Content uploaded by Erdenebat Bataa

Author content

All content in this area was uploaded by Erdenebat Bataa on Apr 23, 2019

Content may be subject to copyright.

Growth and Inflation Regimes in Greater Tumen Initiative Area

Erdenebat Bataa

National University of Mongolia

April 2019

Corresponding author:

Erdenebat Bataa

Department of Economics, School of Social Sciences

Building V, Suite #603

National University of Mongolia

Ikh surguuliin gudamj-1, 14201 Ulaanbaatar, Mongolia

Email: tsors79@yahoo.com

Ph : +976 80343365

Acknowledgements: The author gratefully acknowledges financial assistance from the

Economic Research Institute for Northeast Asia’s Visiting Researcher program, under which

this work was initiated. However, the Institute bears no responsibility for the contents of this

paper. I would like to acknowledge the constructive comments received at the Institute’s

seminar and at University of Niigata Prefecture workshop “International Trade and Innovation

in Global Economization”.

2

Abstract

This paper tests for multiple structural breaks in the mean, seasonality, dynamics and

conditional volatility of Greater Tumen Initiative Countries’ (GTI) growth and inflation, while

also accounting for outliers. It finds a drop in the level of Chinese growth rate in the third quarter

of 2011 and of inflation rate in 1998. There are more volatility regimes than the growth regimes

and most GTI countries are currently enjoying historically low volatility of their growth and

inflation. Two exceptions are the increased growth volatility for Japan since 2006 and inflation

volatility for Russia since 2012. There is an increased importance of seasonality in GTI and

especially in Chinese inflation volatility, constituting at least a half of the total volatility.

Keywords: China slowdown, multiple structural breaks, seasonality, Greater Tumen Initiative,

growth and volatility regimes, growth and inflation.

JEL Classifications: E31, E32, C22, C18

3

1. Introduction

China, Mongolia, Russia and South Korea have agreed to transform the Greater Tumen

Initiative (GTI) into an international organization of economic cooperation in Northeast Asia

during the summit in Yanji, China on Sept. 17, 20141. The Tumen River Area Development

Programme (TRADP) was first formed by the United Nations Development Programme

(UNDP) with the objectives of regional cooperation, economic development, and

environmental management in 1995. In spite of its great potential GTI had been largely inactive

due to several challenges including disharmony of interest among member countries, weak

infrastructure development, and lack of funding to activate the project. However, GTI has

received new stimulus since China adopted it as part of its central economic development plan

in 2009. This paper sheds light on the recent developments in the main macroeconomic

variables of growth and inflation for these four countries and Japan, currently an observer nation

to the initiative.

There are many earlier studies on growth and inflation regimes but none fully focuses

on this important geographical region that produces about a quarter of the World GDP.2 The

moderation in volatility of output has been well documented for the US and other developed

countries, see McConnell and Perez-Quiros (2000), and Gadea, Gomez-Loscos and Perez-

Quiros (2018), among others. Coric (2012) studies 98 countries and finds that almost two thirds

experienced GDP growth volatility decline between 1961 and 2007, implying that the so-called

“Great Moderation” took place in economies at all income levels.

On the other hand, Easterly, Kremer, Pritchett and Summers (1993) find that medium

term growth lacked persistence, and countries transitioned between high and low growth

regimes. Ben-David and Papell (1998), Pritchett (2000), Hausmann, Pritchett and Rodrick

(2005), Jones and Olken (2008), Berg, Ostry, Zettelmeyer (2012) show that the growth regimes

are indeed more important phenomena than the long run average growth rate that masks them.

Yet Kar, Pritchett, Raihan and Sen (2013) criticize that the structural break tests that are used

to identify the growth regimes suffer from low power, due to the presence of high volatility in

shorter annual samples, hence miss some of the “true” regimes.

1 See http://www.tumenprogramme.org for details. Such development is extremely important for Mongolia as

imports from China, Russia, Japan and South Korea constitute 41.8%, 36.8%, 11.9% and 5.6% of the total

respectively, while 96% and 1.3% of the total exports go to China and Russia as of 2018.

2 According to the IMF World Economic Outlook-2018, the World, Chinese, Japanese, South Korean, Russian

and Mongolian GDP were 84835462, 13457267, 5070626, 1655608, 1576488, 12724 million USD respectively.

4

Kar et al. (2013) suggest to refrain from using purely statistical test but to marry it with

an ad hoc filter approach that has been used in earlier studies of Hausmann et al. (2005), and

Aizenman and Spiegel (2010), among others. In particular, they propose evaluating the sample

splits derived from Bai and Perron’s (1998, 2003) dynamic programming approach based on a

priori defined filters and find more breaks. In contrast to the earlier works that often use a

simple model with regime dependent intercept (e.g. Jones and Olken, 2008), Jerzmanowski

(2006) and Kerekes (2012) use Markov-switching AR(1) model for the growth rates, whose

intercept, AR coefficient and volatility depend on four different regimes: growth, stagnation,

crisis and miracle growth. But the condition that the intercept, AR coefficient and volatility are

required to change at the same time can be restrictive.

Although the Great moderation and growth regime literatures both address the economic

development, which makes “hard to think about anything else” (Lucas, 1988) due to its

implication for human well-being, the former often uses the growth rates of quarterly real GDP

while the latter relies on that of the annual real GDP per capita. Therefore, the first contribution

of the paper is to use the recently developed iterative structural break testing methodology of

Bataa, Osborn, Sensier and Dijk (2014) and to identify growth regimes in the GTI using the

longest and most up-to-date quarterly data, to increase power of the test.

Blanchard and Simon (2001) show that while the causes of the decline in US output

volatility are complex, this decline can be linked to changes in the properties of inflation and

particularly to a decline in inflation volatility over the period 1952-2001. Similarly,

Eichengreen, Park and Shin (2012) find that policy instability, measured by high and variable

inflation rates, are precursors to growth slowdowns. Therefore, my second contribution is to

search for coincident changes in inflation and growth properties. This is in line with Jones and

Olken (2008) who ask what the breaks actually entail without making statements about the

direction of causality between the variables.

The paper is organized as follows. Section 2 explains the data and summarizes the

iterative decomposition method of Bataa et al. (2014) to identify and distinguish between breaks

in mean, seasonality (if any), persistence and (conditional) volatility of the growth and inflation

series, while also accounting for the possible presence of outliers. Section 3 provides the results

and compares with previous studies. Section 4 concludes.

5

2. Data and Methodology

I analyse quarterly real GDP growth and CPI inflation rates for each of the GTI countries. The

sample for China, Russia and Mongolia starts later than the other two countries due to the lack

of quality data. The start dates are therefore the second quarters of 1960 for Japan and South

Korea, of 1993 for China and of 1995 for Mongolia and Russia. All end in the last quarter of

20183. Russian and Mongolian growth rates and all the inflation rates are not seasonally

adjusted.

The series analysed are shown in panel a) in each of Figures 1 to 5. A cursory glance at

these graphs indicates the presence of outliers (see, for example, Japanese inflation after the

first oil shock in Figure 1), changes in mean inflation (such as for Russia, Figure 4) and/or

volatility (apparently present for Chinese, and Korean growth and Mongolian inflation, Figures

3, 2 and 5). In addition, many of the series are seasonal, with this perhaps being clearest for

Russian and Mongolian growth (Figures 4 and 5, respectively).

Bataa et al. (2014) consider decomposing a stationary time series Yt into components

capturing level (Lt), seasonality (St), outliers (Ot) and dynamics (yt), where level and seasonality

are deterministic and only the last component is stochastic and represented by means of an

autoregressive (AR) process (although this could include stationary stochastic seasonality, if

appropriate).

The model they consider allows for structural change in each of the level, seasonal and

dynamic components, where breaks in the latter may occur in the AR coefficients or in the

conditional volatility. A crucial feature of the model is that the numbers of structural breaks in

these components do not have to be the same and nor do their temporal locations, hence might

prove more flexible than the Markov-switching framework used in Jerzmanowski (2006) and

Kerekes (2012). The general model specification is given by

ttttt yOSLY

(1)

1

kt

L

11

111 ,...,1 kk TTt , 1,...,1 11

mk (2)

3 Data for China, Japan, South Korea and Russia are obtained from the OECD (www.oecd.org). Chinese growth

rate prior to the first quarter of 2011 is not available there, hence obtained from Bataa et al. (2018). Russian growth

rate is also not available from the OECD prior to the second quarter of 2003, hence seasonally unadjusted series

is obtained from the Federal State Statistical Office of the Russian Federation (www.gks.ru) and both Mongolian

series are from the National Statistical Office of Mongolia (www.1212.mn).

6

s

lltlkt DS

1

2

22

122 ,...,1 kk TTt ; 1,...,1 22

mk (3)

t

p

iitikt uyy

1

,

3

33

133 ,...,1 kk TTt ; 1,...,1 33

mk (4)

,

44

144 ,...,1 kk TTt ; 1,...,1 44

mk (5)

where mj denotes the number of breaks of type j that occur at observations

(kj = 1, ..., mj),

with

0 and

(where T denotes the total sample size), and for s seasons per year (s

= 4 for quarterly data), Dlt (l = 1, …, s) are seasonal dummies equal to unity if the observation

at time t falls in season l and zero otherwise. Note that the coefficient lk2

represents the

deviation of the unconditional mean of Yt in the l-th season (month) from the overall mean level

j

and, for identification purposes, we impose the restriction

s

llk

12

= 0 for all seasonality

regimes k2 = 1, …, m2+1.

I can then define the seasonal share in the total volatility as in (6), where ,

and

are simply standard deviations of the fitted values of (2) and (4) respectively:

1,…,

; 1,...,

1 (6)

Although our principal interest is the possibility of breaks in the components (2) to (5),

outliers are corrected to prevent these distorting inferences concerning other components.

Outliers, Ot in (1), are observations that are abnormally distant from the overall level, defined

as 5 times interquartile range from the median and, when detected, are replaced with the median

of the six neighbouring non-outlier observation. The null hypothesis of no break is tested

against an unknown number of breaks up to M using WDMax test and if rejected the exact

number of breaks are identified using sequential tests Seq(i+1|i), starting with i=1 as in Bai and

Perron (1998, 2003). Bataa et al. (2014) employ an iterative approach using Qu and Perron

(2007) test to examine breaks in each of the components of (2)-(5) and the details of their

methodology are relegated to the original study to conserve space.

There is a well-known trade-off between size and power when choosing the maximum

number of breaks (M) and trimming parameter, that is the minimum fraction of the sample

between any two breaks (see Bai and Perron, 1998, 2003). My choice is to allow for a maximum

of three breaks (20% trimming) except for the autoregressive and seasonal parameters for

7

China, Russia and Mongolia. For these I consider up to two breaks (30% trimming). The results

are quite robust to other sensible parameterization.

3. Empirical Results

Figures 1 to 5 show the empirical results of the iterative decomposition in graphical

form. These charts provide: a) the original unadjusted GDP growth and CPI inflation series; b)

the estimated dynamic component yt (constructed by removing outliers, mean and seasonal

components) together with its estimated persistence, defined as the sum of the autoregressive

coefficients in (4) and corresponding 2

standard error bands (in red), and volatility break

dates (vertical green lines); c) the level component Lt with 2

standard error bands; and d) the

estimated seasonal component for each seasonal regime (again with 2

standard error bands)4.

Standard errors are obtained using the HC covariance matrix in the corresponding regression

over the regime defined by the appropriate estimated break dates. Where relevant, the graphs

showing the seasonal components are colour-coded with the first regime (that is, the sub-sample

to the first break date) in blue, the second in red and the third in pink.

Table 1 provides structural break test results for the mean, seasonality (if not already

seasonally adjusted), autoregressive parameters and volatility in its first four panels. The last

panel reports the convergence statistics of the Bataa et al. (2014); the number of iterations for

outer (and inner) loop. Table 2 shows the break dates and the respective component’s regime-

specific estimates based on the breaks and also the estimates ignoring those breaks. 95%

confidence interval for the break dates and heteroskedasticity robust standard errors are also

reported in brackets. There are five country-columns, each split into further growth and inflation

sub-columns. I discuss the results for seasonality first, then dynamics and finally level and

volatility.

The null hypothesis of no structural break in the seasonal pattern of Japanese inflation

against an alternative of unknown number of breaks is rejected soundly as WDmax statistic of

22.34 is significantly higher than the critical value of 14.55 (panel B of table 1). The sequential

test indicates that there are two seasonal breaks, which occur in the first quarter of 1978 and the

4 The procedure detected the following outliers: 74q2 in Japanese inflation, 98q2 in growth, and 63q4, 64q2, 64q4,

74q2 and 80q2 in inflation for Korea, 95q3-96q1 and 98q4-99q2 in inflation for Russia and 95q4 in inflation for

Mongolia. These outliers are associated with well-known historical events such as th e first oil shock, the transition

related shock therapy consequences and Russian debt crisis of 1998.

8

last quarter of 1999 (panel B of table 2)5. As the right-hand side of panel d) in Figure 1 reveals,

the first quarter decline in prices started in 1978. Since then although the overall pattern of

seasonality is largely intact, the magnitude of the seasonal oscillations has reduced in the new

millennia. Bataa et al.’s (2014) study monthly of G7 inflation found that there are also two

seasonality breaks in Japanese inflation; in September 1984 and May 1999, the latter of which

is very close to the one in this study.

I find no statistically significant structural change in Korean, Russian and Mongolian

inflation seasonality; prices peak in the first quarter and drop subsequently throughout the year

for the former two countries (figures 2 and 4) while inflation is highest in the first half of the

year and declines only in the autumn in Mongolia. As for China, there is a marg inally signif icant

structural break in inflation seasonality; after 2009 prices neither drop in the third quarter, nor

increase in the last quarter, as much as they used to before that (figure 3). In terms of the size

of seasonality, measured by their standard deviations, the countries rank from low to high order

as Japan, Korea, China, Russia and then Mongolia.

Remarkably large seasonal fluctuations for Russian and Mongolian growth in figures 4

and 5 contain two structural breaks each with similar timing, perhaps both reflecting their

dependence on fuel and energy products as their main economic growth. In both countries,

growth drops in the first quarter and recovers in the second quarter (and also third quarter for

Russia’s case). The magnitude of the drop in the first quarter has intensified in Russia, first in

the second quarter of 2003 and again in the third quarter of 2011 while the seasonality is overall

declining for Mongolia, although starting from an extremely high level. The magnitude of the

Russian seasonal fluctuations is a drop of 12.92% in the first quarter, and seasonal recoveries

of 3.92%, 8.04% and 0.96% respectively in the remaining quarters of the year before 2003.

Then the pattern changes into a drop of 15.04% in the first quarter and recoveries of 6.66%,

7.47% and 0.91% in the following quarters. After 2010, the drop is 18.55% and the recoveries

are 6.31%, 6.45% and 5.79%. The comparative seasonal drop and recoveries are -43.56%,

47.34%, -8.49%, 4.71% over the quarters before 2003, -42.02%, 40.21%, -5.31%, 7.13%

afterwards and -39.26%, 36.98%, 0.82% and 1.47% after 2011 in Mongolia.

It is also interesting to find no growth persistence break for all the countries (panel C of

Table 1). This is line with the earlier literature that finds low growth persistence (see e.g.

Easterly et al.). The growth persistence is statistically insignificant for Japan and Korea but

5 Note that the growth rates are seasonally adjusted for Japan, Korea and China, thus indicated with N.A. in the

tables; see OECD and Bataa et al. (2018).

9

significant for the other countries (panel C of Table 2). Inflation persistence, measured by the

sum of autoregressive coefficients declines in China in the third quarter of 2011 and is now the

lowest in the GTI area, while that of Mongolia increased after a quarter6.

The null hypothesis of a constant mean is rejected for all series except in Russian growth

and Mongolian growth and inflation. The sequential tests indicate that there are two breaks for

Japan and one break for China in their levels of growth and inflation. The number of breaks in

growth and inflation does not match for Korea and Russia; 1 and 2 for the former and 0 and 2

for Russia. The 95% confidence intervals for the break dates much tighter than the seaonality

breaks.

Japanese growth declines in the second quarter of 1973, after the first oil shock, from

2.31% per quarter to 1.03%, and again in the last quarter of 1990, after the burst of its asset

price bubble. The growth is mere 0.24% after this break, which is at least 3 times lower than

the post-60 average growth of 0.94%, obtained by ignoring the breaks. Interestingly, Korea

maintained its miracle growth rate of 2.21% up until 2001. This is in contrast to Ben-David and

Papell (1998) who found, using annual real per capita GDP, growth slowdowns, in 1967 for

Japan and in 1979 for Korea using data from 1950 to 1990. However their methodology allows

for only one break. Bai and Perron (1998, 2003) show that when there are more than one breaks

such a procedure can be misleading. Jones and Olken (2008) and Kar et al. (2012) found two

down-breaks in 1970 and 1991 for Japan, which are very close to what I find. For South Korea,

Jones and Olken (2008) found an up-break in 1962 using a sample that ends in mid 2000s. Kar

et al. (2012) reported two up-breaks in 1962 and 1982 and two down-breaks, in 1991 and 2002

for South Korea. But as explained in the Introduction they do not consider the statistical

significance of their breaks.

There is some evidence that the growth regimes precede those of inflation for Japan and

Korea, in contrast to Eichengreen et al.’s (2012) claim. The second Japanese growth decline

occurs less than 3 years before the inflation decline, while 2001 growth slowdown of 1.28

percentage point in Korea is followed within a quarter by 0.52 percentage drop in its level of

inflation. Interestingly, Bataa et al. (2014) also found two down-breaks in Japanese inflation;

the first one is in January 1981 and the second one in 1990s. The level of inflation is breaking

down in both China and Russia but remains at stubbornly high level in Mongolia.

6 When the autoregressive lag in panel C of Table 1 is 1, the AR(1) coefficient itself is the persistence.

10

Eichengreen et al. (2012) note that a special anxiety is attached to the question of how

and when Chinese growth might slow. This study finds evidence that the slowdown might have

already occurred and is dated in the third quarter of 2011. If this break is ignored one would

wrongly calculate the average annual growth is 8.8% per annum since 1995, but as Table 2

indicates the growth has declined from 9.56% before the break to 6.9% afterwards. This could

indeed be associated with the increased power of Bataa et al.’s (2014) testing strategy.

Panel Ds of Tables 1 and 2 indicate that volatility regimes are most common. There are

6 and 7 of them in growth and inflation, respectively. While most other countries’ growth

volatilities are entering more stable regimes, Japanese one is substantially higher in the latest

regime that started in the last quarter of 2006, influenced by the GFC, yet the inflation volatility

is still subdued since 1979. Inflation volatilities are also mostly subdued except in Russia, where

the inflation has become more volatile after the first quarter of 2012, perhaps reflecting the

Western sanctions.

The volatility for inflation first declines in 1981 for South Korea followed by a decline

for its growth in 1989 (panel D, Table 2). The inflation volatility further declines in 2000, which

is again followed by a growth volatility decline after 3 years. Such close relationship applies

for China and to a lesser extent for Russia. Given that inflation volatility is often used as policy

instability (see e.g. Eichengreen, Park and Shin, 2012) it could be that inflation volatility

precedes growth volatility. This hypothesis should be an interesting topic for future research,

perhaps using the multivariate approach as in Bataa et al. (2013).

The share in the total volatility of seasonal origin has increased for all inflation series.

While around a quarter of the total volatility used to be attributed to the seasonality in Japan

and Korea before the 1980s, more than a half is due to such forces in the new millennia. It is

particularly high in otherwise tranquil Chinese inflation where it accounts for 82% of the total

volatility after the third quarter of 2011. As for the Russian and Mongolian growth rates, more

than 80% their volatility is driven by seasonal fluctuations. As the business cycle component

of Russian quarterly growth volatility declines, the seasonal cycle’s share has increased; 92%

of its total volatility is being driven by seasonality since the second quarter of 2013.

11

4. Conclusions

This paper uses a newly developed iterative procedure for the decomposition of GTI growth

and inflation into level, seasonality and dynamic components, together with conditional

volatility, when these components are permitted to exhibit distinct multiple structural breaks

over the sample period and outliers are taken into account. To my knowledge, such a flexible

procedure has not been used previously in the.

The paper delivers evidence that important structural changes occurred not only in the

level (mean) of growth and inflation, but also in their seasonal pattern, and volatility. These

results contribute to the on-going debate (see, for example, Pritchett and Summers, 2014) about

the nature and implication of Chinese growth slowdowns. More specifically, just as did the

growth slow down in the second quarter of 1973 and in the last quarter of 1990 in Japan and

also in the first quarter of 2001 in South Korea, I find a statistically significant growth

slowdown in the third quarter of 2011 for China.

The paper also sheds light on the sources of inflation volatility and documents that

seasonality is emerging as the dominant source of volatility in an era of reduced business cycle

volatility.

12

References

Aizenman, J. and Spiegel, M.M. 2010. Takeoffs. Review of Development Economics 14(2),

177-196

Bai, J.S. and Perron, P., 1998. Estimating and testing linear models with multiple structural

changes. Econometrica 66, 47-78.

Bai, J.S. and Perron, P., 2003. Computation and analysis of multiple structural change models.

Journal of Applied Econometrics 18, 1-22.

Bataa, E., Osborn, D.R., Sensier, M. & van Dijk, D. 2013. Structural breaks in the international

dynamics of inflation. Review of Economics and Statistics 95(2), 646-659

Bataa, E., Osborn, D.R., Sensier, M. & van Dijk, D. 2014. Identifying changes in mean,

seasonality, persistence & volatility for G7 & Euro area inflation. Oxford Bulletin of

Economics and Statistics 76(3), 360-388,

Bataa, E., Izzeldin, M. & Osborn, D.R. 2016. Changes in the global oil market. Energy

Economics 56, 161-176.

Bataa, E., Osborn, D.R. & Sensier, M. 2018. China's increasing global influence: Changes in

international growth spillovers. Economic Modelling 74, 194-206

Ben-David, D & Papell, D.H. 1998. Slowdowns and meltdowns: Postwar growth evidence from

74 countries. Review of Economics and Statistics 80(4), 561-571

Berg, A., Ostry, J., Zettelmeyer, J., 2012. What makes growth sustained? Journal of

Development Economics 98(2), 149-166.

Blanchard, O. and Simon, J. 2001. The long and large decline in U.S. output volatility.

Brookings Papers on Economic Activity, 135-164.

Coric, B. 2012. The global extent of the Great Moderation. Oxford Bulletin of Economics and

Statistics 74(4), 493-509.

Easterly, W, Kremer, M., Pritchett, L. and Summers, L.H. 1993. Good policy or good luck?:

Country growth performance and temporary shocks, Journal of Monetary Economics

32(3), 459-483

13

Eichengreen, B., Park, D. and Shin, K. 2012. When Fast Growing Economies Slow Down:

International Evidence and Implications for China. Asian Economic Papers 11(1), 42-87

Gadea, M.D., Gómez‐Loscos, A. and Perez‐Quiros, G. 2018. Great Moderation and Great

Recession: From plain sailing to stormy seas?, International Economic Review 59(4), 2297-

2321

Jerzmanowski, M. 2006. Empirics of hills, plateaus, mountains and plains: a Markov-switching

approach to growth. Journal of Development Economics 81, 357-385.

Jones, B. and Olken, B. 2008. The anatomy of start-stop growth. Review of Economics and

Statistics 90(3), 582-587.

Hausmann, R., Pritchett, L., and Rodrik, D. 2005. Growth accelerations, Journal of Economic

Growth 10(4), 303-329

Kar, S., Pritchett, L., Raihan, S. and Sen, K. 2013. Looking for a break: identifying transitions

in growth regimes. Journal of Macroeconomics 38, 151-166.

Kerekes, M. 2012. Growth miracles and failures in a Markov switching classification model of

growth. Journal of Development Economics 98, 167-177.

Lucas, R. Jr. 1988. On the mechanics of economic development, Journal of Monetary

Economics 22(1), 3-42

McConnell, M.M. and Perez-Quiros, G. 2000. Output uctuations in the United States: What has

changed since the early 1980s? American Economic Review 90, 1464-1476.

Pritchett, L. 2000. Understanding patterns of economic growth: searching for hills among

mountains, plateaus, and plains. World Bank Economic Review 14(2), 221-250.

Qu, Z.J., and Perron, P. 2007. Estimating and testing structural changes in multivariate

regressions. Econometrica 75, 459-502.

14

Figure 1: Japan Decomposition

Notes: Panels show: a) observed growth and inflation, b) dynamic component, persistence (red line) and

volatility break dates (green vertical lines); c) regime means and d) deterministic seasonal component for

regime 1 in blue, regime 2 in red and regime 3 in pink.

15

Figure 2: South Korea Decomposition

Notes: See Figure 1.

16

Figure 3: China Decomposition

Notes: See Figure 1.

17

Figure 4: Russia Decomposition

Notes: See Figure 1.

18

Figure 5: Mongolia Decomposition

Notes: See Figure 1.

19

TABLE 1. Structural break tests in components of GDP growth and CPI inflation

Japan South Korea China Russia Mongolia

1960q2-2018q3 1960q2-2018q3 1993q2-2018q3 1995q2-2018q3 1995q2-2018q3

growth inflation growth inflation growth inflation growth inflation growth inflation

A. Mean 0

WDmax 102.52* 202.16* 48.19* 110.37* 51.73* 33.60* 8.93 64.31* 7.18 1.50

Seq(2|1) 26.44* 38.04* 2.68 24.84* 0.20 0.19 24.96*

Seq(3|2) 0.45 0.02 0.0 0.0

B. Seasonality

WDmax N.A. 22.34* N.A. 7.04 N.A. 13.29* 87.06* 4.40 41.31* 3.18

Seq(2|1) 19.84* 7.67 18.32* 50.30*

Seq(3|2) 0.0

C. AR lags 0 3 3 4 4 2, 0 1 1 3 0, 4

WDmax 1.83 (9.4) 9.63 (14.6) 13.07 (14.6) 14.04 (16.9) 5.32 (15.5) 14.34*(10.8) 4.4 (8.24) 1.42 (10.8) 10.72 (13.3) 19.8*(13.3)

Seq(2|1) 2.17 (11.70) 4.35 (16.70)

D. Volatility

WDmax 15.68* 64.22* 70.42* 140.29* 58.63* 45.55* 11.44* 23.74* 3.62 24.47*

Seq(2|1) 15.73* 5.98 55.53* 23.40* 2.63 6.78 6.63 21.49* 6.74

Seq(3|2) 4.27 4.65 0.0 1.30

E. # iteration 2 (2) 6 (2) 3 (2) 5 (2) 3 (2) 19 (2) 19 (2) 8 (2) 3 (2) 5 (3)

Notes: Decomposition using the iterative method of Bataa et al. (2014), with breaks detected using Qu and Perron’s (2007) test. * indicates a rejection of the null

hypothesis with 95% confidence. The null hypothesis of WDmax test is no structural break while the alternative is up to M breaks. If the null is rejected then

Seq(i+1|i) test is sequentially applied to determine the exact number of breaks, starting with a null of 1 break against an alternative of 2, until the null is not

rejected. Asymptotic 5% critical values of WDmax, Seq(2|1) and Seq(3|2) tests for the mean and volatility are 9.42, 9.82 and 10.72, respectively with trimming

20% and M = 3. The corresponding values for the seasonality are 14.55, 15.46 and 16.34 respectively, with 20% trimming and M = 3 (13.26 and 14.34 with 30%

trimming and M = 2). Those for the autoregressive parameters (trimming parameter and M are the same with seasonality) are reported next to the test statistics

in brackets in panel C as the lag orders differ across variables. The autoregressive order of the dynamic component is selected by the AIC criterion and is reported

in panel C. Finally, the numbers required to achieve convergence on the main and (sub) loops are shown. If the iteration converges to a two cycle (when 19) it

reports results based on Bataa et al. (2016)’s information criteria.

20

TABLE 2. Regimes in components of GDP growth and CPI inflation and seasonality share in the total volatility

Japan South Korea China Russia Mongolia

growth inflation growth inflation growth inflation growth inflation Growth inflation

A. Mean break dates 73q2 80q4 01q1 81q4 11q3 98q3 01q3

(72q1-74q3) (80q1-81q3) (97q3-04q3) (81q1-82q3) (09q4-13q2) (98q1-99q1) (00q4-02q2)

90q4 93q4 01q2 09q4

(87q3-94q1) (90q1-97q3) (88q3-14q3) (08q1-11q3)

Regime means (s.e.) 2.31 (0.16) 1.70 (0.10) 2.21 (0.16) 3.29 (0.22) 2.39 (0.09) 2.37 (0.31) 4.62 (0.27)

1.03 (0.13) 0.51 (0.05) 0.93 (0.10) 1.13 (0.09) 1.71 (0.03) 0.51 (0.07) 2.81 (0.10)

0.24 (0.09) 0.06 (0.04) 0.61 (0.05) 1.68 (0.18)

[0.94 (0.08)] [0.73 (0.06)] [1.82 (0.12)] [1.77 (0.12)] [2.20 (0.07)] [0.89 (0.11)] [0.69 (0.18)] [2.84 (0.16)] [1.64 (0.42)] [2.55 (0.35)]

B. Seasonality break

dt

N.A. 78q1 N.A. N.A. 09q1 03q2 03q1

(74q1-82q1) (03q3-14q3) (01q2-05q2) (01q1-05q1)

99q4 10q3

11q3

(91q4-07q4) (09q3-11q3) (10q1-13q1)

Seasonal St.Dev. N.A. 0.41 N.A. N.A. 0.90 8.42 34.38

0.52 0.64 9.66 32.19

0.26 11.17

28.00

[0.33] [0.61] [0.76] [9.23] [0.86] [30.10] [2.91]

C. Dynamic break dates 11q3 11q4

(09q3-13q3) (09q4-13q4)

Persistence (s.e.) 0.80 (0.07) 0.46 (0.33)

0.15 (0.24) 0.69 (0.21)

[0.02 (0.09)] [0.63 (0.11)] [-0.15 (0.14)] [0.46 (0.14)] [0.49 (0.19)] [0.78 (0.07)] [0.37 (0.17)] [0.47 (0.13)] [-1.24 (0.25)] [0.46 (0.32)]

Notes: Estimates of the break dates and of the components of (1) in the resulting regimes. 95% confidence intervals for the break dates and

heteroscedasticity robust standard errors are reported in brackets. The quantities that are estimated ignoring the breaks are in square parentheses.

21

TABLE 2. Continued

Japan South Korea China Russia Mongolia

growth inflation growth inflation growth inflation growth inflation Growth inflation

D. Volatility break dates 90q3 79q3 89q2 81q1 11q1 11q1 13q2 00q4 09q3

(84q1-93q3) (65q4-79q4) (80q2-89q4) (67q2-81q2) (07q3-11q2) (07q1-11q2) (01q4-14q1) (96q4-01q1) (05q1-10q1)

07q2 04q1 00q4 12q1

(04q3-11q2) (04q1-05q1) (95q3-03q1) (11q3-16q2)

Regime shock St.Dev 1.09 0.92 2.27 2.06 0.76 0.94 2.01 1.81 4.25

0.66 0.42 1.21 0.80 0.18 0.28 1.07 0.58 1.92

1.24 0.79 0.45 1.31

[1.01] [0.62] [1.73] [1.29] [0.62] [0.78] [1.81] [1.10] [4.09] [3.41]

E. Seasonality share N.A. 0.27 N.A. 0.22 N.A. 0.31 0.80 0.29 0.85 0.40

in total volatility 0.32 0.42 0.38 0.82 0.39 0.87 0.52

0.37 0.55 0.70 0.84 0.56 0.88 0.59

0.54 0.82 0.92

[0.44] [0.56] [0.45] [0.82] [0.51] [0.88] [0.63]

Notes: Estimates of the break dates and of the components of (1) in the resulting regimes. 95% confidence intervals for the break dates and

heteroscedasticity robust standard errors are reported in brackets. The quantities that are estimated ignoring the breaks are in square parentheses.