Investment management across the board has benefitted from advances in machine learning and artificial intelligence (AI), but those which use quantitative-based processes have arguably benefitted more. Better technology and computing have allowed them to analyse more and larger amounts of data and, as a result, to more effectively identify patterns and make predictions.
It is no surprise therefore that the focus on machine learning – and its implications for quant-based strategies in particular – is increasing. Some investment managers, keen to move into the realm of liquid alternatives, may even believe that the power of computing and artificial intelligence is such that investing vast sums in computing technology and hiring mathematicians to analyse data is all that is required to perform.
The reality is not so simple. Machine learning and the ability of AI to interpret and analyse data are a great step forward, but as with any complex area of human endeavour, there are few quick wins. Long-term commitment, investment in both people and infrastructure, experience, trial and error and hard work are required, and these things cannot be achieved or replicated overnight.
And most importantly, human input remains key.
Artificial intelligence is no replacement for the human mind
With all the talk about artificial intelligence, big data and machine learning, it’s easy to think that these concepts are new. In fact, the majority of artificial intelligence techniques we use today, like neural networks, support vector machines and random forests, has been around since the 1990s, which means it is now coming up to its 30th anniversary. Linear forms of support vector machines date back even further to the 1960s.
AI has many applications in financial services, but at the same time, it’s important to recognise that machines, including AI machines, are very efficient tools for doing one thing, but are not yet close to replacing humans for higher level functions. For investment managers, new AI techniques are a welcome addition to existing quant toolkits, but they cannot replace humans in terms of initiating, guiding and providing context.
At the same time, techniques such as deep neural networks, which fall into the machine learning subset of AI, are a step forward for quantitative-based managers, like CFM, because they allow for better predictions to be made from data. This is because machines are now able to learn by practice and feedback, in the way a human brain does, which means it can both calculate effectively, but more importantly, adjust calculations when market characteristics change.
Liquid alternative strategies have been enhanced by advances in AI
In part due to the proliferation of data, and the ability of machines to analyse it, there has been a rise in the number and sophistication of liquid alternative strategies, including alternative beta strategies, over the past few years. These methodical rule-based strategies are designed to provide access to the portion of returns attributable to a market’s overall systematic risk, or beta. They are often described as aiming to offer a transparent way of accessing hedge fund returns which were previously viewed as alpha, but which can actually be attributed to systematic risk. Because they tend to be non-correlated to major market indices or traditional asset, they can provide effective diversification, downside protection and risk-return efficiency in a liquid, low-cost and transparent way.
As more investors begin to see the potential benefits of alternative beta strategies as a mean of enhancing overall portfolio diversification, pioneers in the field, like CFM, have been at the forefront of harnessing AI and machine learning to produce better outcomes for investors.
Better data and better analysis can lead to better outcomes
For example, due to a long-term, substantial investment in infrastructure and human capability, we are now better able than ever before to analyse vast amounts of data that might impact the future price of a financial instrument. These include:
– Price data at a granular level from exchanges all over the world;
– Fundamental data such as financial results, sector and macro-economic trends; and
– Alternative data such as investor sentiment derived from sources as diverse as social media, the state of crops as gleaned from satellite images and usage trends of geolocation data.
However, it’s important to understand that developing new techniques to improve returns is a difficult and time-consuming task. It is far more complex than simply feeding data into a machine and coming up with an answer. In fact, algorithms require continual checking to monitor their performance and then to measure performance in order to decide whether they are performing well, or need to be modified or even switched off.
The challenge for investors looking to invest in technology is that trading results improve only gradually, and require ongoing expertise which takes years and decades to achieve.
Conclusion
AI is certainly a powerful tool for an asset manager, especially for those in the field of quant-based strategies. However, successful implementation requires an investment in people and infrastructure, hard work and patience. And the benefit of trial and errors gleaned over many years.
Machine learning in practice: Using neural networks to improve trading execution
A decision has been made to purchase 1,000 shares in a company today based on a view, generated by an alpha strategy, that the company’s share price is likely to increase over the next week.
The question now is whether all 1,000 stocks should be purchased at once, at the current price, or whether waiting for a lower price might be the better option. Or perhaps splitting the order into multiple and smaller options would be better.
This is the kind of problem which a deep neural network can solve. The order book data of exchanges from around the world, including prices, volumes and timing priorities of buyers and sellers of every traded financial instrument is available. But it is vast and continuously changing.
Algorithms developed by CFM have been trained to analyse this data and determine how to minimise the costs of executing a trade. They “learn how” to do this, through trial and errors given any kind of shape, size or dynamic of order book. And determine the best execution rules for any kind of financial instruments, including stocks, bonds, futures and options.
Once trained, the algorithm interacts with the market. It might decide that the price is likely to decline slightly within the next two minutes, so it delays executing the buy order.
Clearly, at an individual trade level, savings are tiny. But over time, shaving a few percentage points off the trading costs of a large book with a high volume of trades can significantly boost returns.
Laurent Laloux is the chief product officer at Capital Fund Management (CFM).
Disclaimer: Any description or information involving investment process or allocations is provided for illustrations purposes only.
Any statements regarding correlations or modes or other similar statements constitute only subjective views, are based upon expectations or beliefs, should not be relied on, are subject to change due to a variety of factors, including fluctuating market conditions, and involve inherent risks and uncertainties, both general and specific, many of which cannot be predicted or quantified and are beyond capital fund management’s control. Future evidence and actual results could differ materially from those set forth, contemplated by or underlying these statements. There can be no assurance that these statements are or will prove to be accurate or complete in any way. All figures are unaudited.
