International Journal of Advances in Applied Sciences (IJAAS) Vol.
No.
March 2026, pp.
ISSN: 2252-8814.
DOI: 10.
11591/ijaas.
Markov-switching and noise-to-signal ratio approach for early detection of currency crises Sugiyanto.
Muhammad Bayu Nirwana.
Isnandar Slamet.
Etik Zukhronah.
SyifaAo Salsabila Gita Parahita Statistics Study Program.
Faculty of Mathematics and Natural Sciences.
Universitas Sebelas Maret.
Surakarta.
Indonesia
Article Info
ABSTRACT
Article history:
Economic instability can easily lead to a currency crisis.
Therefore, observing a number of crisis indicators is crucial for building an early warning system (EWS).
However, selecting the indicators most responsive to the crisis is the best choice.
For this purpose, the noise-to-signal ratio (NSR) method was used.
Monthly data from 1990-1925 were used in the autoregressive moving average (ARMA), generalized autoregressive moving average with generalized autoregressive conditional heteroscedasticity (GARMACH), and Markov-switching (MS)-GARMACH hybrid models to explain the crisis.
Model interpretation indicates that there will be no crisis from May 2025-April 2026.
Received Sep 1, 2025 Revised Oct 14, 2025 Accepted Nov 4, 2025 Keywords:
Currency crisis Early warning system Markov-switching Noise-to-signal ratio Volatility model This is an open access article under the CC BY-SA license.
Corresponding Author:
Sugiyanto Statistics Study Program.
Faculty of Mathematics and Natural Sciences.
Universitas Sebelas Maret 36 Ir.
Sutami Street.
Kentingan.
Jebres.
Surakarta.
Central Java 57126.
Indonesia Email: sugiyanto61@staff.
INTRODUCTION
From the 1970s to the mid-1990s.
Indonesia recorded solid economic growth, controlled inflation, and a healthy external balance.
Because of this strong performance, the World Bank once called Indonesia an economic miracle .
However, this situation changed drastically when the Asian Financial Crisis struck in The Indonesian economy again faced turbulence during the 2007-2008 Global Financial Crisis .
, and more recently, the COVID-19 pandemic in 2020 triggered a global recession that also affected Indonesia's financial stability .
Ae.
Past financial stability does not always guarantee protection against future disruptions .
Ae.
These conditions have prompted policymakers to design an early warning system (EWS) using macroeconomic indicators to anticipate potential crises.
Previous research has identified around fifteen key indicatorsAisuch as export and import performance, foreign exchange reserve adequacy, interest rate differentials, and monetary aggregatesAithat tend to move ahead of financial stress .
, .
Among various selection techniques, the noise-to-signal ratio (NSR) approach has been widely adopted because lower NSR values imply stronger predictive power for detecting crises .
Ae.
Economic and financial data often display volatility clustering, making it necessary to use models that account for time-varying variance.
The autoregressive conditional heteroskedasticity (ARCH) and generalized autoregressive conditional heteroskedasticity (GARCH) models introduced by Engle .
and Bollerslev .
are well suited for this purpose .
Yet, these models alone are unable to capture structural changes or regime shifts that frequently accompany crises.
The Markov-switching (MS) model proposed by Hamilton .
provides an alternative by allowing the system to switch probabilistically between stable and Journal homepage: http://ijaas.
Int J Adv Appl Sci
ISSN: 2252-8814
crisis states .
, .
Hybrid versions such as Markov-switching-generalized autoregressive conditional heteroskedasticity (MS-GARCH) and Markov-switching dynamic conditional correlation generalized autoregressive conditional heteroskedasticity (MS-DCC-GARCH) extend this flexibility, offering better tools for analyzing nonlinear financial dynamics .
, .
In recent years, several studies have employed such hybrid models to examine financial vulnerability in Indonesia and other Asian economies .
Ae.
Nonetheless, only a few combines NSR-based indicator selection with regime-switching volatility frameworks.
Addressing this gap, the present study integrates both approaches to construct an early-warning system for detecting potential currency and financial crises in Indonesia.
The objective is to offer a more adaptive, data-driven framework that can support macroprudential policy design and enhance the countryAos financial stability monitoring .
Ae.
RESEARCH METHOD
This study selects 15 macroeconomic indicators as potential crisis signals based on their lowest NSR Monthly data from January 1990-April 202 covering trade, reserves, interest rates, exchange rates, money supply, stock prices, output, and domestic credit per gross domestic product (GDP) were obtained from International Financial Statistics (IFS) and Bank Indonesia (BI).
Indicator selection was based on each variableAos ability to detect crises using the exchange market pressure (EMP) index, as shown in .
, calculated as the weighted average of exchange rate and reserve changes .
, .
Ae.
yc OeycycOe1 yaycAycEyc = ( yc ycycOe1 yca OeycaycOe1 ycaycOe1 ) Oe ( y.
( yc Where ycyc is the rupiah exchange rate against the US dollar in month t, ycayc is the foreign exchange reserve in month t, yuayc is the standard deviation of the rupiah exchange rate against the US dollar, dan yuayca is the standard deviation of foreign exchange reserves.
The threshold value representing crisis conditions is calculated using .
yca = ycuI yu yua Where yu is set at 1.
Based on this threshold, crisis periods are identified through .
1, ycnyce EMP > b yaya = { 0, if EMP O b Where 1 denotes a crisis, and 0 denotes no crisis .
, .
, .
Macroeconomic indicators were transformed to improve their sensitivity to crises.
Seasonal variables were converted into annual growth rates, while non-seasonal variables were differenced.
Additionally, the lending-to-deposit rate ratio was log-transformed, and the real exchange rate was split into trend and cycle components using the HodrickAePrescott filter as .
, .
, .
Ae.
ya = Ocycyc=1.
cyc Oe yuayc )2 yuI OcycNOe1 yc=2 [.
uayc 1 Oe yuayc ) Oe .
uayc Oe yuaycOe1 )] .
Where ycyc is the time series observation at t, yuayc is the trend component at t, yuI is the penalty term, set to 129,600 for monthly data.
Signal effectiveness was evaluated using a 24-month signal horizon.
If a crisis signal occurs and a crisis follows within 24 months, it is classified as a correct signal (A).
if no crisis follows, as a false signal (B).
if there is no signal and no crisis, as (D).
and if there is no signal but a crisis occurs, as (C).
The signal matrix is presented in Table 1, and the NSR value is calculated as .
ycAycIycI = yaA/.
aA y.
a y.
Table 1.
Signal indicator matrix Signal No signal Crises occurred No crises occurred The three indicators with the lowest NSR values were selected for further modelling.
The data were divided into in-sample (January 1990-April 2.
and out-of-sample (May 2024-April 2.
Stationarity testing was performed using the augmented Dickey-Fuller (ADF) test, and if non-stationarity was detected, a log-return transformation was applied in .
, .
, .
Markov switching and noise-to-signal ratio approach for early detection of currency crises (Sugiyant.
A ISSN: 2252-8814
ycyc = ycoycu ycyc Oe ycoycu ycycOe1 Granger causality testing was conducted to examine relationships among the indicators .
, .
, .
The optimal lag length was determined using SchwarzAos Criterion (SC) as in .
ycIya = Oe2 ln.
yco ycoycu.
cN) .
Where T is the number of observations, k is the number of parameters estimated in the model, and L is the maximum likelihood value of the model.
The Granger causality test statistic is formulated as in .
, .
, .
yaya = .
ayaycIycI OeyayaycIycOycI )/yco .
yayaycIycOycI /.
cNOeyc.
Where yayaycIycI = yayaycIycOycI = OcycNycn=1.
cUyc Oe ycUCyc )2 , yayaycIycOycI is the residual sum of squares from the unrestricted regression, and yayaycIycI is from the restricted regression.
If no causal relationship was found, univariate autoregressive moving average (ARMA) .
, .
modelling was performed.
The orders p and q were determined from the autocorrelation function (ACF) and partial autocorrelation function (PACF) plots, and the best model was selected based on the Akaike Information Criterion (AIC) as in .
yayaya = Oe2 ln ya 2yco Where yco denotes the number of variables, and ya denotes the maximum likelihood value of the model.
The ARMA model was validated using three diagnostic tests.
Autocorrelation with the Ljung-Box test .
, heteroskedasticity with the Lagrange multiplier test .
Ae.
, and normality with the KolmogorovSmirnov test .
If heteroskedasticity was detected.
ARCH.
and GARCH .
, .
models were applied in .
Ae.
yuayc 2 = yu0 Ocyco ycn=1 yuycn ycaycOeycn yuayc 2 = yu0 Ocyco ycn=1 yuycn ycaycOeycn 1 Ocyc=1 yuyc yuaycOeyc To capture economic regime changes, the MS model was applied.
Combining MS with GARCH produced the MS-GARCH model, which was estimated using maximum likelihood estimation (MLE) as in .
This approach allows volatility to shift between regimes, providing a clearer identification of periods that may signal the onset of a currency crisis.
yuayc,yc = yu0,ycyc Ocyco ycn=1 yuycn,ycyc ycaycOeycn Ocyc=1 yuyc,ycyc yuaycOeyc The probability of a crisis at time t was calculated using smoothed probability as in .
cyc = y.
yuycN ) = Ocya yc=1 ycEyc.
cyc 1 = y.
yuycN ) y ycEyc.
cyc = y.
ycyc 1 = yc, yuyc ) .
Forecasting of crisis probability for the period May 2025-April 2026 was performed using .
Ae.
ycEyc.
cyc 1 = y.
yue ycN ) = Ocya yc=1 ycyycyc ycEyc.
cyc = y.
yue ycN ) .
The model is considered accurate if the forecast status and the actual smoothed probability in the out-of-sample period show consistent results.
The crisis threshold was set as the lowest smoothed probability value observed during past crisis periods, representing the probability of transitioning into a crisis regime.
RESULTS AND DISCUSSION
Selection of crisis signal indicators Financial crisis periods in Indonesia were identified using the EMP threshold with a sigma coefficient of 1.
5, revealing crises in August 1997-June 1998.
October 2008, and March 2020 corresponding to the Asian.
Global, and COVID-19 crises.
Macroeconomic indicators were transformed through annual Int J Adv Appl Sci.
Vol.
No.
March 2026: 42-54
Int J Adv Appl Sci
ISSN: 2252-8814
growth rates for seasonal variables, differencing for non-seasonal ones, logarithmic conversion for the lending to deposit ratio, and Hodrick-Prescott filtering for the real exchange rate.
Using these transformations, a signal matrix was constructed, and NSR values for 15 indicators were computed, as shown in Table 2.
The three indicators with the lowest NSR values, real deposit interest rate.
M2 per foreign exchange reserves, and real exchange rate, were identified as the most crisis-sensitive.
Their transformation and threshold comparisons used in NSR calculation are illustrated in Figure 1, where the real deposit interest rate as shown in Figure 1.
M2 per foreign exchange reserves as shown in Figure 1.
, and the real exchange rate as shown in Figure 1.
Table 2.
NSR values Variables Imports Exports Foreign exchange reserves Stock price
The ratio of lending to deposit interest rates Real deposit interest rate
Gap between real BI rate and real Fed rate
Bank deposits Real exchange rate
Trade exchange rates M2 per foreign exchange reserves M2 multiplier Real output
Domestic credit per GDP
NSR
Ranking The three indicators selected in this study reflect patterns of monetary instability in Indonesia during the period 1990-2024.
During the 1997-1998 Asian currency crisis, all three exhibited significant This fluctuation reflects the dramatic changes in currency conditions.
After 2000, market conditions gradually stabilized.
However, in 2020, a new wave of volatility emerged following the COVID-19 pandemic.
The wave resurfaced in early 2024, likely due to a series of geopolitical disruptions.
These episodes highlight the time-varying nature of volatility and underscore the importance of dynamic modeling frameworks such as ARMA, generalized autoregressive moving average with generalized autoregressive conditional heteroscedasticity (GARMACH), and MS-GARMACH for explaining and predicting currency stability.
Data pattern identification and stationarity Figure 2 displays the movement of the real deposit interest rate, the M2 to foreign exchange reserves ratio, and the real exchange rate over time from January 1990-April 2024.
Figure 2 shows that the three indicators fluctuate considerably over time, indicating that their original series are likely non-stationary.
Once the log-return transformation was applied, the ADF test produced p-values of 0.
These results confirm that the real deposit interest rate, as shown in Figure 2.
, and the M2-to-reserves ratio, as shown in Figure 2.
, and the real exchange rate became stationary after transformation as shown in Figure 2.
Granger causality test The causal relationships between the indicators were examined using the Granger causality test.
The results are summarized in Table 3.
Since all p-values exceed =0.
01, no significant causal relationships were thus, each indicator was modeled univariately using ARMA.
ARMA models The best model for the real deposit interest rate indicator, with significant parameters and the lowest AIC value, is ARMA .
, as expressed in .
Subsequently, for the M2 per foreign exchange reserves indicator, the best ARMA model is ARMA .
, as presented in .
Finally, for the real exchange rate indicator, the best ARMA model selected is ARMA .
, as shown in .
Residuals from each best model were tested for normality (KolmogorovAeSmirno.
, autocorrelation (LjungAeBo.
, and heteroskedasticity (Lagrange multiplie.
The first two tests showed p-values >0.
01, indicating normality and no However, the Lagrange multiplier test produced p-values <0.
01 for all indicators, confirming heteroskedasticity and justifying the use of volatility modeling.
ycyc = Oe0.
0046 Oe 0.
9830 ycycOe1 0.
5613 ycaycOe1 Oe 0.
4387 ycaycOe2 ycayc Markov switching and noise-to-signal ratio approach for early detection of currency crises (Sugiyant.
A ISSN: 2252-8814 ycyc = Oe0.
3545 ycycOe1 Oe 0.
1444 ycycOe2 Oe 0.
3930 ycaycOe1 ycayc ycyc = 0.
0097 Oe 0.
3367 ycycOe1 Oe 0.
7791 ycycOe2 0.
5436 ycaycOe1 0.
8638 ycaycOe2 ycayc Figure 1.
Transformation values and threshold plots of .
real deposit interest rate, .
M2 per foreign exchange reserves, and .
real exchange rate Int J Adv Appl Sci.
Vol.
No.
March 2026: 42-54
Int J Adv Appl Sci
ISSN: 2252-8814
Figure 2.
Time series plots of .
real deposit interest rate, .
M2 per foreign exchange reserves, and .
real exchange rate Markov switching and noise-to-signal ratio approach for early detection of currency crises (Sugiyant.
A ISSN: 2252-8814 Table 3.
Granger causality test results Indicator relationship Real deposit interest rateM2 per foreign exchange reserves M2 per foreign exchange reservesreal deposit interest rate Real deposit interest ratereal exchange rate Real exchange ratereal deposit interest rate M2 per foreign exchange reservesreal exchange rate Real exchange rateM2 per foreign exchange reserves P-value ARMACH and GARMACH models Since the models are based on ARMA, the corresponding volatility models are ARMA-GARCH or ARMA-ARCH.
For the real deposit interest rate, the best model addressing heteroskedasticity is ARMA-ARCH .
, with its variance equation shown in .
For the M2 per foreign exchange reserves indicator, the best model addressing heteroskedasticity in the ARMA .
structure is GARCH .
, with its variance equation shown in .
For the real exchange rate indicator, the best model addressing heteroskedasticity in the ARMA .
structure is GARCH .
, with its variance equation shown in .
Diagnostic tests on the best volatility models for all three indicators showed p-values >0.
01 in the KolmogorovAeSmirnov.
LjungAeBox, and Lagrange multiplier tests, confirming that the residuals are normal, uncorrelated, and free from heteroskedasticity, thus validating the models.
yuayc 2 = 0.
504 ycaycOe1 2 yuayc 2 = 0.
4084 ycaycOe1 2 0.
6537 yuaycOe1 2 yuayc 2 = 0.
7261 ycaycOe1 2 0.
4648 yuaycOe1 2 MS-ARMACH and MS-GARMACH models The silhouette test indicated two optimal clusters for each indicator.
Thus, the appropriate models are MS-ARMA-ARCH .
for the real deposit interest rate and MS-GARCH .
,1,.
for both M2 per reserves and real exchange rate, with their variance equations shown in .
0000 ycaycOe1 , ycycycaycyce 1 1379 0.
0001 ycaycOe1 , ycycycaycyce 2 0000 ycaycOe1 0.
5106 yuaycOe1 , ycycycaycyce 1 4381 0.
0001 ycaycOe1 0.
8886 yuaycOe1 , ycycycaycyce 2 0053 ycaycOe1 0.
9883 yuaycOe1 , ycycycaycyce 1 yua3,ycI yc 6903 0.
7299 ycaycOe1 0.
0027 yuaycOe1 , ycycycaycyce 2 yua1,ycI yc yc yua2,ycI yc yc Where yua1,ycI represents the variance equation of the MS-ARMACH .
model for the real deposit interest ycyc rate indicator, yua2,ycI represents the variance equation of the MS-GARMACH .
,1,.
model for the M2 per ycyc foreign exchange reserves indicator, and yua3,ycI represents the variance equation of the MS-GARMACH ycyc .
,1,.
model for the real exchange rate indicator.
State 1 and State 2 correspond to the low-volatility and high-volatility states, respectively.
The transition probability matrices for the real deposit interest rate, the M2 per foreign exchange reserves, and the real exchange rate indicators are presented in matrices ycE1 , ycE2 , and ycE3 , respectively.
The transition matrix results indicate that all three indicators predominantly remain in low-volatility .
o-crisi.
For ycE1 , stability persists with a 0.
9781 probability, while shifts to high volatility are rare .
In ycE2 , the stability probability is 0.
8441 with a 0.
1559 chance of rising volatility, and in ycE3 , calm conditions persist with 0.
while volatility increases occur with 0.
Overall, transitions tend to revert quickly to stable regimes.
ycE1 = ( ) , ycE2 = ( ) , ycE3 = ( Determining crisis boundaries Using the combined volatility and MS model, smoothed probability values were generated to determine crisis thresholds.
Crisis periods were identified from fluctuations in these probabilities, as shown in Figure 3.
The lowest smoothed probability values in Figure 3, which show instability during the financial Int J Adv Appl Sci.
Vol.
No.
March 2026: 42-54
Int J Adv Appl Sci
ISSN: 2252-8814
crisis period in Indonesia, obtained from the EMP calculation, are summarized in Table 4.
Table 4 presents the smoothed probability thresholds for each indicator, where values below the threshold indicate a crisis, and those above indicate a no-crisis state.
Figure 3.
Smoothed probability plots for .
real deposit interest rate, .
M2 per foreign exchange reserves, and .
real exchange rate Markov switching and noise-to-signal ratio approach for early detection of currency crises (Sugiyant.
A ISSN: 2252-8814 Table 4.
Lowest smoothed probability values during crisis periods in Indonesia Indicator Real deposit interest rate M2 per foreign exchange reserves Real exchange rate Period July 1997 February 2008 February 2020 Smoothed probability Accuracy of the combined volatility and Markov-switching model A forecast from May 2024-April 2025 was conducted to evaluate model accuracy by comparing forecasted and actual smoothed probabilities for the three indicators, as shown in Tables 5 -7.
As shown in Tables 5-7, the predicted and actual conditions are fully consistent, showing that the model can successfully identify crisis periods through the three key indicators.
The full alignment from May 2024-April 2025 highlights the robustness of the MS-ARMA-GARCH framework in separating calm phases from turbulent ones.
Table 5.
Forecasted and actual smoothed probability values for the real deposit interest rate Period May 2024 June 2024 July 2024 August 2024 September 2024 October 2024 November 2024 December 2024 January 2025 February 2025 March 2025 April 2025 Forecast Forecast status No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis Actual Actual status No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis Table 6.
Forecasted and actual smoothed probability values for the M2 per foreign exchange reserves Period May 2024 June 2024 July 2024 August 2024 September 2024 October 2024 November 2024 December 2024 January 2025 February 2025 March 2025 April 2025 Forecast Forecast status No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis Actual Actual status No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis Table 7.
Forecasted and actual smoothed probability values for the real exchange rate Period May 2024 June 2024 July 2024 August 2024 September 2024 October 2024 November 2024 December 2024 January 2025 February 2025 March 2025 April 2025 Forecast Forecast status No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis Actual Actual status No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis Early detection of financial crisis in Indonesia The three indicators were used to forecast smoothed probabilities for May 2025-April 2026 as an early warning of potential financial crises, with results shown in Table 8.
Based on these three indicators, the smoothed probability values are lower than their respective thresholds, indicating no crisis in the period from May 2025-April 2026.
Int J Adv Appl Sci.
Vol.
No.
March 2026: 42-54
Int J Adv Appl Sci
ISSN: 2252-8814
Table 8.
Forecasted smoothed probability values for the real deposit interest rate.
M2 per foreign exchange reserves, and real exchange rate (May 2025-April 2.
Period Forecast .
eal deposit interest rat.
May 2025 June 2025 July 2025 August 2025 September 2025 October 2025 November 2025 December 2025 January 2026 February 2026 March 2026 April 2026 Forecast (M2 per foreign exchange reserve.
Forecast .
eal exchange rat.
Status No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis No-crisis DISCUSSION The central bank monitors the movement of a number of selected indicators in real time for application in the MS-ARMA-GARCH hybrid model.
A crisis warning is issued whenever the smoothed probability exceeds a certain threshold for several consecutive months.
The most significant events, when detected within a month, prompt stakeholders to promptly identify their causes to avoid a crisis.
This study also corroborates the findings of Sugiyanto et al.
and Du et al.
and complements them by applying indicator selection through the NRS to the MS-ARMA-GARCH hybrid model.
Future research could use the Currency Crisis Index (CCI) or Market Pressure Index (MPI) to determine crisis periods.
The results of this study provide a practical framework and focus on Indonesia, so this methodology can be extended to other developing countries facing similar volatility patterns.
CONCLUSION
The stages in this research are always based on data characteristics and the selection of models that match these characteristics, resulting in a very good EWS.
For example, the determination of past crisis periods is carried out using the EMP, and the determination of future crises using smoothed probabilities.
Data fluctuations and regime shifts are analyzed using a hybrid MS-GARMACH.
The smoothed probabilities indicate no signs of crisis risk during the period from May 2025-April 2026.
Although this research framework focuses on Indonesia, the methodology can be extended to other developing countries facing similar volatility patterns.
ACKNOWLEDGEMENTS
We would like to express our gratitude to Universitas Sebelas Maret for providing funds to carry out the research.
FUNDING INFORMATION
The RKAT of Universitas Sebelas Maret funds this research for the 2025 Fiscal Year through the Research Scheme Strengthening the Capacity of the Research Group (PKGR-UNS) B with Research Assignment Agreement Number: 371/UN27.
22/PT.
03/2025.
AUTHOR CONTRIBUTIONS STATEMENT
This journal uses the Contributor Roles Taxonomy (CRediT) to recognize individual author contributions, reduce authorship disputes, and facilitate collaboration.
Name of Author Sugiyanto Muhammad Bayu Nirwana Isnandar Slamet Etik Zukhronah SyifaAo Salsabila Gita Parahita ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue ue Markov switching and noise-to-signal ratio approach for early detection of currency crises (Sugiyant.
A C : Conceptualization M : Methodology So : Software Va : Validation Fo : Formal analysis ISSN: 2252-8814 I : Investigation R : Resources D : Data Curation O : Writing - Original Draft E : Writing - Review & Editing Vi : Visualization Su : Supervision P : Project administration Fu : Funding acquisition CONFLICT OF INTEREST STATEMENT The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
INFORMED CONSENT
Not applicable.
This study does not involve human participants.
ETHICAL APPROVAL
The research did not involve human or animal subjects.
therefore, ethical approval was not required.
DATA AVAILABILITY
All data used in this study are publicly available from the IMF at https://data.
No new data were generated.
REFERENCES