Another day and another ChatGPT story lands in my in-tray. This time it is the announcement from JP Morgan that they have developed an AI model that deciphers the Fed speeches for possible trading signals.
How times have changed, back in the day during an episode of “The Hitchhikers Guide to The Galaxy”, Ford Prefect suggested that if you put a babel fish in your ear then you will be able to decipher any language. In many ways this is how AI is presented to us today and perhaps not so surprisingly raises many questions.
What are you going to do about it?
Let’s just suppose for a minute that with an AI model one knew with 100% certainty what the Fed’s intentions were when they next meet to set interest rate policy. That’s all well and good, but what about investors, not everyone will take the same trading view and very quickly one realizes that as an investor one certainty doesn’t eliminate all the other uncertainties.
The role of an ETF Strategist is to figure out which strategic asset allocation model to select for the core and which strategic and tactical re-balancing decisions to make. What exactly is a strategist meant to do with JP Morgan’s ‘Hawk or Dove’ trading signals?
Remember the days when we recorded an Excel macro?
The issue of how to use trading signals to manage a multi-asset portfolio of ETFs, is one that we have spent much time on over the last 15 years, and I can say for sure there are not any simple answers. There are simple things that one can do, but every now and then these will get you into trouble. Ergo, I’ve lost count of the number of Model Portfolio Services suffered badly during 2022 when most traditional models simply didn’t work.
Don’t forget to diversify your portfolio.
To get to grips with the way one might go about employing trading signals to re-balance one’s portfolio, it is necessary to engage in a slight detour into the realms of statistics. When selecting funds, a buy and hold strategy often works because on a statistical basis asset prices drift up. Of course, the amount of ‘drift’ varies on an asset-by-asset basis, depending on its riskiness, and is often described as its risk premium. We all know that this ability to drift up isn’t universally true and there will be periods of time when an asset’s returns are drifting down. That’s why an investor benefits when they invest in a diversified set of assets, with the view that the weighted average of the return of the portfolio itself drifts up more often than not.
And don’t forget to diversify which trading signals as well.
It’s one thing to identify and construct a trading signal, but it’s another thing altogether to know when to start and stop using it. Any one signal will go through phases when they are just not as predictive as one would hope, and for this reason it also makes sense to diversify the set of trading signals that you use at any one time.
Not forgetting about JP Morgan’s AI model which deciphers the Fed’s future intentions, it’s wrong to think of this as providing a single binary decision tree. For example, if you know interest rates are going up, should I invest in short, mid or long-term treasuries? Instead, this signal needs to be converted into an expected return for any particular investment, along with a handful of other trading signals. These alternative signals come with their own expected return forecasts challenging what you think you already know, but that’s good and that’s the diversification you need. To deal with this just blend all of these separate forecasts into a single forecast.
Still more decisions to make.
For each asset under consideration, with your set of forecasts in hand, then it’s time to bring in your own favourite portfolio construction model – risk parity, mean variance, equal weight, you name it and there’s a research group who will tell you their approach is the best. Yet again, what works some of the time doesn’t work all of the time.
Who said it was going to be simple, if you thought I had reached the end, well not quite, the never-ending quest for the all-weather portfolio has only just begun. Now it’s time to wheel out your favourite machine learning model able to figure out the best way to dynamically blend your full suite of trading signals. Experienced Technical Analysis traders have been doing this for years. In a volatile market they might use a 30-day moving average, whereas under normal conditions it may be preferable to slow down their technical signals by switching to a 90-day moving average or even longer.
It turns out that JP Morgan’s robot is in reality providing an input and not an output, leaving much work to do for the humble investor.
Until next time.
Allan Lane