Suppose you go for passive management. Currently there are about 300 index funds registered in Spain that replicate indices of both fixed income as of renta variable. The offering is extraordinarily broad: it ranges from the euro single-day interest rate index (€STR) – with a volatility close to zero – to thematic indices more volatile than Nasdaq itself. If, in addition to these transferable funds, we broaden our view towards ETFs, the offer available to the Spanish investor is multiplied almost tenfold.
The index funds with the largest volume of assets under management replicate the world stock market index and the global bond index. In both cases, the investor can contract them with the currency hedged to the euro or without hedging. And this is already an important decision: without hedging, much of the exposure will be in currencies other than the euro; but completely hedging the currency can also be very costly.
Within our own region there are funds that replicate the Ibex 35, the Euro Stoxx 50 or broader European indices. Something similar happens in Asia and global markets: in addition to indices where countries and companies overlap, there are also regional versions that exclude certain markets, such as the MSCI Europe ex UK or the MSCI World ex USA.
In fixed income, the offer is also very broad. The investor can index to public or corporate debt, choosing between different regions, countries, maturities and credit quality levels. That is to say, even when one decides to “not decide” and limit oneself to indexing, one ends up having to make important decisions, which is why it is usually advisable to seek advice or delegate it to a professional. In practice, this means replacing active fund management with delegated management of a portfolio of index funds. The great advantage, however, is that these decisions can be adapted to the particular situation of each investor.
Knowing which combination of indices should be replicated and, above all, when portfolios should be reviewed remains an active decision that should be made taking into account financial objectives and the time frame to achieve them.
At this point, it is worth asking: what if we tested artificial intelligence as if it were a professional advisor to see what indexed portfolio it would design for us?
I start by posing as an investor with a short-term vision and I asked Claude. If you can build me an optimal indexed portfolio for the next quarter. Claude responds that he needs more context – thank goodness, I think – and offers me to choose between four risk profiles, from conservative to aggressive, in addition to a personalized response.
So I write: “Earn as much as possible assuming a maximum quarterly loss of 10%.” The algorithm responds by proposing an allocation of between 65% and 68% in equities.
And I ask the reader: does a proportion close to 70% in variable income seem reasonable to you if the investment term were only one quarter? Here the well-known complacency bias of artificial intelligence emerges: that algorithmic tendency to agree with the user and validate their expectations of profitability (in this case the “maximum possible” instead of confronting them with the uncomfortable reality that short-term falls in the stock market can greatly exceed that loss limit.
To force the model to recognize that it cannot know what the optimal combination of indices will be for the future, I posed the following question: “What would you have recommended in the third quarter of 2007?” His literal response is extremely instructive: “The difference between what seemed reasonable then and what would have been optimal ex post is one of the most valuable lessons of recent financial history.”
What happened in 2008 is now history. The S&P 500 and the total US market plummeted 37%, while international stocks fell 43%. With the benefit of historical perspective, the optimal ex post portfolio to weather the storm of 2008 would have been practically unthinkable for any advisor: not a single euro in equities and an enormous concentration in Treasury bonds, which would have generated a return close to 25% in the darkest year of modern markets.
This historical simulation exercise reveals several uncomfortable truths about portfolio management. The first is the fallacy of short-term stop losses: a 10% threshold does not protect during a systemic crisis. A portfolio with two-thirds of equities, like the one that the AI quite happily proposes for the current scenario, would have lost more than 25% in 2008, pulverizing any limit in the first year. Real stress episodes always end up overflowing statistical calibrations based on normal volatilities. But investing thinking about the maximum falls in extraordinary years does not allow us to beat inflation in the medium and long term. And it is precisely in the medium and long term where the probability of having losses of that caliber fades.
In conclusion, the limits of AI are not technical, but ontological: it is not that it cannot analyze millions of data, but that the future does not yet exist as an observable reality. The events that have the greatest impact on the markets depend on future decisions of millions of people facing contexts that have not yet occurred. And perhaps that is the final paradox: if one day an artificial intelligence managed to predict perfectly, it would be enough for investors to know that prediction for it to automatically stop being fulfilled.
Marta Díaz-Bajo is Director of Strategy at Atl Capital
