I was delighted to read this article by David Thorpe in FT Adviser yesterday because ever since the advent of Artificial Intelligence (AI), I have been wondering whether someone might one day use to it to trade stocks? It turns out I didn’t have long to wait, see below:
“An investment process that used artificial intelligence with no human involvement outperformed during a bear market and in the early stages of a recovery, according to an experiment conducted by Ryan Pannell, who runs investment firm Kaiju Worldwide.
Pannell said that while strategies which deploy machine learning tools have long existed, usually branded as “quantitative”, he said the limitation of these strategies was that humans tended to be excessively influenced by previous data and their own experiences, and these inputs can prevent the full value of the artificial intelligence being felt, “as what happens is the human builds a set of static rules for the AI to operate within, but that tends to mean the full potential of the machine learning is not harnessed.
The other issue is that quant strategies tend to be based around back testing, i.e. looking at how they would have done in previous market conditions, and from there, a static set of rules end up being created.
He says the key value that an AI can add is the ability to recognise patterns in how stocks are being bought and sold.
Pannell said the limitation of AI was that it "can’t creatively come up with a solution to something it hasn’t seen before, but it can adjust".
When he ran an AI programme to invest in US stocks, for example, it missed the sell-off in US bank shares caused by the collapse of Silicon Valley Bank.
He said this was because the machine could not understand the reason for the contagion impacting other bank shares, and so could not immediately understand why they were selling off, instead mis-identifying the sell-off as an opportunity to “buy the dip”.
Over the longer time period, Pannell said AI strongly outperformed in bear markets, losing around 20 per cent, when the US index dropped 35 per cent.
He believes the reason for this is that in a profound bear market, “stocks are sold off because sentiment is so negative, so even if there is no reason for a stock to fall, it will do, and that’s something an AI system can’t really understand".
"But the gains it makes will come, just as sentiment hits the bottom, and the market starts to recover, because that’s when, if a stock falls, it is easier to understand why it might be falling for the wrong reason, i.e., it might be artificial over-selling. That is really where the value is added by AI."
I thought this article was really interesting and to some extent it supports the old adage that markets are governed by ‘fear and greed’. An AI bot has no such emotions and consequently it cannot understand things such as sentiment driven selling (or buying). I always scratch my head when say, Apple has a bad day in the US and say, Tesco shares, along with most of the UK market, fall. Why are these two seemingly separate things connected and the answer is of course, they are not, it could simply be that it’s a ‘risk off’ day! The fall in Tesco’s share price in this example clearly has nothing to do with the financial prospects for the company and so how could AI know to sell Tesco when Apple has a bad day? The fact that Tesco will not always fall when Apple has a bad day further will further confuse the AI algorithm.
The result of the experiment seems to suggest however, that AI was successful, but these results can only have been achieved by ‘back testing’. This means the experiment looked at what would have happened if AI had been trading during previous Bull and Bear Markets (Bull markets occur when shares are going up and Bear markets when shares are falling). This type of analysis is notoriously unreliable because altering the time period of the test, even by just a few days, can have a profound impact on the results.
In summary, I remain to be convinced, although I wonder if AI might be a good student of form at the races? Maybe someone will try that too.
I hope you have found this interesting but, if you have any questions about this piece or any other finance related matter, please do not hesitate to get in touch.
Yours sincerely,
Graham Ponting CFP Chartered MCSI
Managing Partner
