Do Inflows or Outflows Dominate? Global Implications of Capital Account Liberalization in China
Author: Mr. Tamim Bayoumi and Ms. Franziska L Ohnsorge

Contributor Notes

Author’s E-Mail Address: tbayoumi@imforg; fohnsorge@imf.org
Publication Date:
28 Aug 2013
eISBN:
9781475532159
Language:
English
Keywords:
WP; bond yield; foreign currency; sovereign bond
Download PDF (4 MB)

This paper assesses the implications of Chinese capital account liberalization for capital flows. Stylized facts from capital account liberalization in advanced and large emerging market economies illustrate that capital account liberalization has historically generated large gross capital in- and outflows, but the direction of net flows has depended on many factors. An econometric portfolio allocation model finds that capital controls significantly dampen cross-border portfolio asset holdings. The model also suggests that capital account liberalization in China may trigger net portfolio outflows as large domestic savings seek to diversify abroad.

Abstract

This paper assesses the implications of Chinese capital account liberalization for capital flows. Stylized facts from capital account liberalization in advanced and large emerging market economies illustrate that capital account liberalization has historically generated large gross capital in- and outflows, but the direction of net flows has depended on many factors. An econometric portfolio allocation model finds that capital controls significantly dampen cross-border portfolio asset holdings. The model also suggests that capital account liberalization in China may trigger net portfolio outflows as large domestic savings seek to diversify abroad.

I. Introduction

China’s 12th Five-Year Plan pledges to accelerate capital account liberalization. This would be a major event in the global financial architecture. Despite steps to ease restrictions and some circumvention over the past decade, much of China’s capital account and, especially, portfolio investment flows remain severely restricted—as can be seen from various cross-country indexes of capital controls (Bayoumi and Sabroski, 2012, Table 2).1 Accordingly, this paper looks into the potential consequences of a full opening of China’s capital markets. We do this by estimating an empirical model that relates the level of gross bilateral international assets and liabilities between a wide range of countries to fundamentals, including the level of capital controls. We then use the results to infer the potential portfolio in- and outflows that might follow liberalization of the Chinese capital account. We find that gross flows both into and out of China would be substantial and that the likely direction of net flows would be outflows. Our estimated orders of magnitude would imply significant repercussions for both Chinese and global financial markets.2

Table 1.

Capital Account Restrictions in China

article image
article image
Source: AREAR (2011).
Table 2.

Capital account restrictions

article image
Source: Schindler (2009), extended to 2010.Note: Red denotes either a restriction beyond reporting requirements. Green denotes no restrictions. Yellow in an aggregate position denotes one out of four categories are restricted; orange in an aggregate position denotes two out of four categories are restricted; red in an aggregate position denotes three or four out of four categories are restricted.

Previous authors have also estimated potential capital flows if China liberalizes its capital account and have found broadly similar results. In a study that this paper builds on but that is limited to U.S. financial markets, Forbes (2010) estimates that capital account opening could raise Chinese exposures to U.S. equity markets by almost half. Based on the experience of 37 developing and emerging markets that liberalized since the mid-1990s, Sedik and Sun (2012) estimate that inflows and outflows could rise by 2–3¼ percentage points of GDP between 2012 and 2016, and outflows would increase by more than inflows. Based on a dynamic panel estimation of portfolio and FDI asset and liability flows in advanced or emerging markets, He and others (2012) estimate that capital account opening could generate a stock adjustment in international portfolio assets and liabilities of 21 and 16 percent of GDP, respectively, by 2020 and in direct investment assets and liabilities of 22 and 11 percent of GDP, respectively.3

The following section describes the main Chinese capital controls. Section III recaps briefly the experience of other advanced and emerging market countries that have opened their capital accounts since the 1970s. Section IV presents the econometric results and Section V concludes with some implications of the econometric results.

II. Background: Restrictions in China

China has gradually been opening its capital account, but capital flows remain subject to considerable restrictions. There are at least three categorizations of capital account openness: Quinn (2003) for capital and current accounts; Chinn and Ito (2008) for FDI and non-FDI financial accounts; and Schindler (2009) for detailed capital account categories. Schindler (2009) has proposed a classification of capital controls based on the main categories of capital flows in the balance of payments: FDI, equity and bond flows; flows through collective investment vehicles (e.g., ETFs or mutual funds); and foreign borrowing. Of course, these measures are indices of the existence of restrictions, not their intensity. They would not capture a relaxation (as opposed to elimination) of existing controls, e.g., by increasing quotas. Bearing in mind this caveat, every category of flow—whether by residents or nonresidents, inflows or outflows—retains some approval requirement or quota in China (Table 1). By comparison with other emerging markets, the existence of restrictions remain pervasive (Table 2). Specifically:

Capital account restrictions

article image
Source: Schindler (2009); Quinn (1997); Chinn and Ito (2007).

Includes ratings for FDI which are not shown here. 1 indicates a restriction that goes beyond registration and notifaction requirements.

Quinn (1997) index takes into account intensity of restrictions on a scale of 0, 0.5, 1, 1.5, and 2. Capital restrictions include FDI and non FDI.

Chinn-Ito (2007) index is principal component of four 0-1 subindices on average over 5 years. Index refers to current account and financial account restrictions.

  • FDI. Inward FDI and its liquidation remain subject to approval requirements in several areas (Table 1).

  • Portfolio investment is controlled by quotas. Inward investment is channeled through Qualified Foreign Institutional Investors (QFII), subject to a 3-month lock-in period for most shares, and an aggregate ceiling of US$150 billion (since July 2013). In 2011, an R-QFII scheme was introduced that allows qualified firms to invest offshore renminbi back into China, subject to an overall ceiling that was raised to renminbi 270 billion by end-2012. Outward portfolio investment—for foreign securities purchased by residents—is channeled through Qualified Domestic Institutional Investors (QDII), subject to institution-specific ceilings that amounted to US$86 billion by end-2012. Cross-border issuance of securities requires approval.

  • Other investment. Foreign borrowing is subject to a ceiling (for short-term borrowing) or approval requirements (for long-term borrowing), but lending abroad is largely unrestricted. The holding of cross-border accounts requires SAFE approval.

uA01fig01

Qualified Foreign & Domestic Institutional Investor

(Approved Investment Fund)

Citation: IMF Working Papers 2013, 189; 10.5089/9781475532159.001.A001

Sources: CEIC; and IMF staff calculations.

Despite these restrictions, there are substantial non-FDI capital flows into and out of China. In particular, “other” investment flows are similar in absolute magnitude to those of a fully liberalized country such as Australia. These other investment flows include the buildup or drawdown of foreign currency deposits at domestic banks. These deposits have tended to fluctuate with exchange rate expectations (Ma and McCauley, 2003) as state-owned enterprises have adjusted their profit repatriation. Many capital flows also pass through the current account. Li (2008) estimated capital flows through mis-invoicing of trade flows on the order of 5.5–5.9 percent of GDP in 2007.

Capital flows, average 2005-2010

(percent of GDP)

article image
Source: Haver Analytics; IMF IFS; staff estimates.Note: Colours reflect the quartile of absolute values in each row, with red the lowest quartile and yellow the highest quartile. Data for Australia, Russia, Malaysia based on BPM6. For all others based on BPM5.

As China’s authorities speed up steps to open the capital account, historical experience in other countries serves as caution. Capital account liberalization has historically often been followed by exchange rate or banking crises (Box 1). That said, the link is not always as close as is sometimes protrayed. For example, the financial crises in the U.K. and Japan occurred about a decade after capital account liberalization and that in Denmark two decades later.

uA01fig02

The next crisis: Financial or exchange rate crisis following capital account liberalization

Citation: IMF Working Papers 2013, 189; 10.5089/9781475532159.001.A001

Sources: Kaminsky and Schmukler (2003), IMF (2012).

Several authors have explored the preconditions for “safe” capital account liberalization and have concluded that they are not yet fulfilled in China. Since Fischer (1998), the preconditions for successful capital account liberalization have generally been accepted to be a stable macroeconomic environment, a sound banking system, and developed financial markets. Already in 2001, Groomberg (2001) listed weak SOEs and banks as well as underdeveloped capital markets as risk factors in capital account liberalization for China.4 Most recently, Lardy (2011) also argued that the preconditions for “safe” capital account liberalization have not yet been fulfilled: without interest rate liberalization it is difficult to assess the soundness of the financial system; parts of the Chinese financial system, and especially the corporate bond market, are profoundly underdeveloped; and the exchange rate remains undervalued. Whether conditions are in place for smooth capital account opening, is a question that is beyond the scope of this paper.

III. Stylized Facts: Capital Account Liberalization Episodes Since the 1970S

Kaminsky and Schmukler (2003) have compiled a database that dates full or partial capital account, financial sector, and stock market liberalization. These dates are used to compare changes in portfolio and other investment capital flows (in percent of GDP) before and after liberalization.

Following liberalization, gross capital flows generally increased substantially. For example, capital account liberalization was followed by a buildup of gross international assets over the subsequent five years of some 60 percent of GDP in the United Kingdom (1979) and about half that amount in Chile (1992) and Italy (1992).

uA01fig03

Increase in gross international assets during five years following capital account liberalization

(Percent of GDP)

Citation: IMF Working Papers 2013, 189; 10.5089/9781475532159.001.A001

Sources: IMF IFS.Note: Data for UK only available for year after capital account liberalization.

The direction of net capital flows after liberalization, however, depended on many factors. Prasad and Rajan (2008) point to a plethora of nonlinearities and threshold effects that make predictions about the eventual impact of capital account liberalization unreliable. For example, capital account liberalization was followed by substantial net portfolio and other investment outflows in Sweden, Finland, and Spain, but inflows in Denmark, Chile, and Colombia.

Here we focus on four factors that could determine net flows: the domestic and global business cycle, the sequencing of ancillary reforms, and growth prospects. For each country a bilateral trade-weighted average real GDP growth of 12 of the largest economies is defined as “world” growth.5 For both domestic and world real GDP growth, business cycle peaks and troughs at least four quarters apart are identified using the algorithm of Harding and Pagan (2002). A country’s or the world’s position in the business cycle is described by a variable that is the fraction of an upswing underway (positive) or the fraction of a downswing underway (negative).6 Since data on both business cycles and capital flows is only available for some 16-21 countries, regression estimation is of doubtful value. However, there are a few correlations that are noticeable in scatter plots; especially for the relationship between net other investment flows and domestic growth and the relationship between portfolio investment flows and world growth and financial sector liberalization.

  • Domestic business cycle: Figure 1 a shows the business cycle position at the time of capital account liberalization against the change in net other investment inflows (in percent of GDP) between the year following liberalization and the year preceding it. Typically, the more advanced a domestic upswing, the greater the net outflows.7 This may have reflected residents seeking to diversify their domestic financial assets in upswings and borrowing in downswings.

  • Growth prospects: Figure 1b shows the correlation between average annual growth in the ten years following capital account liberalization (a proxy for growth prospects) and the change in net other investment inflows between the years following and preceding liberalization. Net inflows increased more in countries with higher growth prospects.

  • World business cycle: Figure 1c shows a similar plot to Figure 1a, of the world business cycle position at the time of capital account liberalization and the change in net portfolio investment inflows. A more advanced upswing in the world business cycle typically increased net portfolio inflows. This may reflect inflows from nonresidents seeking to diversify their securities exposures when the global business growth is buoyant and investor sentiment sanguine.

  • Financial sector liberalization: Figure 1d plots the number of years from financial sector liberalization to capital account liberalization (both dated as in Kaminsky and Schmukler, 2003) against the change in net inflows of portfolio investment (in percent of GDP) in the year following liberalization and the year preceding it. In general, the more recent financial repression, the greater net outflows in portfolio investment tended to be. In contrast, in Japan and the United States, where capital account liberalization preceded financial sector liberalization by several years, net inflows were negligible after capital account liberalization.

Figure 1.
Figure 1.

Change in Net Financial Flows and Business Cycle at Time of Capital Account Liberalization

Citation: IMF Working Papers 2013, 189; 10.5089/9781475532159.001.A001

1/ Business cycle defined as share of real GDP growth upturn completed from trough to peak (+) or share of downturn completed from peak to trough (-). Peak and trough dated using Harding-Pagan (2002) algorithm. Timing of financial sector and capital account liberalization as in Kaminsky and Schmukler (2003).2/ Trend growth is defined as real GDP growth over the ten years following capital account liberalization.

The experiences of capital account liberalization in the U.K. and Japan in 1979 and 1980, respectively, serve as contrasting examples. The U.K. (Box 2) liberalized its capital account in 1979 and removed remaining financial sector restrictions shortly thereafter. The liberalization was followed by substantial net capital outflows in a stock adjustment that has been estimated at up to 10 percent of GDP (Taylor and Artis, 1989).8 Following two decades of alternating liberalizing and restricting measures, Japan (Box 3) eventually liberalized its capital account in 1980 but the financial sector remained heavily restricted. Capital account liberalization was initially not followed by any significant net capital flows. Financial sector liberalization only began in 1985. (Note that despite the different timing of the Japanese and U.K. financial sector liberalizations, both countries experienced a credit boom in the late 1980s and a housing market crash in 1989–90.)

IV. Estimating the Possible Impact of Capital Account Liberalization

The previous section described capital flows over a short two-year window. Such short-term flows would be motivated not only by capital account liberalization but also by short-term factors such as business cycles. A long history of capital controls, however, is likely to cause a stock adjustment that may take longer than two years and is driven by structural factors such as financial market development. This is what we explore next.

The baseline regression equation is the generalized, multi-country version of that used in Forbes (2010) for the U.S. alone:9

wi⁢j=β0⁢wj*+β1⁢ci−(β2⁢ci⁢,⁢j−β3⁢cj)

where wij is the share of country i’s total portfolio investment that is invested in country j, wj* is country j’s share in the world market portfolio, ci are the marginal cost of investing in country i, cij are the marginal cost of an investor in country i investing in country j, and cj are the marginal cost of investing in country j. The first term on the right hand side controls for the share of country j in the global stock market portfolio. The second term captures the fact that higher cost of investing in country i increase the share of portfolio investment abroad, in any country j.

The last term in brackets captures the notion that country i will invest a greater share of its portfolio assets in country j than the world market share if country i’s cost of investing in j are less than world average cost of investing in j.10

This regression specification is similar to that of Bertaut and Kole (2004) and Lane and Milesi-Ferretti (2008) but expanded to focus on capital controls which is a core variable of interest here. Capital controls are not included in the regressions reported in Bertaut and Kole (2004), only source country controls are included in the regressions reported in Lane and Milesi-Ferretti (2008) and only destination country controls in Guo (2011).

The focus on bilateral exposures mitigates some of the concerns about reverse causality with which previous studies have struggled.11 The independent country-level variables are unlikely to be driven by the dependent variable of diversified bilateral exposures to investment partner countries. The size of source country and destination country financial markets are the exception. Since they include foreign holdings, they could in principle be biased by reverse causality. We therefore include an instrumental variables regression as robustness check. Forbes (2010) highlights other econometric challenges of the regression specification above: persistence in the dependent variable that is defined as a stock variable—and hence likely autocorrelation in the error term—and heteroskedasticity in the variance across country pairs. Following Forbes, an FGLS estimation is used that allows for heteroskedasticity and country-pair-specific autocorrelation in the error term with a first-order autoregressive process.

The dependent variable, the share of country i’s total portfolio investment that is invested in country j, is calculated using bilateral portfolio assets (equity and debt separately) reported in the IMF’s CPIS database as a share of country i’s total securities portfolio. Unfortunately, China does not report CPIS data. Constrained by this lack of data, we assume that international experience over the decade is still a relevant—even if only approximate—benchmark for China. For equities, country i’s total securities portfolio consists of domestic stock market capitalization as reported in the World Bank’s Global Financial Development Database plus all international equity assets of country i minus country i’s international equity liabilities. Similarly, for debt, the size of the domestic holdings is defined as outstanding domestic private and public debt securities as reported in the World Bank’s Global Financial Development Database.12 Although data on bilateral international equity and bond assets are available to 2011, the broader cross-country financial development indicators are only available to 2010.13

The cost of investing in any country depend on the depth and size of financial markets, on capital controls, on information asymmetries, and return differentials and correlations. In line with Forbes (2010) but updating the data to 2010, the following variables are included to capture the cost of investing:14

  • Financial market size: A larger financial market is likely to be a more liquid one and hence reduce the cost of investing. For regressions on equity assets, stock market capitalization in percent of GDP from the World Bank’s Global Financial Development Database is used and for regressions on bond assets, outstanding domestic public and private debt securities in percent of GDP from the same database are used.

  • Capital controls: Capital controls can raise the cost of investing cross-border both for residents and nonresidents. Schindler (2009) develops a detailed categorization of equity and bond inflow and outflow restrictions. Using the same methodology, his dataset is updated to 2011.

  • Information asymmetries: Existing trade or other nonfinancial ties and greater proximity are likely to reduce information asymmetries and reduce the cost of investment between any country pair. Such ties are proxied by bilateral trade (exports and imports) in percent of GDP using data from the IMF’s DOTS database. The population-weighted geodesic distance between the largest cities of both partner countries is available from Mayer and Zignano (2011).

  • Return differentials: A greater relative return in the host market makes portfolio investment in the host country more attractive. For regressions on equity portfolio assets, the annual average of monthly stock market returns for the stock market index with the broadest coverage in 59 countries in the sample is used based on Bloomberg data. For regressions on bond portfolio assets, the annual average 5-year sovereign bond yield for 37 countries is used based on Bloomberg data.

  • Return correlations: Risk diversification, e.g., as measured by a minimum variance portfiolio, is a key motivation for cross-border investment. Hence the bilateral correlation coefficient in monthly stock market returns or sovereign bond yields over the past three years is added as a control. Alternatively, high bilateral correlations may proxy fewer informational asymmetries.

  • Governance: Better governance increases the transparency of investment and thus may reduce investment cost. Governance is proxied by the first standardized principle component of indicators of control of corruption, rule of law, and regulatory quality, all available from the World Bank’s World Governance Indicators database.

As expected, the results confirm that international investors seek exposure to deeper financial markets and higher returns. Table 3 shows the baseline regression results for equities in columns I-II and for bonds in column IV-V. There is a concern that the bilateral holdings or the denominator in the dependent variable may be correlated with financial market size in the source and destination country, Columns III and VI show estimation results from an instrumental variables regression. For equity exposures, financial market size is instrumented with stock market value traded in percent of GDP (Column III).15 For bond exposures, financial market size is instrumented with the share of private debt in total domestic outstanding debt (Column VI).16 Table 4 tests the robustness of the results in column I and IV by removing each source and each destination country at a time and showing the range of coefficient estimates.

Table 3.

FGLS regression: Share of bilateral portfolio assets in total portfolio, 2005-2010

article image
pval in parentheses*** p<0.01, ** p<0.05, * p<0.1Note: Regression includes time fixed effects.

Stock market capitalization is instrumented with stock market value traded in percent of GDP.

Total outstanding debt is instrumented with the share of private domestic debt in total domestic outstanding debt (in percent).

Table 4.

FGLS regression: Share of bilateral portfolio assets in total portfolio, 2005-2010

article image
pval in parentheses*** p<0.01, ** p<0.05, * p<0.1, † p<0.11.Note: Regression includes time fixed effects.

A few robust results stand out. First, for both asset classes, larger source country financial markets discourage domestic investors from investing abroad. The effect of larger destination market size is more complicated. Although in the baseline FGLS regressions larger destination markets attract foreign equity exposures, these results are not robust to our instrumental variables specification. In contrast, the counter-intuitive sign on destination bond market size in Columns III and IV corrects in our instrumental variables specification or once Japan or the U.K. are removed from the sample: Japan—despite being the second largest bond market in the world—has only half the average share of global bond holdings; the U.K.—with a bond market one-eight the size of Japan’s—on average accounts for four times the average share of bond holdings. If either of these two countries is removed from the sample, the sign reverses to the expected positive sign and is significant (Table 4).17 Second, higher returns in the destination country (and, for bonds, lower returns at home) also encourage greater exposures of domestic investors in foreign markets. Being significantly the opposite of the expected sign, the correlation in bond yields is clearly picking up effects other than diversification; this could reflect correlation being another proxy for information cost. Third, the other control variables have the expected signs: cross-border portfolio exposures are greater if home and host country have better governance and if there are greater other bilateral ties (greater trade and proximity). In general, bond exposures are more responsive than equity exposures to returns, source country financial market depth, destination country governance, and other bilateral ties.18

Capital controls both in the source and the destination country significantly reduce cross-border portfolio exposures. In general, equity exposures appear less sensitive to capital controls than bond exposures and less sensitive to destination country controls than source country controls. For equity exposures, the coefficients on capital controls on equity outflows from the source country and equity inflows into the destination country (Column II in Table 3) are somewhat stronger than a wider measure of controls that averages in- and outflows (Column I in Table 3).

What if investors, instead of separately deciding on bonds and equities portfolios, view equities and bonds as substitutes in their portfolio decision? Table 5 shows the results for the allocation of portfolio assets in total (bonds plus equities). Both the control variables and capital controls retain their significance and broadly similar magnitudes to those in the regression for bond exposures. However, returns in source and destination country are too poorly measured as simple averages to generate robust and significant coefficient estimates.

Table 5.

Panel regression: Share of bilateral portfolio assets in total portfolio, 2005-2010

article image
pval in parentheses*** p<0.01, ** p<0.05, * p<0.1Note: Regression includes time fixed effects.

The results are broadly robust to including country dummies, a broader or narrower sample, or wider measures of capital controls. Columns I and VI of Table 6 show the baseline regressions of Table 3 as reference. Not surprisingly, adding country fixed effects (Columns II and VII) removes the significance of some of the country-level explanatory variables, including capital controls for both equities and bonds. During the sample period 2005–10, capital controls were mostly unchanged. As a result, their effect is absorbed by country fixed effects. Stock market prices and sovereign bond yields are only available for a subsample of 59 and 37 countries, respectively. Columns III and VIII of Table 6 exclude assets returns and correlations and show results for the resulting larger sample. In this larger sample, the effect of restrictions on bond and equity flows shrink in magnitude but remain significant. Replacing of the measure of capital controls in Schindler (2009) with that of capital account openness as in Chinn and Ito (2008) confirms that capital account openness increases bilateral exposures (except for source country controls in equity markets, Columns IV and IX). Finally, the two remaining large countries with heavily restricted capital accounts are China and India. Results excluding them—and Hong Kong SAR as gateway for investment in and from China—are shown in Columns V and X of Table 6. The coefficient estimates on capital controls remain significant and negative although they fall somewhat for equity exposures.

Table 6.

FGLS regression: Share of bilateral portfolio assets in total portfolio, 2005-2010

article image
pval in parentheses*** p<0.01, ** p<0.05, * p<0.1Note: Regression includes time fixed effects.

V. Implications for China and Conclusions

Using the coefficient estimates of the regressions above, one can speculate about the effects of Chinese capital account liberalization on the accumulation of foreign portfolio assets and liabilities. This speculation is necessarily partial-equilibrium. It does not take into account other changes in the macroeconomic environment that would occur as the capital account is opened: changes in interest rates domestically and abroad; structural changes in financial markets and institutions or governance; offsetting other investment flows; or offsetting changes in reserve accumulation policy.

uA01fig04

Predicted Impact of Capital Account Liberalization on China’s International Portfolio Assets and Liabilities 1/

(Percent of GDP)

Citation: IMF Working Papers 2013, 189; 10.5089/9781475532159.001.A001

Note: Range of changes in gross international debt and equity assets and liabilities predicted by regression coefficients of regressions in Tables 3 and 6.1/ Portfolio assets and liabilities exclude official reserve assets.

The magnitude of the predicted stock adjustment in gross exposures and the resulting net flows is subject to substantial uncertainty, depending on the underlying regression specification. For example, applying the coefficient estimates on capital controls of Columns I through V of Table 3 and Columns III, V, VIII, and X of Table 6 to data for 2010 yields a wide range of estimates of counterfactual stock adjustments in China’s international portfolio assets and liabilities had there been no capital controls.19 These estimates suggest that capital account liberalization may be followed by a stock adjustment of Chinese assets abroad on the order of 15–25 percent of GDP and a smaller stock adjustment for foreign assets in China on the order of 2–10 percent of GDP.

This would imply a net accumulation of Chinese net international assets of 11–18 percent of GDP (Table 7). While the magnitudes differ, the direction is in line with that estimated by He and others (2012) and Sendik and Sun (2012).

Table 7.

Impact of hypothetical capital account liberalization in China on gross portfolio investment, 2010

(percent of GDP)

article image

Applies to total capital inflows and outflows.