AI vs Humans in stock market investing - Can humans compete with AI?
Brief history of evolution of machines
In 1997, IBM's supercomputer Deep Blue defeated the world chess champion Garry Kasparov in a six-game match. Deep Blue was a specialized machine that used hand-coded rules and brute-force search to evaluate millions of possible chess positions per second. The defeat of Kasparov was a landmark event, as it showed that machines could surpass human intelligence in complex domains such as chess. However, Deep Blue was not a general-purpose AI system, and it relied heavily on hand-coded rules and brute-force search. In the following decades with increase in computing power and amount of available digital data, AI research progressed rapidly, with advances in machine learning, natural language processing, computer vision, robotics, and other fields. AI systems became more capable of learning from data and discovering new strategies without human intervention.
In 2017, a new milestone was achieved when AlphaZero, a program developed by Google's DeepMind, taught itself to play chess (as well as Go and shogi) at a superhuman level by playing against itself millions of times. AlphaZero did not use any human knowledge or rules, but instead relied on a deep neural network and a reinforcement learning algorithm to learn from its own experience. AlphaZero demonstrated a novel and creative style of play that amazed chess experts and revealed new possibilities for the game. It also showed that AI could achieve general intelligence across multiple domains by using self-learning methods. The evolution of AI from Deep Blue to AlphaZero reflects the increasing autonomy and generality of AI systems, as well as the potential for AI to surpass human intelligence in more domains in the future.
ChatGPT which was made public in November 2022 is a conversational AI model developed by OpenAI that can generate natural and engaging text responses to any prompt. It is trained using reinforcement learning from human feedback and fine-tuned from a large and advanced language model. ChatGPT can answer questions, converse on various topics, and generate creative content. ChatGPT has amused the world with its impressive and surprising performance in natural language processing.
AI vs Humans in stock market investing
While AI is fast evolving and is surpassing human intelligence in various domains, in this post I can think of four drawbacks in AI when it comes to stock market investing:
Drawback 1 - Garbage in Garbage out
AI systems do not work on their own completely. Prompts are sets of instructions or guidelines that are used to guide AI systems in making decisions or performing tasks. They are designed to help the AI system understand what is expected of it, and how it should go about completing the task at hand. By crafting effective prompts, you can ensure that you get efficient outcomes from AI that meet your needs and expectations. Prompts are provided by Humans. Accordingly, AI can give effective output only if the prompt is effective. So, if the input given is garbage, then output will also be garbage.The efficiency of the AI model depends on the underlying Human working on it.
The weakest link is the underlying human giving prompts. Accordingly, AI has to finally rely on human giving prompts for arriving at stock market investing strategies. Hence efficiency of the investing strategy depends on the human providing inputs.
Drawback 2 - Whole is not greater than sum of parts
Conventional wisdom says that the whole is greater than the sum of its individual parts. For example it's difficult to break a bunch of sticks when tied together whereas the sticks can be broken easily individually.
But in stock market investing, if the majority of the investors follow a similar strategy, then they will become the market and start representing the market. Returns earned will automatically become average returns. Instead to beat the market, we need to think unconventionally or take unpopular strategy.
Even if an AI comes up with a strategy and becomes popular, then if the majority of the investors start investing in the strategy, then returns will automatically become average as they start representing the market. Market is a self correcting mechanism.
There is no single strategy which works for the majority of the investors. If it works, then it automatically becomes an average return earning strategy.
Whole is not greater than the sum of individual parts in the stock market. Instead, the whole is the average of individual parts.
To beat the market, we need to remain small and not be big enough to represent the market. We need to follow the unconventional or least possible strategy.
Drawback 3 - AI is a black box
A strategy developed by AI is like a black box. Due to various reasons like complexity and uncertainty of the market, black swan events or sudden shocks, inaccurate data, etc, even if a strategy is developed by AI, we may not have confidence that the strategy will work. Wealth creation through the stock market is not overnight and instead done over a period of 20 to 30 years of working life of a person. Generating returns in those 20 to 30 years is critical. Lacking confidence in an investment strategy will be a big deterrent.
For example, Nikkei 225 (Japanese index) gave per year returns of 12.5% from 1961 to 1991. But from 1991 till 2023, it gave negative returns. An AI model trained on data provided between 1961 to 1991 could not have predicted the next 30 years outcome.
COVID-19 is another example which disrupted the financials of various companies. An AI model trained on data between 2009 till 2019 could not have predicted the COVID-19 impact.
Drawback 4 - AI works on feedback
AI works on feedback mechanisms. For example, Google search results work on feedback provided by users in the form of clicks i.e. whether we are clicking a link on the 10th page or the first page, etc. Similarly, Amazons recommendations for products works on feedback in the form of purchases made by user, searches done by user, etc.
In stock market investing, what can be the feedback?
Backtesting models - Back testing means testing a strategy on past data. Back testing may not work like we have seen in the case of Nekkei 225 or COVID-19. Testing on past data may not work in the stock market; there is no correlation with how the future pans out.
Future performance of the strategy- Future performance of the AI strategy can be a feedback for AI. But how long does AI need to observe the future performance for? Should it be few months or should it be multiple years? Volatility dominates shorter tenor performances. Optimum tenor will be 10 years to mitigate volatility. After 10 years of feedback, if the strategy developed by AI did not generate market beating returns, then do we need to wait for another 10 years to improve the strategy?
We cannot rule out AI completely
While I highlighted some of the drawbacks of AI with regard to stock market investing. Investors can make more informed and efficient decisions using AI. AI can help in expediting their research and provide timely alerts/updates upon which investors can make a decisions.
If you want to learn more about how to use fundamental analysis for stock market investing, don’t miss out on my newsletter. Every Sunday, I’ll send you a post like this one with valuable insights. To subscribe, just enter your email below. And if you know anyone who might benefit from this newsletter, please share it with them too. Thank you for your support!
I have very different thoughts. Let us speak on phone sometime :)