AI investing has captured the fascination of finance experts for several years. The appealing idea of a computer analyzing markets and identifying trends 24/7 is only matched by many investors’ reservations about giving control of their money to AI, especially in a market as volatile as the crypto space. However, the benefits of AI trading are numerous.
As humans, we simply cannot apply complex trading strategies, we struggle to identify when to buy and when to sell, and we can only work for part of the day. AI trading solves these issues, as it can process and execute complex trading strategies, be strategic with entry and exit points, and works around the clock. AxionV is an AI crypto fund that is undertaking an ICO starting September 16. They’re in their “Momentum Stage” and are harnessing the power of AI for cryptocurrency trading.
To acquire some background on why their approach may work, I explored the current research on AI in finance. Many believe that AI investing methods only relate to the automation of asset selection. However, the uses of AI goes far beyond that. According to “Artificial Intelligence in Financial Markets” by Dunis and Middleton, AI in investing has at least three major use cases all statistically verified by academic research of artificial intelligence algorithms and neural networks:
- Price prediction
- Portfolio management
- Risk management
All three aforementioned elements can make an enormous difference on a fund’s performance. According to Peter Boroykh, AxionV’s Head of Algorithmic Trading Strategies and author of “Blockchain Applications in Finance”, the firm plans to implement AI algorithms in all of the three aforementioned areas of successful asset management.
The Bitcoin split will have massive ramifications on the crypto world for years, and you can read about how and why it happened here.
So far, the research conducted bears this out: According to Eurekahedge, absolute returns for AI hedge funds are clearly higher than for an average hedge fund. Between 2010 and 2016, the average hedge fund returned a staggeringly low 4.3% to investors on an annual basis. The average AI fund however, returned double at 8.6%. (Figure below).
Historically, AI funds underperform in the long run, a well-known critique of the strategy class. However, it’s been found that Sharpe ratios, a well-known measure of performance, only converge for longer-term periods, while short-term AI hedge funds are on a level higher relative to an average hedge fund. According to Dr. Marco Pereira, a lead researcher at AxionV and author of “The Hypergeometrical Universe”
“We’ve seen historically that AI outperforms on measures of performance such as Sharpe by 2:1 over the short run of 1-3 years. Only on a longer timeframe has the advantage been diminished.”
| AI | Hedge Fund Index | |
| 1-year | 1.51 | 0.64 |
| 3-year | 1.53 | 0.89 |
| 5-year | 1.28 | 1.41 |
Pereira further states:
“This research means that AI gives more consistent returns. The reason why it’s important is because if there is bad performance in the short-term, even if the average performance turns out to be above average, investors are unlikely to stick to the fund. So, investors run from one hedge fund to another hedge fund, constantly underperforming the hedge fund index itself.”
This conclusion is backed up by research from JP Morgan Asset Management, which found that the average hedge fund outperformed the average investor by a significant factor due to their proclivity for redemptions and a double-down on timing the market.
Given clear outperformance on both absolute and risk-adjusted bases, what is the difficulty relating to quality, bullet-proof AI algorithms? The answer is that it is often difficult to accumulate a sufficient amount of data needed to create an AI system that could surpass human performance. The data in finance especially can be extremely sparse or noisy, and it may prove difficult to understand fully the meaning of various metadata. To address these challenges, researchers are investigating ways to combine expert rules with statistical learning from data.
Blockchain has wide-reaching applications, but these 4 industries will be impacted the most.
One solution is to apply “transfer learning.” This seems to be the favored approach by AxionV in their methodology. The firm will utilize transfer learning, which teaches computers to take learning from an AI model in one area and apply it in another area.
“Many financial firms already compare cryptocurrency price behavior with the behavior of the dotcom bubble. But AI can take experiences learned from other markets with decades of available data and draw parallels to attempt maximizing returns, while minimizing downswings,” says Borovykh.
This method is particularly useful when extending models from data-rich domains to data-poor domains, and where there’s a customization requirement of general models in areas of sparse personal data.
However, if there is a lot of data in the market (like equities or forex), the market is much more efficient and has a much lower profit potential. Borovykh states that “AxionV takes experts and their models from mature markets that know the tricks there, and places their theories into a new market (cryptocurrencies) with not much information available and gets an outsized return on investment.”
AxionV says that it currently favors neural network use, or more specifically Adaptive Neural Networks (ANN).
“Due to their adaptive nature, ANN can provide solutions to problems such as forecasting, decision making and information processing. In recent years, ANN have proved to be a powerful tool for handling dynamic financial markets in terms of prediction, planning, forecasting and decision making,” Pereira stated.
As an extension of their technology, AxionV is preparing to open an exclusive AI Lab as a center for AI based learning and development to study the cryptocurrency market. According to Borovykh,
“We are currently in partnership talks with the strongest engineering universities in engineer-heavy countries.”
Despite the rigorous work, AxionV seems to agree that human intervention will still be needed to help sort out liquidity, upgrade AI algorithms and help provide the systems and clients with a larger, more appropriate data set of crypto assets.
The whole concept, especially on blockchain, is to create something no other human—and no other machine—is currently doing.
Where will blockchain be in 3 years? You can read the answer to this question from a panel of experts here.
