Neither Google, nor Microsoft, Amazon, or Tesla – none of them beat the S&P 500 each of the last 3 years. Individual stocks just have a hard time consistently beating the market (not to take away from these venerable high flyers – each has, of course, outperformed with incredible long-term returns. Consistent outperformance is different).
Can a machine or AI be used to beat the S&P 500 consistently?
Ask any serious investor or portfolio manager, and they’ll tell you it isn’t easy to beat the S&P 500 – and few, if any, have been able to do it year after year.
What about the last 3 years – through a multi-regime market? The last 3 years were like few others in market history and included  pandemic  stimulus and 0% rates  war, now two  sky-high inflation, and then  a Fed rate-hike cycle with rates now close to 5%.
Fact. Well, the Trefis high-quality portfolio (HQ) was a humble start – no frills, first money invested in the middle of a raging pandemic in September 2020. It has produced more than 1,000 basis points of outperformance* over the last 3 years. HQ has beaten the S&P 500 each year – 2020 partial year, 2021, 2022, and 2023 year-to-date. If you know many funds – in fact, any fund – that achieved this, please do send them along to us, we’re always keen to learn and improve!
Below, we share more in 3 areas: Why HQ worked? What’s been our broader learning, specifics of our approach, our journey? What’s next?
So – why did HQ work? Start with the basics
We started with a simple belief. A belief that a stock’s attributes like defensible revenue growth, profitability, balance sheet discipline, and reasonable valuation – just don’t get old. Curious money managers often wonder if there’s more to it – and of course, there is. However, following even these consistently was hard for us. The temptation to buy more growth in Zoom and Peloton stock at a higher price, or while turning a blind eye to the balance sheet in another value stock was irresistible. We stuck to a systematic, data-driven process.
The practical side of the trade is equally important. Interested in learning more, or want to evaluate if you could invest? Feel free to schedule a chat with Empirical, a local Boston-based firm. They’re happy to share more about how they’ve put into practice HQ and other strategies we’ve produced.
Before we get into more specifics of our approach, our process, and the practice – however, let’s take a step back and reflect on a bigger, more important learning.
There is something simpler, yet more powerful than the machine
We started building our 1-million-analyst platform – thinkHub, almost five years ago in early 2019. While thinkHub uses modern analytical techniques, the machine alone has not been the most distinctive attribute of our process. What has been distinctive, is our team’s ability to learn from the machine, and being able to feed that learning back into the thinkHub platform.
We wanted to learn from the machine’s analytic powers – and, in turn, wanted to teach the engine. Why? We wanted to start a flywheel. One that never stops learning, growing, and improving.
While some of us have experience with analytics and neural networks, fuzzy logic, and all else that gets bundled into AI today – that wasn’t enough for us. With thinkHub, we wanted to do more. We wanted to build an understandable decision-making system – an auditable decision system. One that helps us understand an investment, take it apart, and then put it back together – much like you’d do for a puzzle if you wanted to master it. We call it Understandable-AI**.
As the use and influence of AI grows, we believe everyone can – and should focus on learning from the machines. Teams that cultivate this mindset of “humans && machine” are likely to see outsized rewards – way, way more than a culture that focuses on machines alone – or competing/comparing with machine performance.
OK, so where are the secrets of the process and the journey?
They’re hiding in plain sight. We’ve, in fact, documented much of it through published analyses over many years – including on these pages, but here’s a quick summary.
On our way to a 3-year outperformance, we ran many 1:1 peer comparisons – we think about peers broadly – same industry, similar size, or even profit margin peers, it quickly becomes quite rich computationally. In the same breath, we thought aloud about counterintuitive ideas. We were fearful – yes – literally – not to miss incorporating macro events – the impact of covid, inflation, correlation, and cross-correlations (post-covid, and post-inflation shock trajectory), along with valuation (Buy or Fear), and technicals. Each of these provides a clear view of how we think – our most valuable asset. This and much more content, when stitched together with the learning process described above, helped us build a multi-lens system***.
We’re only getting started
This journey helped us develop, test, and refine our HQ portfolio, but subsequently also the Bayesian portfolio, and recently allowed us to launch the Reinforced Value portfolio – a hedged strategy where we seek to hedge away the market’s idiosyncratic risks.
What we’re doing next?
We realize that even a strategy with an 80% chance of beating the S&P 500 in any year has only about a 51% chance of, in fact, winning three years in a row. Consistency isn’t easy. Expanding our data sources, making thinkHub more transparent, understandable – and easier to use for non-engineers – all of this is important for learning and testing a broader set of strategies across different asset classes.
Above all, we are doing everything we can to preserve the culture, the mindset of learning from the machine!
Schedule a chat – to learn more or evaluate investing in Trefis strategies.
** Understandable-AI: Could our approach mean something inherently less powerful? Maybe. If you think power is solely defined by the number of hidden layers or compute steps in your analytical process. In contrast, we believe our approach – dismantling the building blocks of any modern analytical technique – whether Genetic Algorithm, Neural Networks, Fuzzy Logic, or other – into the fundamental ingredients, in fact, helps build more resilient systems. What fundamental ingredients? Here are some we think about: which data we use, methods to decide which data receive more weight versus less, hurdles used to filter out noise, testing results in the face of shocks or disturbances, and ensuring stability in the results.
*** Multi-lens system: Much like how a chess app doesn’t play like Magnus Carlsen, or Kasparov, or any one single player, we’re creating an investing equivalent