I’ve seen a few forum discussions about robo-advisors like TrueWealth or Selma, but what about other AI tools (e.g Custom ChatGPTs, etc)? Are you using them to analyze your portfolio, get advice, etc?
Which tools, and prompting approach, have you found useful?
Layman answer, but no, at least for the advice part. Wisdom of the crowd is a terrible way to handle investments and trading decisions, you want to either not trade at all (buy and hold very broad diversified fund, in which case there’s only one decision to have, which is your asset allocation, at the start, for which I’m not convinced publicly available AI is really good) or be a contrarian.
Portfolio analyzing could work if you know what you want to have data on. You’d still have to have access to longish series of data so I’m not sure the gain would be very big vs already existing models but it could allow to run up to date analyses pretty quickly. On a personal level, I feel that personally running the analysis is part of understanding the situation and I put a lot of value in personally doing things vs delegating them to AI or an intern but my brain is wired a bit weirdly and that may not apply to everybody.
Same here. I see it a bit like going to the gym to stay fit vs. taking weight-loss drugs. That said, I could see how AI could eventually deliver value for investors if, at some point, it is able to compile and condense information about companies and entire industries, and ideally find correlations between markets and events that would simply be too complex for me to figure out myself within a reasonable amount of time. After all, the advantage of AI should be its ability to turn big data into useful information that gives me an edge. But we’re still very far from that.
There is one thing that neither AI nor natural intelligence can do: foresee the future.
The more tools we use the bigger the illusion that we can. But it is and stays an illusion.
The best we can do is calculate probabilities from past events. Many tools, including AI, can help with that. But every new event will be different.
Have done some interesting research into Kelly sizing with ChatGPT last year, it was 100% on point and able to size things appropriately. That was many models ago and from what I gather using 4o-mini near-daily, the present ones will be even better at it.
Good stuff, we’ll see massive more improvements in the next five years with more specialist applications wired to private, long run time series databases with stock data at tick level, I’m super optimist about this.
I would feel like the village idiot if I invested in what AI said and then it underperformed
Isn’t chatGPT simply reading what others are saying and providing a summary?
Or we believe models have reached a level to do an actual analysis of economic variables , valuations etc and recommend a portfolio?
It likely does have some internal representation of how our reality functions. It is not efficient to do it in more verbose ways (e.g. represent all the training data exactly and manipulate it with explicit if/else logic).
It can solve problems it has not seen before (accuracy diminishing the less familiar with the topic it is).
You absolutely can construct AI systems capable of rather advanced reasoning. But it is still far from perfect and really expensive.