14 Sep That thing called Artificial Intelligence …
I often get caught in discussions where investment process get intertwined with fancy words and phrases such as artificial intelligence, big data, machine learning, neural network for predictive analysis. Being a part of the new world order where technology drives parts of decision making, we keep a close eye in this space.
Artificial intelligence (AI) will continue to bring technological advances in various fields. Established use cases are for companies designing data management tools that deploy machine learning in cybersecurity, healthcare, IOT and autopilot systems for driverless cars. Even the CIA has been using AI for a while now – it currently has 137 AI projects undergoing that use unstructured data.
As we look for data points of valid use cased for artificial intelligence (AI), we take a cue from where VCs are investing. The better known recent investments in the artificial intelligence (AI) space are in NIO (US$600m), which designs autonomous electric vehicles, or Flatiron Health (US$175m), which aims at fighting cancer, thanks to big data. AI has also seen some failures. Auto company Pearl Automation, which raised $50M last year fell flat in Q2’17
The following table from CB insights is a map of CVC deal share in various sub-sectors.
The media rhetoric combines robo advisors and artificial intelligence and all the jargon that goes with it. However, we see no clear evidence of VC and professional investors making systematic or serious investments in firms that combine artificial intelligence (AI) in investments and fintech, though I have seen a few compelling examples in Insurtech firms, which are geared to make an effective use of big data in actuarial analysis.
We are quite certain that artificial intelligence (AI) will revolutionise the way we interact with technologies, and to a further extent the way we work and live. So, naturally, one may believe that using AI-based models to manage your investments it a great idea. Let’s explore this space.
Differentiate AI from algorithmic finance
Before delivering our thoughts on artificial intelligence (AI) for investing, we must first make clear that there exists a distinct frontier between algorithms and artificial intelligence (AI). While both seem edgy and eventually share some common features, they do not overlap. At Bento, we provide our clients with portfolios constructed thanks to rule based algorithmic finance. The model is based on fundamental research led forward-looking data, statistical analysis and our proprietary optimiser. No artificial intelligence (AI) or machine learning is involved … yet! It is something entirely different to model market returns or macroeconomic trends using big data machine learning. Indeed, in this case, you would feed it with historical data and then leave it up to the machine. By attempting to create such technology that efficiently emulates humans and then beats them at their own game, this new breed of algos may simply end up as a complex combination of all of the worst characteristics of human investors that history has seen.
Another issue encountered with AI is the data-mining bias. Data mining consists in searching extensively through historical data to find significant patterns. Such issue can result in strategies that are over-fitted to past data. Considering investments, this would result in models that exhibit strong backtested performance but immediately fail on new data. In fields where history repeats itself, this is less problematic. For instance, in chess, once an artificial intelligence (AI) system identifies the best move to make in a particular situation, it can apply the same conclusion every time this situation occurs with the same end-result. This, unfortunately, does not apply to financial markets, in which market movers vary extensively across time. It is always useful to be reminded that “past performance does not guarantee future results”.
Now here is the conundrum: some of the best investors, the likes of Warren Buffet, openly share their tricks and portfolios with no apparent damage whereas AI-driven investing, among the most-hyped trends in finance, tends to fiercely guard its proprietary knowledge. Therefore, we don’t know exactly what sort of computer-powered magic these firms are devising, although their public pronouncements suggest they seem particularly preoccupied by “beating” humans in markets. We don’t know what is the data and logic that drives the investment engines.
Among many other tricks that AI-driven investment algorithms may deploy, it seems they have the ability to crawl through millions of news articles and identify trends. English is the language of business and even here nuances of the English language are different in parts of the world. The semantics, syntax and context widely vary depending on which part of the world you are in. So how do we crawl through and look for predictive data sets from news articles? Also we need to bear in mind that only 21.84 % of global publications are in English.
Now coming to seemingly binary data points such as dividend pay-out ratios, interest rate coverage ratios can be different dependent on accounting standards.
The sophisticated and promising deep learning algorithms are truly black box, more so than the hedge funds in 90’s that fashionably claimed to be “black box” which meant that managers just wont share investment process. We simply don’t understand how these machines solve problems. Should something go wrong, we might not be able to define the problem, let alone a solution.
The concern regarding artificial intelligence (AI) lies in the fact that no research has proven the benefits it could bring to investing.
This mostly explains why our model sticks to proven rule-based strategies, which are backed by extensive research. These have consistently delivered well-performing portfolios, and are reliable.
Consequently, Bento’s proprietary optimizer that constructs our clients’ portfolios does not incorporate artificial intelligence (AI) or any other unproven theories. We stick to theories that have been put to practice and have stood the test of time, such as the Nobel prize-winning Modern Portfolio Theory designed by Harry Markowitz, and we continuously work to improve upon them in the context of client needs and market evolution. Such rule based asset allocation model go back to the initial formative years of the asset management industry.
Therefore, Bento’s bionic approach is the best of both worlds. The rule based proprietary algorithm with a human touch brings time tested and proven investment theories to individuals at low price point using technology.
As Burton G. Malkiel claimed in his best-seller A Random Walk on Wall Street, a blindfolded monkey throwing darts at a financial newspaper to choose stocks would beat an active human investor. We also believe that, as of today, this same monkey would most certainly beat an AI-driven model.
While we don’t doubt that artificial intelligence (AI)/machine intelligence will be this evolution is inevitable, we need to be mindful of the risks such unproven methods bring to our client portfolios.
Conclusion – For now, we stick to time tested theories …
After much research and analysis, we have come to the conclusion that artificial intelligence (AI) for investments is like a bathtub for my pet cat!