p–ISSN: 2723 - 6609 e-ISSN : 2745-5254
Vol. 4, No. 3, Maret 2023 http://jist.publikasiindonesia.id/
Doi : 10.59141/jist.v4i3.608 382
EARLY DETECTION OF CURRENCY CRISIS IN INDONESIA BASED ON JCI
INDICATOR USING A COMBINATION OF VOLATILITY AND MARKOV
SWITCHING MODELS
Berlyana Ayu Prasasti
1*
, Sugiyanto
2
, Sri Subanti
3
Universitas Sebelas Maret, Indonesia
Email: berlyanaay[email protected]
1
2
,
3
*Correspondence
INFO ARTIKEL
ABSTRACT
Diterima
: 12-03-2023
Direvisi
: 15-03-2023
Disetujui
: 29-03-2023
Currency crises have occurred in Indonesia in 1997-1998 and 2008,
causing significant losses both in terms of economy and social life.
Therefore, a system is needed that can detect currency crises to create
economic and currency stability. Crises can be detected through
economic indicators such as Indonesia Stock Exchange Composite
Index (IDX Composite) or Jakarta Composite Index (JCI). This study
aims to determine the appropriate model and to predict the currency
crisis in Indonesia from November 2022 to October 2023 based on the
JCI indicator. The study begins by forming an AR model, then a
volatility model in the form of an ARCH model, and finally a combined
volatility model and Markov switching two-state model. This combined
model is then used to form a smoothed probability that can detect
crises. The results of the study indicate that the MS-ARCH(2,1) model
is the appropriate model, and from the detection results, it is found that
Indonesia will not experience a currency crisis from November 2022 to
October 2023.
Keywords: Crises detection;
IHSG; AR; ARCH; Markov
Switching.
Attribution-ShareAlike 4.0 International
Introduction
Currency crises have occurred in many countries around the world, including
Indonesia. The Indonesian currency crisis occurred in 1997-1998, which began with the
fall of the Thai baht exchange rate by 27.8%, followed by the weakening of the South
Korean won, Malaysian ringgit, and the Indonesian rupiah. During this period,
Indonesia's economic growth decreased by 13.13%, and the rupiah depreciated by
600%, from Rp2,350 to Rp16,650 per 1 USD (Sri & Suliswanto, 2016). In addition,
Indonesia also experienced the Sub-prime Mortgage crisis in 2008, which originated
from the bankruptcy of the US property business. The rupiah depreciated by 30.9%
from Rp9,840 per January 2008 to Rp12,100 per November 2008 (Sri & Suliswanto,
2016). These two currency crises caused significant losses in terms of the economy and
social life. Therefore, a system is needed to detect currency crises to create economic
and currency stability, especially in Indonesia.
Kaminsky et al. (1998) proposed 15 indicators that can be used as a guide to
detect currency crises in a country. These indicators are imports, exports, trade
exchange rates, foreign exchange reserves, Composite Stock Price Index (CSPI), real
exchange rates, real savings interest rates, bank deposits, loan and deposit interest rate
ratios, real domestic rate differentials and FED real rate differentials, M1 (narrow
Early Detection Of Currency Crisis In Indonesia Based On Jci Indicator Using A Combination
Of Volatility And Markov Switching Models
Jurnal Indonesia Sosial Teknologi, Vol. 4, No. 4, April 2023 383
money), M2 (broad money) multiplier, M2 to foreign exchange reserve ratios, real
output, and domestic credit to GDP ratios. In Indonesia itself, the 1997-1998 crisis was
influenced by exchange rate indicators, interest rates, debt service ratios, and inflation,
while the 2008 crisis was influenced by CSPI indicators, interest rates, and inflation
(Keumala Sari et al., 2016). stated that there is a significant relationship between
currency crises and financial crises, so that in financial crisis modeling, currency crisis
indicators can be used. The CSPI is one of the indicators that can detect currency crises
in a country. The CSPI is defined as the stock price expressed in index numbers used for
analysis purposes and to avoid the negative effects of using stock prices. The stock price
index is an indicator or reflection of stock price movements (Widodo, 2017).
Since 1982, many methods have been developed to build models that can detect
currency crises. Engle (1982) developed the Autoregressive Conditional
Heteroscedasticity (ARCH) model to detect volatility in data that causes
heteroskedasticity effects. Then, Bollerslev (1986) developed the Generalized
Autoregressive Conditional Heteroscedasticity (GARCH) model as a development of
the ARCH model. Both models do not take into account the changes in the economic
variable conditions caused by economic crises, wars, or other causes that cause
significant changes in data values. Then, Hamilton and Susmel (1994) introduced the
Markov Switching Model as an alternative in modeling time series data with fluctuating
data.
The crisis detection model is often developed using a combination of Markov
switching and volatility models. Ananda (2015) conducted research on the detection of
financial crises in Indonesia based on the IHSG indicator using a combination of
volatility and Markov switching models with three states. The study found that the
suitable model was the MRS-ARCH(3,1) model with AR(1) as the mean model. Dina
(2015) conducted early detection of financial crises in Indonesia based on the IHSG
indicator. The IHSG indicator data contained heteroskedasticity, asymmetry, and
structural changes, so it was modeled using a two-state MS-TGARCH model.
Conducted research on forecasting stock returns in 2016 using the Exponential
Generalized Autoregressive Conditional Heteroscedasticity (EGARCH) model. Suwardi
(2017) conducted research on early detection of financial crises in Indonesia based on
import, export, and foreign exchange reserve indicators using the MS-ARCH model.
Pratiwi (2017) also conducted research on early detection of financial crises in
Indonesia using the MS-ARCH model based on the M1 indicator, the M2-to-foreign
exchange reserve ratio, and the M2 multiplier. Sugiyanto and Hidayah (2019) conducted
early detection of financial crises in Indonesia using the MS-GARCH model with the
smallest smoothed probability value during the financial crisis in Indonesia in 1997-
1998.
In this research, a combination of volatility and Markov switching models with
two states will be used to detect currency crises in Indonesia based on the IHSG
indicator. The data used is monthly data from January 1990 to October 2022 obtained
from the official Yahoo Finance website. The aim of this research is to determine the
Berlyana Ayu Prasasti, Sugiyanto, Sri Subanti
Jurnal Indonesia Sosial Teknologi, Vol. 4, No. 3, Maret 2023 384
suitable model and to predict the results of currency crises in Indonesia from November
2022 to October 2023.
Research Method
1. Research method
This study uses monthly data of the Indonesian Composite Index (IHSG) from
January 1990 to October 2022 obtained from the official Yahoo Finance website. The
study begins with creating a time series plot of the IHSG data and then conducting an
Augmented Dickey-Fuller (ADF) test to determine the stationarity of the data. If the
data is not stationary, a log return transformation is performed. The transformed data is
then used to form an AR model, and a Lagrange multiplier test is conducted to test for
heteroscedasticity effects on the residuals. If there are heteroscedasticity effects on the
residuals, a volatility model is formed, and diagnostic tests are conducted on the
residuals. These diagnostic tests include tests for normality, non-autocorrelation, and
heteroscedasticity effects. From the formed volatility model, a sign bias test is then
conducted to examine whether there is an asymmetric effect on the volatility model or
not. If there is no asymmetric effect, there is no need for further volatility modeling, and
the modeling can continue by forming a combined volatility and Markov switching
model with two states. The study then calculates the smoothed probability value and
forecasts the smoothed probability value for the period from November 2022 to October
2023, and performs crisis detection.
2. Indonesia Stock Exchange Composite Index (IDX Composite) or Jakarta
Composite Index (JCI)
Indonesia Stock Exchange Composite Index is the stock price expressed in index
numbers that are used for analysis purposes and to avoid the negative impacts of using
stock prices. The stock price index is an indicator or reflection of the movement of stock
prices (Widodo, 2017). A high stock index value indicates a busy market condition,
while a low stock index value indicates a sluggish market condition. The tendency of
increasing stock prices in the long term indicates rapid economic growth, while in the
short term, stock prices tend to fluctuate (Widoatmodjo, 2009).
3. Autoregressive Model (AR)
The autoregressive process is the process of modeling predictions r_t as a
function of the value in the previous period. The AR(p) model can be written as in
Equation (1).
with
is the log return value of the data in the t-th period formulated as
, with
is
the data of each indicator in the t-th period, ϕ_0 is a constant, ϕ_p parameters on
autoregressive models, and a_t is the residue in the t-th period (Tsay, 2002).
Autoregressive Conditional Heteroscedasticity Model (ARCH)
Early Detection Of Currency Crisis In Indonesia Based On Jci Indicator Using A Combination
Of Volatility And Markov Switching Models
Jurnal Indonesia Sosial Teknologi, Vol. 4, No. 4, April 2023 385
The Autoregressive Conditional Heteroscedasticity (ARCH) model is a type of
volatility model that can overcome the heteroskedasticity effect on the average model.
The ARCH(m) model can be written as in Equation (2).
with α_0 is constant model of ARCH, α_t is parameter model ARCH, and σ_t^2 is
the residual variance in the t-th period.
4. Model Markov Switching-ARCH (MS-ARCH)
The Markov Switching-ARCH (MS-ARCH) model is a combination of the ARCH
and Markov Switching volatility models. According to Hamilton and Susmel (1994),
the MS-ARCH(K,m) model can be formulated as in Equation (3).
with K is the number of states, m adalah orde pada model ARCH, and σ_(t,st)^2 is
the residual variance of a state in the t-th period.
5. Transition Probability Matrix
The markov process is called the stochastic process if the probability of any future
behavior (state) depends only on the behavior (state) in the present and is not changed
by additional knowledge of the behavior (state) in the past. The markov chain process
can be written as in Equation (4).
With P_ij adalah matriks probabilitas transisi berada pada state i at the time n will
go to j at time n+1. The one-step transition probability matrix for an infinite state can be
written as in Equation (5).
dengan
untuk dan
untuk .
6. Smoothed Probability
Smoothed probability is the probability value in a state based on information up to
T. According to Fruhwirth-Schnatter, et al., (2000) the smoothed probability value can
be written as in Equation (6).
Berlyana Ayu Prasasti, Sugiyanto, Sri Subanti
Jurnal Indonesia Sosial Teknologi, Vol. 4, No. 3, Maret 2023 386
With ψ_T is a collection of all information up to the T -th. Crisis detection in the
following year can be detected using smoothed probability forecasting at selected states
in that data period.
In Pr(S_t=j|ψ_T) is the value of smoothed probability at the t -th time for the j -
th regime and p_ij the probability of transition in the regime. If the smoothed
probability value is high, there is a possibility of a crisis and vice versa, if the smoothed
probability value is low, there is a possibility that there will be no crisis. According to
Hermosillo and Hesse (2009) smoothed probability values of 0 – 0.39 indicate
indicators of a financial crisis in a stable state, 0.4 – 0.59 indicates a vulnerable
condition, and 0.6 – 1 indicates a crisis state.
Result And Discussion
1. Identify Data Patterns
To determine the pattern in JCI data, an analysis was carried out on the time series
plot as presented in Figure 1.
Gambar 1 Plot time series IHSG
Figure 1 shows fluctuations in the JCI data, where the data increases over time
and it can be seen that the variance is not constant, thus indicating that the data is not
stationary. To prove this conjecture, an ADF test was carried out and a probability value
of 0.978 was obtained. Since this value is greater than α=0,05, it can be concluded that
the data is not stationary. To solve this problem, a log-return transformation was carried
out on the data and a time series plot was obtained for the transformation result data as
presented in Figure 2.
Early Detection Of Currency Crisis In Indonesia Based On Jci Indicator Using A Combination
Of Volatility And Markov Switching Models
Jurnal Indonesia Sosial Teknologi, Vol. 4, No. 4, April 2023 387
Figure 2 Time series plots of transformed data
Figure 2 shows that the transformed data is already stationary, as the data
fluctuates around the average value. Based on the ADF test, a probability value of 0.01
was obtained, which is smaller than α=0,05, so it can be concluded that the log return
data is stationary. It is this transformation result data that is used for the formation of the
model.
2. Formation of the AR Model
The Autoregressive model identification process is based on the PACF behavior
of the transformed data. From several possible models formed, significance tests were
carried out on each parameter and the model with the smallest AIC value was selected.
For JCI indicators, the best model is AR(1) with the following model.
Furthermore, testing was carried out using the Lagrange multiplier test to
determine whether the residue from the ARMA model contained the effect of
heteroskedasticity or not. From the tests that have been carried out, a probability value
of 4,19×〖10〗^(-9) was obtained. This value is smaller than α=0,05, so it can be
concluded that the residues of the AR model contain the effect of heteroskedasticity.
Therefore, advanced modeling with volatility models is carried out.
3. Formation of a Volatility Model
The formation of the volatility model is based on the ACF plot of the squared
residual AR model that has been formed. From several possible models formed,
significance tests were carried out on each of the parameters and the model with the
smallest AIC value was selected. For the JCI indicator, the best volatility model is
obtained, namely ARCH(1) with the following model.
Next, diagnostic tests were conducted on the residuals of the volatility model to
determine the adequacy of the model. These diagnostic tests included tests for
normality, non-autocorrelation, and heteroskedasticity. The normality test was
conducted using the Kolmogorov-Smirnov test and obtained a probability of 0.8811,
which is greater than α=0,05. This indicates that the residuals of the volatility model are
Berlyana Ayu Prasasti, Sugiyanto, Sri Subanti
Jurnal Indonesia Sosial Teknologi, Vol. 4, No. 3, Maret 2023 388
normally distributed. The non-autocorrelation test was performed using the Ljung-Box
test and obtained a probability value of 0.8457, which is greater than α=0,05. Therefore,
it can be concluded that there is no autocorrelation in the residual. The
heteroskedasticity test was performed to determine the presence of heteroskedasticity
effects on the residual. This test used the Lagrange multiplier test and obtained a
probability value of 0.9897, which is greater than α=0,05. This means that the residuals
generated by the model do not contain heteroskedasticity effects.
4. Sign Bias Test
Furthermore, a sign refractive test is carried out to check whether there is an
asymmetric effect on the volatility model or not. This test was carried out with a sign
bias test and obtained a probability value of 0.5479 which is greater than α=0,05. This
shows that the residue generated in the volatility model does not have an asymmetric
effect (leverage effect), so there is no need for further volatility modeling.
5. Formation of a Combined Model of Volatility and Markov Switching
The already formed volatility model is then combined with a two-state switching
markov model to detect stable and crisis conditions. Changes in conditions that occur in
the model are considered as the result of an unobserved random variable called a state.
This state is divided into two, namely low and high volatility conditions. To describe
such changes in conditions, a transition probability matrix is used. The transition
probability matrix for the JCI indicator is as follows.
Based on the matrix, it can be seen that the probability value of staying in a low
volatility state is 0.9822, the probability of changing state from low to high volatility is
0.0178, the probability of changing state from high to low volatility is 0.7530, and the
probability of staying in a high volatility state is 0.2470.
6. Smoothed Probability Value and Smoothed Probability Forecasting Value
From the MS-ARCH(2,1) model, the smoothed probability value presented in
Figure 3 is obtained.
Figure 3 Plot smoothed probability
Crisis detection is carried out by looking at the minimum value of smoothed
probability results when there was a currency crisis in Indonesia, namely in 1997 - 1998
Early Detection Of Currency Crisis In Indonesia Based On Jci Indicator Using A Combination
Of Volatility And Markov Switching Models
Jurnal Indonesia Sosial Teknologi, Vol. 4, No. 4, April 2023 389
and 2008. For the JCI indicator, crisis conditions occur if the smoothed probability
value is more than or equal to 0.5116. Forecasting the detection of currency crisis in
Indonesia for the period of one year ahead is presented in Table 1.
Table 1 Currency crisis detection prediction
Period
Smoothed Probability Value
Condition
Nov-22
0,140698
Stable
Dec-22
0,050036
Stable
Jan-23
0,029253
Stable
Feb-23
0,02449
Stable
Mar-23
0,023397
Stable
Apr-23
0,023147
Stable
May-23
0,02309
Stable
Jun-23
0,023077
Stable
Jul-23
0,023074
Stable
Aug-23
0,023073
Stable
Sep-23
0,023073
Stable
Oct-23
0,023073
Stable
Table 1 shows that all smoothed probability values are smaller than the threshold
value, so it can be concluded that in the period from November 2022 to October 2023
there was no currency crisis in Indonesia.
Conclussion
Based on the research that has been done, it can be concluded as follows.
a. The combination of volatility and markov switching models that are suitable for
early detection of currency crises in Indonesia based on the JCI indicator is the MS-
ARCH(2,1) model with the AR(1) model as the average model
b. Based on JCI indicators, Indonesia will not experience a currency crisis from
November 2022 to October 2023.
Berlyana Ayu Prasasti, Sugiyanto, Sri Subanti
Jurnal Indonesia Sosial Teknologi, Vol. 4, No. 3, Maret 2023 390
Bibliografi
Frey, M., Taylor, H. M., & Karlin, S. (1994). An Introduction to Stochastic Modeling.
Technometrics, 36(4). https://doi.org/10.2307/1269970
Fruhwirth-Schnatter, S., Kim, C.-J., & Nelson, C. R. (2000). State-Space Models with
Regime-Switching: Classical and Gibbs Sampling Approaches with Applications.
Journal of the American Statistical Association, 95(452).
https://doi.org/10.2307/2669796
González-Hermosillo, B., & Hesse, H. (2009). Global Market Conditions and Systemic
Risk. IMF Working Papers, 09(230), 1.
https://doi.org/10.5089/9781451873771.001
Hamilton, J. D., & Susmel, R. (1994). Autoregressive conditional heteroskedasticity and
changes in regime. Journal of Econometrics, 64(1–2), 307–333.
https://doi.org/10.1016/0304-4076(94)90067-1
Tsay, R. S. (2002). Analysis of Financial Time Series. In Analysis of Financial Time
Series. https://doi.org/10.1002/0471264105
Abdushukurov, N. (2019). The impact of currency crises on economic growth and
foreign direct investment: The analysis of emerging and developing economies.
Russian Journal of Economics, 5, 220–250.
Ananda, A. (2015). Pendeteksian Krisis Keuangan di Indonesia Berdasarkan Indikator
Harga Saham Menggunakan Gabungan Model Volatilitas dan Markov Switching
Tiga State.
Bollerslev, T. (1986). Generalized autoregressive conditional heteroskedasticity. Journal
of Econometrics, 31(3), 307–327.
Dina, A. (2015). Model Krisis Pasar Modal di Indonesia Menggunakan Markov
Switching TGARCH (MS-TGARCH) Dua State Berdasarkan Indikator IHSG.
Engle, R. F. (1982). Autoregressive Conditional Heteroscedasticity with Estimates of
the Variance of United Kingdom Inflation. Econometrica, 50(4), 987.
Frey, M., Taylor, H. M., & Karlin, S. (1994). An Introduction to Stochastic Modeling.
Technometrics, 36(4).
Fruhwirth-Schnatter, S., Kim, C.-J., & Nelson, C. R. (2000). State-Space Models with
Regime-Switching: Classical and Gibbs Sampling Approaches with Applications.
Journal of the American Statistical Association, 95(452).
González-Hermosillo, B., & Hesse, H. (2009). Global Market Conditions and Systemic
Risk. IMF Working Papers, 09(230), 1.
Early Detection Of Currency Crisis In Indonesia Based On Jci Indicator Using A Combination
Of Volatility And Markov Switching Models
Jurnal Indonesia Sosial Teknologi, Vol. 4, No. 4, April 2023 391
Hamilton, J. D., & Susmel, R. (1994). Autoregressive conditional heteroskedasticity and
changes in regime. Journal of Econometrics, 64(1–2), 307–333.
Kaminsky, G., Lizondo, S., & Reinhart, C. M. (1998). Leading Indicators of Currency
Crises. IMF Staff Papers, 45(1).
Keumala Sari, P., (2016). Identifikasi Penyebab Krisis Moneter dan Kebijakan Bank
Sentral di Indonesia: Kasus Krisis Tahun (1997-1998 dan 2008). Jurnal Ilmiah
Mahasiswa Ekonomi Pembangunan, 1(2), 377-388.
Pratiwi, E. S. (2017). Penerapan Gabungan Model Volatilitas dan Markov Switching
dalam Pendeteksian Dini Krisis Keuangan di Indonesia Berdasarkan Indikator
M1, M2 per Cadangan Devisa, dan M2 Multiplier.
Sri, M., & Suliswanto, W. (2016). TINGKAT KETERBUKAAN EKONOMI DI
NEGARA ASEAN-5. Neo-Bis, 10(1), 33–48.
Sugiyanto, & Hidayah, A. Z. (2019). Indonesian Financial Crisis Prediction Using
Combined Volatility and Markov Regime Switching Model Based on Indicators
of Real Interest Rate on Deposits and Lending Interest Rate/Deposit Interest Rate.
Journal of Physics: Conference Series, 1373(1), 012045.
Suwardi, V. R. A. (2017). Pendeteksian Krisis Keuangan di Indonesia dengan
Gabungan Model Volatilitas dan Markov Switching pada Indikator Impor,
Ekspor, dan Cadangan Devisa.
Tsay, R. S. (2002). Analysis of Financial Time Series. In Analysis of Financial Time
Series.
Wibowo, N. M., Sugito, & Rusgiyono, A. (2016). Pemodelan Return Saham Perbankan
Menggunakan Exponential Generalized Autoregressive Conditional
Heteroscedasticity (EGARCH). Jurnal Gaussian, 6(1), 91–99.
Widoatmodjo, S. (2009). Pasar Modal Indonesia: Pengantar & Studi Kasus. Ghalia
Indonesia.
Widodo. (2017). Analisis Pengaruh Indeks Harga Saham Gabungan Regional Asia
Terhadap Indeks Harga Saham Gabungan Indonesia. EkBis: Jurnal Ekonomi Dan
Bisnis, 1(2), 148–164.
