Machine Learning Algorithms vs Economic Understanding

Do machine-learning algorithms truly understand economic concepts, or do they mimic patterns without grasping underlying incentives, causality, or theory? In recent years, machine learning algorithms  have become widely used in economics for forecasting, credit scoring, labour market analysis and more (Mullainathan & Spiess, 2017; Varian, 2014). It is a branch of artificial intelligence that aims to make predictions without being programmed to do so by learning from past observations and training data. However, do machine learning algorithms truly understand economic concepts or just detect patterns? 

[AI and machine-learning illustration] [Online image]. (n.d.). The Economic Times. https://img.etimg.com/thumb/width-1200,height-900,imgsize-1531109,resizemode-75,msid-123297883/tech/artificial-intelligence/ai-and-machine-learning-driving-a-global-tech-revolution-worth-trillions.jpg 

What does understanding mean in economics?

Economic understanding relies greatly on interpreting mechanisms and not merely predicting outcomes. Economics is a study of how people respond to incentives, constraints and trade-offs (Pearl, 2019). It involves comprehending why agents behave the way they do, recognising causal relationships and making use of theory-driven frameworks (e.g. supply and demand). Economics also includes evaluating counterfactuals. For example, how would economic outcomes differ if a specific policy had not been implemented, holding all else constant?

What machine learning algorithms actually do

Machine learning algorithms learn statistical patterns from historical data, focusing on correlation-based predictions (Agrawal et al., 2019; Athey & Imbens, 2019) and not explanations, being optimised for tasks like classification, regression and forecasting. They lack the knowledge of underlying economic theories, awareness of incentives and motivations, as well as the ability to think, evaluate and reason why economic agents behave in certain ways. 

Why do machine learning algorithms do well in economic tasks?

Machine learning algorithms have the capability to handle huge and messy datasets better than traditional econometrics, capturing non-linear relationships (Varian, 2014; Mullainathan & Spiess, 2017) that the majority of simple models cannot. They have proven useful in performing tasks that require more prediction than explanation. These tasks include credit default predictions, demand forecasting, inflation trends (Rajkomar et al., 2019) and detecting anomalies in financial data. It's clear to say that it has its advantages in providing economists with more accurate forecasts and faster data processing. 

Where do Machine learning algorithms struggle in economics?

Machine learning algorithms do not understand causality (Pearl, 2019; Bareinboim & Pearl, 2016), failing to explain behaviour, incentives, change and shift in policies, particularly struggling with reverse causality and missing variables (Athey & Imbens, 2019). Certain scenarios, such as counterfactual questions (e.g. What would happen if taxes were cut?) require economic reasoning, which machine learning algorithms do not acquire. 

Why do machine learning  algorithms not have economic understanding?

This branch of AI does not have the basic knowledge of core economic concepts, mainly utility, equilibrium, marginal cost (Agrawal et al., 2019) and rational behaviour; neither does it have the ability to generate theories, only being able to detect patterns. Furthermore, they lack interoperability and do not understand time consistency, strategic behaviour or expectations. Unlike symbolic AI, which relies on explicit programmed logic and casual rules, machine learning  algorithms does not encode concepts such as utility maximization and equilibrium conditions. For example, machine learning algorithms predict that unemployment rates drop when interest rates rise but don’t have the ability to explain why, therefore showing that machine learning algorithms knowledge are  surface-level and not conceptual. 

Hybrid approaches: Machine learning algorithms + economic theory 

To maximise understanding, economists have started to combine Machine Learning algorithms  with theory-based models (Athey, 2018; Athey & Imbens, 2019), also known as “Hybrid models”. Hybrid Models leverage the predictive power of machine learning algorithms while keeping the economic theory needed for casualty. This approach has been seen to be promising, utilising the strengths from both fields. 

Bank of Canada. (2023). Published research papers that use machine learning, 2018–22 [Online chart]. Bank of Canada. 

Real-world examples 

Good machine learning algorithms performance examples 

• Banks using machine learning algorithms for credit scoring 

• Retailers using machine learning  algorithms  for demand forecasting 

• Job platforms using machine learning algorithms  to match candidates to employers 

Bad machine learning algorithms performance examples 

• Failing to predict the 2008 world crisis (Brynjolfsson & McAfee, 2014) 

• Stock markets collapse when unexpected events occur

• Models incorrectly estimate the effects of policy changes because incentives have shifted (Athey & Imbens, 2019) 

This shows how machine learning algorithms  can be powerful but fragile without theory. 

Ethical and practical concerns 

Another significant limitation of machine learning algorithms in economics is its reliance on biased historical data. This can create existing inequalities(Kleinberg et al., 2018). Furthermore, machine learning algorithms lacks the needed transparency, often operating as "black boxes" where the reasoning behind its predictions can be very complicated to understand. This increases the ethical concerns of using machine learning algorithms  for policy decisions as it can unfairly cause harm to certain groups without clear accountability. More or less, a lack of transparency increases the difficulty for policymakers in decision making, increasing the risk of failures. Since policy requires both accuracy and understanding, relying too much on machine learning  algorithms can lead to undesired outcomes (Mullainathan & Spiess, 2017).

In conclusion, machine learning  algorithms are excellent at identifying patterns and creating predictions; however, it's also important to consider that they do not understand economic concepts as they lack reasoning, theory and causal interpretation. Economics requires an understanding of how people behave, how incentives shape decisions and how policy changes can affect certain outcomes, which machine learning  algorithms cannot perform on their own. The best output has been seen from using machine learning algorithms to supplement the work of human economists.

References 

Athey, S., & Imbens, G. (2019). Machine learning methods that economists should know about. Annual Review of Economics, 11(1), 685–725. https://doi.org/10.1146/annurev-economics-080217-053433 

Bareinboim, E., & Pearl, J. (2016). Causal inference and the data-fusion problem. Proceedings of the National Academy of Sciences, 113(27), 7345–7352. https://doi.org/10.1073/pnas.1510507113 

Chalfin, A., Danieli, O., Hillis, A., Jelveh, Z., Luca, M., Ludwig, J., & Mullainathan, S. (2016). Productivity and selection of human capital with ML. American Economic Review, 106(5), 160–165. https://doi.org/10.1257/aer.p20161029 

Kleinberg, J., Lakkaraju, H., Leskovec, J., Ludwig, J., & Mullainathan, S. (2018). Human decisions and machine predictions. Quarterly Journal of Economics, 133(1), 237–293. https://doi.org/10.1093/qje/qjx032 

Mullainathan, S., & Spiess, J. (2017). Machine learning: An applied econometric approach. Journal of Economic Perspectives, 31(2), 87–106. https://doi.org/10.1257/jep.31.2.87 

Rajkomar, A., Dean, J., & Kohane, I. (2019). Machine learning in medicine. New England Journal of Medicine, 380, 1347–1358. https://doi.org/10.1056/NEJMra1814259 

Varian, H. R. (2014). Big data: New tricks for econometrics. Journal of Economic Perspectives, 28(2), 3–28. https://doi.org/10.1257/jep.28.2.3