“No Idea” Investing: How I Built a Concentrated Fundamentals-Based Portfolio Using Nothing but Algorithms

[This article is adapted from a talk I gave at The Zurich Project.]

I’m a quant.

But unlike most quants, I run a concentrated fundamentals-based portfolio. More than 50% of my fund is invested in only eight companies, and they’re the kinds of stocks that Peter Lynch and Charlie Munger would have favored: safe, boring, inexpensive firms that are primed to outperform.

How did I get here?

Well, I came late to the investing game. It was one of those mid-life switches: I was fifty years old. Until then I didn’t have much disposable income, and my retirement savings were invested in the usual mix of mutual funds. But when I caught the investing bug—or maybe it was the trading bug at that point—I became totally obsessed.

Math had been my favorite subject in high school, but I had abandoned it altogether to work as a writer and editor. Now I went back to it with a passion. I read everything I could get my hands on about statistics, accounting, and investing. I totally lost interest in things that I had earlier been obsessed with, like music and film. At first I played with ETFs, then individual stocks. And over the next two years I made every mistake in the book, losing about 25% of my savings.

But late in 2015 I had a “eureka moment.” I had been playing with stock screeners and backtesting them. I discovered that if I made a stock screen for wonderful companies and a very different stock screen for horrible companies—both of them backtesting superbly—at some point I’d find a company that ended up passing both screens. My eureka was the realization that ranking stocks—on as many factors as practicable—was much more effective than screening them based on a few rules. And with screens, you can only apply a few rules or else you’ll end up screening everything out.

So this became my guiding principle: evaluate every stock you buy from every angle possible.Graham and Dodd’s famous definition from eighty years earlier, in Security Analysis, comes to mind: “An investment operation is one which, upon thorough analysis, promises safety of principal and a satisfactory return.” Upon thorough analysis was the key for me.

And then I was off to the races. I slowly went from using a dozen factors to over 200, many of them gleaned from my readings, and a few developed on my own.

I’m a writer, and I learn best by writing about what I’m doing, so I started investigating investment conundrums, and over the years I published more than a hundred pieces on Seeking Alpha and other forums. Every one of them taught me a lesson.

From the get-go, I maintained a concentrated fundamental value portfolio, dominated by a handful of positions, because I knew from backtesting that this was the most profitable way to manage a portfolio of thoroughly analyzed stocks. I was gobsmacked by the growth of my portfolio. After about six years, I’d made enough for my wife and I to retire early and start a private foundation. And then, a little over two years ago, I started a small hedge fund, primarily so that I could increase my personal contributions to charity and better fund my foundation, but also because I had followers, friends, and family who were asking me to handle some of their money.

Even though I’ve written all those articles on investing, I have never once published an “idea” piece. I have never talked to management, have rarely paid attention to earnings calls, and never consider whether a company has a “moat” or not. Everything I do is algorithmic. (Some people might think that there’s an algorithm for evaluating a company’s “moat.” If so, I haven’t found it.)

As far as I can tell, most quantitative fund managers are either of the AQR ilk—they hold investments for a long time, but none of their funds are very concentrated, and they don’t talk about digging deep into company fundamentals, instead relying on some rather general metrics—or of the Renaissance ilk—they do a lot of short-term trading, use enormous leverage, and don’t really invest in the Graham-Dodd sense. As for non-quantitative or only partially quantitative fund managers, most would probably agree with Ian Cassell of the Microcap Club, who recently wrote, “The ultimate edge is still simply being present with management in varying ways that most other investors aren’t doing. These are not sophisticated activities. There is no algorithm involved. And yet they consistently produce insights that no machine can replicate.”

Although we both invest in value-oriented microcaps, Cassell and I take the exact opposite approach. Is mine better? I couldn’t say. But by avoiding “being present with management,” I avoid the possibility that they’ll pull the wool over my eyes. I avoid the possibility that I’ll be reluctant to sell when it’s time to sell because I’ve gotten too attached to management. I avoid the possibility of rejecting a stock priced for perfection with a turnaround in view just because the management are jerks. And even if no machine can replicate the insights gleaned from “being present with management,” I’m pretty happy with the insights one can glean from taking a comprehensive and comparative look at stock fundamentals while rigorously avoiding emotional biases. By using multifactor ranking systems I can cover thousands of stocks at once instead of only those I’ve had time to fully research. And I can quickly compare stocks on every single metric and assign a score to each metric for each stock.

Here’s an analogy. Let’s say you were putting together a fantasy basketball team based on statistics. Among the things you want to consider when choosing players are their points, rebounds, assists, blocks, 3-pointers, and steals per game; you also want to consider their heights, age, field goal percentage, and games played. How are you going to put all those things together? You could assign each player a number between 0 and 100 on each of those things corresponding to her rank, multiply each one according to the factor’s importance, and then add all the results together. That’s exactly what multifactor ranking does.

I suppose this is the Moneyball approach. It’s a lot more complicated, though, since assigning an importance to each factor involves a lot of backtesting, something you’d never do when assembling a fantasy basketball team.

Too little attention is paid to the idea that factors have to be considered in combination, not tested individually. The number of games played has little to do with a player’s future ability, but it heightens the relevance of the other factors. The same thing goes for factors like company size. Smaller companies do not tend to outperform larger companies; but small size amplifies the effect of other factors.

Now this may not be the typical quant approach. For one thing, quants seem to dislike accounting. I’m no accounting expert myself—I know a lot more about quantitative analysis than I do about accounting. But I’ve dealt with hundreds of quantitative investors on Portfolio123 and in my Seeking Alpha investing group, and few of them seem to grasp why net operating assets, as a portion of total assets, should be relatively low, or why there are two entirely different ways to think of accruals, one based on the balance sheet and the other on the cash flow statement, or even why debt is not an asset. When it comes to discounted cash flow analysis, quants are generally uninterested in coming up with a calculation of cost of debt and cost of equity that makes sense to them, or evaluating whether to base their numbers on free cash flow or NOPAT. If you were to try to discuss the differences between Damodaran’s, Mauboussin’s, and Penman’s approaches to intrinsic value, they’d be lost immediately. As for me, I’d be totally fascinated, and would be trying to figure out an algorithmic way to test them all.

Now all my quant buddies have turned to “AI.” To be clear, there are two extremely different things that are called AI these days: LLMs and machine learning. Both are terrific. But I use LLMs as a sounding board, not to help me pick stocks. Doing the latter is, to me, no different from consulting your neighbor’s teenage children. Matt Levine at Bloomberg recently wrote, “Training a model on a corpus of books and Reddit to write coherent prose, and then asking it to predict future financial data, is a non sequitur.”

Also, LLMs are terrible at accounting. For example, I wanted to figure something out about mining companies’ financial statements. The number of blatant errors that ChatGPT made would make an accounting professor pull her hair out, and all the while it pretended that it was an absolute authority. To name a few:

• Using Compustat codes, it wanted to add XRD to XSGA when XSGA already includes XRD.
• It thought that exploration costs for mining companies would be classified as R&D expenses. That’s almost never the case.
• It suggested that I use operating cash flow as part of a formula for expenses. Since when is cash flow an expense?
• It suggested subtracting investing cash flow from that, but since investing cash flow is usually negative, that would give the opposite result from what it suggested.
• It suggested that D&A is always included in COGS, while in many income statements it’s a separate line item: it’s what distinguishes EBIT from EBITDA.

VALS, an independent platform committed to advancing the future of Gen AI, tests various LLMs on core financial analyst tasks. The models are best in simple quantitative retrieval tasks. Most of the tests involve asking models to research SEC filings. Accounting measures aren’t tested. Yet even with stuff this simple, none of the models have accuracy rates greater than 58%.

As for machine learning, I’m not nearly as comfortable with ML algorithms as I am with ranking systems. But perhaps that will change.

I’ve learned a lot from my mistakes along my investing journey. A few examples (and believe me, this is the short list):

• I learned the hard way that you can’t take financial statements at face value without looking at the likelihood that they’ve been manipulated by unscrupulous accountants. I lost a lot of money on a company that was pegged as a complete fraud, and I could have spotted that if I’d known how to look.
• I learned the hard way that you can’t look at growth without looking at financial stability—that sometimes middling growth numbers are better than high ones, and that including measures of the stability of sales, receivables, and the cash conversion cycle is really important.
• I learned the hard way that what Compustat and FactSet mean by “special items” is quite different, and that they have opposite signs; gross profit and D&A also are very different in the way that they’re standardized by data providers.
• I learned the hard way that you have to compare most measures to other companies in the same sector or industry rather than to the universe as a whole, even if the latter seems to backtest better, simply because of the mechanics of portfolio construction.
• I learned the hard way that when backtesting it’s vital to segment your universe instead of using the whole thing, and it’s vital to correct for outliers. I use “bootstrapping,” as do most experienced quants, but newbies usually don’t.
• I learned the hard way that selling a position when you don’t need the cash to buy something better is usually a cause for regret.

Now if I were to add to this mix non-algorithmic judgments about a company’s moat or my takeaway from a visit to its offices and my conversations with its CFO about its runway, I’d have to do that for about eight thousand companies per quarter, going back twenty years, to get all the data I need. To stick with the basketball analogy, sometimes you just have a feeling from watching a player, regardless of her stats, that she’s going to make a great addition to your team, or instead that she’d be impossible to work with. But how do you integrate that into the ranking system? You’d have to watch every single basketball player you’re considering. Some of my best investments have been in what I call “stinky stocks,” stocks that I would steer miles clear of if I were investing based on qualitative factors.

So that’s why I’m not very good at generating “ideas.” Every now and then I’ve toyed with the notion of integrating qualitative judgments into my ranking systems. If my hedge fund got really, really big, maybe I could hire a bunch of qualitative analysts to do just that.

But I don’t think I will. I’d be sacrificing the comprehensiveness and the insurance against behavioral pitfalls that a purely quantitative approach enjoys. So I’m afraid that, despite my late start, I’m probably too far along in my career for me to join the ranks of “idea” investors.