Artificial Intelligence: Using AI in entire investment process

This article first appeared in Personal Wealth, The Edge Malaysia Weekly, on July 16, 2018 - July 22, 2018.
-A +A

Quant funds, or investment funds that select securities using advanced quantitative analysis, were the fastest growing segment of the hedge fund industry last year. As more of them expand their artificial intelligence (AI) strategies, this may pose a challenge for human fund managers in the future, according to US-based quant fund Taaffeite Capital Management.

“There are so many techniques you can use to analyse data and make 24-hour forecasts. There are just so many statistics, AI and machine learning technologies,” says Taaffeite co-founder and CEO Howard Siow.

“These are large evolving fields that people are using to find inefficiencies in the markets. We think it will be very difficult for [human fund] managers to compete against that. That is not to say humans cannot do it as well, but it will certainly be more challenging to beat a computer at a quantitative exercise.”

Last December, in collaboration with Taaffeite, Kenanga Investors Bhd launched the Kenanga Global Multi Asset Fund in Malaysia. The wholesale fund feeds into Taaffeite’s TCM Global Index Fund, which trades index futures and currencies.

Computer programmes are able to sift through large amounts of data quickly, detect patterns and take positions based on analysis rather than sentiment. Many quant funds are already using AI, but to varying degrees, with some only using the technology to assist in the information-gathering process or to create a basket of securities.

For instance, Two Sigma — a large US-based quant fund — hired an AI expert from Google Brain in April to expand its AI strategies to go beyond just machine learning and big data. Well-known London-based quant fund Man AHL began incorporating AI into some of its strategies in 2015.

Taaffeite uses AI for the entire process, from choosing investments to automated trading. “We believe that something like trading is a quantitative exercise and it is something computers should be able to do in a less biased way than humans. But it needs to be carefully designed by a human. I think that is where the industry is at the moment,” says Siow.

He co-founded Taaffeite in 2014 with Desmond Lun, a professor of computer science and plant biology at Rutgers University in the US. Lun had created algorithms to predict how cells would behave for research purposes. Utilising his experience in building models, the duo started to build an AI model that could provide consistent returns.

“Anyone can claim to use AI but ultimately, you need to generate returns. So, what we have been working on for the last three to five years is proving that and using our own capital,” says Siow.

“I seeded the fund. We started to develop a strategy and refine the system to what it is today. We are one of the few firms that not only say we do AI but have actually got it to work. We are providing fairly consistent returns at 20% per annum.”

After building the model, the team recruited experienced traders. Siow says all of the company’s chief operating officers have been trading for more than a decade. For instance, Martin Redgard — who joined the team as chief investment officer this year — started trading in 2002 and had run his own proprietary trading firm in Stockholm.

“I believe that is where you start having an understanding of how the markets work. When you have a trading technique, especially in markets that trade 24/7, you would be better off in the long run doing what you know works, but automating it to make it quantitative. We started trading manually 15 years ago, but we moved into coding these strategies very quickly because I had a firm belief that we would have an edge in doing so,” says Redgard.

According to Siow, Taaffeite has a five-year track record and currently has US$35 million under management. Its target was to achieve a return of 15% to 20% per annum after fees over a two to three-year rolling period.

Siow says the fund has seen a return of 20% per annum since it launched. The company’s clients include a Japan-based pension fund and a few institutional investors in Europe and the US.

“These are statistical techniques refined over many years to get them working. Our idea is to have a portfolio where the long and short positions are negatively correlated and are able to offset large left tail events. It is always looking for opportunities,” says Siow.

 

Looking for mispricing

The company has adopted a strategy called the TCM Liquid Alpha programme to power its fund. It scans the global markets and gathers price data from equity, fixed-income, commodity and currency markets on a tick basis.

“The programme looks for short-term patterns and relationships in the data to make some kind of forecast of what the global markets will do over the next 24 hours. It holds a basket of global indices and rebalances the portfolio once a day based on the 24-hour forecast,” says Siow.

After analysing the relationships between all the data sets, the programme detects any divergence from the norm that could present an arbitrage opportunity among the financial instruments. “For example, it may see the S&P 500 and FTSE 100 having some kind of relationship over a long period of time. But it won’t be just these two instruments. It will be a very long formula with 200 variables to predict what the S&P 500 will do in the next 24 hours,” says Siow.

“But the function is so sophisticated that only AI can determine what the actual formula is. So, that is what it does: It looks at all your previous price actions, develops the actual formula and makes a forecast around the S&P 500.”

With more information, the programme is able to make better predictions because it understands the movements and relationships between the different asset classes on a historical basis. “The system is designed not to have an opinion. It only looks at the statistics and makes a decision based on that,” says Redgard.

One of its best trades was during the Brexit referendum, when the strategy went short on the FTSE, pound sterling and euro stocks going into the vote. “It is fairly typical for strategies like ours to do well when there is a large mispricing that results from uncertainty or stress in markets. When those kinds of mispricing correct, those are typically the periods in which we do well,” says Siow.

Periods of uncertainty tend to result in the mispricing of assets, thus presenting opportunities for quant strategies. But there are different kinds of uncertainty. The strategies do not perform as well during unpredictable events such as when US President Donald Trump’s tweets affect the markets or the 9/11 tragedy.

In the case of Brexit, there was a long period between the announcement and the vote. “People knew there was going to be a vote, so it did not create any gaps. While I do not have an opinion on this, Trump’s tweets are often about some completely new information that no one was prepared for — no one could guess that unless it was pure luck. So, no one could make money from these market gaps,” says Redgard.

Not every event during periods of uncertainty warrant a trade as well. “If there is a large mispricing in the market due to heightened market uncertainty, the programme may not be able to take a position because the potential return does not justify the risk. Preserving capital should always take priority and over the long run, it is better to target high-quality opportunities with high risk-adjusted return potential,” says Siow.

That kind of market uncertainty has been observed this year. According to the Eurekahedge indices, while the AI hedge funds outperformed their conventional peers last year, their returns this year are in negative territory compared with the conventional hedge funds’ 0.11%. This comes on the back of a potential trade war, geopolitical tensions and rising interest rates.

Siow says short-term events can impact returns, but the key is having a robust system. “Short-run returns are effectively random, meaning that there may be periods where you may have a string of losses. However, the expectation is that if you are careful with your risk management, you should do extremely well from the greater activity in the market.

“The assumptions that the system makes are also important. If an investment manager does not have a significant edge by being able to generate alpha in excess of 10% or good risk management, greater market volatility could be very dangerous for the strategy.”

He emphasises that the system is not predicting whether events such as Brexit will occur, but identifying mispricing in the markets. “There are some events that are just unpredictable and those kind of black swan, left tail large events can hit a portfolio in a nasty way. But what you want to do mathematically is create a portfolio that is robust enough for such last-hour shocks,” he says.

Taaffeite does this by testing its strategies against out-of-sample data. That is because Siow believes that the managers who are developing strategies should be careful to not over-fit or be over-trained to historical data.

“That is like a strategy telling you to buy Apple, Amazon and bitcoin 10 years ago. We ensure that our strategies are robust when they are tested against out-of-sample data,” he says.

“For instance, a strategy for trading GBP/USD should be tested against the last 20 years of gold prices. We also test the strategy against a number of ‘what if’ scenarios to see whether there is a general negative skew or short left tail of returns.

“It is also important to have strict concentration and tight stop-losses on positions. As hedge funds are typically 5 to 10 times leveraged, these are general risk management parameters that institutional investors expect to protect against unknown tail risks.”

 

Misconceptions about AI

A common question Siow gets from investors is whether he is able to profit from every kind of big market movement. “[My response to that is] this is not a crystal ball. Our system is not able to predict every single event that will occur. All we are saying is that our system is like having a coin that has a bias on one side, and we know that if we repeat it over a long period we will be profitable,” he says.

“Our system is able to create a probability forecast that gives us an edge and we repeat that over a long period of time. That is why I say to investors, whether I can predict the most recent event occurring is immaterial to us. Stay invested for one to three years and allow the edge to repeat itself.”

Since quantitative trading strategies are built on complex algorithms and systems, it is essential to choose a firm that has a robust model, he adds. “Designing a quantitative system is very difficult to do. The normal problem that occurs is people over-optimise the strategy or they start extrapolating the past, like misusing back-tests.

“There was not enough discipline in the way they were built. I think a lot of it just comes down to experience. You need to get people who have deep experience in designing systems.”

Taaffeite does not practise high-frequency trading (HFT), which was blamed for the flash crash in 2010. HFT is an automated trading platform that carries out transactions at extremely high speeds. It uses algorithms to detect emerging trends and transacts millions of orders in seconds.

“There are also theories that it was derived from a fat finger order placed by one human being, which speaks to why you should not trade as a human more than a machine. We do not derive our returns from speed, we derive them from a robust strategy, in the sense of finding opportunities,” says Redgard.

 

 

Different from other AI funds in Malaysia
The Kenanga Global Multi Asset Fund feeds into the TCM Global Index Fund, which was established in 2013. Kenanga Investors Bhd CEO Ismitz Matthew de Alwis says it chose this strategy to provide its clients with an alternative investment.

“We believe that the strategy — once we have established a longer period of performance — will eventually be the way to go, where clients have a choice. Do they want an emotionless investment? Would they like to put their money in something that invests across multiple asset classes?” he adds

“We are fairly confident that this is the direction of the market. The local market has matured in its own way. Over the years, there have been just so many strategies that you could employ in the market.

“So, I think it is an opportune time for a different style of investment. We expect some investors to wait and see how all this turns out, but I think people like this type of emotionless investment in a multi-asset strategy in the current environment.”

There are several other AI-powered funds in Malaysia, but they are equity funds. Kenanga’s solution is the first multi-asset strategy fund in the country, says De Alwis.

He points out that the fund is thoroughly driven by AI instead of just using machine learning to select stocks. Quant strategies use complex mathematical models to detect investment opportunities.

“It should not be mistaken for machine learning. A lot of people say they are doing AI, machine learning and so on, but if you look at it, a lot of it comes out of a quant strategy. When we do back-testing on data and everything else, there should not be any intervention by humans,” says De Alwis.

The Kenanga Global Multi Asset Fund is only open to sophisticated investors. It is suitable for those with an aggressive risk appetite, according to the fund fact sheet. There are two share classes — in ringgit and the US dollar. The minimum initial investment amount is RM20,000 or US$5,000. As at last month, the fund’s assets under management stood at RM25 million.

According to the fact sheet, the fund achieved a return of -2.4% after fees and expenses in May. The most significant gains came from the FTSE 100 and the Dow Jones Industrial Average. Throughout May, the target fund was short at various levels in the FTSE 100, which reaped significant profit as the index declined significantly in US dollar terms. The most significant losses were from The Euro Stoxx 50 Index, Euro-Bund and Euro-Buxi futures.

“This is an interesting time, with many things going on. I am not too worried even though it went down a couple of months ago. We are not overly concerned because it did recover. The pattern seems to be very consistent and consistency is the first thing you want to look at,” says De Alwis.