Journal Issue
Share
Article

The Evolution of Asian Financial Linkages

Author(s):
Selim Elekdag, Phurichai Rungcharoenkitkul, and Yiqun Wu
Published Date:
November 2012
Share
  • ShareShare
Show Summary Details

I. Introduction

Uncertainty continues to shroud the global economic and financial outlook. Downside risks, most obviously stemming from the euro area, remain a concern for Asia, because developments in major global financial centers tend to have large effects on Asian financial markets. In particular, equity returns in Asian economies seem generally to move in tandem with those in systemic economies, including, for example France, Germany, Japan, the United Kingdom, and the United States (Figure 1).

Figure 1.Equity Returns: Selected Systemic Economies and Asia

(In percent over 12 months)

Sources: Thomson Reuters Datastream; and authors’ calculations.

This paper attempts to provide a positive analysis of the underlying determinants of these financial linkages, or financial “betas”, in Asia. The objective is to measure how the strength of these linkages varies across time and countries, and to understand the causes of these changes. Normative implications of movements in beta however are not obvious a priori. For example, a secular uptrend in beta (if any) may bring welfare gains associated with greater trade and financial integration, while at the same time increasing exposure to financial shocks. Likewise, movements in betas over the business cycle could reflect (welfare-enhancing) international risk sharing through financial markets, (welfare-reducing) financial contagion, or a combination of both – see Rungcharoenkitkul (2011) for an empirical evaluation of the tradeoffs between the risk-sharing benefits and the spillovers costs of financial integration. Such welfare discussion lies beyond the scope of this paper.

Against this backdrop, this paper focuses on the following questions:

  • How has the sensitivity of Asian financial markets to systemic economies varied across economies and over time?
  • How important are real and financial linkages with systemic economies in explaining Asian financial market fluctuations?
  • To what extent can macroeconomic policies help mitigate financial market spillovers?

To explore the evolution of Asian financial linkages, an extended version of the Capital Asset Pricing Model (CAPM) is used. Building upon the studies of Bekaert and others (2011), Forbes and Chinn (2004), and Balakrishnan and others (2009), two complementary approaches are utilized. One approach primarily focuses on the time averages of betas, and provides an analysis of more permanent differences in betas across Asia. The other econometric approach delves deeper into understanding time series fluctuations of betas using a larger set of countries. These two approaches provide different perspectives on the interplay between the key determinants (including the role of policies) and the changing nature of financial linkages in Asia.

The analysis yields three main conclusions:

  • In line with Asia’s growing role in the global economy—including through deeper financial integration—regional financial markets have become more sensitive to systemic economies.
  • Asian financial sensitivities to systemic economies exhibit cyclical fluctuations which correspond to tranquil and turbulent periods across international capital markets. These financial sensitivities reached historically high levels during the latest global financial crisis.
  • Macroeconomic policies—including those designed to lower government debt and increase international reserves (up to a limit)—can reduce Asia’s sensitivity to financial spillovers from systemic economies even after global factors and other economy-specific characteristics are accounted for. While macroeconomic policies can limit financial sensitivities during both tranquil and turbulent periods, they cannot completely insulate Asian financial markets against major global financial shocks.

The organization of the rest of the paper is as follows. Section II presents the conceptual framework and the econometric methodology. Section III discusses the main results, which is followed by the conclusion.

II. Model and Estimation Framework

This section describes the modeling framework used to estimate the strength of financial linkages between Asian and systemic economies, as well as their key determinants, including the role of policies. To do so, two complementary approaches are used, both which build on the Capital Asset Pricing Model (CAPM). While one of the approaches focuses more on cross-country differences within Asia, the other approaches concentrates primarily on how the empirical relationships have changed over time taking into consideration a larger set of economies. These differing perspectives highlight the role of policies, but also serve to assess the robustness of the main findings.

A. Approach I: Focusing on Cross-Sectional Variation

The first approach is based on a two-stage modeling framework used to estimate the importance of Asian financial linkages with systemic economies. In the first stage, a factor model of equity market returns is estimated, controlling for global factors. This first stage yields estimates of Asian financial (equity market) sensitivities to systemic economies. In other words, the first stage generates estimated factor loadings, also known as “betas”. In the second stage, the estimated cross-country factor loadings (betas) are decomposed into trade and financial bilateral linkages between Asian and systemic economies. In addition, the importance of macroeconomic policies is assessed.

This first approach builds on the literature in several directions. The empirical strategy extends the framework developed by Forbes and Chinn (2004), most importantly by seeking to investigate the role of macroeconomic policies. It is closely related to other studies focusing on financial spillovers including Balakrishnan and others (2009) and IMF (2009). It builds upon other studies by controlling for various global factors including world economic activity, given its importance for Asian trade, and proxies for international investor risk aversion because of its relevance for capital flows. Furthermore, in addition to the use of country-specific fixed effects (to account for differences across economies that may not be captured by the observable variables considered), standard errors are corrected for any potential clustering owing to the fact that equity returns across systemic economies may not be independent of each other (even after controlling for global equity returns).

The First Stage

Equity returns in two economies could move in tandem for at least two reasons. First, returns in these economies may be synchronized because shocks to a systemic economy are transmitted to Asian economies through cross-country linkages, such as for example bilateral trade, bank lending, or investment flows. Second, returns in both economies could be affected by global shocks, such as changes in global economic activity, world interest rates, global stock markets, risk aversion of international investors, and commodity prices.

Therefore, in order to isolate the importance of cross-country equity market linkages, the factor model used in the first stage—which could be thought of as an extended CAPM—controls for several global factors. The model also allows market returns in each economy to be determined by a economy-specific effect. For each Asian economy i, equity market returns ri,t at each time t is thus expressed as follows:

where the intercept is a economy-specific effect, and while the error term is assumed to be normally distributed, these errors are not necessarily independent across economies. In this approach, therefore, the factor loadings, βi,c, are assumed to be constant for each economy (over the designated sample period), but can vary across economies. The goal of this approach is to exploit this cross-sectional variation within Asia as much as possible. In what follows, five systemic countries are considered: France, Germany, Japan, the United Kingdom, and the United States (C=5). In addition, five global factors are incorporated in the specification above (G=5): global commodity price fluctuations, global growth, global equity market returns, international interest rates, and the Chicago Board Options Exchange Market Volatility Index (VIX). The factors try to capture global business cycle and financial conditions, as well as risk aversion.2

Data

The data used in the first stage are relatively standard. Equity returns are computed using the equity indices compiled by Morgan Stanley Capital International (MSCI) and computed as monthly logarithmic differences. These returns are calculated for the five systemic economies, and 12 Asian economies (Japan is not included since it is treated as a systemic economy, and Vietnamese data were deemed insufficiently long for robust statistical inference): Australia, New Zealand (sometimes grouped together), Indonesia, Malaysia, the Philippines, Thailand, and Singapore (sometimes grouped as “ASEAN”), Hong Kong SAR, Korea, and Taiwan, Province of China (sometimes grouped as East Asia, excluding China), China, and India. As for the global factors, to exploit as much as possible the information contained in the monthly return series, the global business cycle was proxied using a measure of global industrial production (the PPP-weighted average of industrial production of the five systemic economies). World interest rates are constructed analogously. The global commodity price index is standard, and the VIX, despite being U.S.-oriented, is nonetheless taken to be a global factor. For world stock returns, the residual world stock return after controlling for the five systemic economies’ returns is used to minimize multi-collinearity. Specifically, global stock market returns (using the MSCI world index) is regressed on the returns of the five systemic economies. The estimated residual is then the measure of global stock market returns. For the monthly data used, the sample starts in 1990 (but later for some countries), and ends in March 2012.

Robustness

The estimated first-stage factor loadings are robust to alternative specifications. To assess the sensitivity of the estimated factor loadings, specifications which consider only four (omitting France), and three (omitting France and the United Kingdom) systemic economies were also considered. In addition, specifications with various combinations of the global factors were also estimated, including a version without any. Overall, the estimated factor loadings and their relative rankings do not change in a way that would markedly affect the main results (Appendix Table). What does affect the estimated factor loadings, however, is the sample period considered—an important aspect of this paper which is discussed further below.

The Second Stage

In the second stage, the goal is to uncover the key determinants of the estimated factor loadings i,c). Because there are five systemic economies and 12 Asian economies, a panel of 60 estimated factor loadings is generated in the first stage. This panel has an Asian dimension to it, and a systemic economy dimension, as previously indicated with subscripts i, and c, respectively.

The second stage focuses on two main channels through which shocks to a systemic economy c could affect an Asian economy i, namely bilateral trade and financial linkages. First, shocks to systemic economy c could affect country c’s demand for imports from country i. Second, shocks to systemic economy c could affect bank lending or flows of foreign investment from economy c to Asian economy i. There are obviously other channels that could link systemic and Asian economies such as portfolio investment, trade credit, or multinational exposure, but owing to data limitations, it is not feasible currently to explore these other channels more rigorously.

Sound macroeconomic policy frameworks could mitigate the impact of shocks from systemic economies. For example, international reserves could be drawn down in the case of a sudden reversal of bank lending or investment flows. In addition, fiscal stimulus could be deployed to support the economy that is hit be a severe trade shock originating from a systemic economy. At the same time, the degree of capital account openness and the exchange rate regime would not only determine the extent of how shocks are transmitted across economies, but the flexibility of the policy response.

In order to estimate the importance of the bilateral linkages and the role of macroeconomic policies, the following specification is initially considered:

The first three explanatory variables capture trade and financial linkages between the Asian economy i and the systemic economy c. At first glance, it might be expected that declines in systemic economies’ equity markets would have negative effects on Asian economies’ equity markets through each of the three bilateral linkage variables. However, this needs not be the case. Consider financial linkages: negative news about a systemic economy’s growth outlook could cause negative returns in that country’s equity market. This could generate a contraction in lending and investment by banks and firms based in that systemic economy as they strengthen balance sheets and prepare for the expected slowdown. The resulting contraction in lending and investment into Asian economies would be expected to have a negative effect on Asian equity markets, implying δ2,δ3 > 0. By contrast, if in response to the unfavorable outlook, the banks and firms in the systemic economy keep their lending and investment levels stable, but instead shift exposure away from the systemic economy and into Asia, this may have a positive impact on Asian equity markets, yielding δ2, δ3 < 0. In other words, the importance and signs of the coefficient estimates for the bilateral trade and financial linkage variables need to be ascertained empirically.

The equation above also tries to assess the role of macroeconomic policies for each individual Asian economy. International reserve metrics and fiscal policy indicators (including various debt and deficit metrics) are included as the main macroeconomic policy measures. The idea is to capture the notion of policy space. Intuitively, economies with lower debt levels and deficits, and higher international reserves (but up to point, owing to, among other things, quasi-fiscal costs, as discussed further below), have greater room to respond to shocks. This perceived or actual room for maneuver would likely imply that sensitivities to financial shocks stemming from systemic economies remain broadly stable. Owing to the existence of fixed exchange rate regimes and also fundamental changes to policy frameworks, measures of monetary policy space that would be consistent across Asian economies are difficult to design, and are therefore not explicitly considered, although some indirect and partial account for them is through the exchange rate regime variable.

Data

The data used in the second stage is also quite standard, but available at an annual frequency. The variable TRADE captures real linkages between Asian and systemic economies. Using the IMF’s Direction of Trade Statistics (DOTS) database, total imports by a systemic economy c from Asian economy i, measured in U.S. dollars, is scaled by GDP for economy i. Two variables capture financial linkages between Asian and systemic economies: FDI and BANK.3 The variable FDI is based on the OECD’s International Direct Investment Statistics, and is measured as the total stock of foreign investment from systemic economy c in Asian economy i as a share of country i’s GDP. The variable BANK is based on lending data reported by the BIS, and measured as the total stock of bank lending from systemic economy c to Asian economy i as a share of i’s GDP (consolidated international claims of BIS-reporting banks in c vis-a-vis i). For policies, the stock of international reserves is scaled by GDP or M2, and fiscal variables are in percent of GDP. For the type of exchange rate regime in place and degree of capital account openness, data from Reinhart and Rogoff (2004), and Chinn and Ito (2008), respectively, are used. Higher values of the Reinhart-Rogoff and Chinn-Ito indices indicate more exchange rate flexibility and greater capital account openness, respectively.

Robustness

In the second stage, the full sample period is from 1992 to 2011. However, to both assess the robustness of the results and explore how these relationships have changed over time the sample is split into two sub-periods: 1992–01, and 2002–11. Moreover, given the bankruptcy of Lehman Brothers in 2008, the 2002–11 sample was further split into (pre- and post-bankruptcy periods) 2002–07 and 2008–11.

B. Approach II: A Deeper Investigation of Time Series Fluctuations

Financial betas can also portray important fluctuations over time at different frequencies. Potential drivers of changes in betas include, for example, low-frequency changes in the degree of financial integration and high-frequency shocks corresponding to episodes of stress in financial markets and the business cycle fluctuations. Disentangling the relative influences of these factors on the time series of financial betas is important for understanding the realtime transmission of financial shocks, fluctuations in financial linkages, and the role of policy. The second approach aims to address these issues, and also considers a larger set of economies (as discussed below).

To structurally allow for betas’ changes over time, the CAPM model is modified to take the form:

The first equation is the basic CAPM relationship, where rit denotes excess return of equity market i, and rtus is the excess return of the US equity market both in period t. The term αoi + α1rit−1 is used as a simple proxy for the expected excess return Et-1(rit), or expected risk premium which can differ across economies. The financial beta for country i is taken to be its sensitivity to the US market alone, a simplification that allows a closer examination of its determinants. The second beta equation expresses the scalar βit as a function of observable country-specific factors Xit and global factors Yt, both of which fluctuate over time.

The model’s structure is similar to that of Bekaert and others (2011), which specifies a parametric model for beta. Unlike Bekaert and others (2011), this paper does not measure “contagion”, defined by the authors as changes in beta that are triggered by events but cannot be explained by other observables. Here, the objective is to describe the evolution of beta purely in terms of what can be observed, which, as will be seen, already generates significant variations in beta over time. In the terminology of Forbes and Rigobon (2002), the paper’s focus is to understand the sources of interdependence.

Estimation

Combining the return and beta equations, the parameters {αoi, α1, b0, b1, b2} can be estimated from the reduced-form panel specification

by regressing rit on its own lag, the US equity return, and its interactions with observables Xit and Yt. The estimate for financial beta is then simply obtained by

The model selection follows a simple general-to-specific strategy, a variant of which is also employed in Bekaert and others (2011).4 The general unrestricted model with the full set of candidate factors Xit and Yt is first estimated. Using 10 per cent confidence level as an acceptance threshold, the most insignificant variable is successively removed from the specification until a specification is obtained that features only significant variables. After the procedure settles on a specification, each previously rejected variable is retested for significance, to arrive at the final parsimonious specification. The selection procedure therefore lets the data both identify the determinants of financial beta and quantify their impacts.

Several restrictions on the role of expected risk premium can be considered. When both αoi and α1 are set to zero, the model is reduced to the standard “Pure CAPM” equation, with rtus representing the market portfolio return. Under this restriction, each market’s excess return is only driven by the market (US equity) return. Alternatively, αoi could be constrained to be constant across i, to be estimated along with α1. The restriction allows for the expected risk premium to vary over time, but in a functionally-equivalent way across economies (“Nonzero Expected Risk Premium”). In both of these restrictions, the estimation is simply OLS on stacked-up data. Finally, the unrestricted model allows for the expected risk premium to exist and vary across economies via constant αoi, which can be estimated in fixed effect panel regression (“Heterogeneous Expected Premium”).5

Data

To construct a good-sized panel, the data set is extended to include non-Asian economies from OECD (Germany, Great Britain, France, Austria, Belgium, Canada, Denmark, Finland, Greece, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, and Switzerland), and other emerging markets (Argentina, Brazil, Chile, Colombia, Mexico, Peru, South Africa, Turkey, Poland, Hungary, Czech Republic, and Russia). The monthly panel comprises 42 economies, with 13 from Asia, covering the period 1990–2011.

All excess returns are computed using the monthly returns of equity indices, less the return on the 3-month US treasury bills. Equity indices used to calculate returns are the same as explained in the previous section. Exchange rate movements are not taken into account, under the assumption that equity investors hedge away their FX risks.

The country-specific factors Xit include measures of economic ‘linkages’, through the channels of trade, banking, and FDI. The same variables used in the previous section, namely TRADE, BANK and FDI (describing the bilateral linkages between the US and country i) are adopted and supplemented by the ratio of seasonally-adjusted exports plus imports to GDP as an additional indicator of trade openness. Xit also includes POLICIES, namely fiscal balances, the exchange rate regime, capital account openness, and foreign reserves-to-GDP also described in the previous section. Vulnerability to capital outflows is also included, measured by the ratios of short-term external debt to GDP, and reserves to short-term external debt. Lastly, the cyclical component of real GDP (HP-filtered series) is added to Xit to capture the effect of the business cycle.

The common factor Yt consists of two standard measures of global financial market conditions, VIX and the slope of the US treasury yield curve (10- minus 2-year yields). During financial stresses, VIX is expected to increase, while the treasury slope should flatten, indicating a flight to safety and higher chance of a US (and hence global) recession. A time trend is also included to pick up any longer-term structural changes in financial beta not captured by other included variables, but is found to be insignificant and is therefore dropped from all regressions.

III. Results

In this section, the results from the second panel regressions (Approach II) are first presented, which are used to derive the implied time series of financial beta in Asia. The fitted model is also exploited to explain the contributions to betas by the underlying determinations vary over the recent episodes of global financial turmoil. Then, the results from the first approach (Approach I) are discussed, emphasizing that sound macroeconomic policy frameworks can play an important role in explaining cross-country variations in financial beta.

Understanding How Betas Change Over Time

Table 1 through Table 3 show results using Approach II (panel regressions focusing on how betas vary over time). The most general specification is shown under the “General” column. After the general-to-specific selection procedure is applied (see Campos and others, 2005), the model is reduced down to the final specification as shown under the “GS search” column. The data availability for BANK and short-term external debt is however limited for a number of countries. To expand the number of observations and also as a robustness test, two restricted models are considered. The first drops BANK and external debt before the general-to-specific selection is applied. The number of observations roughly doubles, and the result of the procedure is shown in the “GS 2” column. In the second restricted model, only debt variables are dropped, with the result shown in the “GS 3” column. Table 1 through Table 3 show the results for pure CAPM 0i = α1 =0), risk premium (constant α0i), and heterogeneous risk premium specifications, respectively.

Table 1.Pure CAPM
VariablesGeneralGS searchGS 2GS 3
rtus0.758***

(0.000)
0.635***

(0.000)
0.544***

(0.000)
0.616***

(0.000)
Trade × rtus0.001

(0.164)
0.001**

(0.038)
0.002***

(0.000)
0.001***

(0.000)
Imports × rtus-0.001

(0.824)
Banking × rtus-0.030***

(0.003)
-0.026***

(0.001)
-0.018***

(0.000)
FDI × rtus0.031*

(0.130)
0.021**

(0.018)
Fiscal × rtus0.014**

(0.019)
0.014**

(0.017)
0.006*

(0.076)
Reserves × rtus0.001

(0.748)
-0.003***

(0.006)
Capital Openness × rtus0.000

(1.000)
ER Regime × rtus0.048

(0.160)
0.072**

(0.035)
Debt/GDP × rtus0.001

(0.125)
0.002***

(0.000)
Reserve/Debt × rtus-0.000**

(0.022)
-0.000**

(0.025)
GDP × rtus-0.008

(0.275)
-0.013**

(0.017)
VIX × rtus0.006**

(0.014)
0.007***

(0.002)
0.007***

(0.006)
0.006***

(0.000)
Yield Slope × rtus-0.054

(0.101)
-0.057*

(0.066)
R20.270.240.270.27
Observations4573488497138082
Source: Authors’ calculations.

Column “General” shows the most general specification; “GS search” shows results under the general-to-specific specification search algorithm; “GS 2” drops Banking, Debt/GDP and Reserve/Debt to increase observations before applying general-to-specific search; “GS 3” similarly drops Debt/GDP and Reserve/Debt.

***, ** and * denote significance at 1, 5, and 10 percent confidence levels respectively. P-values are shown in parentheses.

Source: Authors’ calculations.

Column “General” shows the most general specification; “GS search” shows results under the general-to-specific specification search algorithm; “GS 2” drops Banking, Debt/GDP and Reserve/Debt to increase observations before applying general-to-specific search; “GS 3” similarly drops Debt/GDP and Reserve/Debt.

***, ** and * denote significance at 1, 5, and 10 percent confidence levels respectively. P-values are shown in parentheses.

Table 2.Non-Zero Expected Risk Premium
VariablesGeneralGS searchGS 2GS 3
Constant0.417***

(0.000)
0.473***

(0.000)
0.178***

(0.007)
0.297***

(0.000)
rit−10.045**

(0.036)
0.052***

(0.010)
0.068***

(0.000)
0.051***

(0.001)
rtus0.706***

(0.000)
0.590***

(0.000)
0.538***

(0.000)
0.564***

(0.000)
Trade × rtus0.001

(0.130)
0.001**

(0.033)
0.002***

(0.000)
0.002***

(0.000)
Imports × rtus-0.001

(0.807)
Banking × rtus-0.029***

(0.004)
-0.025***

(0.002)
-0.012***

(0.004)
FDI × rtus0.031

(0.133)
Fiscal × rtus0.013**

(0.034)
0.013**

(0.027)
0.006*

(0.070)
Reserves × rtus0.0003

(0.836)
-0.003***

(0.005)
-0.003**

(0.020)
Capital Openness × rtus0.001

(0.958)
ER Regime × rtus0.049

(0.158)
0.069**

(0.042)
Debt/GDP × rtus0.001

(0.158)
0.002***

(0.001)
Reserve/Debt × rtus-0.0002**

(0.022)
-0.0002**

(0.027)
GDP × rtus-0.006

(0.423)
-0.010*

(0.082)
VIX × rtus0.007***

(0.002)
0.009***

(0.000)
0.007***

(0.000)
0.007***

(0.000)
Yield Slope × rtus-0.068**

(0.040)
-0.073**

(0.020)
R20.270.250.280.27
Observations4573487996898075
Source: Authors’ calculations.

Column “General” shows the most general specification; “GS search” shows results under the general-to-specific specification search algorithm; “GS 2” drops Banking, Debt/GDP and Reserve/Debt to increase observations before applying general-to-specific search; “GS 3” similarly drops Debt/GDP and Reserve/Debt.

***, ** and * denote significance at 1, 5, and 10 percent confidence levels respectively. P-values are shown in parentheses.

Source: Authors’ calculations.

Column “General” shows the most general specification; “GS search” shows results under the general-to-specific specification search algorithm; “GS 2” drops Banking, Debt/GDP and Reserve/Debt to increase observations before applying general-to-specific search; “GS 3” similarly drops Debt/GDP and Reserve/Debt.

***, ** and * denote significance at 1, 5, and 10 percent confidence levels respectively. P-values are shown in parentheses.

Table 3.Heterogeneous Expected Premium (Fixed Effects)
VariablesGeneralGS searchGS 2GS 3
Constant0.418***

(0.000)
0.454***

(0.000)
0.273***

(0.000)
0.294***

(0.000)
rit−10.041*

(0.099)
0.051**

(0.040)
0.074***

(0.000)
0.045**

(0.028)
rtus0.708***

(0.001)
0.671***

(0.000)
0.515***

(0.000)
0.577***

(0.000)
Trade × rtus0.001

(0.358)
0.002**

(0.020)
0.001***

(0.004)
Imports × rtus-0.001

(0.866)
Banking × rtus-0.028**

(0.026)
-0.017**

(0.017)
FDI × rtus0.028

(0.399)
0.020**

(0.028)
Fiscal × rtus0.013**

(0.036)
0.019***

(0.001)
Reserves × rtus0.0003

(0.866)
-0.003*

(0.076)
Capital Openness × rtus0.0005

(0.990)
ER Regime × rtus0.049

(0.313)
Debt/GDP × rtus0.001

(0.367)
0.001***

(0.000)
Reserve/Debt × rtus-0.0002*

(0.059)
GDP × rtus-0.005

(0.254)
-0.011*

(0.064)
VIX × rtus0.007***

(0.003)
0.009***

(0.000)
0.008***

(0.000)
0.007***

(0.000)
Yield Slope × rtus-0.067*

(0.065)
-0.068*

(0.093)
R20.270.240.240.27
Observations45734949105388075
Source: Authors’ calculations.

Column “General” shows the most general specification; “GS search” shows results under the general-to-specific specification search algorithm; “GS 2” drops Banking, Debt/GDP and Reserve/Debt to increase observations before applying general-to-specific search; “GS 3” similarly drops Debt/GDP and Reserve/Debt.

***, ** and * denote significance at 1, 5, and 10 percent confidence levels respectively. P-values are shown in parentheses.

Source: Authors’ calculations.

Column “General” shows the most general specification; “GS search” shows results under the general-to-specific specification search algorithm; “GS 2” drops Banking, Debt/GDP and Reserve/Debt to increase observations before applying general-to-specific search; “GS 3” similarly drops Debt/GDP and Reserve/Debt.

***, ** and * denote significance at 1, 5, and 10 percent confidence levels respectively. P-values are shown in parentheses.

The variable VIX is a significant determinant of betas across all specifications considered, suggesting that financial stresses are associated with an increase in financial betas.6 The yield curve slope is found to be significant for half of the specifications, with negative sign as expected (that is, beta picks up as the yield curve flattens). The other cyclical factor, the deviation of output growth from trend, is found to be important with an expected negative sign under all “GS 3” models.

Trade openness (as measured by the sum of exports and imports relative to GDP) is significant for most specifications, with greater openness associated with a higher beta. Higher FDI, when significant, also tends to raise beta through greater financial exposure. Banking linkages, on the other hand, appear to exert negative influences on beta. While larger banking flows may be associated with a greater exposure to the banking channel of financial linkages, the results suggest that international bank financing may serve as a substitute for capital markets, thereby resulting in the negative sign. In other words, to the extent that banking inflows can help ease financing stresses, they should also help lower beta. Lower fiscal balances, when significant, tend to decrease beta, perhaps reflecting the benefits of counter-cyclical fiscal policy during severe economic slowdowns.

Higher levels of foreign exchange reserves, by providing more insurance against disruptions from sudden outflows, help lower financial beta as expected. While the magnitude of the coefficient, when significant, is almost identical across all specifications, there is also some evidence of nonlinearity, namely there are diminishing ‘insurance’ benefits from ever increasing reserves (see below). Short-term external debt raises financial beta as might be expected, whether measured relative to foreign reserves or GDP. Lastly, while the general specification controls for exchange rate regime and capital account openness, there is only scant evidence that either of these matters in a robust way.

A single model needs to be chosen to compute an estimate of a beta time series. The fitted model with fixed effects leaving out short-term debt variables is adopted for this purpose (Table 3, GS 3). This model nests the pure CAPM in Table 1 and the simple premium model in Table 2. Also, by leaving out short-term debt, more observations are available for the estimation, and a longer beta time series can be generated. Nonetheless, the fitted beta does not differ significantly across models in any case.

How Have Asian Financial Betas Evolved Over Time?

Estimates from the panel regressions (Approach II) suggest that Asian financial markets’ sensitivity to US equity performance has followed a steady upward trend over the last two decades, reflecting Asia’s increasing financial and trade integration with the world (Figure 2). This trend, however, has been associated with strong cyclical fluctuations, linked to developments across international capital markets. In particular, the betas spike in all Asian economies during episodes of global financial turbulence, including the bursting of the technology bubble (and the associated NASDAQ crash) in 2001 and, more recently, the Lehman Brothers bankruptcy and the turmoil in the euro area.7 For all Asian economies, the financial betas reached unprecedented levels during 2008–11, surpassing those prevailing during the Asian crisis in 1997 (Figure 3).

Figure 2.Asian Financial Betas and Global Financial Shocks1

Sources: Thomson Reuters Datastream; and authors’ calculations.

1 Chart depicts estimated financial beta with respect to the United States from panel regressions.

Figure 3.Financial Betas across Asian Economies1

Sources: Thomson Reuters Datastream; and authors’ calculations.

1 Chart depicts estimated financial beta with respect to the United States from panel regressions (averages over indicated periods).

While generally synchronized, individual Asian financial markets tend to react differently to shocks in the U.S (Figure 3). In general, ASEAN and East Asia (which include Singapore and Hong Kong SAR, respectively) appear to have the largest financial betas. In the case of the ASEAN economies, the panel regressions generate an estimated beta of about 0.75 on average over the full sample. This estimate suggests that a 10 percent increase in the U.S. stock market is associated with a 7½ percent increase across the ASEAN stock markets, all else equal. At the other end of the spectrum, economies that pursued a more gradual pace of international capital market integration, such as China, generally had lower financial sensitivities to systemic economies.

The significant increase in financial betas during the latest stress episode since 2008 is driven largely by the global financial shock, with other country-specific determinants (particularly the business cycles) playing only a minor role. Figure 4 shows the difference between average financial betas during 2008–11 relative to the more tranquil 2002–07 period (solid black line), together with the breakdowns of contributions by sources shown in bars. In contrast, during the 1997–2001, business cycles variations contribute much more to cross-country differences in financial betas. As Figure 5 shows, those countries heavily affected by the turbulences that ensued during 1997–2001, such as Indonesia and Thailand, also experienced an exceptional pickup in betas relative to a calmer period, that mirrored their output contractions.

Figure 4.Contributions to Differences in Betas between 2008–2011 and 2002–2007

Source: authors’ calculations.

Figure 5.Contributions to Differences in Betas between 1997–2001 and 2002–2007

Source: authors’ calculations.

Can Macroeconomic Policies Reduce Financial Sensitivities in Asia?

Macroeconomic policies play a notable role in determining Asia’s financial betas, after controlling for bilateral linkages to systemic economies and other economy-specific characteristics. Results from the econometric analysis (Approach I), shown in Table 4, suggests that bilateral trade, FDI, and banking exposures to systemic economies help explain the diversity of financial betas across Asia. Even after taking into consideration the importance of economy- specific factors that are difficult to account for explicitly (including structural and institutional difference across economies), the regressions indicate that higher bilateral trade, FDI, and banking exposures to systemic economies are positively correlated

Table 4.Determinants of Asian Financial Betas1,2,3
Dependent VariableBetasBetasBetasBetasBetasBetasBetas
FullFullFull1992-1992-2002-2008-
SampleSampleSample2001201120072011
Trade40.01160.01160.01160.00240.00440.0260-0.0050
8.2%8.2%8.2%79.1%68.1%12.1%78.2%
FDI40.62400.62400.62401.89400.0909-0.25400.0288
3.7%3.7%3.7%0.4%69.9%319%716%
Bank40.00600.00600.00600.0103-0.00090.0009-0.0034
0.2%0.2%0.2%2.2%37.5%48.5%0.6%
Debt40.00090.00110.00040.00300.0052-0.0003
0.3%0.1%28.6%0.0%0.0%57.7%
Reserves5-0.0033-0.0042-0.0093-0.0040-0.00880.0004
0.4%0.0%0.2%2.4%0.2%79.1%
Financial openness60.01810.05660.02530.04290.0110
2.2%0.0%2.4%0.8%215%
Exchange rate regime7-0.0118-0.0257-0.0215-0.0146-0.0377
0.2%0.6%5.5%13.4%12.5%
Constant-0.09370.07320.12100.19300.18300.10400.3270
11.5%0.9%0.0%0.0%0.0%3.8%0.2%
Observations60606060606060
R-squared12.4%12.4%12.4%15.5%2.6%6.7%3.3%
Source: Authors’ calculations.

Dependent variable: estimated CAPM betas.

Country fixed-effects included, but not reported, in regressions.

Robust p-values (accounting for clustering with respect to Asian economies) in parentheses (in percent).

Trade, FDI, Bank, and (government) debt refer to bilateral linkages to systemic economies (in percent of GDP).

Reserves (international) scaled by M2.

Financial openness measured using Chinn-Ito (2008) metric.

Exchange rate regime classification based on Reinhart-Rogoff (2004).

Source: Authors’ calculations.

Dependent variable: estimated CAPM betas.

Country fixed-effects included, but not reported, in regressions.

Robust p-values (accounting for clustering with respect to Asian economies) in parentheses (in percent).

Trade, FDI, Bank, and (government) debt refer to bilateral linkages to systemic economies (in percent of GDP).

Reserves (international) scaled by M2.

Financial openness measured using Chinn-Ito (2008) metric.

Exchange rate regime classification based on Reinhart-Rogoff (2004).

with higher financial betas.8 Macroeconomic policies also matter. Specifically, the empirical results indicate that a lower government debt-to-GDP ratio and a higher stock of international reserves are associated with lower financial betas. For example, a 10 percentage-point increase in the reserves-to-GDP ratio is characterized by financial betas that are lower by 0.03–0.04, which is also corroborated by the results using panel econometrics. However, the panel regression (Approach II) also suggests that the marginal benefit of holding reserves may diminish after a certain threshold. Table 5 reproduces regression result of specification GS 2 from Table 3 together with two modified models, one using the logarithm of reserves and the other using a power function of reserves (which nests GS 2 and can be estimated by non-linear least squares). Both alternative models imply diminishing effect of reserves in lowering beta. IMF (2012) also documents similar findings elsewhere.

Table 5.Nonlinear Effects of Reserves
VariablesLinearLogNonlinear least

squares
Trade × rtus0.002**

(0.020)
0.001***

(0.009)
0.001***

(0.000)
Reserves × rtus-0.003*

(0.076)
log(Reserves) × rtus-0.044*

(0.082)
Reservesγ × rtus-8.069***

(0.000)
γ0.005***

(0.005)
VIX × rtus0.008***

(0.000)
0.008***

(0.000)
0.008***

(0.000)
R20.240.240.24
Observations105381053810538
Source: authors’ calculations.Note: ***, ** and * denote significance at 1, 5, and 10 percent confidence levels respectively. P-values are shown in parentheses.
Source: authors’ calculations.Note: ***, ** and * denote significance at 1, 5, and 10 percent confidence levels respectively. P-values are shown in parentheses.

The role of bilateral linkages and macroeconomic policies has changed over time (Table 4). When restricting the sample to the first half of the period (1992–2001), bilateral FDI and banking linkages to systemic economies become at least twice as influential. At the same time, although the role of debt is less prominent, measures of macroeconomic policies such as the stock of reserves—but up to a limit—become more important. In the second half of the sample period (2002–11), while the role of the bilateral linkages diminishes, the empirical results suggest that sound policies remain correlated with lower financial betas. To understand the latter result, recall from Figure 4 that the global financial shock explains a majority of the pickup in financial betas across Asia from 2002–07 to 2008–11 even after accounting for macroeconomic policy measures and other economy-specific characteristics. The occurrence of the global financial crisis is the main reason why the empirical relationships between financial betas and the fundamentals of Asian economies weaken during the last decade. While macroeconomic policies still appear to be useful in lowering financial betas during this period, most of the variation across Asian financial betas during the crisis years is accounted for by global factors, with a relatively more limited role for economy-specific characteristics.

How does the role of macroeconomic policies vary across Asian economies?

The contribution of macroeconomic policies toward reducing financial betas varies across Asian economies and over time. Regression analysis covering the Asian economies can be used to disentangle the importance of macroeconomic policies, bilateral linkages to systemic economies, and other country-specific factors that are difficult to explicitly account for (including structural and deeper institutional differences across economies) in terms of explaining how betas vary. Decompositions shown in Figure 6 depict the predicted values of the betas abstracting from the global controls (in line with two-stage methodology of Approach I), which explains why these betas are lower than those shown earlier. The main advantage of this approach is to highlight the important of sound macroeconomic policies and the strength of the bilateral linkages in determining the estimated betas. These breakdowns indicate that country-specific characteristics that cannot be explicitly accounted for in the regressions raise betas, for example, in the case of the NIEs, which include financial centers. As expected, higher bilateral trade, FDI, and banking exposures to systemic economies also yields higher beta estimates. For most economies, macroeconomic policy measures (including lower government debt, higher international reserves, or greater exchange rate flexibility) seem to lower estimated betas.9 Interestingly, the contribution of policies seems to have increased in the latter half of the sample and thereby helped reduce the estimated betas across some of the Asian economies.

Figure 6.The Decomposition of Financial Betas across Asia over Time

Source: authors’ calculations.

IV. Conclusion

Overall, the analysis suggests that sound macroeconomic policies help keep Asia’s financial betas stable during both tranquil and turbulent periods, but they cannot completely insulate Asian financial markets against major global financial shocks. To be sure, these policies may also limit the impact of major global shocks on the real economy (Balakrishnan and others, 2009), and while beyond the scope of this paper, is an interesting topic for further investigation. But, given how strong spillovers can become in the event of major global shocks, an appropriate response may require a more comprehensive mix of policies, including the various individual and collective financial measures adopted in Asia in response to the global financial crisis.

Appendix Table. Estimated Factor Loadings (Betas) Over the Full Sample
Panel A. Factor model with five systemic economies
AUSNZLHKGKORSGPTWNIDNMYSPHLTHACHNIND
Multi5
USA0.250.090.530.340.40-0.010.440.270.270.450.330.07
JPN0.130.210.080.420.220.290.230.070.120.240.060.02
DEU0.080.000.260.040.300.650.370.550.450.570.010.06
GBR0.320.320.450.760.340.060.150.300.310.750.170.05
FRA-0.030.05-0.14-0.30-0.200.06-0.11-0.40-0.25-0.75-0.080.48
Multi5_rw
USA0.250.090.530.340.40-0.010.440.270.270.450.310.07
JPN0.130.210.080.420.220.290.230.070.120.240.070.02
DEU0.080.000.260.040.300.650.370.550.450.570.020.06
GBR0.320.320.450.760.340.060.150.300.310.750.170.05
FRA-0.030.05-0.14-0.30-0.200.06-0.11-0.40-0.25-0.75-0.080.48
Multi5_IC
USA0.240.120.500.370.38-0.020.400.290.260.410.21-0.08
JPN0.130.200.080.400.210.260.190.060.090.200.04-0.01
DEU0.08-0.010.260.020.300.640.360.540.440.560.030.08
GBR0.320.310.450.750.320.020.120.290.280.710.190.03
FRA-0.040.06-0.13-0.28-0.180.11-0.06-0.38-0.21-0.69-0.110.51
Multi5_ICY
USA0.240.120.500.350.37-0.020.390.270.240.390.20-0.09
JPN0.130.210.080.400.210.260.190.050.090.190.04-0.02
DEU0.08-0.010.260.030.300.640.360.540.440.560.030.08
GBR0.320.310.450.760.330.020.130.300.290.730.200.04
FRA-0.040.06-0.13-0.27-0.180.11-0.06-0.37-0.21-0.69-0.100.52
Multi5_ICYV
USA0.240.120.500.350.37-0.020.400.270.250.390.20-0.09
JPN0.130.190.090.410.210.270.160.040.070.200.03-0.02
DEU0.07-0.040.270.050.310.670.310.520.410.570.020.08
GBR0.320.310.450.760.330.020.130.300.290.730.210.04
FRA-0.030.07-0.14-0.28-0.180.10-0.03-0.37-0.19-0.69-0.100.52
Multi5_rwICYV
USA0.250.130.520.360.38-0.020.400.290.250.390.20-0.09
JPN0.130.190.090.410.210.270.160.040.070.200.04-0.02
DEU0.07-0.040.270.040.310.670.310.520.400.570.020.08
GBR0.330.310.450.760.330.020.130.300.300.730.210.04
FRA-0.030.07-0.14-0.28-0.180.10-0.03-0.37-0.19-0.69-0.100.52
Panel B. Factor model with three systemic economies
AUSNZLHKGKORSGPTWNIDNMYSPHLTHACHNIND
Multi3
USA0.390.250.700.620.510.030.480.320.350.630.390.20
JPN0.150.240.110.450.230.300.230.060.120.230.060.06
DEU0.150.120.290.050.260.710.330.370.370.280.010.39
Multi3_rw
USA0.390.250.700.620.510.030.480.320.350.630.370.20
JPN0.150.240.110.450.230.300.230.060.120.230.080.06
DEU0.150.120.290.050.260.710.330.370.370.280.010.39
Multi3_IC
USA0.380.270.680.650.490.010.440.340.340.580.270.05
JPN0.150.230.100.430.210.270.200.040.090.190.040.04
DEU0.150.120.300.050.270.720.350.370.380.300.020.43
Multi3_ICY
USA0.380.270.680.640.490.010.440.330.330.570.270.04
JPN0.150.230.100.430.210.270.190.040.090.180.040.04
DEU0.150.110.300.060.280.720.360.380.390.310.020.43
Multi3_ICYV
USA0.380.280.680.640.480.010.450.330.340.570.270.04
JPN0.150.220.110.440.220.290.160.030.070.190.040.04
DEU0.150.090.310.070.280.740.320.360.360.320.010.44
Multi3_rwICYV
USA0.390.290.690.650.490.010.450.340.350.580.270.05
JPN0.150.220.110.440.220.290.160.030.070.190.040.04
DEU0.140.090.300.070.280.740.320.360.360.310.010.43
Source: authors’ calculations.
Source: authors’ calculations.
References

    BalakrishnanR.S.DanningerS.Elekdag and I.Tytell2009The Transmission of Financial Stress from Advanced to Emerging EconomiesIMF Working Paper No. 11/242 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    BekaertG.M.EhrmannM.Fratzscher and A. J.Mehl2011Global Crises and Equity Market ContagionNBER Working Paper No. 17121 (Cambridge, Massachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    CamposJ.N. R.Ericsson and D. F.Hendry2005General-to-Specific Modeling: An Overview and Selected BibliographyInternational Finance Discussion PapersFederal Reserve Board.

    • Search Google Scholar
    • Export Citation

    ChinnM.D. and H.Ito2008A New Measure of Financial OpennessJournal of Comparative Policy Analysis Vol. 10 No. 3 pp. 30922.

    ForbesK.J. and M.Chinn2004A Decomposition of Global Linkages in Financial Markets Over TimeThe Review of Economics and Statistics Vol. 86 No. 3 pp. 70522.

    • Crossref
    • Search Google Scholar
    • Export Citation

    ForbesK. and R.Rigobon2002No Contagion, Only Interdependence: Measuring Stock Market ComovementsJournal of Finance Vol. 57 No. 522232261.

    • Crossref
    • Search Google Scholar
    • Export Citation

    International Monetary Fund2009World Economic Outlook (WashingtonApril).

    International Monetary Fund2012Regional Economic Outlook: Asia and Pacific (WashingtonApril).

    JudsonR. A. and A. L.Owen1999Estimating Dynamic Panel Data Models: A Guide for MacroeconomistEconomic Letters Vol. 65 pp. 915.

    • Crossref
    • Search Google Scholar
    • Export Citation

    NickellS.1981Biases in Dynamic Models with Fixed EffectsEconometrica Vol. 49 pp. 141726.

    RungcharoenkitkulP.2011Risk Sharing and Financial Contagion in Asia: An Asset Pricing PerspectiveIMF Working Paper No. 09/133 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation

    ReinhartC.M. and K.S.Rogoff2004The Modern History of Exchange Rate Arrangements: A ReinterpretationQuarterly Journal of Economics Vol. 119 No. 1 pp. 148.

    • Crossref
    • Search Google Scholar
    • Export Citation
1We thank Roberto Cardarelli, Romain Duval, Olaf Unteroberdoerster, and seminar participants at the Asian Development Bank, the Hong Kong Institute of Monetary Research, and Gateway House (Mumbai, India) for their insightful comments.
2The model focuses on the effect of shocks to systemic economies on Asian equity markets, rather than on estimating simultaneous equations between all economies. Feedback effects are harder to verify empirically, given limited time series data on bilateral linkages and other asset prices and the predominance of systemic financial markets. Moreover, existing correlation across equity returns in the five systemic markets is another limitation of the methodology.
3Note that the general uptrend seen in the BANK and FDI time series across countries reflects in part greater financial integration and financial deepening.
4The general-to-specific methodology owes its development to David Hendry. For a recent survey, see Campos, Ericsson, and Hendry (2005).
5A potential concern is the standard downward dependent variable bias associated with the estimation of dynamic fixed effects models (Nickel, 1981). However, the downward lagged dependent variable bias is known to fall as the time span of the sample increases. Furthermore, it is less of a concern when the time span is large and of the same order of magnitude of the number of countries, as is the case here (Judson and Owen, 1999).
6Because an explicit asset-pricing factor model is used here, the linkage between volatility (VIX) and correlation (beta) is not subject to critiques of Forbes and Rigobon (2002). In other words, the result is not a statistical artifact of how correlation is calculated, but points towards a genuine link between financial stresses and financial “spillovers”.
7The upward trend in betas existed even before the onset of the Lehman crisis in 2008, at least for East Asia excluding China and ASEAN. Therefore a pickup in betas over time is likely secular, and it is not driven by the global financial crisis.
8The effect of banking exposure may appear at odds with the estimate from Approach II, which suggests that the impact on beta is negative. However, banking exposure can have an effect on both the secular trend and the cyclical fluctuations in beta. Results from Approach I, which looks at the determinants of period-average beta, point to a positive impact of banking exposure on the secular trend, while Approach II suggests a negative counter-cyclical effect on beta.
9The exceptions being Australia and New Zealand. In these two cases, the role of policies is likely being accounted for by country-specific factors that are not directly captured by in the regressions, including, for example, that both are commodity-exporting industrialized economies.

Other Resources Citing This Publication