The findings show that the best statistical model for the pandemic period was different from the model that best described the post-pandemic period. The result suggests that forecasting models for stock indices may need to be reassessed when market conditions undergo major disruptions such as a global health crisis.
Why BISNIS-27 Matters
BISNIS-27 is an Indonesia Stock Exchange index consisting of 27 selected blue-chip companies. The constituents are jointly selected by Bisnis Indonesia and the Indonesia Stock Exchange based on factors including fundamental performance, corporate governance and trading liquidity.
The index provides an important view of how selected large and relatively liquid companies respond to changing economic conditions. The COVID-19 crisis created an unusual environment for financial markets as restrictions, declining economic activity, supply disruptions and uncertainty affected companies and investor behavior.
Indonesia's capital market experienced substantial volatility during the early stages of the pandemic. For an index such as BISNIS-27, understanding how historical price movements changed during and after the crisis can help researchers evaluate whether forecasting models remain appropriate when market conditions shift.
1,449 Daily Observations Analyzed
Untoro and Sumarto conducted the research between May and July 2025 using 1,449 daily closing-price observations for BISNIS-27 obtained from Investing.com.
The data were divided into two periods:
- Pandemic period: January 2, 2020 to December 30, 2022, covering 735 trading days.
- Post-pandemic period: January 2, 2023 to December 30, 2025, covering 714 trading days.
The researchers used the Box-Jenkins approach and the Autoregressive Integrated Moving Average (ARIMA) method. In simple terms, ARIMA examines how current values are related to previous observations and past shocks in a time series.
The analysis included descriptive statistics, the Augmented Dickey-Fuller test to determine whether the data were stable enough for modeling, correlation analysis to identify relevant time lags, model estimation and residual diagnostics. The researchers conducted the analysis using EViews 12.
BISNIS-27 Reached a Higher Average After the Pandemic
The descriptive results show a noticeable difference between the two periods.
During 2020–2022, the average BISNIS-27 index level was 509.05, with a minimum of 329.29 and a maximum of 617.97. Its standard deviation was 57.96.
During 2023–2025, the average increased to 560.03, while the minimum and maximum were 431.37 and 635.09, respectively. Standard deviation declined to 43.18.
The combination of a higher average index level and lower standard deviation indicates that BISNIS-27 moved at a higher level with less dispersion after the acute pandemic period.
The statistical tests also showed that the index was non-stationary in both periods when analyzed in its original form. After the researchers applied first differencing, the series became stationary. Both periods therefore required a differencing order of one, or d = 1, in the ARIMA framework.
Different ARIMA Models Emerged
The most important finding was the difference between the best-fitting models for the two periods.
For the 2020–2022 pandemic period, the selected model was:
ARIMA([3,9],1,[15])
The model included autoregressive components at lags 3 and 9 and a moving-average component at lag 15. It produced an AIC value of 6.6875 and an SIC value of 6.7188. Its residuals passed the white-noise diagnostic, indicating that the model adequately captured the remaining time-dependent structure in the data.
For the 2023–2025 post-pandemic period, the selected model was:
ARIMA([2],1,[35])
This model contained an autoregressive component at lag 2 and a moving-average component at lag 35. Its AIC was 6.3414 and its SIC was 6.3671, both lower than those of the selected pandemic-period model. The residual diagnostics also indicated white-noise behavior.
The comparison reveals that the pandemic-period model required a more complex autoregressive structure, while the post-pandemic model had a simpler autoregressive component but a substantially longer moving-average component.
What the Model Differences Mean
The researchers interpret the difference between the two models as evidence that the pandemic altered the temporal structure of BISNIS-27.
During the pandemic, the index showed significant relationships with observations from several earlier time points, represented by the AR components at lags 3 and 9. After the pandemic period, the autoregressive structure became simpler, with the main autoregressive relationship occurring at lag 2.
At the same time, the moving-average component extended from lag 15 during the pandemic to lag 35 afterward. This suggests that the statistical pattern of how past shocks were reflected in subsequent index movements also changed.
Importantly, the research does not mean that the ARIMA model can predict stock prices with certainty. Instead, it demonstrates that different market periods can require different model specifications.
Implications for Financial Analysis
The findings have practical implications for analysts who use historical stock-price data to develop forecasting models.
A model calibrated during an exceptional crisis may not remain appropriate once market conditions change. Conversely, a model developed during relatively stable conditions may not adequately represent the behavior of a market during another major disruption.
For BISNIS-27, the authors therefore emphasize the need to periodically reassess ARIMA specifications rather than treating one model as permanently applicable.
The researchers also recommend future comparisons involving ARIMA-GARCH, ARIMAX and nonlinear machine-learning approaches such as LSTM. These methods could incorporate volatility, external economic variables or nonlinear relationships that a basic univariate ARIMA model does not capture.
The current research is limited because it uses only daily closing prices and does not include external variables such as interest rates, inflation, exchange rates or other macroeconomic indicators. The authors also note that linear ARIMA may not fully capture the conditional volatility commonly found in financial markets.
Researchers' Perspective
Untoro and Sumarto of Universitas Mercu Buana conclude that the COVID-19 period and the subsequent recovery produced different statistical patterns in the BISNIS-27 Index. Their analysis indicates that the same forecasting specification should not automatically be applied across substantially different market regimes.
For financial analysts and researchers, the key message is methodological: forecasting models need to be evaluated against the characteristics of the period being analyzed. Major economic disruptions can change the relationships embedded in historical market data.
Author Profile
Hamizan Azh-Zhafif Untoro is affiliated with the Study Program of Management, Faculty of Economics and Business, Universitas Mercu Buana. His research in this article focuses on financial market analysis, stock-index movements and time-series forecasting using ARIMA.
Agus Herta Sumarto is also affiliated with the Study Program of Management, Faculty of Economics and Business, Universitas Mercu Buana. The research focuses on the statistical analysis of BISNIS-27 and the changing dynamics of Indonesia's capital market before and after the COVID-19 pandemic.
The article was published in 2026 in the International Journal of Sustainable Applied Sciences (IJSAS).
Source
Article: “Analysis of the BISNIS-27 Index Movement Using the Autoregressive Integrated Moving Average (ARIMA) Method During and After the COVID-19 Pandemic”
Journal: International Journal of Sustainable Applied Sciences (IJSAS)
Volume: 4, No. 9, 2026, pp. 1217-1228
Authors: Hamizan Azh-Zhafif Untoro and Agus Herta Sumarto
Affiliation: Study Program of Management, Faculty of Economics and Business, Universitas Mercu Buana
DOI: https://doi.org/10.59890/ijsas.v4i9.68
Journal: http://ijsasjournal.my.id/index.php/ijsas
0 Komentar