Transforming risk management in 2025: How AI is bringing calm clarity in volatile times

Keep reading to find out:

• How the boring but important bits of human due diligence are being replaced
• Why AI can be an unexpected facilitator of creativity and novel ideas
• The historical weaknesses of risk management modelling that AI is already solving

AI can outperform even the most advanced human teams when it comes to the speed and scale of predictive data set analysis. While this is unlikely to replace quantitative analysis as a discipline, it will support human decision making in times of tension and uncertainty. By simulating a theoretically limitless range of scenarios, from economic shocks to political strife, it can bring new levels of clarity to highly emotional situations before they happen.

The most pressing current challenge for traders and investors is finding the right ways to integrate the immense analytical power of AI into daily workflows in a way that supports, rather than disrupts established human-driven processes.

Let’s explore the ways savvy traders and investors are putting AI to work.

Due diligence - eliminating the grunt work

Typically, a task for junior analysts, due diligence is repetitive and labour-intensive. Successfully eliminating the ‘grunt work’ of due diligence without hampering the actual ‘diligence’ part of the process has obvious advantages.

Rogo^[1], an artificial intelligence start-up developed by former Lazard analyst Gabriel Stengel, aims to automate as much of it as possible.

The firm has built a chatbot that can create basic valuation reports for analysts. Investment banks are already using it, and Stengel says the tool can perform in minutes tasks that would have normally taken him days.

Whether that will free up junior analysts to do more interesting things or just make it even harder for young investors to access those positions to begin with will remain to be seen.

JPMorgan Chase^[1] has gone a different route. They’ve developed their own large language model (LLM) to look at deals and assess potential investor value.

LLMs handle vast quantities of unstructured data and information like social media posts, news stories, press releases and regulatory filings.

The sheer density and variety of this kind of unstructured information make it time-consuming to analyse for humans. Think of NLP as an engine for extracting meaning from disparate sources of conflicting information.

Volatility forecasting - bigger picture thinking

Volatility is a complex and inexact science with a variety of established traditional models. One of the most common is GARCH. It stands for Generalized Autoregressive Conditional Heteroskedasticity and is a statistical modelling technique for analysing and forecasting market volatility. It’s highly complex, but in the simplest terms possible, it looks for clusters of high or low volatility.

GARCH is best used in tandem with other models. Where it falls short is its capacity to capture what experts call the asymmetric effects of volatility, where negative returns increase volatility more intensely than equally positive returns.

A potential AI-driven solution could lie in something called long short-term memory (LSTM). These are deep learning networks that aim to predict volatility in a different way, overcoming the limitations of models like GARCH. They use ‘gates’ to regulate the flow of information.

This means they can learn complex and non-linear patterns that the more rigid GARCH formula may miss.

In practical terms, they can take a ‘bigger picture’ perspective, incorporating sentiment analysis and identifying patterns over longer periods of time.

Liquidity stress testing - going beyond hypotheticals

Liquidity stress testing exists to determine how well, or not, an institution handles a financial shock within a pre-defined time window. Traditional stress testing models deal in hypotheticals, based on historical or simulated data.

Machine learning is helping to advance the accuracy and scope of these models. Although more expensive to run than traditional methods, machine learning makes it possible to work with larger and more complex datasets with higher accuracy.

Transformer-based models show particular promise. Using neural network architecture, they are able to discern financial risk signifiers from unstructured data like earnings calls and regulatory filings.

Portfolio management and optimisation - driving operational efficiency and smarter targeting

AI is not yet at the level to generate direct portfolio risk management instructions and inputs. Where it can be useful is generating ideas and starting points for further research and highlighting potential risks.

Perhaps this is the work the junior analysts not tied up with due diligence will be doing?

A McKinsey report^[2] found that, on top of productivity cost-saving worth as much as 40%, asset managers are finding useful ways to deploy AI for “optimised portfolio construction” and better client targeting. The report also highlights how AI is driving operational risk management through automated compliance monitoring and codification of institutional knowledge.

If an AI tool that can deliver institutional-level insights and curated analysis when you need them sounds like a useful addition to your toolkit, check out Tradu’s Analyst AI. Users benefit from sentiment alerts and insights tailored to their own trading activity.

Can AI automate financial trading strategies?

Probably one day. But not yet.

Traders are already using AI to automate stop losses. But even then, the AI needs to be trained on accurate data at scale. The need for reliability intensifies when trading with leverage. To learn more about stop losses and how they can help you manage risk while trading with leverage, check out our guide to risk management.

The most constraining limitation of AI’s current capacities is nuance. Yes, markets are all about data. But without experienced human oversight, there is an unignorable risk of overfitting. Automated models drive efficiency, but they are always based on past data.

There is no such thing as future data. Only analysis and predictions. Models remain prone to overfitting, where the strategies they produce are too closely tied to past data and when tested in the markets, they fail badly as soon as conditions evolve.

Experienced traders are trained to spot overfitting and react accordingly.

The bottom line

Traders and investors pinning their hopes on a fully automated risk management system for growing and preserving wealth are going to be disappointed. It’s not something we can automate with a few ChatGPT prompts.

The value of AI lies in its capacity for handling data at scale, tightening up processes, optimising decision making, identifying risks that hide in unstructured and complex information, and surfacing novel and non-obvious ideas.

There may come a time when AI is all we need to manage risk, but we’re not there yet.