In a study published in a Nature Portfolio journal, researchers have demonstrated that artificial neural networks—a form of advanced artificial intelligence—can predict the movements of a local Islamic stock index with near-perfect accuracy by learning from global Sharia-compliant markets. The findings offer investors a powerful new tool for risk management and portfolio optimization.
Imagine having a crystal ball that could tell you, with 99% accuracy, whether your investments will rise or fall tomorrow. For investors in Islamic finance, that crystal ball may now be within reach—not through magic, but through mathematics.
A ew study, titled “Exploring the predictive power of artificial neural networks in linking global Islamic indices with a local Islamic index,” has demonstrated that a specific type of artificial intelligence called a Nonlinear Autoregressive with External Input Neural Network (NARX) can predict the Dubai Financial Market Sharia Index (DFMSI) with an astonishing coefficient of determination (R²) of 0.9905.
In plain English, that means the model explains over 99% of the variability in the index’s price movements. This is not a small improvement. It is a leap.
Why This Matters: The Growth of Islamic Finance
Islamic finance is one of the fastest-growing sectors in global finance, with assets estimated at over $3 trillion worldwide. Sharia-compliant (SC) indices track companies that adhere to Islamic law, which prohibits interest-based transactions (riba), excessive uncertainty (gharar), and investments in industries such as alcohol, gambling, pork, and conventional financial services.
These indices serve a vital purpose. They allow observant Muslim investors to participate in financial markets without compromising their religious principles. Major global index providers—Dow Jones, Standard & Poor’s, and FTSE Russell—all maintain Islamic indices. And many national markets, including Dubai, have their own local Sharia indices.
But predicting how these indices move has always been a challenge. Financial markets are notoriously complex, driven by a chaotic interplay of macroeconomic variables, investor sentiment, geopolitical events, and global shocks. The COVID-19 pandemic, for example, sent markets around the world into unprecedented volatility.
The new study addresses this challenge head-on. The researchers asked a simple but powerful question: Can global Islamic indices predict a local Islamic index? And if so, which predictive model works best?
The Data: Four Indices, Four Years
The study analyzed daily closing prices of four Sharia-compliant indices over a period from October 27, 2019, to January 3, 2024—a span that included the entire COVID-19 pandemic and its aftermath.
The dependent variable (the one being predicted) was the Dubai Financial Market Sharia Index (DFMSI) , which tracks Sharia-compliant companies listed on the Dubai Financial Market.
The three independent variables (the predictors) were global Islamic indices:
- Dow Jones Islamic Market World Index (DJIMWI) : Based in the United States, maintained by S&P Dow Jones Indices.
- Standard & Poor’s Global Sharia Index (SPGSI) : Also U.S.-based, part of the S&P Global Shariah Index Series.
- Financial Times Stock Exchange Sharia All World Index (FTSESAWI) : Based in the United Kingdom, created and managed by FTSE Russell.
The researchers chose these three global indices to diversify data sources, capture global market insights, and enhance the robustness of their findings. Rather than relying on a single international benchmark, they used three, providing a more comprehensive picture of the factors influencing the Dubai market.
The Models: From Simple to Sophisticated
The research team tested five predictive models, ranging from traditional statistical methods to advanced artificial intelligence. Each model was trained on the first 70% of the data (from October 2019 to September 2022) and tested on the remaining 30% (from September 2022 to January 2024).
The models were:
- Linear Regression: A classic statistical method that assumes a straight-line relationship between variables.
- Least Squares Regression (LSR) Kernel: A more flexible model that uses kernel functions to capture non-linear relationships.
- Fine Tree (Decision Tree): A machine learning model that makes predictions by learning simple decision rules from the data.
- Back Propagation Neural Network (BPNN): An artificial neural network that learns by adjusting its internal weights through repeated iterations.
- Nonlinear Autoregressive with External Input Neural Network (NARX): A specialized recurrent neural network designed for time series prediction, particularly effective with noisy, volatile data like financial markets.
The researchers evaluated each model using three standard metrics:
- R-squared (R²): The proportion of variance in the dependent variable explained by the model. Higher is better (maximum 1.0).
- Mean Absolute Error (MAE): The average magnitude of prediction errors. Lower is better.
- Mean Squared Error (MSE): Another measure of prediction error that penalizes larger errors more heavily. Lower is better.
The Results: A Clear Winner Emerges
The results were striking. While all models showed some predictive power, the NARX model dramatically outperformed every other approach.
Table 1: Comparative Performance of Predictive Models
| Model | R-squared (R²) | Mean Squared Error (MSE) | Mean Absolute Error (MAE) |
|---|---|---|---|
| Linear Regression | 0.7226 | Not reported | Not reported |
| LSR Kernel | 0.8172 | Not reported | Not reported |
| Fine Tree (Decision Tree) | 0.8391 | Not reported | Not reported |
| Back Propagation Neural Network (BPNN) | 0.8661 | Not reported | Not reported |
| Nonlinear Autoregressive Neural Network (NARX) | 0.9905 | 0.000388 | 0.0226 |
Source: Boulanouar et al. (2024), Humanities and Social Sciences Communications
The linear regression model, while useful, showed significant limitations. During the volatile COVID-19 pandemic period (approximately time steps 300 to 350 in the dataset), it failed to accurately predict the DFMSI, with several points deviating significantly from actual values.
The LSR Kernel and Fine Tree models showed improvement, with R² values of 0.8172 and 0.8391 respectively. These are respectable figures, indicating that the global indices do have a meaningful relationship with the Dubai index.
The BPNN model pushed the R² to 0.8661, demonstrating the power of neural networks to capture complex, non-linear relationships that traditional models miss.
But the NARX model was in a league of its own. With an R² of 0.9905, it explained over 99% of the variance in the DFMSI. The MSE was an astonishingly low 0.000388, and the MAE was just 0.0226. In practical terms, the model’s predictions were almost indistinguishable from the actual data.
Visual Proof: Seeing the Accuracy
The paper includes compelling visual evidence of the NARX model’s performance. Figure 4 (referenced in the paper) shows the original time series of the DFMSI (blue dots) alongside the NARX model’s predictions (yellow dots). The two lines are nearly superimposed, even during the chaotic COVID-19 period.
Figure 5 presents an error analysis, plotting the original data against the predicted data and the error between them. The NARX model consistently demonstrates minimal error, indicating its robustness across various market conditions—bull markets, bear markets, and the unprecedented volatility of a global pandemic.
As the authors write: *”The NARX model significantly outperformed other models, with an R2 of 0.9905 and MSE of only 0.000388, indicating its superior predictive accuracy for DFMSI. This finding underscores the potential of ANNs in capturing complex, non-linear relationships within financial data, which is especially crucial during periods of heightened market volatility such as the COVID-19 pandemic.”*
Why the NARX Model Works So Well
Financial time series data are notoriously difficult to predict. They are noisy (random fluctuations), non-linear (relationships change over time), and volatile (sudden large movements). Traditional models like linear regression assume that relationships are stable and linear—assumptions that are almost always violated in real markets.
The NARX model is specifically designed for such challenges. It is a recurrent neural network that uses previous values of the time series (autoregressive component) along with external input variables (the three global indices) to make predictions. The network is first trained in an “open loop” configuration, where actual target values are used to optimize the training process. Once trained, it transitions to a “closed loop” configuration, where its own predictions become inputs for future predictions.
This architecture allows the NARX model to learn the underlying dynamics of the system—the hidden patterns and relationships—rather than simply fitting a line to the data.
The paper’s methodology section explains: “The NARX network was trained to predict the DFMSI time series. The network is initially constructed and trained in an open loop manner, where the target variables are used as responses to optimize the training process and achieve high-quality results. Once the training is complete, the network transitions into a closed loop configuration, where predicted values are employed as inputs to continuously provide new responses, ensuring ongoing and dynamic functionality.”
Implications for Investors and Portfolio Managers
The practical implications of this research are substantial.
For individual investors: The ability to accurately predict a local Islamic index using global indices means better timing of entries and exits, reduced risk, and improved returns. An investor who knows, with 99% confidence, where the market is heading can make decisions with far greater confidence.
For portfolio managers: The findings offer a powerful tool for risk management. By incorporating predictions from the NARX model, managers can adjust their portfolios in anticipation of market movements, hedge effectively, and optimize asset allocation.
For Islamic financial institutions: The model can be integrated into trading algorithms, robo-advisors, and risk management systems. As the authors note, “The integration of advanced machine learning models can enhance decision-making processes and improve risk management strategies.”
For policymakers: Regulators and central banks can use such models to monitor market integration, detect emerging risks, and formulate policies that ensure market stability and resilience.
Table 2: Benefits of Using Global Islamic Indices for Local Prediction
| Benefit | Explanation | Practical Application |
|---|---|---|
| Diversification of Data Sources | Using three global indices instead of one expands the information base. | More robust predictions, less reliance on a single dataset. |
| Global Market Insights | Captures trends and factors relevant to SC investments worldwide. | Identifies interdependencies and correlations between markets. |
| Risk Management | Enables identification and analysis of risks across different markets. | Assess potential impact of global events on local index. |
| Performance Evaluation | Allows comparison of local SC index performance against global benchmarks. | Better understanding of relative strengths and weaknesses. |
| Enhanced Decision-Making | Provides accurate predictions for timely investment actions. | Improved entry/exit timing, hedging, and portfolio allocation. |
| Source: Adapted from Boulanouar et al. (2024) |
Theoretical Contributions: Beyond Practical Applications
The study also makes important theoretical contributions.
First, it supports the Efficient Market Hypothesis (EMH) in a nuanced way. EMH posits that asset prices reflect all available information. The study shows that market behavior can indeed be understood and predicted using advanced models that capture the complexities and irrationalities inherent in financial markets. The EMH is not rejected; it is refined.
Second, it underscores the relevance of financial contagion theory—the idea that financial shocks can spread from one market to another. By demonstrating strong linkages between global and local Islamic indices, the study provides empirical evidence for information flow across Sharia-compliant markets. During the COVID-19 pandemic, for example, shocks to global Islamic indices were rapidly transmitted to the Dubai market.
Third, it incorporates insights from behavioral finance, which recognizes that psychological factors influence market behavior. The NARX model, by learning patterns from actual data, implicitly captures these behavioral elements without needing to model them explicitly.
Limitations and Future Research
The authors honestly acknowledge the limitations of their study. The sample period, while including the COVID-19 pandemic, is limited to approximately four years, constrained by the launch date of the DFMSI (October 27, 2019). Future research should test the model over longer time horizons and across different market conditions.
The study also focuses on a single local index (Dubai). Future research could extend the analysis to other national Islamic indices in Saudi Arabia, Malaysia, Pakistan, Bahrain, and Turkey, examining whether the NARX model’s predictive power generalizes.
Additionally, the study does not incorporate other potential predictors such as oil prices (critical for Gulf economies), geopolitical events (e.g., regional conflicts), or macroeconomic variables (interest rates, inflation). Future models could integrate these factors for even greater accuracy.
The authors also note that while ANNs are promising, they require further validation across different contexts: “Future studies should aim to refine these models and test their robustness in various market conditions.”
Conclusion: A New Era for Islamic Finance
The integration of artificial intelligence into finance is not a distant future—it is the present. This study demonstrates that advanced neural networks can predict Islamic stock indices with near-perfect accuracy, offering investors a powerful tool for navigating increasingly complex and volatile markets.
For the observant Muslim investor, this is particularly significant. Sharia-compliant investing already involves navigating additional layers of screening and compliance. Having an accurate predictive model reduces uncertainty, lowers risk, and allows for more confident decision-making—all while remaining fully compliant with Islamic principles.
As the authors conclude: “The implications of this study are profound for both practical and theoretical perspectives. For investors and portfolio managers, the superior performance of the NARX model implies that they can rely on ANNs for more accurate predictions of SC indices, enhancing decision-making processes, improving risk management strategies, and optimizing portfolio allocations.”
The crystal ball, it seems, is not magic. It is mathematics. And it is now available to anyone willing to embrace the power of artificial neural networks.
Reference: here
Other Articles:
