The Network Effects of Risk
Risk is the result of certain events and the cause of other events. We propose a graph of cause-and-effect relationships that serves as a mathematical model exposing the factors responsible for the risk in question through time and geography. A model that serves as the foundation for Artificial General Intelligence (AGI).
Before Meteorologist Edward Lorenz left for a coffee break while a weather simulation was running on his computer, most scientists believed the universe behaved predictably. Lorenz’s simulation was very similar to many he had run before. Except, this time, he had rounded off one variable from .506127 to .506.
After returning from coffee, Edward was shocked that such a seemingly minor change dramatically altered the results. Lorenz then realized that small changes can lead to significant consequences. Famously called the “butterfly effect,” his work produced the founding principle of chaos theory. It is often quoted as “A butterfly flapping its wings in Brazil can lead to a hurricane in Texas one month later.”
Inner Workings of Chaos
Lorenz postulates that future events are impossible to predict since any small change to the initial conditions can have dramatic consequences later in the timeline. His work concluded that predictions are useless without a perfect idea of initial conditions.
To all the complexity of the physical world of weather, crops, ores, and factories
However, it is not just the initial conditions that are important. As events unfold, they create or change other conditions that impact what we want to forecast. French Mathematician Benoit Mandelbrot explains, “To all the complexity of the physical world of weather, crops, ores, and factories, you add the psychological complexity of men acting on their fleeting expectations of what may or may not happen-sheer phantasms.”
Causal Networks
Marketplaces are chaotic systems that are influenced by a multitude of tiny changes. While the statistical physics of Mandelbrot does well to forecast future prices with many unknowns, there is a more comprehensive method — a model based on Network Science and Causality.
While it is impossible to account for every event, we can graph many contributing factors using causal detection. This produces a network of cause-and-effect relationships that is the basis of our AI model. Estimand uses thousands of datasets with proprietary causal detection to discover the causal affinities, their weight, and their direction.
For example, home prices have apparent causal relationships with current inventory, mortgage rates, labor costs, and lumber prices. However, the number of precipitation days from up to 20 years ago also plays an important role. This is the critical time component.
Cascades and Time
Mandelbrot explains his price forecasting, “My model redistributes time. It compresses it in some places, stretches it out in others.” He concludes, “Price is a function of trading time, which in turn is a function of clock time.” To simplify, the speed at which contributing events occur changes.
For example, Apple stock may move slightly upward and down over a month. Then, breaking news occurs, and buyers start buying or selling rapidly. This is the rationale behind Mandelbrot’s trading time and an example of a cascade.
Cascade events are specific to networks. These are based on individual decision-making but culminate into mass changes. The influencing factors for a cascade are the state of the world, the individual payoff, and private signals.
Lorenz’s initial condition in our example is the breaking news. This is the current state of the world. From here, each investor decides their expected payoff. Perhaps the information is terrible; some panic and sell, while others see it as an opportunity to buy more stock. Last are private signals. While this could include insider trading, it is usually much more subtle. For example, consider the behavior of a person with millions in Apple stock versus someone with less than a hundred shares.
Knowing the Risks
Surprisingly, cascades are often based on very little information. They can also be wrong. This makes cascades fragile. Meaning they can form and disappear quickly. Following the psychology of crowds is a messy business and impossible to model.
It feels like cheating, but we have simplified the modeling of this phenomenon with causal detection. Understanding the causal relationship between contributing factors makes it possible to know the weight, direction, and relationships between many influencers. This removes the direct human behavior and instead considers the likelihood of the effect one event has on another.
The reason this method works is the fact that markets, and thus risk, behave like networks. Their movement is not linear. Instead, it is chaotic. A collection of small stimuli influencing a significant outcome.
Conclusion
Networks must account for popularity, which is responsible for extreme imbalances. For example, most nodes make decisions based on their nearest neighbors in a social network. Each influences and is influenced by the other participants. However, specific nodes have more significant weight when it comes to influencing.
The strange thing about these influencers is that their number of followers is significantly greater than the majority. For example, the average YouTube channel has less than 1000 followers. However, seven channels have over 10 million followers. Even stranger is that the same types of imbalances occur for risk factors.
So, the question remains as to what influences the influencers. Be it social media stars or risk factors, a small percentage represent the majority. However, these influencers are influenced by the long tail. The content for their podcast, stunts for their videos, and even the deciding factors of peril come from small but significant events.
For example, ocean temperature is the leading cause of hurricanes. However, additional factors from the long tail influence the ocean temperature. For example, air temperatures, cloud coverage, wind, and currents contribute. Thus, each of these factors has factors, and each of those factors has factors. The network of risk goes deep through the long tail.
About Estimand
Founded in April 2023, Estimand Inc. is a Delaware Corporation with principal offices in Raleigh, North Carolina. We began as a paper on causal AI that evolved into the first company to instantly reveal the factors behind global financial risk. You can learn more about Estimand by visiting our website: https://estimand.ai.
References
- Dizikes, P. (2011) When the Butterfly Effect Took Flight. MIT Technology Review. technologyreview.com/2011/02/22/196987/when-the-butterfly-effect-took-flight/
- Easley, D. & Kleinberg, J. (2010) Networks, Crowds, and Markets. Cambridge University Press
- James, H., Borscheid, P., Gugerli, D., & Straumann T. (2013) The Value of Risk. Oxford University Press
- Mandelbrot, B. & Hudson, R. (2004) The (Mis)Behavior of Markets. Basic Books
- Pearl, J. (2009) Causality — Models, Reasoning, and Inference. Cambridge University Press
