University researchers in the UAE have found that an AI tool – known as the nonlinear autoregressive with external input neural network (NARX) – can potentially predict movements of the Dubai Financial Market Sharia Index (DFMSI).
Given the study’s finding, “regulators should consider establishing guidelines for the use of AI and machine learning models in financial forecasting,” wrote lead researcher Zakaria Boulanouar, an assistant professor at the Department of Mathematics and Statistics of Abu Dhabi University.
Zakaria explained the study was conducted as Shariah compliant indices exhibited resilience compared to conventional indices during the COVID-19 market downturn.
“Further, geopolitical events have a significant but short-lived impact on Shariah compliant indices, highlighting their relative stability compared to conventional indices during such periods.”
Together with UAE colleagues Ghassane Benrhmach, Rihab Grassa, Sonia Abdennadher and Mariam Aldhaheri, this study was conducted using five predictive models of:
- Linear regression;
- Least-square regression (LSR) kernel;
- Fine tree;
- Back propagation neural network (BPNN); and
- NARX.
These machine learning techniques, or artificial neural network models, are still relatively new and “offer the potential to capture complex, non-linear relationships that traditional models might miss” – thereby improving prediction accuracy and aiding stakeholders in making more effective decisions.
Specifically, the study aimed to check for correlations and information flow between a local Shariah compliant index and global Islamic indices – to potentially anticipate the direction in which other stock exchanges will move following the decline or rise of a particular market.
Reference data came from daily closing prices covering the period from the 27th October 2019 until the 3rd January 2024, with a base date for market capitalization on the 31st December 2009 for the Dow Jones Islamic Market World Index, the S&P Global Sharia Index and the FTSE Sharia All World Index together with the DFMSI.
“The study sample is limited by the availability of data since DFMSI was launched only on the 27th October 2019, which determines the start date of the data sample.”
The researchers used the initial 70% of the data – from the 27th October 2019 to the 29th September 2022 – to train the five models. The remaining 30% – from the 30th September 2022 to the 3rd January 2024 – became the basis for testing the performance of the five models.
“The choice of 30% as a testing set was to ensure a robust evaluation of the model’s performance,” where the coefficient of determination (R2) indicates how well the observed data fit the models tested – with a higher R2 indicating a better fit.
The tests quickly eliminated the linear regression model, which didn’t align with data during the COVID-19 pandemic period. As for the other models, the R2 coefficients were:
- LSR : 0.8172
- Fine tree : 0.8391
- BPNN : 0.8661
- NARX : 0.9905
The NARX model’s consistency in demonstrating minimal error – indicating its robustness in various market conditions to capture complex, non-linear relationships within financial data – “is especially crucial during periods of heightened market volatility such as the COVID-19 pandemic”.
The researchers concluded the study shows that the right AI tool can capture the complexities and irrationalities inherent in financial markets – enabling easier prediction of market movements and informed decision-making by investors and policymakers.
“It also underscores the relevance of financial contagion theory, highlighting how financial shocks and information flow between interconnected global markets can be effectively analyzed using these models.
“Insights from behavioral finance are reinforced, showing that psychological factors influencing market behavior can be better incorporated into predictive models through machine learning techniques.”
Despite the upbeat prognosis, the researchers also cautioned that “research specifically examining the performance of AI-based forecasting models during the Russian-Ukraine war is limited. Future studies could explore how AI techniques can capture the impact of geopolitical events on Islamic financial markets and identify profitable trading and investment strategies.”
