data that maximize variance. 3 The goal of the method is to capture most of the observed variability in the data in a lower-dimensional object and, thereby, filter out noise. Often, a single component summarizes most of the variation of the original data. I argue that the “first” principal component (with the components ordered according to how much of the data-variation they capture) has a natural interpretation when the PCA is applied to a comparable series (prices, returns, etc.) across markets. When the observed economic variables have a high signal-to-noise ratio, which would be the case under economic integration, a few principal components with larger variance would capture the dynamics that will be informative of the extent of market integration. This result should also be consistent with the standard no-arbitrage theory that implies if the majority of countries are integrated into the world financial markets, their interest rates move together with the world rate. The proportion of total variation in individual returns explained by the first principal component serves as an index of integration. I estimate the index of integration over the long-run dynamically via rolling windows. Using the dynamic PCA in the context of market integration is an innovation of this paper as it reveals important trends in integration and country- or group-specific shocks hidden when a single estimate corresponding to the entire period is reported. 4 Furthermore, using individual country weights (called “loadings” in PCA) on the first principal component, I develop two complementary indices of segmentation, which summarize country or group-specific effects, to investigate possible reasons for the changing degree of integration. Being a price-based measure, the PCA-based index of integration has several practical advantages. The first advantage is the reliance on historic price data, which has a better quality compared to the volume of capital flows used to construct quantity-based measures (Obstfeld and Taylor, 2004). Second, issues that plague price-based measures over the short-term would not affect the results as much when we look at very long time periods.5 Finally, the long historic context allows researchers to compare the relative degree of integration in the present and during previous periods rather than to test for “full integration” or “no integration” which necessitates the choice of some questionable benchmarks and creates difficulty when the absolute values of these measures have to be interpreted. My primary data are monthly series of sovereign bond yields from 1875 to 2009 available in the Global Financial Database (GFD) ( Global Financial Data Inc, 2002). The sample includes 15 economies whose sovereign debt was continuously traded in a major international financial center (London) and was available in other locations as early as the mid-19th century. 6 Historically, sovereign bonds have been the most actively traded segment of financial markets. 7 In contrast to the stock market indices, the characteristics of the underlying instruments in bond markets (maturity, coupon payments, the identity of the issuer, etc.) are similar across countries and over time. This comparability makes these data attractive for long-term study of the dynamics of financial integration. I prefer using bonds payable in national currency and do not convert the data into a single currency or constant prices because I take the prospective of an investor in international financial markets and want to analyze all possible reasons for changes in comovement (currency and country risk, cross-border frictions, and other limits to arbitrage). Prior to World War II, the database reports multiple bond series, while the unified series became available from the 1920s onward. I carefully select those early bond series that are most comparable with the subsequent series in order to minimize breaks and make the long-term series consistent. I estimate the index of integration over 1875–2009 using a relatively wide rolling window of 156 months (13 years), which makes the results relatively immune to short-term noise and to conditional heteroskedasticity in returns. Over the very long-term, the evidence points to higher financial market integration at the end of the 20th century compared to earlier periods. Therefore, the integration followed a J-shaped trend with a trough as early as the 1920s rather than a U-shape documented by Obstfeld and Taylor (2004) and others. This pattern is also confirmed by time-series regression analysis. According to the indices of segmentation, around the turn of the 20th century countries frequently drifted apart while in the recent years the most countries move together. This uncovers the changed nature of shocks prevalent in the two eras of globalization. Finally, I find that both policy variables (average inflation, government deficit, and, in Bretton Woods period, the exchange-rate regime) and the global market environment (proxied by average trade openness) play a role in explaining the time variation in the index of integration, while “unexplained” changes in overall level of country risk are also empirically important. For illustrative purposes, Fig. 1 presents a smoothed estimate of the index of integration over 130 years. As illustrated in this figure, and consistent with the literature, the dynamics of integration have not been even throughout history. Specifically, integration grew from the late 19th century up to 1914, when World War I began. Following that period, the trend in integration turned negative and reached a historic low around the time of the Great Depression in the 1930s. The partial recovery of international financial linkages in the 1920s is invisible in the graph since it was short-lived. After World War II, the trend in integration turned positive and continued its upward crawl almost uninterrupted to a historic high by the end of the 20th century. It is clear that integration reached levels comparable to the Gold Standard era as early as in the late 1960s. This evidence is consistent with the view expressed by Kose et al. (2009) that evaluating integration based on the de jure government restrictions on capital flows may be misleading because capital controls could be avoided, as was the case in the 1960s, which was characterized by rapid expansion of trade, offshore banking, and Eurocurrency markets.8 This result casts a serious doubt on the Macroeconomic Trilemma paradigm: clearly, during the Bretton-Woods era of tight controls some capital flows were possible in spite of fixed exchange rates and monetary policy that was designed to keep rates fixed. Full-size image (25 K) Fig. 1. Long-run trend in bond market integration, 1900–2008. Notes: Estimates of the proportion of variation in bond returns explained by the first principal component smoothed using the uniformly weighted moving average smoother. Government bond returns are in levels. In-sample countries are Belgium, Denmark, France, Germany, Italy, Netherlands, Norway, Spain, Sweden, United Kingdom, and United States. The estimation of the component is performed as the centered rolling window with the bandwidth of 156 months. Figure options My measure of integration does not rely on any particular underlying model; nevertheless this measure has a clear theory-based interpretation. In the case of countries’ interest rates, the first principal component may naturally be interpreted as the “world” rate. Fig. 2 illustrates the individual country yields used in this study with the values on the left axis and the estimated world return on the right axis. This figure demonstrates that the world return captures the dynamics of individual rates remarkably well while it is not affected by country-specific shocks. This is because the world return uses country weights based on features of the data and not subjectively asserted by the researcher. This proxy also reflects the continuously changing individual counties’ influence on the world return. For example, when a country shuts down its capital markets, its domestic financial market behavior diverges from the world market. First principal component will immediately capture this decoupling, which will show up as a lower loading of this country. Full-size image (39 K) Fig. 2. Yields on sovereign long-term bonds and the estimated “world” return, 1875–2002. Notes: The graph depicts historical monthly series for the yields on long-term government bonds issued by industrialized economies (thin solid lines, left axis) and the estimated \world" return (thick dashed line, right axis). The following abbreviations for the country names are used at the graph: AUT for Austria, BEL for Belgium, DNK for Denmark, FIN for Finland, FRA for France, DEU for Germany, ITA for Italy, JPN for Japan, NLD for the Netherlands, NOR for Norway, ESP for Spain, SWE for Sweden, SWI for Switzerland, GBR for United Kingdom, and USA for the United States. World is the estimate of the first principal component using all countries (the \world" return). The estimation of the component is performed as the centered rolling window with the bandwidth of 156 months. Figure options The remainder of this paper is organized as follows. The following section discusses related literature. In Section 3, I lay out a conceptual framework that motivates the empirical analysis and helps interpret the results. Section 4 discusses the methodology used to quantify integration in various markets. Section 5 describes the data and the pattern of integration. Section 6 offers some explanations of the observed pattern. Finally, Section 7 concludes.

I propose a systematic methodology based on the method of principal components to quantify economic integration, explore its dynamics, and capture the episodes of market segmentation. This method overcomes the limitations of conventional approaches. Despite its computational simplicity, the suggested methodology is quite general and applicable to a variety of markets. I explore why the existing empirical literature, which relies on comovement of economic variables, lacks consensus on whether the highest degree of integration was achieved before World War I under the Gold Standard or by the late 20th century. I argue that a conventional measure of comovement, the coefficient of correlation, has limited applicability as a measure of economic integration. Based on the suggested methodology I find clear evidence of higher financial integration at the end of the 20th century compared to the earlier periods. Time-series regressions show that both policy variables (average inflation, average government deficit, and the fixed exchange-rate regime during Bretton Woods) and the global market environment (approximated by the average trade openness) played a role in explaining the time variation in the index of integration. I also find that “unexplained” changes in overall level of country risk are also empirically important, which warrants further research on the factors behind the unexplained country risk.