7 Learnings Using Equity Market Data for Early Warning

In this blog, we explore the challenges and learnings from adapting our early warning system, Risk Alert, to include equity market-related data. Beyond its classic use, our goal in using this data is to build forward-looking predictions of future direction, operating environment, and fundamental financial indicators for companies. We then tie these factors back to an assessment of fair value and expected market price movements. Ultimately, we are looking for early signs of financial deterioration so creditors and investors can make informed decisions to mitigate emerging risks.

Here are the lessons we’ve learned, along with tips for those embarking on a similar journey:

1. Ensure your IT infrastructure can handle high-frequency data. This data requires various transformations and calculations for effective use, which adds complexity to the necessary IT infrastructure. For example, one year of tick data for a single listed company could require 12-15TB of storage space, while one year of daily share price data requires only 1MB. It is important to define the appropriate data frequency for your analysis and prepare your infrastructure accordingly.
2. Be specific about what you are predicting. In early warning systems, we aim to predict financial deterioration as early as possible. However, applying this logic to the outcome variable directly would be a mistake. Predicting an event 10 years in advance might be a theoretical achievement but holds little practical value. A 6 to 18-month outcome period is usually suitable for credit risk early warning. For asset management portfolios, 60-180 days has shown the best results in our tests. The optimal definition depends on the use case, type of exposure, business cycle, industry, and other factors; considering these will significantly improve prediction accuracy.
3. Use trend variables rather than point-in-time variables. Point-in-time information provides insight into the current standing, and unless it indicates obvious deterioration (which would be too late for early warning anyway), its value is limited. The direction of travel is more important: we want to anticipate the trajectory of financial development and, more importantly, its slope. This gives us an understanding of how soon the deterioration will directly impact our exposure.
4. Peer analysis sets the baseline. Understanding movements against peers and the movements of the peer group itself over time helps answer a crucial question: Is there a problem with individual exposure or a systematic issue in the portfolio driven by external factors? This ultimately informs decisions about appropriate actions – mitigating risk at the exposure level or reviewing the entire portfolio.
5. Invest effort in designing your aggregation logic for alerts. In early warning systems, we strive to diversify the nature of our data as much as possible. More diverse data means a higher chance of spotting problems at their origin. However, this also means receiving alerts of different natures – some on a company's financials, some on its reputation from news, and some on market movements. Certain combinations of alerts can amplify risk, while others can reduce it. Using defined rules to handle combinations, rather than a simple scorecard approach, will reduce both false positives and false negatives. This topic deserves its own spotlight, and we plan to dedicate a separate blog post to it.
6. Rather than using traditional logistic regression, try non-parametric methods such as decision trees, for example. They work well when modelling non-traditional use cases and do not require data to comply with distributional or statistical assumptions which provides more flexibility in our case. More importantly, they provide clear, interpretable predictions, which is a must-have from transparency point of view.
7. Generative AI models provide quick, textual summaries of identified risks, powerfully guiding human analysts toward potential areas to investigate. Use of Gen AI in early warning is an exciting topic on it’s own and we will discuss it in a separate blog.

Adding equity market data to early warning systems increases prediction accuracy when integrated effectively. The key considerations we have identified focus on data quality, method appropriateness, and result interpretation. Considering these when integrating market risk data into early warning systems will help you unlock the powerful potential to predict risk events well in advance.

Several other aspects deserve attention, such as the precision-recall balance, the influence of individual indictors, and, of course, the use of generative AI. We will discuss these and more in future blog posts.

Risk Alert webpage: Risk Alert: The Early Warning System of the Future | Deloitte UK