Are there Rational Speculative Bubbles in ASEAN Stock Markets?

Download Report

Transcript Are there Rational Speculative Bubbles in ASEAN Stock Markets?

ARE THERE RATIONAL
SPECULATIVE BUBBLES
IN ASEAN STOCK
MARKETS?
Gilbert V. Nartea
Bo Hu and Baiding Hu
Faculty of Commerce
Lincoln University
New Zealand
Introduction

Bubbles: long run up in price followed by a
crash. (e.g., ++++++++++++ -)
Stock prices diverge from fundamental values
 Often attributed to irrational behaviour and
evidence of inefficient markets.


Rational bubble is a special case

investors realise a bubble exists, but they find it
rational to stay in the market because they believe
the growth rate of the bubble compensates them
for the possibility of a crash.
Introduction



Episodes of price run ups followed by large
drops in ASEAN stock markets.
Led observers to suggest presence of
bubbles
No conclusive evidence of whether or not
rational speculative bubbles exist in Asian
stock markets.
• Important issue in relation to the
efficient allocation of investment
resources and asset pricing.
Introduction

Common tests for bubbles


Cointegration (relation between fundamentals and stock
prices)
Descriptive stats (autocorrelation, skewness, and kurtosis)
in stock returns






These attributes necessarily derive from the two characteristics
of bubbles (i.e price run-up and crash) .
++++++++++++ The long run of positive returns (positive autocorrelation) and the
singular negative return during the crash means that the bubble
process must be skewed.
Leptokurtosis (fat-tailed distributions) is consistent with the
occasional large deviation in price characterising the crash.
Problem is these could also be caused by other factors other than
bubbles.
Duration dependence (McQueen and Thorley, 1994).
Introduction
We use descriptive stats and duration
dependence tests for Indonesia,
Malaysia, and the Philippines (1991 to
2009).
 Preliminary results

Introduction
McQueen and Thorley (1994)
 Testable implication of rational
speculative bubbles:
the longer is the run of positive
returns, the smaller is the probability
that the run will end (negative
duration dependence)

Introduction



Rational bubble process :
UPC = ΔFV + ΔBubble
UPC = unexpected price changes
ΔFV = changes in fundamental value
ΔBubble = changes in the bubble
As the bubble component grows it dominates the fundamental
component



in such a way that negative shocks to fundamental value will have
minimal impact on total returns
hence as the bubble grows, the longer is the run of positive returns
and the probability of getting a negative total return decreases (ie.,
negative duration dependence)
(as to what causes the bubble to grow and what causes it to
eventually burst, unfortunately we do not yet have a coherent
theory about how this comes about)
Introduction

Duration dependence test does not require
 specification
of the underlying fundamental
value relationship model.
 normality of the time series behaviour under
investigation
Literature Review

Chan et al. (1998)
Duration dependence tests (1975- 1994)
 Rational speculative bubbles in Thailand but not in
Hong Kong, Japan, Korea, Malaysia, Taiwan.


Sarno and Taylor (1999)
Cointegration tests (1989-1997)
 Bubbles in China, Indonesia, Malaysia, Philippines,
Singapore, South Korea, Taiwan and Thailand


Ali et al. (2009) and Mokhtar et al. (2006)

Rational speculative bubbles in Malaysia
Literature Review

Jirasakuldech et al. (2008)
 cointegration
and duration dependence tests
(1975- 2006)
 Rational speculative bubbles in the Thai stock
market consistent with those of
Watanapalachaikul and Islam (2007).

Rangel and Pillay (2007)
 excess
volatility tests, cointegration tests,
duration dependence tests, and the intrinsic
bubbles model (1975-2007)
 Rational speculative bubbles in Singapore
Data and Methods

McQueen and Thorley (1994) suggest
monthly data contains less noise
 but short data series from monthly data could
result in lack of power of tests used
 we use both monthly and weekly index values.


Monthly and weekly closing prices from
Datastream (February 1991- December 2009)
Jakarta Composite Index,
 Kuala Lumpur Composite Index and the
 Philippine Stock Exchange Index

Data and Methods



We use real returns
First, transform real returns into run
lengths of positive and negative observed
returns, i.e., ++++++---+++++++++Run is a sequence of returns with the same
sign.
Data and Methods


Next we compute the probability of ending
a run or the sample hazard rate for each
length i
hi = Ni/(Mi + Ni)
where
 Ni
is the number of completed runs of the
length i, and
 Mi the number of completed runs with length
greater than i.
Data and Methods

Examine the relation between the
hazard rate or probability of the run
ending (hi) and length of the run i.
A

negative relationship indicates the
presence of rational speculative bubbles.
Focus only on positive runs
 since there can be no bubbles in runs of
negative returns because this would imply
that stock prices can be negative as the
bubble grows over time.
Data and Methods

Formal test
We estimate a Log-logistic hazard function.
hi = {1+ exp[-α -β ln(i)]}-1
where
hi is the hazard rate (probability of ending a run)
i is the length of the run and
β is the slope parameter.

A negative β signals the presence of rational bubbles






probability of a positive run ending is a decreasing function
of the length of the run (ie., negative duration dependence)
Empirical Results



Index plots
Descriptive statistics
Duration dependence tests
1/04/2008
1/05/2006
1/06/2004
1/07/2002
1/08/2000
1/09/1998
1/10/1996
1/11/1994
1/12/1992
1/01/1991
1000
500
0
Philippine (Philippine Composite Index)
4000
3500
3000
2500
2000
1500
1000
500
0
1/12/1998
1/07/2000
1/02/2002
1/09/2003
1/04/2005
1/11/2006
1/06/2008
1500
1/05/1997
2000
1/10/1995
2500
1/03/1994
3000
1/08/1992
Indonesia (Jakarta Composite Index)
1/01/1991
1/01/1991
1/10/1992
1/07/1994
1/04/1996
1/01/1998
1/10/1999
1/07/2001
1/04/2003
1/01/2005
1/10/2006
1/07/2008
Figure 1. Index Plots
Malaysia (Kuala Lumpur Composite Index)
1600
1400
1200
1000
800
600
400
200
0
Figure 1. Index plots



Evident from plots
Long run-up in price followed by a sudden
drop
1993 to 1997 (Asian financial crisis)
JCI reached an historic low in September 1997;
 PSE CI dropped 1000 points from a high of some
3000 points in the same year.
 KL CI lost more than 50%, from above 1,200 points
to fewer than 600 points during May to December.


2001 (U.S. housing bubble) to 2007 (Sub-prime
loan financial crisis)
Table 1a. Real monthly returns, 02:1991-12:2009
Monthly returns
T
Mean
Maximum
Minimum
Standard Deviation
Skewness
(SE)
Kurtosis
(SE)
Jarque-Bera
ρ1
ρ2
ρ3
ρ4
ρ5
ρ6
ρ12
Q(6)
Q(12)
Indonesia (Jakarta
Composite Index)
Malaysia (Kuala
Lumpur Composite
Index)
Philippines (Philippine
Composite Index)
227
-0.000809
0.286522
-0.401982
0.089032
-0.818110
(0.162578)
6.451037
(0.325156)
137.9675
0.229
0.000
0.051
0.102
0.002
0.048
-0.074
15.680
20.708
227
0.001691
0.334583
-0.386219
0.079253
-0.275717
(0.162578)
7.841069
(0.325156)
224.5411
0.104
0.145
-0.119
-0.061
-0.071
-0.074
-0.066
13.978
36.509
227
0.001758
0.336165
-0.294769
0.083648
0.183939
(0.162578)
5.222854
(0.325156)
48.01441
0.109
0.074
-0.049
-0.044
0.011
-0.023
0.073
5.1811
11.048
Table 1b. Real weekly returns, 02:1991-12:2009
Weekly returns
T
Mean
Maximum
Minimum
Standard Deviation
Skewness
(SE)
Kurtosis
(SE)
Jarque-Bera
ρ1
ρ2
ρ3
ρ4
ρ5
ρ6
ρ12
Q(6)
Q(12)
Indonesia (Jakarta
Composite Index)
Malaysia (Kuala
Lumpur Composite
Index)
Philippines (Philippine
Composite Index)
987
-0.000379
0.218598
-0.259975
0.041312
-0.288350
(0.077968)
8.298191
(0.155936)
1168.090
-0.025
0.081
0.119
0.087
0.061
0.065
0.020
36.498
50.057
987
0.000347
0.285299
-0.210479
0.035344
0.143651
(0.077968)
12.22985
(0.155936)
3503.283
-0.018
0.005
0.082
-0.006
0.060
0.044
-0.061
12.585
25.569
987
0.000276
0.180984
-0.216020
0.039632
-0.043448
(0.077968)
5.551043
(0.155936)
267.9447
-0.011
0.064
0.079
-0.020
0.035
-0.016
-0.006
12.218
24.632
Descriptive statistics

Mean Monthly returns
Similar for Malaysia and the Philippines
 Indonesia had a negative average monthly real
return


Maximum Monthly returns
Similar for Malaysia and the Philippines
 Lower for Indonesia


Minimum Monthly returns


Indonesia experienced the lowest real return,
followed by Malaysia then Philippines.
ALL monthly and weekly return distributions
are Non-normal (Jarque-Bera test statistics).
Descriptive statistics


Negative coefficients of skewness imply the
presence of bubbles.
Skewness of Monthly and Weekly returns
 Significant
negative skewness in Indonesia
(consistent with bubbles)
 BUT NOT in Malaysia and the Philippines.
Descriptive statistics
• Leptokurtosis of returns also imply the
presence of rational speculative
bubbles.
•
•
Monthly and weekly returns in ALL markets
exhibit leptokurtosis
Consistent with the presence of rational
speculative bubbles.
Descriptive statistics
Autocorrelated returns (serial
dependence) could also indicate the
presence rational speculative bubbles.
 Sample autocorrelation coefficients

 Monthly
returns: seriaI dependence in
Indonesia and Malaysia but NOT in
Philippines
 Bubble
 Weekly
series.
 Bubble
in Indonesia and Malaysia
returns: serial dependence in ALL
in ALL three markets
Descriptive statistics



We also conducted two Ljung-Box (LB) tests, one that
included the first six and twelve-order autocorrelation
coefficients (Q(6)) , and (Q(12)).
Monthly series: the two LB tests agree with the
conclusions derived from the individual autocorrelation
coefficients that returns are serially correlated in
Indonesia and Malaysia but not in the Philippines.
Weekly series: slight disagreement between the LB tests
and the individual autocorrelation coefficients in the case
of the Philippines. The Q(6) values suggest serial
independence in the Philippines while the individual auto
correlation coefficients suggest otherwise.
Descriptive statistics
• Autocorrelation
• On balance, the autocorrelated returns
in Indonesia and Malaysia and to some
extent in the Philippines suggest the
presence of rational speculative
bubbles in these markets.
Descriptive statistics
In summary
 The characteristics of the return
distributions indicate (in varying
degrees) the presence of rational
speculative bubbles in these markets.
 However, these characteristics could
also be due to factors NOT related to
rational speculative bubbles.
 Hence we conduct duration dependence
tests next.

Duration dependence tests



Monthly returns
Weekly returns
Sub-period analysis
Table 2. Tests of duration dependence for positive runs of
monthly indices’ returns for the full period (1991-2009)
Run
Length
Indonesia
Actual
Run Counts
1
24
2
9
3
9
4
2
5
2
6
1
7
2
8
1
9
10
Total
50
Log-logistic test
α
β
LRT of H0:
β=0
( p-value)
Sample
Hazard
Rates
0.4800
0.3462
0.5294
0.2500
0.3333
0.2500
0.6667
1.0000
Malaysia
Actual
Run Counts
25
9
9
3
4
2
1
Sample
Hazard
Rates
0.4717
0.3214
0.4737
0.3000
0.5714
0.6667
1.0000
53
-0.17166
-0.14474
Philippines
Actual
Sample
Run Counts Hazard
Rates
26
0.4643
12
0.4000
11
0.6111
5
0.7143
2
1.0000
56
0.27139
0.03684
0.28014
0.55549
0.2390
0.0145
2.2331
0.6249
0.9041
0.1351
Monthly returns

Indonesia
longest run of positive returns : 8 months
 If monthly returns are independent, the
probability of getting 8 consecutive positive runs is
4 in one thousand.
 The fact that we have 1 run of 8 consecutive
positive returns out of 227 observations seem to
indicate the presence of bubbles. The question is,
are they rational bubbles?
 “Eyeballing” sample hazard rates do not reveal any
discernible patterns (no rational speculative
bubbles over the sample period?)

Monthly returns

Malaysia
 longest
run: 7 months
 sample hazard rates also do not reveal any
pattern
 no rational speculative bubbles over the
sample period.
Monthly returns

Philippines
 Longest
run: 5 months
 Hazard rates appear to be increasing (opposite
to the pattern we would expect in the presence
of rational speculative bubbles).
Monthly returns

Log-logistic test on the level of beta.
 Indonesia
is negative at -0.14474, BUT
statistically insignificant (p-value of 0.6249)
 Malaysia and the Philippines are of the wrong
sign and statistically insignificant.
 Therefore, No rational speculative bubbles in
Indonesia, Malaysia and the Philippines
Table 3. Tests of duration dependence for positive runs of weekly
indices’ returns for the full period (1991-2009)
Run
Length
Indonesia
Actual
Run Counts
1
107
2
47
3
27
4
16
5
8
6
6
7
2
8
2
9
4
10
1
11
0
12
1
13
Total
221
Log-logistic test
α
β
LRT of H0:
β=0
( p-value)
Sample
Hazard
Rates
0.4842
0.4123
0.4030
0.4000
0.3333
0.3750
0.2000
0.2500
0.6667
0.5000
0.0000
1.0000
Malaysia
Actual
Run Counts
Run
Length
107
54
29
14
7
4
8
3
0
0
0
1
0.4714
0.4500
0.4394
0.3784
0.3043
0.2500
0.6667
0.7500
0.0000
0.0000
0.0000
1.0000
227
Philippines
Actual
Run Counts
122
59
24
11
12
7
3
Sample
Hazard
Rates
0.5126
0.5086
0.4211
0.3333
0.5455
0.7000
1.0000
238
-0.09387
-0.26994
0.11663
1.3875
0.026118
-0.07468
4.0593
0.0440**
1.3875
0.2388
0.2230
0.6367
Weekly returns









Indonesia
the longest run: 12 weeks
sample hazard rates appear to be decreasing
consistent with the presence of rational
speculative bubbles.
Malaysia
the longest run: 12 weeks
sample hazard rates appear be decreasing
Philippines
longest run: 7 weeks but its
hazard rates also appear to be decreasing
Weekly returns

Log-logistic test on level of beta
 Indonesia
is negative at -0.26994 and is
statistically significant (p-value of 0.0440).
 Malaysia and the Philippines have negative beta
coefficients BUT not statistically significant.
 Therefore there is evidence of rational
speculative bubbles in Indonesia BUT not in
Malaysia and the Philippines.
Sub-period analysis
• We divide the sample into four subperiods:
•
•
•
•
1991-1997 (run-up towards Asian Fin. Crisis
(AFC))
1998-2001 (immediate aftermath of AFC)
2002-2007 (run-up towards Global Fin.
Crisis (GFC))
2008-2009 (immediate aftermath of the
GFC)
Table 4a. Tests of duration dependence for positive runs of
monthly indices’ returns for sub-periods.
Monthly
returns
1991-1997
1998-2001
2002-2007
2008-2009
α
β
LRT
p - value
α
β
LRT
p - value
α
β
LRT
p - value
α
β
LRT
p - value
Indonesia
Malaysia
Philippines
-0.4947
0.511646
0.7373
0.3905
1.113205
-0.84681
0.6741
0.4116
-1.35115
0.510639
1.1668
0.2801
0.890085
-1.22916
1.2510
0.2634
0.006562
-0.31352
0.4122
0.5209
0.365165
-0.39236
0.4310
0.5115
-0.83989
0.563952
1.1639
0.2807
-1.22568
0.949599
0.7055
0.4009
-0.39061
0.869177
1.7131
0.1906
0.351353
-0.32962
0.1599
0.6892
-0.36268
0.5304
0.7969
0.3720
-1.10348
1.892696
1.9281
0.1650
Table 4b. Tests of duration dependence for positive runs of weekly
indices’ returns for sub-periods.
Indonesia
Malaysia
Philippines
-0.06359
-0.42708
3.4558
-0.31126
0.037624
0.0255
0.8732
0.384873
-0.54013
2.6310
0.1042
-0.21155
-0.21222
0.8985
0.3432
-0.4558
0.321495
0.4595
0.4979
-0.0772
0.105562
0.1523
0.6964
0.174027
-0.08751
0.0509
0.8216
0.097248
-0.26455
1.0574
0.3038
-0.29281
0.118261
0.0680
0.7943
Weekly returns
1991-1997
1998-2001
2002-2007
2008-2009
α
β
LRT
p - value
α
β
LRT
p - value
α
β
LRT
p - value
α
β
LRT
p - value
0.0630*
0.202392
0.255737
0.3527
0.5526
-0.37766
-0.18976
0.8418
0.3589
-0.30705
0.458229
0.7715
0.3798
Sub-periods
• Panel A (Monthly returns)
• No significant beta in any sub-period in ALL markets
• No rational speculative bubbles consistent with full
sample
• Panel B (Weekly returns)
 No rational speculative bubbles in any sub-period for
Malaysia and Philippines.
• Evidence of rational speculative bubbles in Indonesia
(1991-1997); none in other sub-periods.
• Rational bubble detected in Indonesia in the full sample of
weekly returns is due to at least one bubble episode in the
run up towards the AFC of 1997.
Concluding remarks



Apparent episodes of long price run ups
followed by large drops in ASEAN stock
markets
prompted the popular press to conjecture the
presence of asset bubbles in these markets
causing stock prices to deviate from
fundamental values.
Duration dependence to formally test for the
presence of rational speculative bubbles in the
stock markets of Indonesia, Malaysia, and the
Philippines (1991-2009).
Concluding remarks


Results suggest that stock prices have
deviated from fundamental values in
Indonesia likely caused by rational
speculative bubbles.
Sub-period analysis further reveals that
the bubble might have occurred over the
period from 1991 to 1997.
Concluding remarks





We do not detect rational bubbles in Malaysia and
the Philippines.
This implies that the long run up in prices and the
subsequent drop seen in the months leading up to the
AFC and GFC could have been justified by
fundamental value changes (ie, no bubble).
However also possible that there was a bubble but
caused by irrational investor behaviour.
Or our tests may not be powerful enough to detect
bubbles.
We suggest that further research into this area is
warranted.
Thank you