Applying Artificial Intelligence Tools in Economic Models: A Comprehensive Analysis
The application of artificial intelligence (AI) and machine learning (ML) in economic models is gaining significant traction due to their unique capabilities in prediction and causal inference. This article explores how leading economists have integrated these powerful tools into economic studies and the potential for transformative change within the field.
1. Prediction vs Causal Inference
According to Hugh Dance and John Hawksworth from PWC, ML excels in prediction problems by minimizing forecasting errors through a balance between bias and variance. In contrast, traditional econometric models are more adept at capturing causal relationships but may over-fit samples and generalize poorly to new data (Dance Hawksworth, 2021).
Susan Athey, in her assessment, emphasizes that ML is particularly useful for semi-parametric estimation and handling a large number of covariates relative to observations (Athey, 2021). ML also offers a systematic approach to model selection, which contrasts with traditional econometric methods where researchers typically choose a model based on principles and estimate it once.
2. ML in Economic Research
David McKenzie, a prominent development economist at the World Bank, highlights the potential of ML for development interventions and impact evaluations. ML can be used in basic measurements in countries with limited data, as well as in targeting interventions to specific groups (McKenzie, 2021).
However, McKenzie also identifies several challenges, including the lack of a gold standard for evaluating these methods. Supervised ML relies on labeled training datasets and metrics for evaluation, which can be difficult to assess in data-poor environments. Additionally, there are concerns about the stability of predicted relationships and the potential behavioral responses that may affect the reliability of ML for treatment selection.
3. Policy Targeting with ML
Monica Andini, Emanuele Ciani, Guido de Blasio, and Alessio D’Ignazio present two examples of how ML can be applied to policy targeting. The first case study examines a tax rebate scheme introduced in Italy in 2014, aimed at boosting household consumption. The Italian government targeted employees with annual incomes between €8145 and €26000, but ML analysis suggests that targeting consumption-constrained households would have been more effective (Andini et al., 2021).
In the second application, Andini et al. address the "prediction policy problem" of assigning public credit guarantees to firms. Traditional guarantee schemes often exclude borrowers with low creditworthiness, even if these firms face credit rationing. The authors propose an ML-based mechanism that considers both creditworthiness and credit rationing, demonstrating significant performance improvements through a comparison of disbursed bank loans growth rates (Andini et al., 2021).
4. Potential Transformative Impact
Athey predicts that the combination of ML and new data sets will fundamentally change economics, leading to new questions, new approaches, and interdisciplinary collaboration. ML can help policymakers generate more data-driven and systematic models while providing confidence intervals that account for the entire algorithm. This process could also involve larger teams of economists and engineers working together to implement policies effectively (Athey, 2021).
Conclusion
The integration of AI and ML in economic models promises to revolutionize how we understand and intervene in the economy. While challenges remain, the potential benefits of improved prediction, causal inference, and targeted policy-making make the case for continued exploration and application of these tools.
References:
Athey, S. (2021). Machine Learning and Economic Models. Research in Economics, 75(2), 151-171. Dance, H., Hawksworth, J. (2021). Big Data, AI and the Predictive Future of Economics. PWC Blog. Andini, M., Ciani, E., de Blasio, G., D’Ignazio, A. (2021). Using Machine Learning for Policy Targeting: Two Applications. VoxEU. Mckenzie, D. (2021). Machine Learning in Development Interventions and Impact Evaluations. World Bank Blog.