I spent more than 15 years sitting at trading desks and treasury tables, and that’s where I learned to respect a rule nobody teaches at university: 60% in stocks and 40% in bonds isn’t a brilliant strategy, it’s a truce. It’s what the market agreed to do once it accepted that nobody —not even the most sophisticated manager— knows for sure when to move from one side to the other. That truce, the same one that holds up a large share of the world’s pension funds, Colombia’s included, has just received the first serious evidence that an algorithm can break it better than a human can.
What’s real
JPMorgan revealed this year that its researchers built artificial intelligence agents that rotate between stocks and bonds depending on market conditions, and tested them against two decades of historical data. The best system beat a traditional 60/40 portfolio by 0.7 percentage points a year —with lower volatility, not more risk— (Bloomberg Law). We’re not talking about a chatbot offering opinions on stocks: we’re talking about a quantitative system with decades of methodology behind it, now paired with the ability to read and react to regime changes faster than an investment committee. And the timing couldn’t be more fitting: we expand on it today in Daily Briefing #0041, on the same day the 10-year Treasury yield touched its highest level in 24 years and the Nasdaq closed at a record.
What nobody tells you out loud
A twenty-year backtest isn’t a promise: it’s a snapshot of a past that already happened, tuned —sometimes without the researcher even noticing— until it looks good. No model trained on historical data has “lived through” a government shutdown that switches off official statistics in the middle of a Fed cycle, or a Treasury yield at generational highs on the same day stocks set a record. That is, literally, what’s happening this week. The most disciplined algorithm in the world is only as good as the market regimes that have existed before; when a new one shows up, nobody —not the model, not the manager— has a real edge. And there’s a second, more local item on the bill: Colombia’s mandatory pension funds operate under an investment regime that the Superintendencia Financiera reviews and approves line by line. A dynamic allocation run by an AI agent, which can change its stance several times a week, still has no comfortable place to fit within that regulatory framework —however excellent the model may be.
The opportunity, properly understood
None of this means you should ignore the result. It means you should use it for what it is: evidence that fixed asset allocation is no longer the only serious option, not that it should be handed over blindly. If you manage your own portfolio, the practical move isn’t chasing the next robo-advisor with “AI” in its name; it’s asking whoever manages your voluntary pension or balanced fund whether their strategy follows a fixed rule or responds to market conditions, and with what discipline. If you are that manager, JPMorgan’s experiment is the signal that it’s worth setting aside budget to test —not to adopt yet— this kind of tool as a second criterion, never as the only one.
Technology informs; judgment decides. An algorithm can beat the 60/40 over twenty years of history that has already happened. The question you should be asking today is a more uncomfortable one: do you know what rule your pension money is following, right now, when the market gets as strange as it did this week?
If you want to follow this story in real time —including where the Fed moves on October 27 and 28, and what Treasury yields do in the meantime—, subscribe to the Daily Briefing. We cover it there every day, before it becomes a headline.