In earlier articles regarding future expectations, it was clear that the 2020’s are likely to offer much lower long term average returns. But that was based on a static asset allocation strategy. The learning machines are dynamic and they do seem to improve matters. It seems worth revisiting expectations with that hindsight. Trying to predict the near term future is a stupid idea. So let’s climb out on a limb.
In the first quarter 2021 update it became apparent that the pandemic has given us a glimpse into a future that may bifurcate into 2 possible paths. One better than the other and dictated by monetary and fiscal policy going forward. The dynamic asset allocation indicated by the learning machines have provided some hope for an improved future (higher and more consistent average returns).
Everything boils down to signal to noise. Imagine trying to see a snowman in a blizzard. The snowman is the signal and the blizzard is the noise. The idea is to wait for the blizzard to clear and/or get closer to the snowman. So I will use the least noisy signal (the learning machine output) and get closer to the signal (by building a more nuanced model of what a future may look like).
The best signal to noise data is from the 4 engine 2 layer stop loss machine learning portfolio shown below. This is a 23 year history of the rolling 12 month return of the learning machine managed portfolio. It is a measure taken every day, so its noisy.
There are two things that stand out. The long term average return is dropping and it seems to have a cyclical component that may be accelerating. If we smooth out the signal by taking a rolling 1 year average and superimposing the best fit amplitude declining and period accelerating sinusoidal predictor things begin to look a bit clearer. Below is that comparison.
The blue line is the rolling 1 year average return of the learning machine from 1998 to 2021. The red line is the predictor model. It represents an RSQR of .73 (which means it's a fairly good predictor). But that might be too sunny a view. So lets super-impose the predictor onto the daily (noisy) data. It still delivers a surprisingly helpful view into short term expectations and is shown below.
The takeaways from the above plot are: There is a strong oscillation of annual returns, which appear to be accelerating (getting quicker) and long term average returns are declining. (Note how the peaks are getting closer together over time. A full market cycle took a bit over 5 years in the late 90's and a market cycle only takes a bit over 3 years in the 2020's. The straight line is the best linear fit to the data over all 25 years from 1998 to 2025. The average return has declined from 15% to between just under 13% in those 20 plus years.) The red line, which is the actual portfolio performance ends in April, 2021 (time of this writing). The blue line is the predictor model and extends out until 2025. It indicates the next market low may be around the end of 2022 and the next market high may be around mid 2024. Obviously, there is still a lot of noise and on any given day, the past 12 month return can deviate a lot. But perhaps this is a helpful guide on what the near term future may have in store and how best to position your investments, how aggressive to be and when to keep your powder dry. Stay tuned.
Comments