Causal AI: Going Beyond Correlation in Finance

Today, Artificial Intelligence (AI) has become a buzzword across industries. 

In finance, AI is used to identify fraud, automate lending decisions, predict stock price movements, and, in some cases, optimize portfolios. However, there is a limitation to contemporary AI approaches—they primarily lack a clear basis in causality, instead favouring correlation.

In other words, AI has the capability of identifying patterns in vast quantities of data, but, for the most part, cannot determine the reason(s) for the recognition of any particular pattern in the underlying data.

In finance, understanding "why" is of paramount importance in determining the efficacy of a given investment decision versus the cost associated with being incorrect in the determination.

This is where Causal AI steps in. It’s an emerging field of artificial intelligence that moves beyond prediction to explain cause-and-effect relationships.

For financial institutions, regulators, and investors, this shift could be revolutionary. 

Correlation versus Causation: Why It Matters in Finance

We have all heard it before: "correlation does not imply cause-and-effect."

• Correlation: Two things move together. Example: During warmer months, ice cream sales and attacks from fish may be correlated in that all occur more or less in, or result from, warmer months. Now, during warmer months, people eat more ice-cream as well as they go more for swimming and hence have more chances of being attacked by fish.

• Cause-and-effect: One or the other instrument provokes or has the direct cause or effect, resulting in the other. Example: The central bank raised interest rates, so lending fell.

In finance, making decisions based purely on correlation can be misleading. Take, for example, an algorithm that predicts loan defaults because, coincidentally, a lot of loan defaulters live in a particular neighbourhood.

There is a correlation in the data for loan defaulters and neighborhood; however, that doesn't mean the neighborhood caused defaults. The real cause of defaults could include income instability or job market conditions. In the absence of understanding causation, financial AI contributes to unnecessary bias and unfair decisions.

What is Causal AI? Causal AI is a type of artificial intelligence that deals with understanding, modeling, and reasoning about causality.

Instead of simply asking, “What relationships exist?”, Causal AI asks:

• “What is causing these outcomes to occur?”

• “What might happen if we intervene/create a change?”

• “What if the conditions had been different?”

Causal AI employs mathematical and statistical tools like causal graphs and structural equations to examine relationships between all factors.

For example, consider a causal AI model for credit scoring that demonstrates how income stability leads to repayment reliability, whereas social media engagement, despite being correlated, does not have the same effect.

The Case for Causal AI in Finance

Finance is about explicit decision-making with consequences, not just predictions about risk! Here’s why Causal AI is an urgent priority for the finance industry:

1. More Effective Risk Assessment

Standard AI can predict a customer's risk of defaulting on a loan—but Causal AI can tell you why they might default. Was it because they lost a job? Interest-rate hikes? Market downturns? Knowing what might have caused a loan default empowers banks to make more sensible lending decisions.

2. Compliance with Policy and Regulations

Regulators, whether it's the Reserve Bank of India (RBI) or the US Federal Reserve, are looking for explainable AI in lending and credit scoring processes and demanding more transparency. Causal AI allows for some transparency, as it is rooted in establishing the reasons for a root cause rather than systematic black-box conclusions.

3. Investment Decisions

Traders understand markets have many spurious correlations. For example, stock prices may appear proportional to oil prices at one point, but the question to ask is whether oil really drives those prices. Causal AI helps identify real reasons for asset price fluctuations rather than spurious correlations.

4. Fraud Detection

Many fraud detection approaches result in false alarms simply because they rely on correlations. Causal AI can enhance accuracy by isolating the transaction patterns that are obstacles to fraudulent outcomes.

5. Fairness and Ethics

Financial AI is often criticized for being biased—rejecting loans based on gender or credit reputation by ethnicity, or geography. Causal AI eliminates these filtered-out non-causal predictors, leading to decisions that are fair and ethically compliant.

Illustration of Causal AI in Finance

1. Credit Scoring

Traditional AI may determine that individuals who frequently spend at restaurants are more likely to default on loans. Do people default on loans because they go to restaurants? Likely no.

Causal AI has the ability to uncover the underlying reasons for defaults, which are often unpredictable income patterns. They provide banks with knowledge that should shape their repayment policies and not vilify certain lifestyle choices.

2. Investment Strategies

Imagine a fund manager using predictive AI to identify correlations between social media trends and stock price movements. While this approach has merit for short selling, it poses significant risks when used as a long-term strategy.

In contrast, Causal AI can identify the underlying factors that drive stock performance—interest rates, earnings growth, regulatory changes, etc.—making it more reliable for long-term strategies.

3. Decreasing Fraud

Suppose AI identifies customers who withdraw cash late at night as "high risk." This correlation does not mean causation.

Causal AI would show you that the driver of the fraudulent outcomes may likely be account takeovers, not that it was a transaction that occurred at 10:00 pm, requiring fewer false positives and making fraud more manageable.

Causal AI helps uncover the real reason behind fraud (stolen accounts) rather than wrongly blaming something coincidental (night-time withdrawals).

4. Regulatory Stress Testing

Stress testing is how banks see how they will react to shocks such as a recession. Causal AI allows testing with simulations, such as "If unemployment increases by 3%, what do the mortgage defaults look like?" The "what if" ability makes the models more realistic and more useful for regulators.

Barriers to Causal AI Adoption

Like all innovations, Causal AI has barriers to adoption in finance:

• Data Need: Causal AI requires structured and high-quality data.

• Complexity: The financial markets have many variables that interact with one another.

• Cultural Change: Banks have always worked with predictive analytics; the move to causal thinking requires retraining in new ways of thinking.

• Computational Cost: Running simulations and counterfactual models is often more resource-heavy than running simple predictions.

The barriers to causal AI are outweighed by the benefits—especially in a sector where one mistake can end up costing billions.

Causal AI vs. Predictive AI: A Simple Analogy

A useful way to think about Causal AI is like a weatherman: "It is going to rain tomorrow, because 10 previous times the humidity rose, it rained!"

Causal AI is a climate scientist: "Rain happens because rising humidity leads to clouds, which leads to precipitation."

In finance, predictive AI would be—"This borrower looks risky because their profile fits the prior defaulters."

Causal AI would be—This borrower is risky because income is likely to decline due to sudden unemployment, which causes the problem with repayment of the loan.

That difference, understanding the "why," was powerful.

The Future of Finance with Causal AI

As financial institutions begin to understand the need for explicable, reliable, and fair AI, more and more financial players are already exploring and working with causal AI platforms.

• A number of startups, such as CausaLens, are focused solely on developing causal AI for business decision-making.

• Global technology leaders like Microsoft, IBM, and Google are demonstrating their eagerness to investigate causal approaches in enterprise AI.

• Banks and regulators of all types are exhibiting a genuine interest in models that focus on causality above all, which should be promoted to make decisions with the intention of enforcing consistent fairness and compliance.

Causal AI systems are expected to become relevant in the institutional and regulatory sectors by enhancing transparency and compliance, similar to how credit scoring models were foundational to finance in the past.

Conclusion

Finance is all about risk and reward. Predictive AI indicates the risks you face, but it does not explain the underlying reasons for those risks. In a sector where billion-dollar decisions can be made, you are not going to rely on simple correlations.

Causal AI addresses that gap. By using cause and effect, it will be easier to understand what a risk is and always keep fairness and the best decisions in mind.

Moving forward for banks, investors, regulators, customers, etc., transparency with trustworthy AI offers a future for transparency where not only is the AI smart, but it is also a transparent, trustworthy advisor!

In short, financial AI needs to evolve from predictive AI into causal AI, where we not only want to search for patterns, but we also want to understand the causes to move finance beyond correlations.