The Feldstein–Horioka Puzzle and Structural Breaks: Evidence from the Largest Countries of Asia

Abstract

The purpose of this article is to investigate the level of capital mobility in the largest economies of Asia by testing the Feldstein–Horioka puzzle. Panel estimations using quarterly data for the period from 1995 to 2011 have been made for the seven largest economies of Asia, specifically Russia, Japan, South Korea, Turkey, India, Indonesia and China. This group of countries has gained significant economic power in the world over the last decade. Specifically, the growth rates of the sample has for a long period of time exceeded the growth rates of most developed countries. The total GDP adjusted for PPP is far above of the GDP of the European Union and NAFTA groups and very close to the G7 group. The article examines changes in investment savings relationships when the presence of structural shifts—where such exist—is taken into account. Recently developed panel techniques are employed to examine the investment–savings relationship and estimate saving–retention coefficients. As a result of these estimations, countries were divided into two groups consisting of stable and unstable economies. This division of countries allows for more precise estimates of capital mobility. The empirical findings reveal the existence of the moderate level of capital mobility in the group. Results indicate that countries with relatively higher capital mobility are exposed to the negative effects of international market fluctuations.

JEL Classification: F32

Keywords

Feldstein–Horioka puzzle saving–investment association capital mobility cointegration structural breaks Asia

1. Introduction

Of late, the level of financial integration in the world has significantly increased. The spread of the effects of economic crises throughout the world is evidence of this. Many studies investigating capital mobility have applied the work of Feldstein and Horioka (1980) on the saving–investment relationship and found that the investment and savings ratios are highly correlated in developed countries, which is a sign of low capital mobility (see, for example, Chen & Shen, 2015; Ho, 2002; Hussein, 1998; Jansen, 1996). These findings are contrary to the expectation of a low correlation between investment and savings, particularly in developed OECD countries. Due to a relatively established financial infrastructure, developed countries are expected to have high international capital mobility, which is determined by a low correlation between investment and savings. An alternative explanation of the high correlation between the investment and savings ratios is suggested in the literature as well. Hence, Coackley, Kulasi and Smith (1996) proposed that the widely discussed high value of the saving retention coefficient in developed countries does not represent a puzzle, but indicates the solvency constraint existence, without connection to the degree of capital mobility. This interpretation of the saving retention coefficient has been supported by many other studies (see, for example, Ma & Li, 2016; Murthy, 2007; Nell & Santos, 2008; Pelgrin & Schich, 2004). Since then, a great deal of the literature has focused on the Feldstein–Horioka puzzle (FHP) with particular attention to the European or OECD countries (see, for example, Fouquau, Hurlin, & Rabaud, 2008; Herwartz & Xu, 2010; Ketenci, 2012, 2013; Kollias, Mylonidis, & Paleologou, 2008).

Many studies have been devoted to Asian countries, as well (see, for example, Huang & Guo, 2006; Jiranyakul & Brahmasrene, 2009; Kaya-Bahçe & Özmen, 2008; Kim, Kim, & Wang, 2007). However, less attention has been paid to Asian countries in panel research (see, for example, Guillaumin, 2009; Kim et al., 2007; Wahid, Slahuddin, & Noman, 2008), and there are no examples of panel studies of the largest Asian economies. This group of countries is worth studying due both to their dynamic development over the last years and also for the importance of their combined market. Table 1 shows GDP information for selected countries, as well as for major economic groups, for comparison. Note that the current and real GDP of these countries exceed the GDP levels of large markets such as the European Union (EU) and NAFTA, and is very close to those of the G7 countries. This, even though the regional average of real GDP per head is significantly lower compared to leading economies. However, from the last two columns of the table it may be determined that the share of the world’s real GDP for all major regions—EU, G7 and NAFTA—has decreased since 2005, but the share of the world’s real GDP for the Asian countries under consideration has increased from 28 per cent to 33 per cent. From Table 2, it can be seen that the growth of the largest Asian countries significantly outpaces the growth of the world’s leading economic markets. For example, 2009 is characterised by a negative change in GDP of about 4 per cent in all leading economies—EU, G7 and NAFTA. The average growth of the largest economies of Asia in 2009 was positive, though these positive growth rates are attributable to high growth in China, India and Indonesia, while other countries experienced a decline in output. Years 2010 and 2011 are characterised by a considerable difference in the growth rates of the largest economies of Asia¹ in comparison with the EU, G7 and NAFTA. In 2010, the lowest observed growth among these economies was in Russia and Japan at about 4 per cent, and the highest rates of growth were in China, Indonesia and Turkey, averaging 9–10 per cent. In the same year, the EU experienced only 2 per cent growth and NAFTA about 4 per cent. In 2011, growth rates declined across the board, but the relative trend remained the same. This tendency favouring Asian economies is not new; with the notable exception of the Asian financial crisis period, it has existed for decades.

Table 1
GDP of Major Asian Countries and Major Economic Groups, 2011

Country GDP^a GDP PPP^b GDP PPP per Head^c GDP PPP Share of World Total (per cent) GDP PPP Share of World Total (per cent) (2005)

China 5,878,257 10,085,708 7,518 14.32 9.45

India 1,537,966 4,060,392 3,339 5.65 4.28

Indonesia 706,735 1,029,884 4,394 1.42 1.24

Japan 5,458,872 4,309,432 32,817 5.62 6.85

Russia 1,465,079 2,222,957 15,807 3.02 2.99

South Korea 1,007,084 1,459,246 30,200 1.97 1.93

Turkey 729,051 960,511 13,392 1.36 1.32

Total 19,434.18 26,333.89 16,349.77 33.37 28.05

EU 17,577.69 15,821.26 31,607.39 20.05 23.03

G7 33,670.02 30,355.27 40,891.57 38.47 45.03

NAFTA 17,985.68 18,151.80 34,512.52 23.01 26.50

Country	GDP^a	GDP PPP^b	GDP PPP per Head^c	GDP PPP Share of World Total (per cent)	GDP PPP Share of World Total (per cent) (2005)
China	5,878,257	10,085,708	7,518	14.32	9.45
India	1,537,966	4,060,392	3,339	5.65	4.28
Indonesia	706,735	1,029,884	4,394	1.42	1.24
Japan	5,458,872	4,309,432	32,817	5.62	6.85
Russia	1,465,079	2,222,957	15,807	3.02	2.99
South Korea	1,007,084	1,459,246	30,200	1.97	1.93
Turkey	729,051	960,511	13,392	1.36	1.32
Total	19,434.18	26,333.89	16,349.77	33.37	28.05
EU	17,577.69	15,821.26	31,607.39	20.05	23.03
G7	33,670.02	30,355.27	40,891.57	38.47	45.03
NAFTA	17,985.68	18,151.80	34,512.52	23.01	26.50

Source: Based on statistical data from the IMF.

Note: Figures in ^abillions of current U.S. dollars, ^bbillions of current international dollars, ^ccurrent international dollars.

Table 2

GDP Growth Rates, 2005–11 (per cent compared to the previous year)

Country	2005	2006	2007	2008	2009	2010	2011
China	11.310	12.677	14.162	9.635	9.214	10.447	9.237
India	9.033	9.530	9.991	6.186	6.579	10.623	7.241
Indonesia	5.693	5.501	6.345	6.014	4.629	6.195	6.457
Japan	1.303	1.693	2.192	–1.042	–5.527	4.435	–0.748
Russia	6.388	8.153	8.535	5.248	–7.800	4.300	4.300
South Korea	3.957	5.179	5.106	2.298	0.319	6.320	3.634
Turkey	8.402	6.893	4.669	0.659	–4.826	9.006	8.460
Average¹	6.584	7.089	7.286	4.143	0.369	7.332	5.512
EU¹	2.186	3.6	3.395	0.512	–4.208	2.003	1.618
G7¹	2.282	2.608	2.243	–0.38	–4.042	3.036	1.379
NAFTA¹	3.090	3.543	2.452	0.513	4.177	3.929	2.721

Source: Based on statistical data from the IMF.

Note: ¹ Average calculations.

This study differs from others on capital mobility in the following respects: First, it contributes to the literature on international capital mobility by providing robust estimation results using the latest econometric techniques. Second, the research investigates the relationships for the seven largest Asian countries by GDP and ascertains levels of capital mobility within the group. Third, the Hansen (1992) stability test has allowed for analysis of countries in different panels according to their relative stability, and such subdivision of the countries under consideration provides detailed results that are distinct from estimations that do not take stability into account.

The remainder of the article consists of the following sections: Section 2 outlines the empirical methodology adopted in the article. Section 3 presents the empirical results and Section 4 draws conclusions on the data.

2. Methodology

This study investigates the degree of capital mobility of the seven largest Asian countries, taking into account identified structural breaks. In order to estimate the level of capital mobility in OECD countries, Feldstein and Horioka (1980) used the following equation:

{(\frac{I}{Y})}_{i} = α + β {(\frac{S}{Y})}_{i} + e_{i},

(1)

where I is gross domestic investment, S is gross domestic savings and Y is the gross domestic product of the country under consideration, i. Coefficient β—known as the saving–retention coefficient—measures the degree of capital mobility. If a country possesses perfect capital mobility, the value of β should approach 0. As the value of β approaches 1, it suggests the immobility of the country’s capital. The results of Feldstein and Horioka’s analysis showed that the value of β for 21 open OECD economies varied from 0.871 to 0.909, demonstrating a relative immobility of international capital in the countries considered. Such controversial results set off widespread debate in the economic literature, and while numerous studies have corroborated the results, at the same time, contradictory results exist in the literature along with an array of possible interpretations. The findings of Feldstein and Horioka (1980), which are indeed contrary to economic theory, have subsequently been referred to as ‘the mother of all puzzles’ (Obstfeld & Rogoff, 2000, p. 9).

2.1 Unit Root Tests

In this article a variety of tests for the panel unit root are employed. The first group consists of tests that do not allow for structural changes in series, constituted by the Levin, Lin and Chu (LLC) test (Levin, Lin, & Chu, 2002), the Breitung (Breitung, 2001) test, the Im, Pesaran and Shin (IPS) test (Im, Pesaran, & Shin, 2003), Fisher-type tests that employ Augmented Dickey–Fuller (ADF) and PP tests (Choi, 2001; Maddala & Wu, 1999) and Hadri tests (Hadri, 2000). The LLC test is based on orthogonalised residuals and on correction by the ratio of the long-run to short-run variance of each variable. Although the test has become a widely accepted panel unit root test, it has a homogeneity restriction allowing for heterogeneity only in the constant term of the ADF regression. The Breitung test assumes that all panels have an autoregressive parameter and a unit root process in common. The IPS test is a heterogeneous panel unit root test based on individual ADF tests and was proposed by Im et al. (2003) as a solution to the homogeneity issue. It allows for heterogeneity in both the constant and slope terms of the ADF regression. Maddala and Wu (1999) and Choi (2001) proposed an alternative approach employing the Fisher test, which is based on combining the p values from individual unit root test statistics such as ADF and PP. One of the advantages of the Fisher test is that it does not require a balanced panel. Finally, the Hadri test is a heterogenous panel unit root test that extends the KPSS (Kwiatkowski–Phillips–Schmidt–Shin) test—outlined in Kwiatkowski, Phillips, Schmidt and Shin (1992)—to a panel with individual and time effects, as well as deterministic trends. It takes as its null hypothesis the stationarity of the series.

Altogether, unit root tests do not take into account the presence of structural shifts in series. The LM unit root test, proposed by Im, Lee and Tieslau. (2005), confronts this issue. It is a panel extension of the Schmidt and Phillips (1992) test that allows for one and two structural shifts in the trend of a panel as well as of every individual time series. Im et al. (2005) illustrated that in a series where structural shifts do not exist, the size of distortions and the loss of power in the panel unit root tests remain insignificant when structural shifts are accommodated. However, size distortions and power loss in the tests are significant when unit root tests were applied to the time series without taking into account existing structural shifts. The break date in the Im et al. (2005) test is chosen using the minimum LM statistics of Lee and Strazicich (2003, 2004), that is to say, when the t-statistic of possible break points is minimised.

2.2 Stability Test

In order to apply panel cointegration tests that allow structural shifts, it is necessary to examine a series for stability. Hansen’s (1992) stability test has been employed to estimate parameter stability in cointegration relationships. The test is based on fully modified OLS residuals proposed by Phillips and Hansen (1990), and a prerequisite of the test is that the series be non-stationary. The stability test produces three test statistics: supF, meanF and Lc. The supF statistic tests for the null hypothesis of cointegration with no structural shift in the parameter vector against the alternative hypothesis of cointegration in the presence of sudden structural shifts. The meanF and Lc statistics test for cointegration with constant parameters against an alternative hypothesis of gradual variance in parameters with no cointegration. Particularly, the meanF statistic is used to capture the overall stability of the model.

2.3 Cointegration Tests

Cointegration tests were employed to determine whether long-run relationships exist between investment and savings. Two of them are the Kao (1999) and the Pedroni (1999) cointegration tests, which do not allow for structural shifts in series. This is followed by the Westerlund (2006) panel cointegration test, which does allow for multiple structural breaks in series. The following system of cointegrated regressors is considered for estimation in cointegration tests:

y_{i t} = α_{i} + β x_{i t} + ε_{i t},

(2)

where i = 1, …, N, and t = 1, …, T, α_i are constant terms, β is the slope, y_it and x_it are non-stationary regressors, and ε_it are stationary disturbance terms. Kao (1999) proposed two types of panel cointegration tests: the Dickey–Fuller and the ADF tests. The statistics of these tests can be calculated using the following formula:

{\overset{?}{ε}}_{i t} = ρ {\overset{?}{ε}}_{i t - 1} + \sum_{j = 1}^{p} θ_{j} Δ {\overset{?}{ε}}_{i t - j} + u_{i t},

(3)

where the residuals derived in the system (2) are used to calculate the test statistics (3) and tabulate the distributions. The null hypothesis of the test is H₀: ϕ = 1 versus the alternative H₁: ϕ < 1.

Pedroni (1999) developed a panel and group cointegration test where seven residual-based tests (with four panel statistics and three group statistics) were introduced in order to test the hypothesis of no cointegration in dynamic panel series with multiple regressors. The first four panel cointegration tests, which are defined as within-dimension-based statistics, use the null and alternative hypotheses: H₀: ϕ = 1 and H₁: ϕ < 1, respectively, and assume the homogeneity of coefficients under the null hypothesis. The other three groups of statistics, which are defined as between-dimension-based statistics, use H₀: ϕ_i = 1 versus H₁: ϕ_i < 1 for all i and assume a heterogeneous slope across countries under the alternative hypothesis.

In the long run, macroeconomic series such as investment and savings may contain a variety of structural changes at the domestic or international levels. Therefore, in order to examine the regression model (1) in the case when structural breaks are detected, the methodology of Westerlund (2006) has been employed. This is a panel cointegration test that accommodates multiple structural breaks in the level as well as in the trend of cointegrated regression. It is based on the panel cointegration residual-based LM test proposed by McCoskey and Kao (1998), which does not allow for structural shifts. The advantage of Westerlund’s test is that it can allow for the possibility of multiple known, a priori structural breaks or, alternatively, allow for breaks the locations of which are determined endogenously from the series. At the same time, the test allows for the possibility of structural breaks that may be placed at different locations in different individual series. To estimate the location of breaks, Westerlund (2006) applied the approaches of Bai and Perron (1998, 2003), which are based on the global minimisation of the sum of squared residuals. He thus showed that the test is free of nuisance parameters under the null hypothesis and that the number and location of structural shifts do not affect the limiting distribution. The null hypothesis of the test is H₀: ϕ_i = 0 for all i = 1, ..., f, N, versus the alternative hypothesis H₁: ϕ_i ≠ 0 for i = 1, ..., N₁ and ϕ_i = 0 for i = N₁ + 1, ..., N. One of the important advantages of the test is that the alternative hypothesis is not merely a general rejection of the null, as in the commonly used LM panel cointegration test of McCoskey and Kao (1998), but instead allows ϕ_i to differ across individual series.

2.4 Saving–retention Coefficient

Finally, in order to estimate saving–retention coefficients for groups of countries the dynamic ordinary least squares (DOLS) technique was employed. The DOLS estimator was proposed by Kao and Chiang (2001) for heterogeneous panels. They illustrate that DOLS outperform ordinary least squares and fully modified ordinary least squares estimators in estimating cointegrated panel regressions.²

3. Empirical Results

3.1 Unit Root Tests

The integration order of panel series must be investigated in order to test the cointegration relationships between investment and savings panel series and to estimate saving-retention coefficients for the panel of the selected Asian countries. The results of six alternative unit root tests are presented in Table 3. All tests provided sufficient evidence to conclude that investment series are non-stationary. While most tests also provide evidence for the presence of the unit root in savings series, the Breitung and PP tests rejected this hypothesis. Based on the results of these alternative unit root tests, it may therefore be concluded that savings series are generated by a non-stationary stochastic process.

Table 3
Unit Root Tests

Variable Level Δ Variable Level Δ

Investment Savings

LLC^a –0.47 –22.41* LLC 1.82 1.40

I(1) I(0) I(1) I(1)

Breitung^a –0.40 –12.35* Breitung –2.23* –4.39*

I(1) I(0) I(0) I(0)

IPS^b –0.39 –21.39* IPS –0.27 –6.49*

I(1) I(0) I(1) I(0)

ADF^b 16.46 169.87* ADF 10.85 66.26*

I(1) I(0) I(1) I(0)

PP^b 16.03 192.17* PP 28.38 187.00*

I(1) I(0) I(0) I(0)

Hadri^c 6.14* –0.58 Hadri 6.11* –0.69

I(1) I(0) I(1) I(0)

Variable	Level	Δ	Variable	Level	Δ
Investment			Savings
LLC^a	–0.47	–22.41*	LLC	1.82	1.40
	I(1)	I(0)		I(1)	I(1)
Breitung^a	–0.40	–12.35*	Breitung	–2.23*	–4.39*
	I(1)	I(0)		I(0)	I(0)
IPS^b	–0.39	–21.39*	IPS	–0.27	–6.49*
	I(1)	I(0)		I(1)	I(0)
ADF^b	16.46	169.87*	ADF	10.85	66.26*
	I(1)	I(0)		I(1)	I(0)
PP^b	16.03	192.17*	PP	28.38	187.00*
	I(1)	I(0)		I(0)	I(0)
Hadri^c	6.14*	–0.58	Hadri	6.11*	–0.69
	I(1)	I(0)		I(1)	I(0)

Source: Author’s calculations.

Notes: Estimations are made with the inclusion of constant and trend. Estimations are made with maximum 4 specified lag. With the increase of lag, the length of the power of tests increases in favour of the unit root presence in level estimations.

* Significance at a 5 per cent level.

^a Tests the hypothesis that the common unit root process is present.

^b Tests the hypothesis that the individual unit root process is present.

^c Tests the hypothesis that there is no unit root in the common unit root process.

The purpose of this article is to investigate changes in investment savings relationships in the largest economies of Asia, taking structural shifts into account when they exist. To obtain stronger evidence for the presence of a unit root in both unstable and stable series, panel unit root tests that allow for one and two structural shifts in series as proposed by Im et al. (2005) were applied, and the results are summarised in Table 4. All forms of the LM unit root tests—with no structural shifts, with one, and with two shifts—provide strong evidence for the presence of the unit root in investment and savings panel series. With regard to individual countries, the LM statistics failed to reject the stationarity hypothesis only in the case of Indonesia when no shifts were allowed. When one and two structural shifts were allowed, the tests provided strong evidence of non-stationarity in all countries.

3.2 Stability Test

To examine series for stability, Hansen’s (1992) stability test was applied to non-stationary series, and the results are presented in Table 5. In the cases of Russia, Turkey and China, all the statistics reject the null hypothesis of the stability of the model parameters, while in the case of Indonesia, only the supF statistic supports the stability hypothesis; the MeanF and the Lc statistics suggest the instability of model parameters. On the other hand, all statistics support the null hypothesis in the cases of Japan, South Korea and India. Further estimations of panels have to be made on the basis of series stability. Taking into account the results of the stability test, the considered countries may be divided into two groups. The first includes Japan, South Korea and India, where no evidence of structural changes was found. The other group would include Russia, Turkey, Indonesia and China, where at least one of the stability test statistics suggested instability.

3.3 Cointegration Tests

Table 6 presents the results of Pedroni (1999) and Kao (1999) panel cointegration tests that were conducted on the stable group: Japan, South Korea and India. Most statistics reject the null hypothesis of no cointegration. The estimations of the cointegration tests provide strong evidence for the presence of cointegration relationships between investment and savings series in the panel.

Table 7 presents the results of the Westerlund (2006) panel cointegration test with multiple structural breaks, which was conducted on unstable series: Russia, Turkey, Indonesia and China. The test was applied with a parameter to detect a maximum of five structural breaks. Panel A shows the results of the test where structural shifts are allowed in constant; Panel B illustrates the results where structural shifts are allowed in both constant and trend of the regression. For these countries, the estimations of the tests detected different numbers of breaks and different break locations. Breaks in Russia, Turkey and Indonesia were identified for the period 1997–98, years characterised by the Asian financial crisis and its contagion effects. Further, 1998 was the year of the Russian financial crisis, which led to the devaluation of the rouble. The test showed breaks in the fourth quarter of 2000 and the first quarter of 2001 for Turkey. These dates correspond to a stock market crash after which the Turkish economy spiralled into turmoil. The global financial crisis of 2008 and its broad effects are also captured by the test in the cases of Indonesia, China and Turkey.

Table 4
Panel Unit Root Test with Structural Shifts

Investment
No Shifts
One Shift
Two Shifts

Country LM Lag LM Break Lag LM Break1 Break2 Lag

Russia –8.06* 1 –11.07* Q4-2009 5 –11.72* Q3-1997 Q1-1999 5

Japan –6.10* 4 –9.48* Q4-1998 0 –10.89* Q3-2006 Q4-2008 0

South Korea –8.38* 1 –10.09* Q1-2010 5 –13.01* Q4-2004 Q1-2010 5

Turkey –8.03* 1 –11.38* Q2-1998 5 –13.35* Q4-2005 Q2-2007 5

India –9.24* 0 –10.15* Q2-2004 0 –13.44* Q4-2003 Q3-2005 5

Indonesia –1.83 3 –10.61* Q1-1997 5 –12.16* Q1-1997 Q3-1999 5

China –10.52* 0 –10.11* Q3-2005 5 –11.63* Q3-2007 Q1-2009 5

MinLM –10.11* Q3-2005 5 –11.63* Q3-2007 Q1-2009 5

LM statistic –24.412* –37.763* –46.232***

Savings
No Shifts
One Shift
Two Shifts

Country
LM
Lag
LM
Break
Lag
LM
Break1
Break2
Lag

Russia –7.63* 0 –9.63* Q4-2009 5 –11.27* Q3-1999 Q1-2010 5

Japan –8.11* 4 –11.49* Q4-2009 5 –10.52* Q1-2008 Q4-2008 5

South Korea –8.35* 4 –9.43* Q3-2008 5 –10.20* Q2-2002 Q1-2010 5

Turkey –6.16* 4 –9.92* Q3-2006 5 –10.89* Q3-2006 Q1-2010 5

India –8.64* 4 –11.80* Q2-2000 5 –14.91* Q2-2000 Q1-2010 5

Indonesia –6.36 0 –8.24* Q1-2009 0 –13.05* Q1-1997 Q3-2001 5

China –8.13* 4 –12.32* Q1-2003 5 –12.50* Q3-2001 Q1-2003 5

MinLM –12.32* Q1-2003 5 –12.50* Q3-2001 Q1-2003 5

LM statistic –25.410* –37.740* –44.441***

Investment	No Shifts	One Shift	Two Shifts
Russia	–8.06***	1	–11.07***	Q4-2009	5	–11.72***	Q3-1997	Q1-1999	5
Japan	–6.10***	4	–9.48***	Q4-1998	0	–10.89***	Q3-2006	Q4-2008	0
South Korea	–8.38***	1	–10.09***	Q1-2010	5	–13.01***	Q4-2004	Q1-2010	5
Turkey	–8.03***	1	–11.38***	Q2-1998	5	–13.35***	Q4-2005	Q2-2007	5
India	–9.24***	0	–10.15***	Q2-2004	0	–13.44***	Q4-2003	Q3-2005	5
Indonesia	–1.83	3	–10.61***	Q1-1997	5	–12.16***	Q1-1997	Q3-1999	5
China	–10.52***	0	–10.11***	Q3-2005	5	–11.63***	Q3-2007	Q1-2009	5
MinLM			–10.11***	Q3-2005	5	–11.63***	Q3-2007	Q1-2009	5
LM statistic	–24.412***		–37.763***			–46.232***
Savings	No Shifts	One Shift	Two Shifts
Country	LM	Lag	LM	Break	Lag	LM	Break1	Break2	Lag
Russia	–7.63***	0	–9.63***	Q4-2009	5	–11.27***	Q3-1999	Q1-2010	5
Japan	–8.11***	4	–11.49***	Q4-2009	5	–10.52***	Q1-2008	Q4-2008	5
South Korea	–8.35***	4	–9.43***	Q3-2008	5	–10.20***	Q2-2002	Q1-2010	5
Turkey	–6.16***	4	–9.92***	Q3-2006	5	–10.89***	Q3-2006	Q1-2010	5
India	–8.64***	4	–11.80***	Q2-2000	5	–14.91***	Q2-2000	Q1-2010	5
Indonesia	–6.36	0	–8.24***	Q1-2009	0	–13.05***	Q1-1997	Q3-2001	5
China	–8.13***	4	–12.32***	Q1-2003	5	–12.50***	Q3-2001	Q1-2003	5
MinLM			–12.32***	Q1-2003	5	–12.50***	Q3-2001	Q1-2003	5
LM statistic	–25.410***		–37.740***			–44.441***

Source: Author’s calculations.

Notes: For the minimum LM test with one break, the critical values for the 1, 5 and 10 per cent levels of significance are −5.11, −4.50 and −4.21, respectively (Lee & Strazicich, 2003).

***, ** and * denote the 1, 5 and 10 per cent levels of significance, respectively.

Table 5

Stability Tests in Cointegrated Relations

Country	SupF		MeanF		Lc
Country	Test	p Value	Test	p Value	test	p Value
Russia	0.53	0.08	9.43	0.01	58.55	0.01
Japan	0.14	0.20	1.05	0.20	1.36	0.20
South Korea	0.20	0.20	2.55	0.20	10.47	0.20
Turkey	1.07	0.01	12.26	0.01	40.66	0.01
India	0.19	0.20	2.73	0.20	9.46	0.20
Indonesia	0.16	0.20	12.34	0.01	24.26	0.01
China	0.39	0.17	4.42	0.16	12.95	0.11

Source: Author’s calculations.

Notes: p values are obtained from the GAUSS program and are associated with the computed statistics taken from Hansen (1992). A series is said to be stable if the estimated probability is 20 per cent. If the p value is smaller than 20 per cent, the null hypothesis that the model parameters are stable is rejected.

Table 6

Panel Cointegration Tests (for stable countries)

Stable		c	c&t
Pedroni
Panel v-Statistic	2.26**	0.72	–0.23
Panel rho-Statistic	–5.83**	–3.66**	–0.83
Panel PP-Statistic	–3.34**	–2.75**	–0.99
Panel ADF-Statistic	–3.67**	–3.02**	–1.39
Group rho-Statistic	–4.68**	–3.27**	–0.45
Group PP-Statistic	–3.56**	–2.61**	–0.46
Group ADF-Statistic	–4.08**	–2.97**	–1.30
Kao
ADF		–3.37**

Source: Author’s calculations.

Notes: The hypothesis for the Pedroni and Kao cointegration tests is the absence of cointegration.

** and * reject the hypothesis of no cointegration at the 1 and 5 per cent levels of significance, based, respectively, on critical values of 2.326 and 1.644. The critical values are retrieved from Pedroni (2004). Lag selection is based on the SIC with maximum of 3 lags.

Table 7

Estimated Structural Breaks Using the Approach of Westerlund (2006) (for unstable countries)

Country	Panel A: Breaks in Constant
Country	Breaks	Year
Russia	3	Q2-1997	Q2-2000	Q4-2006
Turkey	2	Q4-2000	Q3-2003
Indonesia	4	Q3-1998	Q2-2003	Q1-2006 Q3-2008
China	2	Q4-2004	Q1-2009
Lm	1.484
Country	Panel B: Breaks in Constant and Trend
Country	Breaks	Year
Russia	3	Q2-1998	Q4-2001	Q4-2006
Turkey	5	Q2-1998	Q1-2001	Q3-2003 Q2-2006 Q1-2009
Indonesia	3	Q2-1997	Q4-1999	Q4-2008
China	1	Q4-2005
Lm	3.581*

Source: Author’s calculations.

Notes: The null hypothesis of the test is cointegration.

*Rejects the hypothesis of cointegration based on the bootstrap p values at the 5 per cent level of significance. The breaks are estimated using the Bai and Perron (2003) procedure with a maximum of five breaks.

The statistics of the LM panel test support the null hypothesis of cointegration in the case when breaks are allowed in constant. However, when a break is allowed in constant and trend, the LM statistics reject the null hypothesis, providing no evidence for cointegration. It can be concluded that the investment and savings series in the panel of unstable countries are cointegrated only around a broken intercept. Moreover, the cointegration tests that were applied provide sufficient evidence for the presence of cointegration between investment and savings variables in stable as well as unstable countries, which in turn indicates the solvency of current accounts for these countries.

3.4 Saving–retention Coefficient

The saving–retention coefficient β from Equation 1 has been estimated in order to investigate the level of capital mobility in the panels. Table 8, Panel A, presents coefficients employing DOLS estimators with random and fixed effects. Saving–retention coefficients are estimated for three samples of the Asian countries under consideration: full, stable and unstable. The full sample includes all the countries. The second consists of countries found to be stable: Japan, South Korea and India. And the third sample includes only the unstable countries: Russia, Turkey, Indonesia and China.

Table 8
DOLS Estimations of the Saving–retention Coefficient

Sample 1995–2011
2000–11

RE FE RE FE

Panel A

Full 0.87 (44.12) 0.44 (4.89) 0.86 (41.44) 0.38 (4.39)

Stable 1.03 (44.36)** 0.46 (2.41)* 0.99 (69.94) 0.51 (2.85)

Unstable 0.79 (34.14) 0.58 (6.59) 0.79 (28.46) 0.35 (3.44)

Panel B

Full sample without China 0.87 (33.52) 0.25 (2.04)* 0.86 (32.80)** 0.37 (2.15)*

Unstable sample without China 0.75 (21.78)** 0.38 (1.98)* 0.74 (18.13)** –0.15 (0.82)

Sample	1995–2011	2000–11
Panel A
Full	0.87 (44.12)**	0.44 (4.89)**	0.86 (41.44)**	0.38 (4.39)**
Stable	1.03 (44.36)**	0.46 (2.41)*	0.99 (69.94)**	0.51 (2.85)**
Unstable	0.79 (34.14)	0.58 (6.59)**	0.79 (28.46)**	0.35 (3.44)**
Panel B
Full sample without China	0.87 (33.52)**	0.25 (2.04)*	0.86 (32.80)**	0.37 (2.15)*
Unstable sample without China	0.75 (21.78)**	0.38 (1.98)*	0.74 (18.13)**	–0.15 (0.82)

Source: Author’s calculations.

Notes: RE, random effect; FE, fixed effects.

t-statistics are given in brackets. Saving–retention coefficients β are estimated for three sets of countries: total, unstable and stable. The total set includes all countries of a given group, the second set includes only the unstable countries, while the last set includes only stable countries. In order to test the hypothesis that β = 0, critical values from the normal standard distribution are used. The critical values at the 1 and 5 per cent levels of significance for rejecting the hypothesis are 2.575 and 1.96, respectively.

In all samples, in both cases with random and fixed effects, the saving–retention coefficient was deemed significant and determined to be positive, as expected. The estimated value of the coefficient in random effects model exceeds 0.8 in all samples, indicating a low level of capital mobility; however, the results for the fixed effects model present significantly lower value of the coefficients indicating a medium level of capital mobility. In their study on saving–investment relationships in East Asian countries, Bautista and Maveyraud-Tricoire (2007) found that saving–retention coefficients changed from a pre-crisis high value to a lower value in the period following the Asian economic crisis. Similar results were found by Jun (2011), where declining saving–retention coefficients indicate increasing capital mobility for 19 Asian countries over the period 1960–2006. In the current study, post-crisis saving–retention coefficients were examined, as well (Table 8, 2000–11 period), and a declining trend was observed in the model with fixed effects for the full sample and for unstable countries. The saving–retention coefficient for stable countries in the fixed effects model stayed nearly similar in the post-crisis period. The decline in the level of the fixed effects saving–retention coefficients indicates the presence of individual country-specific effects that are overlooked by the random effects model. The best estimator in the case of the heterogenous panels is the DOLS, whose characteristics are discussed in details in Kao and Chiang (2001).

Panel B of Table 8 presents saving–retention coefficient estimates for panel samples that exclude China.³ The saving–retention coefficient for this sample for the 1995–2011 period has not changed in the random effects model; however, the fixed effects model demonstrated values at 0.25 and 0.38 levels for the full sample and for the sample with unstable countries, respectively, which are almost half the values in the case when China is included. This suggests high capital mobility in the considered region. The exclusion of China did not change results for the post-crisis period. The estimates for the sample of stable countries including Japan, India and South Korea did not change. China has the largest economy in the world after the United States,⁴ but the exclusion of China from the panel decreases the saving–retention coefficient by half in the full period. The reason is the low capital mobility in China, where up until the 1990s, the central bank, its municipal branches and commercial banks were subject to persistent intervention by local governments, and sometimes even had dual leadership (Li, 2010). Government intervention in China has led to inefficient capital allocation, while the dominant role of state-owned enterprises, the lack of a social safety net and artificially low interest rates have created distortion problems (Jen, 2012).

When considering the six next-largest Asian countries excluding the very largest economy, China, it is important to analyse the countries in groups according to their stability. Otherwise results may be incorrectly interpreted. This study estimates the saving–retention coefficient for the full sample (excluding China) at 0.25 for the full period and at 0.37 post crisis. These estimates demonstrate relatively high capital mobility in the considered countries for both periods, but further division of the sample into stable and unstable countries significantly changes the value of the coefficient.

The saving–retention coefficient of the stable countries is estimated at 0.46 and 0.51 for the two periods in the fixed effects model, respectively, suggesting moderate capital mobility in India, Japan and South Korea. Over the full period, the saving–retention coefficient of unstable countries (again, excluding China) was estimated at 0.38, demonstrating a higher level of capital mobility indeed in Indonesia, Russia and Turkey. Similar results were found by Ketenci (2013) in the case of OECD members, where the subdivision of panels into stable and unstable countries altered the value of the saving–retention coefficient, which was higher in stable countries and lower in unstable ones.

This study finds evidence of relatively higher capital mobility in countries whose economies are marked by structural shifts. This means that countries with higher capital mobility are relatively more likely to experience economic instability compared to economies where the level of capital mobility is relatively low. High levels of openness and particularly of capital mobility increase the risk that a country will be exposed to instabilities channelled through international capital flows.

4. Conclusion

This article examined the validity of the FHP for the panel sample of the largest Asian countries. Recently developed econometric methods were applied to annual series in order to estimate the saving–retention coefficient and investigate the cointegration relationships of investment and savings variables, taking into account the presence of structural shifts, whenever relevant. To detect series where structural shifts took place, Hansen’s (1992) stability test was employed. Four of the seven Asian countries under consideration—Russia, Turkey, Indonesia and China—were determined to be unstable. The Westerlund (2006) cointegration test was applied to the sample of unstable countries, allowing for a maximum of five breaks, and evidence of cointegration was found only when structural breaks were allowed in constant. No evidence for cointegration was found when constant and trend were considered. The Pedroni and Kao panel cointegration tests were applied to stable countries—Japan, South Korea and India. The results provided strong evidence for cointegration between investment and savings series.

Finally, saving–retention coefficients were estimated for three samples, the full group, unstable and stable subgroups. Results using DOLS estimators with random effects indicate a low level of capital mobility in all three samples, where saving–retention coefficients were estimated at levels above 0.8. Various studies on the FHP in Asian countries have suggest that saving–retention coefficients for periods following the Asian crisis of 1997 differ significantly from coefficients estimated for the full or pre-crisis period. The estimations of the current study suggest that the differences are insignificant, indicating a low level of capital mobility when random effects are estimated. However, in the case of the fixed effects model, the saving–retention coefficient in all three groups significantly declines, demonstrating lower coefficients in the post-crisis period.

On the other hand, estimates of saving–investment coefficients when China is excluded from the groups indicate relatively high capital mobility in the region in the fixed effects model for the 1995–2011 period. Indeed, the saving investment coefficients for the complete period are estimated at 0.25 and 0.38 for the full sample and the sample of unstable countries, respectively. The study found evidence of moderate capital mobility in stable countries and relatively higher capital mobility in countries that experienced structural shifts, proving that countries with high capital mobility level are susceptible to the negative ramification of international market fluctuations.

Footnotes

Notes

References

Bai

, & Perron

(1998). Estimating and testing linear models with multiple structural changes. Econometrica, 66(1), 47–78.

Bai

, & Perron

(2003). Computation and analysis of multiple structural change models. Journal of Applied Econometrics, 18(1), 1–22.

Bautista

C. C.

, & Mavyraud-Tricoire

(2007). Saving–investment relationship, financial crisis and structural changes in East Asian countries. Economie Internationale, 111(3), 81–99.

Breitung

(2001). The local power of some unit root tests for panel data. In Baltagi

B. H.

, Fomby

T. B.

& Hill

R. C.

(Eds), Nonstationary panels, panel cointegration, and dynamic panels: Advances in econometrics, Volume 15 (pp. 161–77). Bingley: Emerald Group Publishing Limited.

Chen

S. W.

, & Shen

C. H.

(2015). Revisiting the Feldstein–Horioka puzzle with regime switching: New evidence from European countries. Economic Modelling, 49(September), 260–69.

Choi

(2001). Unit root tests for panel data. Journal of International Money and Finance, 20(2), 249–72.

Coackley

, Kulasi

, & Smith

(1996). Current account solvency and the Feldstein-Horioka puzzle. Economic Journal, 106(436), 620–27.

Feldstein

, & Horioka

(1980). Domestic saving and international capital flows. Economic Journal, 90(358), 314–29.

Fouquau

, Hurlin

, & Rabaud

(2008). The Feldstein-Horioka puzzle: A panel smooth transition regression approach. Economic Modelling, 25(2), 284–99.

10.

Guillaumin

(2009). Financial integration in East Asia: Evidence from panel unit root and panel cointegration tests. Journal of Asian Economics, 20(3), 314–26.

11.

Hadri

(2000). Testing for stationarity in heterogenous panel data. Econometrics Journal, 3(2), 148–61.

12.

Hansen

E. B.

(1992). Tests for parameter instability in regressions with I(1) process. Journal of Business and Economic Statistics, 10(3), 321–35.

13.

Herwartz

, & Xu

(2010). A functional coefficient model view of the Feldstein–Horioka puzzle. Journal of International Money and Finance, 29(1), 37–54.

14.

(2002). The Feldstein-Horioka puzzle revisited. Journal of International Money and Finance, 21(4), 555–64.

15.

Huang

, & Guo

(2006). An empirical examination of capital mobility in East Asia emerging markets. Global Economic Review, 35(1), 97–111.

16.

Hussein

K. A.

(1998). International capital mobility in OECD countries: The Feldstein-Horioka puzzle revisited. Economic Letters, 59(2), 237–42.

17.

Jansen

W. J.

(1996). Estimating saving-investment correlations: Evidence for OECD countries based on an error correction model. Journal of International Money and Finance, 15(5), 749–81.

18.

Jen

S. L.

(2012). The return of the Feldstein-Horioka puzzle. Global Connections, Itau BBA. Retrieved from https://www.itau.com.br/_arquivosestaticos/itauBBA/contents/common/docs/GLOBAL_CONNECTIONS_105.pdf

19.

Jiranyakul

, & Brahmasrene

(2009). An exploratory inquiry of the Feldstein-Horioka puzzle in selected Southeast Asian countries. Journal of Transnational Management, 14(4), 259–76.

20.

Jun

(2011). The savings-investment correlation in Asian countries. Pacific Science Review, 12(3), 321–30.

21.

K. S.

, Lee

, & Tieslau

(2005). Panel LM unit-root tests with level shifts. Oxford Bulletin of Econometrics and Statistics, 67(3), 393–419.

22.

K. S.

, Pesaran

M. H.

, & Shin

(2003). Testing for unit root in heterogenous panels. Journal of Econometrics, 115(1), 53–74.

23.

Kao

(1999). Spurious regression and residual-based tests for cointegration in panel data. Journal of Econometrics, 90(1), 1–44.

24.

Kao

, & Chiang

M. H.

(2001). On the estimation and inference of a cointegrated regression in panel data. In Baltagi

Badi H.

, Fomby

Thomas B.

, & Hill

R. Carter

(Eds) Nonstationary panels, panel cointegration, and dynamic, panels (Advances in Econometrics, 15) (pp. 179–222). Emerald Group Publishing Limited. Bingley: Emerald Group Publishing Limited.

25.

Kaya-Bahçe

, & Ö zmen

(2008). Exchange rate regimes, saving glut and the Feldstein-Horioka puzzle: The East Asian experience. Physica A: Statistical Mechanics and its Applications, 387(11), 2561–64.

26.

Ketenci

(2012). The Feldstein-Horioka puzzle and structural breaks: Evidence from EU members. Economic Modelling, 29(2), 262–70.

27.

Ketenci

(2013). The Feldstein-Horioka puzzle in groupings of OECD members: A panel approach. Research in Economics, 67(1), 76–87.

28.

Kim

, Kim

S. H.

, & Wang

(2007). Saving, investment and international capital mobility in East Asia. Japan and the World Economy, 19(2), 279–91.

29.

Kollias

, Mylonidis

, & Paleologou

S. M.

(2008). The Feldstein-Horioka puzzle across EU members: Evidence from the ARDL bounds approach and panel data. International Review of Economics and Finance, 17(3), 380–87.

30.

Kwiatkowski

, Phillips

P. C.

B, Schmidt

, & Shin

(1992). Testing the null hypothesis of stationarity against the alternative of a unit root: How sure are we that economic time series have a unit root? Journal of Econometrics, 54(1–3), 159–78.

31.

Lee

, & Strazicich

(2003). Minimum Lagrange multiplier unit root test with two structural breaks. Review of Economics and Statistics, 85(4), 1082–89.

32.

Lee

, & Strazicich

(2004). Minimum LM unit root test with one structural break. Working Paper No. 04–17, Department of Economics, Appalachian State University.

33.

Levin

, Lin

C. F.

, & Chu

C. S. J.

(2002). Unit root tests in panel data: Asymptotic and finite-sample properties. Journal of Econometrics, 108(1), 1–24.

34.

(2010). Savings, investment, and capital mobility within China. China Economic Review, 21(1), 14–23.

35.

& Li

(2016). Time-variying saving-investment relationship and the Feldstein-Horioka puzzle. Economic Modelling, 53(February), 166–178.

36.

Maddala

G. S.

, & Wu

(1999). A comparative study of unit root tests with panel data and a new simple test. Oxford Bulletin of Economics and Statistics, 61(S1), 631–52.

37.

McCoskey

, & Kao

(1998). A residual-based test of the null of cointegration in panel data. Econometric Reviews, 17(1), 57–84.

38.

Murthy

V. N.

(2007). Feldstein-Horioka puzzle in Latin American and Caribbean countries: Evidence from likelihood based cointegration tests in heterogeneous panels. International Research Journal of Finance and Economics, 11, 112–22.

39.

Nell

K. S.

, & Santos

L. D.

(2008). The Feldstein-Horioka hypothesis versus the long-run solvency constraint model: A critical assessment. Economics Letters, 98(1), 66–70.

40.

Obstfeld

, & Rogoff

(2000). Perspectives on OECD economic integration: Implications for U.S. current account adjustment. UC Berkeley. Center for International and Development Economics Research. Retrieved from http://escholarship.org/uc/item/16z3s2s2

41.

Pedroni

(1999). Critical values for cointegration tests in heterogeneous panels with multiple regressors. Oxford Bulletin of Economics and Statistics, 61(S1), 653–69.

42.

Pelgrin

, & Schich

(2004). National saving-investment dynamics and international capital mobility. Working Paper No. 2004–14, Bank of Canada.

43.

Phillips

P. C. B.

, & Hansen

B. E.

(1990). Statistical inference in instrumental variables regression with I(1) processes. Review of Economic Studies, 57(1), 99–125.

44.

Schmidt

, & Phillips

P. C. B.

(1992). Testing for a unit root in the presence of deterministic trends. Oxford Bulletin of Economics and Statistics, 54(3), 257–87.

45.

Wahid

A. N. M.

, Slahuddin

, & Noman

A. M.

(2008). Saving investment correlation in South Asia: A panel approach. European Journal of Economics, Finance and Administrative Sciences, 11, 153–59.

46.

Westerlund

(2006). Testing for panel cointegration with multiple structural breaks. Oxford Bulletin of Economics and Statistics, 68(1), 101–32.