Authors: Dietmar Peetz, PhD, Qunzhi Zhang, PhD, and Daniel Schmitt, PhD
Financial history is there for all to read, yet market participants have short memories. Additionally, each generation of investors seems to believe it is different this time. This inevitably leads to each investor generation repeating the mistakes of former generations. This is because the collective behavior of investors, which is based on human nature, does not change over time as it goes through cycles of greed, exuberance and fear, which predictably lead to asset price booms and busts. Financial crises are interrelated and interdependent phenomena; they share the same root causes and thus constitute an integral part of the system and its dynamics.
But why is that? Specific financial conditions, e.g., lax credit standards or low interest rates can create an impulse for investors to start investing in segments of financial markets (housing, high yield, hedge funds, commodities, equities, bitcoins, etc.). As investors start monetizing these profit opportunities based on the initial impulse they trigger, it elicits an impulse with other market participants who decide to participate in this opportunity and follow the money. In doing so, they intensify the signal, which in turn convinces even more investors to join the party. In natural science parlance, the positive feedback dynamic has created an “attractor” where investor decisions affect other investors – think about those investors as nodes in a highly interlinked network with hierarchical structures. The key to understanding these attractors is that prices are increasing in a non-random way, i.e., frequently triggered by the collective behavior (herding) of investors divorced from fundamentals, news or corporate events.
As more investors pile in, the profit opportunity gets exhausted. This is also usually the point when investors begin to increase leverage to squeeze out the last pennies. Typically, this is the time when the system has reached an unstable equilibrium which reacts very sensitively to both exogenous and endogenous shocks, like the famous wing beat of a butterfly.
When such a shock occurs, leveraged investors must answer their margin calls by selling assets, regardless of the price. This, in turn, sets in motion a cascading effect because falling prices motivate an ever-increasing number of investors to liquidate their holdings to either lock in gains or to limit their losses. Even if they are aware that they are contributing to the selling pressure, they will not stop because they have their own utility function.
The history of market crashes is well documented but rarely is there a consensus about bubbles until they have burst. However, mathematically we can categorize and hence identify crashes before they burst based on aggregate investor behavior. The critical behavior of the market manifests itself in a so-called power law singularity signature. We denote “tc” as the time at which a power law singularity dovetails with the termination of an exuberant S&P 500 regime, which is tantamount to an imminent phase transition, typically a crash. Advances in computational power and artificial intelligence help us to detect, in a very timely fashion, attractors and phase transitions not only for the market, but for individual stocks as well.
Customized deployment of Machine Learning 2.0. instead of black box
When it comes to Artificial Intelligence (AI) in managing funds, the investment manager must still make a lot of decisions about what data to use and how to translate outputs into investment decisions. Those decisions will have a substantial impact on how AI will ultimately generate outperformance for the investor. Historically, neural networks and machine learning 1.0 models tended to overfit the training set. Once the models were exposed to live data, their performance dropped off a cliff. However, given advances in computational power and algorithms, the new models can learn better from their mistakes.
Nonetheless, the danger remains that machine learning and AI could suffer from herding, and there is always the behavioral tendency to gravitate towards an attractor. As more AI comes to market, profits in specific areas will get competed away. History repeats, the contestants’ names just change. Eventually, all the machines may reach sell decisions at the same time.
Usually, to train a black box AI system, one has to provide big data and label the data manually, i.e., tell it that this picture is a cat, that picture is a dog, and so on, and then the black box system will work as a black box, which really means that no one understands how it works. In financial investments, it is usually hard to label data such as fear, greed, herding, etc. In many cases, data related to those labels is not available, and sometimes the labels themselves are not even well-defined.
Consequently, the first problem of such a black box system is that it does not have the same amount of information as humans possess and it is also very hard to train. Human scientists can build models to identify patterns by synthesizing all available knowledge. That is an impossible mission for the current black box AI generation.
The second problem of black box AI is that it lacks humans’ ability to reason and to make trading decisions. As discussed in the AI section, portfolio managers must make decisions with a system that is difficult to understand, or code their decision-making methods in the AI system. Accordingly, a black box AI with less information attempting to mimic human behavior may make even faster or more mistakes than humans.
At SIMAG® we have a different approach: We use AI/machine learning as an overlay after putting substantial common sense yet proprietary intelligence in place. This is done by structuring and by deciphering sense in the data. We do not ask AI to find the patterns for us.
In a related sense, former chess grandmaster Garry Kasparov explains the outcome of a famous freestyle online chess tournament: The chess machine Hydra, which is a chess-specific supercomputer like Deep Blue, was no match for a strong human player using a relatively weak laptop. Human strategic guidance combined with the tactical acuity of a computer was overwhelming. The surprise came at the conclusion of the event. The winner was revealed to be not a grandmaster with a state-of-the-art PC but a pair of amateur American chess players using three computers at the same time. Their skill at manipulating and coaching their computers to look very deeply into positions effectively counteracted the superior chess understanding of their grandmaster opponents and the greater computational power of other participants. Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process.
At SIMAG® we use AI for robust optimization. And it is not a black box as the resulting algorithm is interpretable and is accepted only after it has passed the “smell test”. We combine the advantages of humans and machines as shown in Kasparov’s example. We develop scientific models and identify patterns with human intelligence by incorporating all available information. At the same time, we automate the statistical tests of the models and patterns with the real data by using the computational power of machines. We have our high-performance computing software running on computer clusters. This makes data analysis hundreds of times more efficient than usual.
In our differentiated approach, all the processes are transparent, and all results are well understood by us. We pick the robust results to make trading decisions. In a word, our solution is a human/machine combination system in which we capitalize on the innate strengths of both sides while avoiding the black box pitfalls. This enables us to front run or outsmart machine learning/AI funds because our approach is much more robust. This is SIMAG’s edge. It is also why we are convinced that we constitute a better choice for our clients.