Agent Based Simulation in Volatile Markets
Recent events surrounding GameStop have shown that traditional financial forecasting is ripe for disruption
The 1,700% surge in GameStop’s share price last month represented one of the most significant developments in Wall Street’s history; certainly its biggest since the crash and subsequent recession of 2008. The rise of GameStop represents a watershed moment, where the democratisation of trading platforms led to enthusiastic Redditors pitching themselves against the hedge funds that have for so long ruled the roost. In such a rapidly changing world, financial analysts must reevaluate current models and predictive analysis.
In the case of GameStop, discussion and buzz in an online forum in Reddit ended up becoming a formidable market force. In doing so, it served as an excellent example of the increasing variation of influences that can determine change. How the collective actions of agents can affect markets can be put down to the complexities of emergent phenomena, which has been a puzzling topic for science and philosophy for a great number of years. In essence, emergence occurs when an entity holds a property only through the sum of its parts. A classic example in nature may be the overall intelligence conveyed by an ant colony, despite the ants being simple creatures on their own. Seeing one action on its own would not allow you to understand how the eventual situation occurred, only when seeing them altogether is this possible.
The concept can also be used to help explain unusual outcomes relating to game theory, including ones that seem to have a counter intuitive logic to them. The Braess Paradox for example, is an observation concerning traffic described by the German mathematician Dietrich Braess. Intuitively, it would seem a way to decrease traffic on a road would be to add another lane. However, what actually occurs when this happens is that people are able to swap lanes more in order to try and get ahead, slowing down the traffic even more. How people behave on an individual level collectively causes a counterproductive result. This effect has been shown with numerous real world examples, where the flow of traffic has actually been improved by removing roads.
Consider the GameStop situation by looking at a single agent’s action. The demise of GameStop meant that its stock price was forecast to go down significantly, so much so that traders were betting heavily against it. It’s difficult to see the motivation of an individual citizen buying shares of stock that are bound to lose value. But when you observe this action repeated by many, achieving the goal of reversing the effect of a hedge fund’s shorting practices, then an emergent phenomenon appears! What could be considered an active economic agent is rapidly expanding; by recognising and monitoring the agents across these new variables, like in this case social media, analysts could gain a more accurate picture of the state of play.
While it includes examples like this, emergence doesn’t have to involve a shared goal, it merely requires collective actions to cause something to happen. It could also be used to explain how certain prices are set within the free market or how simple clerical errors in a brokerage can lead to million dollar losses. Essentially, they are driven by two key factors: how the agents act and the complex scale of these agents.
Computational Challenges Facing Agent Based Simulation
Although agent based models have existed in finance for some thirty years, they were called into focus more intently following the stock market crash just over a decade ago. Agents include individuals, companies, biological entities and so on – each behaving according to its own set of complex, pre-defined rules.
Often financial modeling techniques, such as regression analysis, work by taking particular macroeconomic factors and equations and then applying them to the situation. In this sense, they are ‘top down’ methods and are often unsuccessful when applied to emergence. Agent based modelling, in contrast, works by starting from the bottom and analyses the actions of individual entities. In this sense, it is far better suited to tackling emergent phenomena, due to them always finding their origin in the individual actions.
In regard to the scale, the defining features of emergence means that any attempt at analysis would require looking at the myriad of agents involved. With our increased connectivity through technology and urbanisation, the number of interactions we have with each other has undoubtedly increased. It follows that with more individual actions, there is an increased likelihood of emergence occuring.
Agent based modeling at scale is unsurprisingly a key opportunity for high performance computing. The collection and connectivity of data spearheaded by the Internet of Things has given a huge boost to the resources available for computer modeling. Interpreting and using it effectively is however a challenge and ultimately software that succeeds in simulating at scale will better represent the complexities behind emergence.
Agent based simulation tends to be most valuable in unpredictable environments, offering insight that is not apparent using more traditional methods. However, providing an engine and infrastructure that can handle vast quantities of data, often from disparate sources, provides a computational and infrastructural problem. Moreover, solving it will unlock the key to predicting markets more accurately and getting ahead of the competition.