The Asian Crisis and Exchange Rate Volatility Spillovers

Laura Citron, Alex Pick, Mathieu Vital

Aims of the study

Review the evolution of the exchange rate in Thailand, Indonesia and Malaysia
Estimate their correlation before, during and after the crisis Investigate the presence of volatility spillovers between Thailand and Malaysia

Methodology
Split the sample covering the 1996 to 1999 period in three subsamples: before (1/02/96 to 6/30/97), during (7/01/97 to 8/31/98) and after the crisis (9/01/98 to 12/09/99). This is slightly arbitrary but based on historical facts and fits time series. Evaluate the correlations of exchange rate using simple statistics and by calculating a VAR(p) Investigate for spillovers using three different GARCH(1,1) models

The Data
60 55 45 40 35 30 25 20 1996 1997 1998 THAILAND 1999

4.8 4.4 4.0

MALAYSIA

50

3.6 3.2 2.8 2.4

18000 16000 14000 12000 10000 8000 6000 4000 2000 1996 1997 1998 INDONESIA 1999

20

25

30

35

40

45

50

55

60

THAILAND

18000 16000 14000

4.8 4.4 4.0 3.6 3.2 2.8 2.4 1996 1997 1998 MALAYSIA 1999

INDONESIA

12000 10000 8000 6000 4000 2000 20 25 30 35 40 45 50 55 60 THAILAND

Qa) Correlations
Estimated as: In E-views: select your variables, click on Quick, Group Statistics, and Correlation
Exchange Rate Before During After MalaysiaThailand -0.353464 0.905411 0.218336 MalaysiaIndonesia -0.382315 0.896988 0.19079 Indonesia Thailand Medium 0.431710 0.714530 0.378336 Large Smaller

Interpretation of coefficients
Before the crisis, we can observe some positive correlation between the Thai and Indonesian currencies. We can think of this as being the result of common flows of investments made towards the tigers, or maybe common regional monetary policies. The negative relationships between the Malaysian currency and the other two might be a result of changes in global (and especially US) export demand for goods produced by these countries. During the crisis, the correlation coefficients increased dramatically, indicating the fact that the crisis was a regional one. After the crisis, the correlation between the three exchange rates become very small and positive. This can be due to the implementation of different recovery strategies (such as for instance the pegging of Malaysia's currency to the Dollar).
6

Qb) Estimated VAR(p)

A VAR(p) takes the form:

In Eviews: Quick, Estimate VAR and then select your options (unrestricted VAR or ECM, endogenous variable lag length) We estimated two VAR(2) and one VAR(3) using lag length information criteria and obtained the coefficients presented hereafter (VAR(3) used due to unit root in VAR(2) after crisis) We then tested the validity of each of these model by performing residual autocorrelation tests (for coefficients) and normality tests (for validity of confidence intervals using asymptotic theory)
7

Testing the VAR models

VAR before Lag length selection Serial autocorrelation 2 using SIC Reject H0 VAR during 2 using SIC Reject H0 VAR after 3 using AIC Reject H0

Normality test
Cointegration

Reject H0
Accept H0

Reject H0
At least one

Reject H0
At least one

All of the VAR models presented serial autocorrelation. Furthermore, none of the VAR models passed the normality test. So we decided to use a bootstrap method to estimate the confidence intervals of the impulse response functions. Finally, the VAR models presented cointegration during and after so the VAR model is not appropriate (ECM). Altogether, the VAR models performed very poorly.
8

Estimated VAR
Before Thai -1 Thai -2 Indo -1 Indo -2 Malay -1 Malay -2 During Thai -1 Thai -2 Indo -1 Indo -2 Malay -1 Malay -2 Thailand 0.715099 0.316806 -0.040687 -0.021149 -0.167714 0.122000 Thailand 0.907258 0.035167 0.096310 -0.019094 0.026618 -0.038777 Indonesia 0.003368 0.21986 0.661633 0.320853 0.016330 -0.002294 Indonesia -0.026482 0.028952 1.029388 -0.037186 0.039696 -0.041243 Malaysia 0.045599 -0.05814 -0.012639 0.019658 0.929098 0.038749 Malaysia -0.092073 0.079273 0.681249 -0.576516 0.922863 0.050431 The lag length has fallen from 2 to 1 as the volatility has increased Were not sure how to interpret these coefficients! After Thai -1 Thai -2 Thai - 3 Indo -1 Indo -2 Indo - 3 Malay -1 Malay -2 Malay -3 Thailand 1.017929 -0.184533 0.144699 0.023826 -0.012966 -0.012113 0.028887 0.200123 -0.055742 Indonesia 0.469291 -0.675955 0.149486 1.031677 -0.040802 -0.022046 2.216963 0.050597 0.332890 Malaysia 0.001897 0.044270 -0.038171 0.004561 -0.008131 0.005245 -0.00882 -0.119152 0.059926

Impulse Response Functions (Before)

To a change in
.8

Thailand
R e s p o n s e o f B E F O R E L N TH A I t o B E F O R E L N TH A I .4

R e s p o n s e to C h o le s ky On e S .D . In n o va ti o n s 2 S.E.
R e s p o n s e o f B E F O R E L N TH A I t o B E F O R E L N I N D O .8

Indonesia

Malaysia
R e s p o n s e o f B E F O R E L N TH A I t o B E F O R E L N M A L A Y .8 .4

Response of ...

.4

Thailand

.0

.0

.0

-.4

-.4

-.4

-.8 10 20 30 40 50 60 70 80 90 100

-.8 10 20 30 40 50 60 70 80 90 100

-.8 10 20 30 40 50 60 70 80 90 100

R e s p o n s e o f B E F O R E L N I N D O t o B E F O R E L N TH A I .08 .06 .04 .02

R e s p on s e o f B E F O R E L N I N D O t o B E F O R E L N I N D O .08 .06 .04 .02 .00 -.02 -.04 -.06

R e s p o n s e o f B E F O R E L N I N D O t o BE F O R E L N MA L A Y .08 .06 .04 .02 .00 -.02 -.04 -.06

Shocks die out very quickly if not instantly

Indonesia

.00 -.02 -.04 -.06 10 20 30 40 50 60 70 80 90 100

10

20

30

40

50

60

70

80

90 100

10

20

30

40

50

60

70

80

90 100

R e s p o n s e of B E F O R E L N MA L A Y t o B E F O R E L N TH A I .06 .04 .02 .00 -.02 -.04 -.06 10 20 30 40 50 60 70 80 90 100

R e s p o n s e o f B E F O R E L N MA L A Y t o B E F O R E L N I N D O .06 .04 .02 .00 -.02 -.04 -.06 10 20 30 40 50 60 70 80 90 100

R e s p o n s e o f B E F O R E L N M A L A Y t o B E F O R E L N MA L A Y .06 .04 .02 .00 -.02 -.04 -.06 10 20 30 40 50 60 70 80 90 100

Malaysia

10

Impulse Response Functions (During)

To a change in
.03

Thailand
Response of DURI NGLNTHAI t o DURINGLNTHAI
.02 .01

R e s p o n s e to C h o le s ky On e S.D . In n o va tio n s 2 S.E.

Response of DURINGLNTHAI to DURINGLNMALAY
.03 .02 .01 .00 -.01 -.02

Malaysia

Indonesia
Response of DURINGLNTHAI t o DURINGLNI NDO

.03 .02 .01 .00 -.01 -.02 10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100

Response of ...

Thailand

.00 -.01 -.02 10 20 30 40 50 60 70 80 90 100

Response of DURINGLNMALAY to DURINGLNTHAI

.03 .02 .01

Response of DURI NGLNMALAY to DURI NGLNMALAY

.03 .02 .01 .00 -.01 -.02

Response of DURI NGLNMALAY t o DURINGLNI NDO

.03 .02 .01 .00 -.01 -.02

The IRFs are much greater during the crisis than before the crisis. Shocks are more persistent

Malaysia

.00 -.01 -.02 10 20 30 40 50 60 70 80 90 100

10

20

30

40

50

60

70

80

90 100

10

20

30

40

50

60

70

80

90 100

Response of DURINGLNI NDO to DURINGLNTHAI

.08

Response of DURI NGLNINDO t o DURINGLNMALAY

.08

Response of DURINGLNI NDO to DURINGLNINDO

.08

Indonesia

.04

.04

.04

.00

.00

.00

-.04 10 20 30 40 50 60 70 80 90 100

-.04 10 20 30 40 50 60 70 80 90 100

-.04 10 20 30 40 50 60 70 80 90 100

11

Impulse Response Functions (After)

To a change in
.008

Thailand
R e s p o n s e o f A F TE R L N TH A I t o A F TE R L N TH A I

R e s p o n s e to C h o l e s ky On e S.D . In n o va ti o n s 2 S.E.
R e s p o n s e o f A F TE R L N TH A I t o A F TE R L N MA L A Y .008 .006 .004 .002 .000 -.002 -.004

Malaysia

Indonesia
R e s p o n s e o f A F TE R L N TH A I t o A F TE R LN I N D O

.008 .006 .004 .002 .000 -.002 -.004 10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100

Response of ...

.006 .004

Other responses much smaller note change of scale

Thailand

.002 .000 -.002 -.004 10 20 30 40 50 60 70 80 90 100

R e s p o n s e o f A F TE R L N MA L A Y t o A F TE R L N TH A I .0025 .0020 .0015

R e s po n s e o f A F TE R L N MA L A Y t o A F TE R L N M A L A Y .0025 .0020 .0015 .0010 .0005 .0000 -.0005 -.0010

R e s p o n s e o f A F TE R L N MA L A Y t o A F TE R L N I N D O .0025 .0020 .0015 .0010 .0005 .0000 -.0005 -.0010

Malaysia

.0010 .0005 .0000 -.0005 -.0010 10 20 30 40 50 60 70 80 90 100

Capital controls in Malaysia

10

20

30

40

50

60

70

80

90 100

10

20

30

40

50

60

70

80

90 100

R e s p o n s e o f A F TE R L N I N D O t o A F TE R L N TH A I .03 .02

R e s p o n s e o f A F TE R L N I N D O t o A F TE R L N MA L A Y .03 .02 .01 .00 -.01 -.02

R e s p o n s e o f A F TE R L N I N D O t o A F TE R L N I N D O .03 .02 .01 .00 -.01 -.02

Indonesia

Autocorrelation still high for Indonesia and Thailand

.01 .00 -.01 -.02 10 20 30 40 50 60 70 80 90 100

10

20

30

40

50

60

70

80

90 100

10

20

30

40

50

60

70

80

90 100

12

Qc) Volatility spillovers

GARCH models look like:

The model assumes an ARMA process for the square error terms and is covariance stationary if the root of the lag polynomial equation lie outside the unit circle. In E-Views: click on Quick, Estimate equation, scroll down to ARCH. In the dialog box, specify GARCH and select lag length. In the top box specify the model for y. In the right hand box add any desired additional regressor in the GARCH. We used lag length (1,1) in the GARCH but more formal selection procedures exist (portmanteau, Ljung Box Test) with which we are not familiar with (yet).

13

Qc) Volatility spillovers

We used three models to estimate volatility spillovers First built a Garch(1,1) for Thailand with no specification, extract the conditional variance series, and build a Garch(1,1) for Malaysia with no specification but including the Thai conditional variance series as a regressor . First built a Garch(1,1) for Thailand with a constant, extract the conditional variance series, and build a Garch(1,1) for Malaysia with a constant and including the Thai conditional variance series as a regressor First built a Garch(1,1) for Thailand with an AR(3) (BIC) specification, extract the conditional variance series, and build a Garch(1,1) for Malaysia with AR(3) (BIC) specification but including the Thai conditional variance series as a regressor The comparative results are shown hereafter

14

Output of GARCH Models

Compare
Model Form Thailand

Model One
LNTHAI = 0 GARCH = C(1) + C(2)*RESID(1)^2 + C(3)*GARCH(-1) LNMALAY = 0 GARCH = C(1) + C(2)*RESID(1)^2 + C(3)*GARCH(-1) + C(4)*CONVARLNTHAI

Model Two
LNTHAI = C(1) GARCH = C(2) + C(3)*RESID(-1)^2 + C(4)*GARCH(-1) LNMALAY = C(1) GARCH = C(2) + C(3)*RESID(-1)^2 + C(4)*GARCH(-1) + C(5)*CONVARLNTHAICST

Model Three (best)

LNTHAI = C(1)+ [AR(1)=C(2),AR(2)=C(3),AR(3)=C(4)] GARCH = C(5) + C(6)*RESID(-1)^2 + C(7)*GARCH(-1) LNMALAY = C(1) + [AR(1)=C(2),AR(2)=C(3),AR(3)=C(4)] GARCH = C(5) + C(6)*RESID(-1)^2 + C(7)*GARCH(-1) + C(8)*CONVARLNTHAIARTHREE

Model Form Malaysia

Serial Auto.

Yes

Yes

No

Normality
Coefficient on H

No
-0.031305 (p=0.99)

No
0.0018 (p=0.0008)

Yes
0 (p=0.89)

No sign of volatility spillovers coefficient on H significantly close to 0 for almost all models
15

Graphs of Conditional Variances

The conditional variance of a RV Y given the value of a random variable X is:
20

.005
16 12

.004
8 4

.003
0 1996 1997 1998 1999 CONVARLNMALAY CONVARLNTHAI

.002
.24 .20 .16 .12 .08 .04 .00 1996 1997 1998 1999

.001

.000 1996 1997 1998 1999

CONVARLNMALAYARTHREE

CONVARLNTHAIARTHREE

CONVARLNMALAYCST

CONVARLNTHAICST

Does this seem to indicate spillovers?

16