Rise of The Machines

Howard Siow

No, we’re not all going to die. Yet. Artificial Intelligence (AI), or more specifically machine learning, is hardly a new idea. So why the sudden arms race amongst investment managers to include AI in their investment decks?

When Dr Desmond Lun (an MIT PhD) and I founded Taaffeite Capital, we planned to avoid using the term AI, because we felt it would feel too much like science fiction for investors. How the world has changed!

There is a vanguard of Australian investors who see the AI arms race taking over financial markets.

Quant funds like Renaissance Technologies, Citadel, DE Shaw, Two Sigma, etc have never really been offered down here. Australia is the third largest market in the world for managed funds, but until recently, raising capital in USA, Europe and Japan was far easier and Australia was lamented as a market that likes to wait for others to be first.

What is Quant investing?

Technically all investing is quantitative. Macro traders interpret economic data to forecast future asset prices, and value investors interpret market and company data to forecast future company prices. But quants use computers to systematically identify and take advantage of profitable patterns in data.

Technical analysis (e.g. head-and-shoulder patterns) was very popular in the 1980s, however these simplistic easily-described patterns became less useful once computers became ubiquitous in the 1990s. Then there was the charge of the momentum or trend following strategies, which now account for over 70% of Quant AUM. Trend and momentum strategies are attractive because they seem to make sense to investors, and have a great sales narrative about how they manage risk. As AUM grew for trend strategies, the funds raced to find PhDs to use more sophisticated techniques to data mine for parameters for their trend based strategies.

Since 2015 this crowding into trend patterns has caused returns to suffer. They have used AI in their marketing presentations, referring to things like neural networks, but often it is just the use of these techniques to data mine for trend based patterns.

The real opportunity for AI is vastly more ambitious.

What’s the opportunity?

The real opportunity is to develop general learning systems that are not focused on specific, known patterns. So instead of just finding trend-based patterns in market data, it is looking for all profitable patterns in data. That is, you are no longer limited to profiting from simple patterns that you can easily describe and understand.

The hypothesis is that the price of an instrument in the near future is some kind of function of historical and available information. The key to deriving this dynamic function is finding the predictive or causal relationships in huge data sets of financial data. For example, Bridgewater uses 8 factors in their core strategies, however what you want to use AI for (if you’re able) is building a function that uses unlimited factors.

Another way to think about it is your personal future has some relationship to your past and current circumstances around you. The most recent past is most predictive, however even your childhood has some bearing on your future. The key is finding all these causal or predictive relationships, and creating a probabilistic forecast of the future, and then finding a way to “bet” on that in a way that is resilient to tail risks.

Training an AI system to find these very obscure and subtle relationships in large data sets is extremely difficult and is driving the AI arms race. However, like getting an AI computer to play Go or Chess, it takes enormous skill to design the AI computer.

Back in the 1980s many people in academia felt a computer should be able to beat a human at Chess, because despite human ability to innovate, a computer should be able to do specific quantitative tasks better than humans. Computer Chess programs in the 1980s were laughable, they had some success in the 1990s against Kasprov, but it wasn’t until 2006 that it was accepted in the academic and Chess community that no human player can beat a good Chess computer.

I should be clear that under time constraints a computer with no human intervention works best, where human intervention can easily make blunders to weaken the system (and if you can assume trading decisions have time constraints don’t believe all this human + machine is better than machine only and quantamental sales pitch nonsense).

What does a real AI fund do?

Dr Lun has spent 15 years pioneering, at the forefront of academia, developing AI techniques to be able to predict the behaviour of large complex interaction networks, like biological systems. He has applied these techniques to financial markets with great success. The original problem he tried to solve was, rather than buying and holding a stock, like BHP Biliton, can you rebalance that position daily to significantly enhance the returns and reduce the drawdowns or negative volatility? The end result was the TCM strategy, which periodically rebalances a long-short basket of global equity and fixed income indices, with the objective of generating significant returns for investors particularly when markets get volatile rather than just passively holding a basket of indices.

"Our long-short quant hedge fund, which I co-founded with my Melbourne Grammar mate Howard Siow, is an outgrowth of my academic research on using artificial intelligence in computational biology," Lun says in an interview with the Australian Financial Review while in New Jersey.

This liquid alternatives investment might add considerable value to your portfolio if you think we are getting late in this current cycle.

We live in terrific times, when a few people in a dorm room can get together and develop technology to disrupt industries. People with the determination and AI knowhow have this opportunity in financial markets now more than ever.

Disclaimer: This article is not investing advice and all investments can lose value, including TCM.


Comments

Please sign in to comment on this wire.