The rise of artificially intelligent systems has given birth to several industries which have taken the world by storm.
The cryptospace has more innovations per project than industries.
This has created gaps that need filling.
Automation in several areas helps fill these gaps on the fly.
AI systems are often deployed and provide dispassionate and accurate angles and analysis to overcome volatility, human emotions, and data overload when trading digital assets, which enables efficient and profitable trading decisions.
Mikkel Morch, the Chairman and Non-Executive Director of ARK36, explored these concepts in detail and gave us unique insights into the AI-crypto trading universe.
This followed the recent launch of their algorithmic machine-learning trading software.
Mikkel Morch, the Chairman and Non-Executive Director of ARK36
Understanding the Foundations of Trading the Financial Markets
During our sit-down, Mikkel gave us a basic rundown of some operational concepts that drive investment and finance ecosystems.
In his words:
Financial Markets are complex systems. Recent scientific literature – including Stephan Wolfram’s book A New Kind of Science explores the characteristics of “complex”.
The result is a set of principles and hypotheses:
- Markets – especially markets that are open 24/7/365 and highly volatile — are “complex systems”. This means that influences on all prices (market influences) and on the price of a single security cannot be predicted with confidence. That is not to say that none of the influences or price behaviors can be predicted; many aspects are structural – a price may not go below zero, and changes in market pricing regimes will influence most security prices, etc. In complex systems, the exogenous and endogenous influences are (a) not linear in form, (b) the influences are not interrelated in a predictable fashion, and (c) the set of influences itself is unstable – new influences appear, and old influences disappear without warning. These attributes are visible in abundance in crypto markets.
- Predictions of any kind are exponentially more uncertain as the horizon of prediction expands, and the exponent is large. We may have a pretty good idea of a security price one minute on (the price is “persistent” to an extent). But an hour on, or a day from now, or a week from now? Uncertainty approaches 100% in part because over time the price will land on every possible price (allowing for the fact that the price rarely approaches zero, and rarely achieves high multiples of the current price). When we are enmeshed in complex markets, time is the enemy: “Long-term investing” is an illusion.
- If the price is unpredictable, what is a portfolio manager to do? The answer is counterintuitive: although the price is not predictable, perhaps some aspect of the price is predictable. This is not a new idea, and it is not an idea specific to financial markets – it is an idea inherent in complex systems. Every high school student is familiar with Brownian Motion: the movement of a single atom in a beaker cannot be predicted, but the rate of change in location – the amount of movement – can be predicted in every region of the beaker. It is related to energy (usually heat). In markets, the rate of change in price is called volatility, and volatility is a synonym for risk. In other words, uncertainty itself can be predicted over short horizons (it too is exponentially unstable over time, but far less so than price).
- Risk is somewhat predictable, and therefore the change in risk is also somewhat predictable. This “second order of risk” will naturally vary most when the price is reversing trend – from rising to falling, and vice versa. That’s not a market characteristic, it is just math.
So what is a portfolio manager to do? Search for securities that are trending in a desired direction, and have low and lowering risk. Conversely, the portfolio should shed exposures that display rising risk before the risk level reaches an unacceptably high absolute level.
Things Aren’t Really as They Seem
While it may seem straightforward, it really isn’t. There is so much that goes into detecting trading conditions and executing them profitably.
Sound simple? It is decidedly not simple, because ideal conditions are hard to detect and manage by humans, given the breadth and depth of analysis, the need to act swiftly, and the need to act again in a very short time for fear of falling into the “exponential uncertainty” trap.
AI/ML science connected to automated trading is the solution. Combined as a single integrated process, the system can frequently consider and reconsider, and act swiftly on the detected opportunities and aversions.
In essence, this is how we operate: risk first, returns second. Said another way, we are obsessed not with return, but with portfolio return related to (divided by) portfolio risk.
Portfolio return divided by portfolio risk is called the Information Ratio (“IR”). IR is measured by twelve-month rolling returns divided by the standard deviation of daily returns. For portfolios that are managed relative to an index or benchmark security, the excess return is used instead of the absolute return. We relate our portfolio to Bitcoin, in that we seek to deliver an IR that is better than BTC. Our investors do not seek better returns than BTC – although that may happen many days – they seek better risk-adjusted returns than Bitcoin.
We then asked him questions that explored everything he discussed and more.
Related: Invest in Cryptocurrency Safely
Do you take advantage of multi-dimensional analysis and nonlinearity, or is it merely the challenge?
Yes, we exploit rather than just overcome nonlinearity. All conventional AI systems have limitations, so we use a proprietary kernel science to better measure risk.
Nonlinearity and “many dimensions” analysis are crucial.
We have extensive experience using “space determinate” approaches, and we prefer a Bayesian approach. We measure information entropy; a statistical, not a spatial, metric.
What kind of AI/ML do you use, and why?
There are indeed myriad AI/ML technologies – neural nets, random forest, cluster analysis, information entropy, large language models, semantic models, and on and on. They fall into three broad categories: conversational interface AI (ChatGPT), parametric prediction systems that predict a number in a range (“the temperature tomorrow will be 73 degrees”), and classification systems (“the weather tomorrow will be hotter than today”, or “tomorrow will be really hot!”). We use a sophisticated classification system called a Support Vector Machine. This is an established science but a very complex technology. Our proprietary SVM is vastly capable for our purpose.
What is different about how you use AI/ML in comparison to other firms using the science?
Virtually all less experienced portfolio managers use AI to seek returns while using traditional methods to predict risk.
They will usually fail (luck is always a factor!) because (a) returns are virtually impossible to predict in complex systems, and (b) short-term risk (and change in risk), while easier to predict, cannot be predicted accurately for active trading by using traditional approaches.
How do you recognize and assess risk?
By defining “risk parameters” we can determine the degree to which they are present in price action right now. How exactly we do this is proprietary but since complex systems oscillate between chaos order, and the impact of the oscillation (sometimes called “perturbation”) is visible in security risk. We use the oscillations to predict risk in the first and second order.
How do you select and weigh exposures to construct a portfolio?
We structure our portfolios from the set of exposures identified as having lesser disorganization. Selecting all such securities with equal weight would result in a haphazard array of returns despite minimized volatility; resulting in minimized return. Instead, we select a subset expected to deliver a return in the direction required by the portfolio. Without going into detail, the portfolio is implicitly weighted to minimize portfolio volatility.
What is the range of instruments and markets your science can assess?
Our approach allows us to measure the amount of disorganization between and among markets as well as within a market. In the crypto ecosystem, we can use this capability to invest across the sub-markets (such as NFTs and new trading instruments).
Can the science be used in traditional markets?
Perhaps. We are testing it in US equity markets, and will eventually explore international equity markets. In time, we may be able to invest not only within these markets but across them, to provide risk-aware, risk-adjusted returns that satisfy institutional and family clients seeking results from conventional markets. However, as scientists, we are skeptical by nature and empirical by practice; traditional markets represent challenges (such as market closures). It may take years to achieve satisfactory results in new contexts. In the interim, we apply our science and our practices to the challenging crypto markets, achieving long-term portfolio performance competitive with Bitcoin, at far lower risk, by investing in a risk-first return second process.
Will AI trading systems take over?
The truth is automation is here to stay, and with the right hands, such ecosystems will benefit ecosystem players and the industry as a whole.