Transcript An Empirical Investigation of Structural Breaks

Effects of the Asian Financial Crisis on the Korean
Economy: Some Further Empirical Evidence
Charles Harvie
University of Wollongong
and
Mosayeb Pahlavani
University of Sistan and Baluchestan,
This paper provides empirical evidence on the significance of the Asian financial crisis for the key
macro-economic variables of the Korean Economy. This paper employs quarterly time series data to
endogenously determine the timing of structural breaks for various macroeconomic variables in Korean
economy. The Zivot-Andrews(ZA,1992) and Lumsdaine-Papell(1997) testing procedures are then used
to test for the non-stationarity of the Korean macroeconomic data in the presence of potential structural
breaks.
Few countries in history have attained economic development as rapidly as Korea (World Bank,
1993). In a single generation this poor nation, consisting primarily of subsistence farmers in the 1950s
and early 1960s, had been transformed by 1996 into the world’s largest producer of home appliances,
the second largest producer of semi-conductor chips, the second largest ship builder, the fifth largest car
maker, the eleventh largest economy in the world and the third largest in Asia, and the twelfth largest
exporter and trading nation (Harvie and Lee, 2003a, 2003b). Per capita income doubled every eight
years from the early 1960s rising from US$80 to US$10,000 by 1996 (Australian Department of
Foreign Affairs and Trade, 1999; Harvie and Lee, 2003a, 2003b).
Korea GDP Growth Rate
Figure 1
Real GDP Growth Rate (%)
12
10
8
6
Percent
4
2
0
1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
-2
-4
-6
-8
Years
Source: Korea National Statistical Office: http://kosis.nso.go.kr.
Zivot-Andrews Unit root test in the presence of a potential structural break

As is well known the issue of structural change, and its consequential implications for
structural breaks, in macroeconomic time series data must be robustly addressed in order
to ensure non spurious results of unit root tests of such data. There can, of course, be
many reasons for structural change, and these can include such diverse circumstances as
economic crises, policy changes or regime shifts. For this reason it is extremely
important to test the null hypothesis of structural stability against the alternative of a
one-time structural break. If potential structural changes are not allowed for in the
specification of an econometric model but are, in fact, present, the results may be
spurious because they can be biased towards the erroneous non-rejection of the nonstationarity hypothesis (Perron, 1989, Perron, 1997, Leybourne and Newbold, 2003,
Pahlavani, Valadkhani and Worthington, 2005, Harvie and Pahlavani, 2006).
–
–




KEY MACROECONOMIC VARIABLES OF INTEREST TO THIS STUDY :
real GDP (GDP), real Gross National Income (GNI), real Gross Fixed Capital Formation
(GFCF), real Exports (X), real Imports (IM), real Private Consumption Expenditure (PC), real
Government Consumption Expenditure (GC), Consumer Price Index (CPI), Exchange Rate
(ER), Broad Money (BM), Currency in Circulation (CUR), and the Interest Rate (IR).
Time series data for the korean economy is likely to be subject to the existence of structural breaks.
Issue is addressed by:
Analyzing the time series properties of the data using the ZA and LP models.
-Applying these methodologies to the data to check whether the endogenously determined structural
breaks coincide with the Asian financial crisis or not.
Perron(1989) argued that the usual tests such as ADF or PP test for unit root could be biased toward nonrejection of unit root if there exists a one-time permanent change in the data.



Perron’s (1989) key assumption is that the break date of the trend function is fixed (exogenous), and
chosen independently of the data. In fact, traditionally, the time of any structural break was always
assumed to be known a priori in accordance with the underlying asymptotic distribution theory.
However, Christiano (1992) and others have criticised this approach, arguing that considering the
timing of the break as an exogenously known event invalidates the distribution theory underlying
conventional testing (Vogelsang and Perron, 1998). In response, a number of studies, including Zivot
and Andrews (1992), Perron (1994, 1997), Lumsdaine and Papell (1997) and Bai and Perron (2003),
to name just a few, have proposed different ways of estimating the time of the break endogenously.
These studies have shown that this endogenous approach lessens the bias in the usual unit root tests.
Zivot and Andrews(1992) and also Perron & Vogelsang(1992) and Perron (1997) allowed that the
date of the change in the series to be unknownn.
They have shown that bias in the usual unit root tests can be reduced.
by endogenously determining the time of structural breaks.
Zivot and Andrews (hereafter ZA) (1992) propose a testing procedure where the time of the break is
estimated rather than assumed as an exogenous phenomenon. By endogenously determining the time
of structural breaks they argue that the results of unit root hypotheses previously suggested by earlier
conventional tests, such as the widely-employed Augmented Dickey-Fuller (ADF) (Dickey and
Fuller, 1979, 1981) test or the Perron (1989) methodology, may be reversed.
The Zivot and Andrews (1992) model endogenises one structural break in a series
(such as yt) as follows:
H0:
.
yt    yt 1  et
k
yt  ˆ  ˆ DU t (Tˆb )  ˆ t  ˆ DTt (Tˆb )  ˆ yt 1   cˆ j yt  j  eˆt
j 1
Equation (2),
which is referred to as model C by ZA, accommodates the possibility of a change in
the intercept as well as a trend break. ZA also consider two other alternatives where a structural
break impacts on the intercept only (model A) or trend only (model B). Model C is the least
restrictive compared to the other two models; we thus base our empirical investigation on this
model. In equation (2) DUt is a sustained dummy variable capturing a shift in the intercept, and DTt
is another dummy variable representing a break in the trend occurring at time Tb where DUt=1 if t
> Tb, and zero otherwise and is equal to (t-Tb) if (t > Tb) and zero otherwise. The null hypothesis is
rejected if the coefficient is statistically significant. The computations presented in this study were
conducted by means of the RATS program.
More
specifically, the ZA test asserts that TB is endogenously estimated by running the above
equation (2) sequentially in order to allow for TB to be in any particular year/quarter with the
exception of the first and last years/quarters. The optimal lag length is determined on the basis of the
Schwartz Information Criterion (SIC), AIC or t-test (the use of the most significant t ratio in the
literature is referred to as the general to specific approach).
The theoretical underpinning of the Zivot-Andrews model is now described and then the empirical
results are reported. In this methodology, Tb (the time of break) is chosen to minimize the one-sided tstatistic of α=1 in equations 1 and 2. In other words, a break point is selected which is the least favourable
to the null hypothesis
Figure 1. Plots of the Estimated Timing of Structural Breaks by The ZA Procedure
Zivot-Andrews Unit Root Tests for LGNI
Zivot-Andrews Unit Root Tests for LGDP
-2.5
-2.5
-3.0
-3.0
-3.5
-3.5
-4.0
-4.0
-4.5
-4.5
-5.0
-5.0
-5.5
-5.5
-6.0
-6.0
-6.5
-6.5
1991
1993
1995
1997
1999
2001
2003
1991
2005
Zivot-Andrews Unit Root Tests for LGFCF
1993
1995
1997
1999
2001
2003
2005
2003
2005
Zivot-Andrews Unit Root Tests for LX
-1.6
-2.00
-2.25
-2.4
-2.50
-3.2
-2.75
-3.00
-4.0
-3.25
-4.8
-3.50
-3.75
-5.6
-4.00
-6.4
-4.25
1991
1993
1995
1997
1999
2001
2003
2005
1991
1993
1995
1997
1999
2001
Zivot-Andrews Unit Root Tests for LIM
Zivot-Andrews Unit Root Tests for LPC
-2.0
0.5
-2.5
0.0
-3.0
-0.5
-3.5
-1.0
-4.0
-1.5
-4.5
-2.0
-5.0
-2.5
-5.5
-3.0
-6.0
1991
1993
1995
1997
1999
2001
2003
2005
-3.5
1991
Zivot-Andrews Unit Root Tests for LGC
1993
1995
1997
1999
2001
2003
2005
Zivot-Andrews Unit Root Tests for LNCPI
-2.4
-1.5
-2.8
-2.0
-3.2
-2.5
-3.6
-3.0
-4.0
-3.5
-4.4
-4.0
-4.8
-4.5
1991
1993
1995
1997
1999
2001
2003
2005
1991
1993
1995
1997
1999
2001
2003
2005
Zivot-Andrews Unit Root Tests for LER
Zivot-Andrews Unit Root Tests for LNBM
-1
1
-2
0
-3
-1
-4
-2
-5
-3
-6
-7
-4
-8
1991
1993
1995
1997
1999
2001
2003
2005
-5
1991
Zivot-Andrews Unit Root Tests for LNCUR
1993
1995
1997
1999
2001
2003
2005
Zivot-Andrews Unit Root Tests for LNIR
-1.0
-2.50
-1.5
-2.75
-2.0
-3.00
-2.5
-3.25
-3.0
-3.50
-3.5
-3.75
-4.0
-4.00
-4.5
-4.25
-5.0
-4.50
-5.5
1991
1993
1995
1997
1999
2001
2003
2005
-4.75
1991
1993
Note: the numbers on the vertical axis are t ratios for tˆ .
Source: Authors’ calculations.
1995
1997
1999
2001
2003
2005
Table 5. The Zivot-Andrews test results: break in both intercept and trend
Variable
K
LGDP
1998:01
1
-6.02*
LGNI
1998:01
1
-6.37*
1998:01
1
-6.19*
Symbol
Real GDP
Real Gross national Income
Real gross fixed capital
formation
tˆ
TB
Description
LGFCF
Inference
Corresponding break time
Sta
Asian financial crisis – economic downturn
Sta
Asian financial crisis – economic downturn
Sta
Asian financial crisis – economic downturn
Real exports
LX
2001:01
2
-4.19
UR
Tech wreck, global slowdown, semi-conductor
price crash.
Real Imports
LIM
1998:01
2
-5.78*
Sta
Asian financial crisis – economic downturn
Real private consumption
LPc
1997:04
1
-3.02
UR
Asian financial crisis – financial turbulence
Real Government consumption
LGc
2001:04
2
-4.66
UR
Tech wreck, global slowdown, semi-conductor
price crash aftermath
Consumer price Index
LnCPI
1998:04
3
-4.28
UR
Recovery from the Asian financial and
economic crisis
Exchange rate
LER
1997:04
0
-7.91*
Sta
Asian financial crisis – financial turbulence
Broad Money (M2)
LnBM
1997:04
3
-4.89*
Sta
Asian financial crisis – financial turbulence
Currency in Circulation
LnCUR
1997:04
1
-5.17*
Sta
Asian financial crisis – financial turbulence
Interest Rate
LnIR
1998:04
3
-4.71
UR
Recovery from the Asian financial and
economic crisis
Notes: (1) Critical Values at 1, 5 and 10 per cent levels are -5.57, -5.08 and -4.82, respectively (Zivot and Andrews, 1992).
(2) The optimal lag length (k) is determined by the general to specific test, AIC and SBC.
(3) * the corresponding null is rejected at 5 per cent or better. (4) The computations were done with RATS program.
Sources: The Bank of Korea (2006), Korea National Statistical Office: http://kosis.nso.go.kr and authors.

As indicated by the empirical evidence based on the ZA approach the financial and economic crisis
of 1997-98 was the most traumatic event to affect the Korean economy during the period of this study.
In fact it has been the most traumatic economic event to affect the country since its period of rapid
economic growth and development from the early 1960s.

The ZA test results are somewhat mixed as some of the variables under
investigations are not stationary.

Innovational and Additive Outlier Models

Perron (1997) introduced 2 methods for determining the time of the break endogenously:
The Additive Outlier (AO) model that captures sudden change and the Innovational Outlier (IO) model that permits changes to occur
gradually over time.
Harvie, C., and M. Pahlavani(2006), “Testing for Structural Breaks in the Korean Economy 1980-2005: An Application of the Innovational
Outlier and Additive Outlier Models”, Journal of the Korean Economy, Vol 7, No 2, pp.1-29.
Our empirical finding based on these two methodology shows that…

by applying the Innovational Outlier (IO) and Additive Outlier (AO) models for the same variables,
Harvie and Pahlavani (2006), concluded that, after allowing for one structural break, the majority of
endogenously determined structural breaks coincided with the effects of the financial and economic
crisis of 1997-98. They find that GDP, GNI, GFCF, IM, CPI, ER, CUR and IR all experienced
structural breaks during 1997Q4 and 1998Q1
. Taking into account the existence of gradual
effects of a potential structural break, two of the variables under investigation (IM and
ER) do not have a unit root.

The results of this paper can be useful for analysts who will undertake any
future research which involves the use of time series data on the Korean
macroeconomy. However it should be noted that ZA(1992) & Peron (1997) IO
and AO tests capture only one (the most significant) structural break in each
variable. What if, there have been multiple structural breaks in a series?
Considering only one endogenous break may not be sufficient and it could
lead to a loss of information particularly when in reality there is more than one
break (LP, 1997). On this same issue, Ben-David et al (2003) argued that:

“just as failure to allow one break can cause non-rejection of the unit root null
by the Augmented Dickey –Fuller test, failure to allow for two breaks, if
they exist, can cause non-rejection of the unit root null by the tests which only
incorporate one break” (2003: 304).

LP introduced a new procedure to capture two structural breaks. They argued
that unit root test that account for two structural breaks (if significant) is more
powerful than those, which only allows for one single break.
 New Procedure by Bai and Perron(1998;2003) introduced for Multiple
structural break
LP(1997) Empirical Results
Based on the LP test the empirical results indicate that five out of the ten
macroeconomic variable time series become stationary at the 5 per cent
significance level. In other words the LP test results are somewhat mixed as five
out of the 10 variables, i.e. Ln(Pc) Ln(Gc) and Ln(X), Ln(CPI), and Ln(Ms) are
not stationary, while the remaining variables contain a unit root. Given the fact
that all of the estimated coefficients for the indicator and dummy variables are
statistically significant for all of the variables under investigation, one can argue
that the two endogenously determined structural breaks identified are indeed
statistically significant. We also find that the reported t statistics in Table 1 for α
are significant in all case.
 It is interesting to note that the most significant endogenously determined
structural breaks for the variables under investigation closely correspond to the
Asian financial and economic crisis of 1997-98[1]. It can be observed that one of
the most significant structural breaks for all of the variables under investigation,
with the exception of total exports, occurred in 1997Q4, the height of the
financial crisis, or 1998Q1 which can be characterized as the quarter in which
there was a severe economic downturn triggered by the financial turmoil, or in
1998Q4 when the economy was experiencing signs of recovery from the
economic crisis. There is no obvious second major structural break that
commonly impacted upon all or most of the variables under investigation, but
rather this is variable dependent.
Concluding Remarks-1

This paper has used quarterly time series data for ten major
macroeconomic variables for the Korean economy covering the period
1980Q1-2006Q4, to endogenously determine the two most important
years when structural breaks occurred for these variables. For this
purpose we used both the ADF, ZA,IO, AO and the LP (Lumsdaine
and Papell, 1997) test to make robust conclusions about the time series
properties of the data. It was found that according to the ADF test none
of the variables under investigation is stationary. The ZA,AO and IO
tests, which taking into account the existence of one and most
significant structural breaks, results are somewhat mixed as some of
the variables under investigations are stationary, for example, Taking
into account the existence of gradual effects of a potential structural
break, two of the variables under investigation (IM and ER) do not
have a unit root.

However, after allowing for two structural breaks, the LP test results indicate
that the unit root null hypothesis is rejected at the 5 per cent level for majority
of variables under investigation. Most of the structural breaks are associated
with the 1997 financial and economic crisis, plus variable specific events that
have been influenced by specific government policies or other relevant events.
Concluding remarks-2

This study sheds some important light on the issue of structural breaks in key macroeconomic time
series data for the Korean economy, which, during the period of the 1990s, was subject to major
economic turbulence. This needs to be allowed for in the construction and estimation of
macroeconomic models of the Korean economy, or for the estimation of relationships between
specific macroeconomic variables in the economy. As the results from the LP test have indicated, non
allowance for structural breaks in such time series, the exogenous imposition of dummy variables to
capture structural breaks in the data, or indeed the endogenous determination of only one structural
break, can lead to model misspecifcation and spurious results. Since non-stationarity testing with
multiple structural breaks may yield conflicting results to conventional ADF tests, future work
should, therefore, concentrate on such a clearer refinement.
Gregory-Hansen (1996) Cointegration technique in the presence
of structural break
Saikkonen-Lutkephol (2000a,b,c) approachs for cointegration
analysis in the presence of structural breaks
Future Research: the effects of recent World financial crisis on
Korean or other economies in the world?

THANK YOU FOR YOUR ATTENTION