INTERNATIONAL JOURNAL OF COMPUTING SCIENCE AND APPLIED MATHEMATICS.
VOL.
NO.
AUGUST 2024
VAR Model Estimation And Application Of IRF And FEVD On Currency Exchange Rates.
COVID-19 Cases.
And WHO Twitter Information In Southeast Asia Matthew Axel Darmawan.
Helena Margaretha.
Ferry Vincenttius Ferdinand, and Yohan Chandrasukmana AbstractAiThis paper examines the impact of the COVID-19 pandemic and World Health Organization (WHO) information dissemination through Twitter on the exchange rates of Southeast Asian countries.
The study utilizes a VAR model for analysis, incorporating daily positive cases and the percentage of tweets with positive sentiment as proxies for the pandemic and WHO information, respectively.
The VAR models are employed for forecasting and estimating impulse response functions (IRF) and forecast error variance decomposition (FEVD).
The forecasting performance is evaluated using mean absolute error (MAE), root-mean-square error (RMSE), and R2 metrics, revealing that only Cambodia possesses a reliable forecasting model.
The IRF analysis demonstrates varying effects of the pandemic and WHO information across different countries, while the FEVD results indicate distinct contributions of the pandemic and WHO information in each Southeast Asian country.
Additionally, the FEVD analysis reveals that exchange rates are mostly influenced by their own past behavior.
Overall, this study provides insights into the economic impact of the COVID-19 pandemic and WHO information on exchange rates in Southeast Asia.
This study aims to analyze the effect of the COVID-19 pandemic and WHO information on the economic conditions of countries in Southeast Asia using daily positive cases and WHO information data processed into a percentage of tweets with positive sentiment.
This study will use multivariate time series analysis to estimate parameters and forecast exchange rate returns.
Impulse response functions (IRF) and forecast error variance decomposition (FEVD) will be used to analyze the impacts experienced by exchange rates.
The currencies of Southeast Asian countries used in this study are those of Brunei Darussalam (Brunei Dollar / BND).
Philippines (Peso / PHP).
Indonesia (Rupiah / IDR).
Cambodia (Riel / KHR).
Laos (Kip / La.
Malaysia (Ringgit / MYR).
Singapore (Singapore Dollar / SGD).
Thailand (Baht / THB), and Vietnam (Dong / VND), with the USD as the reference currency.
II.
METHODS
Keywords: Exchange Rates.
VAR.
Forecasting.
IRF.
FEVD.
INTRODUCTION
N March 11, 2020.
Dr.
Tedros Adhanom Ghebreyesus as Director-General of WHO determined the spread of the COVID-19 virus as a global pandemic because the spread of COVID-19 continues to occur throughout the world .
The World Bank stated that the COVID-19 pandemic caused the global economy to experience its worst recession since the Second World War .
The dissemination of information during the pandemic is crucial, and the World Health Organization (WHO) has been responsible for distributing information about COVID-19 on various types of platforms such as official websites, news, social media, conferences, and The information received by countries has played an important role in shaping policies to combat the pandemic, which can have direct and indirect impacts on a countryAos economy, reflected by its exchange rate.
The foreign exchange market is one of the largest financial markets in the world and is sensitive to unexpected events like the COVID-19 pandemic, making it a useful tool to study the pandemicAos effects on Southeast Asian countries.
Darmawan.
Margaretha.
Ferdinand, and Y.
Chandrasukmana are with the Department of Mathematics.
Universitas Pelita Harapan Indonesia e-mail: ferry.
vincenttius@uph.
Manuscript received July 18, 2023.
accepted September 14, 2023.
In this study, three types of data were used to estimate VAR models for Southeast Asian countries such as exchange rates.
WHO Twitter accountAos positive tweet percentage, and incidence rate of new COVID-19 cases.
Daily time series datasets from April 2020 to October 2021 were used for this This study aimed to analyze the impacts experienced by exchange rates with respect to WHO positive tweet percentage and incidence rate.
The overall methodology of this study to obtain the results is illustrated in Fig.
All processes are done using RStudio and the vars package in R for VAR modeling, forecasting, and calculating IRF and FEVD values .
Data Collection IRF and FEVD Data Transformation Model Diagnostics and Forecasting Stationarity Test VAR Model Estimation Fig.
1: Research methodology flowchart.
INTERNATIONAL JOURNAL OF COMPUTING SCIENCE AND APPLIED MATHEMATICS.
VOL.
NO.
AUGUST 2024
Different types of VAR models were estimated, classified by its variable determination, that is only constant as its determination variables, only trend as its determination variables, both variables, and none of those variables.
The mentioned VAR modelsAo equations are defined as:
Const zt = I1 ztOe1 I2 ztOe2 .
I p ztOep I0 at .
Both zt = I1 ztOe1 .
I p ztOep I0 IT at .
Trend zt = I1 ztOe1 I2 ztOe2 .
I p ztOep IT at .
None zt = I1 ztOe1 I2 ztOe2 .
I p ztOep at .
The stationary condition of the datasets was checked with the Augmented Dickey-Fuller test, and data transformations were applied to fulfill the stationarity assumption.
Four types of VAR models were estimated for each country based on optimal lag selection using AkaikeAos Information Criterion (AIC).
Bayesian Information Criterion (BIC) or Schwarz Criterion (SC).
Hannan and Quinn (HQ), and Final Prediction Error (FPE).
The chosen model was used for forecasting and estimating impulse response functions (IFR) and forecast error variance decomposition (FEVD).
The accuracy of the forecasting results was measured with mean absolute error (MAE), root-mean-square error (RMSE), and R-squared (R2 ).
The goal of estimating IRF and FEVD was to analyze the impacts of exchange rates in relation to COVID-19 incidence rate and WHO positive tweet percentage.
RESULT
This study has conducted several steps, including testing for stationarity.
VAR model estimation.
VAR model diagnosis, forecasting, and lastly IRF and FEVD estimation.
The stationary condition is important to make statistical inferences about the structure of a stochastic process.
We see that all datasets used in VAR model estimation have fulfilled the stationary condition, shown in Table I below.
It has to be confirmed that the data used for VAR model estimation have to reach stationarity concerning that stationarity is needed to make statistical inferences about the structure of a stochastic process on the basis of an observed record of that process .
model selection.
Four criterions mentioned (AIC.
BIC / SC.
HQ.
FPE) are used to determine the best lag for each type of VAR model.
The chosen lag is the one with the highest vote and the highest value if there are ties.
The results of these steps are shown in Tables II, i.
IV, and V.
TABLE II: Optimal Lag For The Type Const.
Country Brunei Darussalam Philippines Indonesia Cambodia Laos Malaysia Singapore Thailand Vietnam AIC.
HQ.
SC.
FPE.
TABLE i: Optimal Lag For The Type Both.
Country Brunei Darussalam Philippines Indonesia Cambodia Laos Malaysia Singapore Thailand Vietnam AIC.
HQ.
SC.
FPE.
TABLE IV: Optimal Lag For The Type Trend.
Country Brunei Darussalam Philippines Indonesia Cambodia Laos Malaysia Singapore Thailand Vietnam AIC.
HQ.
SC.
FPE.
TABLE I: Stationarity Test.
Dataset Positive Tweet Perc.
Laos IR
USD/BND
USD/MYR
Brunei IR
Malaysia IR
USD/PHP
USD/SGD
USD/IDR
USD/THB
Indonesia
Thailand IR
USD/KHR
USD/VND
Cambodia
Vietnam IR
USD/LAK
P-Value Stationarity < 0, 01
Stationary < 0, 01
< 0, 01
< 0, 01
< 0, 01
< 0, 01
< 0, 01
< 0, 01
< 0, 01
< 0, 01
Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary < 0, 01
Stationary < 0, 01
< 0, 01
< 0, 01
Stationary Stationary Stationary < 0, 01
Stationary < 0, 01
< 0, 01
Stationary Stationary The VAR model estimation consists of three steps, namely optimal lag selection, model parameter estimation, and optimal TABLE V: Optimal Lag For The Type None.
Country Brunei Darussalam Philippines Indonesia Cambodia Laos Malaysia Singapore Thailand Vietnam AIC.
HQ.
SC.
FPE.
After determining the optimal lag for each type of VAR model, the next step is to estimate the parameters.
This involves estimating the parameters for all observed countries in Southeast Asia, and producing a log-likelihood score that measures the goodness of fit of each estimated model.
Using the log-likelihood score for every type of VAR model, the optimal model will be selected based on the highest score.
Table VI highlights the models selected for each observed country in Southeast Asia.
INTERNATIONAL JOURNAL OF COMPUTING SCIENCE AND APPLIED MATHEMATICS.
VOL.
NO.
AUGUST 2024
TABLE VI: Log-Likelihood Score and Optimal Model.
Country Brunei Darussalam Philippines Indonesia Cambodia Thailand Malaysia Singapura Laos Vietnam Optimal Model Const Both Trend None Both Both Both Both Both Both Both Both Trend The selected models will undergo diagnosis using residual analysis, which consists of four steps.
The first step, called serial correlation test or Portmanteau test, checks for serial correlation in the residuals.
The second step is the autoregressive conditional heteroscedasticity test or ARCH-LM, which checks for multivariate ARCH effects among the residuals.
The third step is the residual normality test or Jarque-Bera test, which checks if the residuals are normally distributed.
Lastly, the structural stability test or OLS-CUSUM tests if there are any structural changes within the estimated models.
Results from each test are provided in Table VII.
After completing the model diagnosis, the estimated models are ready for use in forecasting and estimating IRF and FEVD.
Even though the results obtained are not as expected, forecasting will still be done considering that significance test is not necessarily needed for business forecasting .
This study applies a method of forecasting which writer called AyN-RollAy because of mean-reverting behavior that a stationary VAR models have .
The results of this method are evaluated using MAE.
RMSE, and R2 .
The Table Vi shows a comparison between mean, standard deviation.
MAE.
RMSE, and R2 , and it is concluded that only CambodiaAos VAR model provides good forecasts.
The impulse response functions (IRF) illustrate the effect of a shock on one variable on the movement of another variable, while the forecast error variance decomposition (FEVD) shows the contribution of variables influenced by themselves and other variables.
In this study, the exchange rate returns are used as response variables and both positive tweet percentage and respective incidence rate are used as the impulse variables to analyze the impact of COVID-19 and WHO on the economic condition of each country in Southeast Asia.
The results of IRF and FEVD are presented in Table IX and Table X below, showing that all the exchange rates in the observed Southeast Asian countries were mostly affected by their own past performance.
INTERNATIONAL JOURNAL OF COMPUTING SCIENCE AND APPLIED MATHEMATICS.
VOL.
NO.
AUGUST 2024
TABLE VII: Test Results.
Country Brunei Darussalam
Philippines Indonesia
Cambodia
Laos Malaysia
Singapore Thailand
Vietnam
Portmanteau
Serial
P-Value Correlation Multivariate ARCH-LM
ARCH
P-Value Effect Present 055 y 10Oe4 Absent Absent Present Absent Absent Present Absent Present 896 y 10Oe2 < 2.
2 y 10Oe16 696 y 10Oe6 154 y 10Oe4 039 y 10Oe13 843 y 10Oe5 744 y 10Oe5 414 y 10Oe9 Jarque-Bera Residuals P-Value Normality Present < 2.
2 y 10Oe16 Absent Absent Present Present Present Present Present Present Present < 2.
2 y 10Oe16 Absent Absent Absent Absent Absent Absent Absent Absent < 2.
2 y 10Oe16 < 2.
2 y 10Oe16 < 2.
2 y 10Oe16 < 2.
2 y 10Oe16 < 2.
2 y 10Oe16 < 2.
2 y 10Oe16 < 2.
2 y 10Oe16 TABLE Vi: Comparison Between Mean.
Standard Deviation.
MAE.
RMSE, and R2 .
Country Brunei Darussalam
Philippines Indonesia
Cambodia
Laos Malaysia
Singapore Thailand
Vietnam
Mean
StdDev
MAE
RMSE
INTERNATIONAL JOURNAL OF COMPUTING SCIENCE AND APPLIED MATHEMATICS.
VOL.
NO.
AUGUST 2024
TABLE IX: IRF.
Positive Tweet Percenatge Incidence Rate Brunei Darussalam Indonesia Laos Singapore INTERNATIONAL JOURNAL OF COMPUTING SCIENCE AND APPLIED MATHEMATICS.
VOL.
NO.
AUGUST 2024
Vietnam Philippines Cambodia Malaysia Thailand INTERNATIONAL JOURNAL OF COMPUTING SCIENCE AND APPLIED MATHEMATICS.
VOL.
NO.
AUGUST 2024
TABLE X: FEVD Return.
Period (Day.
Period (Day.
Period (Day.
USD/BND (Brunei Darussala.
Return PosTweet Inc.
Rate USD/IDR (Indonesi.
PosTweet Inc.
Rate USD/VND (Vietna.
Return PosTweet Inc.
Rate Return USD/PHP (Philippine.
Return PosTweet Inc.
Rate Return USD/KHR (Cambodi.
PosTweet Inc.
Rate USD/LAK (Lao.
PosTweet Inc.
Rate Return USD/SGD (Singapor.
PosTweet Inc.
Rate Return INTERNATIONAL JOURNAL OF COMPUTING SCIENCE AND APPLIED MATHEMATICS.
VOL.
NO.
AUGUST 2024
IV.
CONCLUSIONS
This study utilized an autoregressive vector (VAR) model to analyze the effect of the COVID-19 pandemic and WHO information through Twitter on the exchange rates of Southeast Asian countries.
The VAR models were estimated and used for both forecasting and estimating impulse response functions (IRF) and forecast error variance decomposition (FEVD).
The forecasting process was evaluated using mean absolute error (MAE), root-mean-square error (RMSE), and R2 , and it was found that only Cambodia had a reliable model for the forecasting process.
The IRF analysis showed that the effects of the COVID-19 pandemic and WHO information were different for each country, while the FEVD results showed that the COVID19 pandemic and WHO information had a different proportion of contributions in each Southeast Asian country.
The FEVD results also showed that the exchange rate of a currency was most selfinfluenced in the past.
APPENDIX
Appendixes should appear before the acknowledgment.
ACKNOWLEDGMENT