Preparing for a black swan event in the financial markets

Global markets are more complex and interconnected than ever. Since the 2008 financial crisis, it’s become increasingly difficult to predict how the behaviour of market participants will affect the broader financial system.

Meanwhile, increased automation and the use of high-frequency trading have contributed to flash crashes.

We struggle to manage the rapid market movements that rest in the hands of fast-moving black boxes working at nanosecond speeds.

With this in mind, traditional risk modelling, which historically relies on forecasts, is no longer capable of capturing the dynamics of electronic markets. As such, it is no longer suitable for mitigating risk.

Traditional limitations

We are currently in a period of unprecedented geopolitical uncertainty, and the need for improved risk modelling techniques is clear.

Traditional risk modelling depends on historic data to help firms prepare for the future.

Yet the truth is that past performance is not a guarantee of future results. Strategies driven by historic data may fail to accurately represent future market dynamics.

It’s the difference between watching a film about flying a plane versus learning to fly a plane using a dynamic flight simulator. The simulator captures how the plane will respond to the pilot’s commands under any number of plausible scenarios, such as heavy turbulence, malfunctions, and even a black swan event.

Within electronically traded markets, traders respond to each other’s activity and adjust their actions – and their algorithms – accordingly.

Just like the game of poker, participants adapt their trading strategies to the stream of events coming at them.

At a more macro level, market participants buy and sell based on economic news and company announcements. These interactions form what’s known as a “complex adaptive system”.

Simulators are used extensively in artificial intelligence to understand and manage complex adaptive systems. And just like flying a plane, trading requires a robust strategy that has been refined in a simulation environment.

These simulators produce realistic synthetic data that represent the market under certain scenarios.

This data reflects the actions and responses of real market participants. That way, the algorithms represent a variety of realistic market conditions.

Recent advances in machine learning have led to improvements in predictive analytics. Yet, they are limited by historic data.

This leaves traders driving at a hundred miles an hour with only their rear-view mirror to navigate.

This is where agent-based modelling and simulation comes in.

Irrational behaviour

An agent is a computer representation of something from the real world – like a broker, a trader, or a bank.

They act independently, and can behave irrationally from time to time – think of the quick-fire sales we witnessed during the financial crisis. Asset prices collapsed as fear and uncertainty permeated the system.

Good computer agents can simulate all sorts of behaviour, whether rational or irrational.

By using agent-based simulation, businesses are able to create a virtual environment to test drive their decisions safely. They can fail fast without consequences, creating ways to safely mitigate risk.

Improvements in these simulation tools are helping financial institutions model complex futures in a virtual environment, before committing capital in the real world.

In the past, organisations would need to build hugely expensive high-performance computers to run these kinds of calculations. Now, the cloud provides the capacity to execute large-scale simulations on inexpensive infrastructure.

Real life scenarios

Strategies to manage credit risk and prevent fraud and money laundering all benefit from agent-based simulation. With the Bank of England releasing its proposed scenarios for 2019’s stress test earlier this month, institutions should use this opportunity to test potential outcomes in simulators.

Many sell-side trade execution teams are already using this type of simulation to model the performance of their algorithms before launching them to market.

In the longer term, as simulation becomes the gold standard of risk modelling, the range of industries it can be applied to are endless.

They not only improve business decisions, but make the markets a safer place by providing institutions with better insight into global risk and structural changes.

And ultimately, simulation tools can help avoid errors from the past, as well as those in the future.