If you have just 2 minutes - check out the Discussion section below for a quick digest of how backrunner sees the current market situation.

Quick reminder on what we do.

As of 18 August we have 123 funds filed complete 13Fs, 2 funds had confidential filings, 1 fund did not file, 1 fund filed with intention to later correct it’s filings (we noticed Cantillon does that).

backrunner.io tracker is now in a preliminary state and you can view Q2 results on most of the pages - funds, stocks, consensus insights and latest filing pages.

The report is aimed to provide information on how backrunner views the current consensus, which stocks are in the spot-light and look supported by other smart money investors.

Assets dynamics

The total long-book value of the funds we track is around $1.6T.

If you are wondering how much of that is Berkshire - it’s roughly 18%. Heavyweight, no doubt. But we are most interested in consensus where even the small players holding a stock at high conviction matter.

Q2 was pretty good for our universe adding $117B in value. Last 4 quarters decomposition is below.

Just $6B estimated net of new money deployed which is rather tame. Wall Street is anxious and talk of rich market valuation is still the main topic of discussion (on par with AI circular financing lol).

Consensus is stable

Table below includes all stocks that we count as consensus holdings - stocks with over 5% weight in each of holding funds portfolios.

Q2 is short on surprises here in overall rating, but lets look at the details.

We see couple of themes here developing.

Notable consensus stories

Alphabet

Alphabet (GOOG + GOOGL) can be seen as one of the most dramatic ones. With a total of 15 selling funds across both classes vs just 5 buys in high-conviction holders it has seen the largest inflow through Berkshire increasing it’s holding of GOOGL and GOOG from 5.9% of portfolio to 9.4% and 0.4% to 3.2% respectively. A decisive bet indeed putting it in 3rd place in Berkshire portfolio - higher than Coca-Cola (KO) at 10.9%.

Berkshire is not alone in this. But overwhelming majority is going in the other direction.

See below Mag7 section to put that move into context.

S&P Global

SPGI is quickly becoming a go-to value choice for our funds. In Q2 it saw $2.8B of net inflows. It never materialised in any price movement during Q2 so now we see that it is as good a time as any to look deeper into it.

SPGI is still down 33% from what it was a year back.

A lot of new investments from our universe at high conviction with current tally at 8 - a strong signal of emerging consensus according to our research.

To illustrate this below is all significant portfolio actions on SPGI from our universe.

Veritas, Pershing initiated new positions at <5% of their respective portfolios, and Triple Frond slightly below the threshold. We view SPGI as a strong candidate for deeper research.

Visa and Mastercard

V and MA held steady with no significant changes in consensus holders.

What is more interesting is that both of them added almost $770 million each with almost no selling or trimming action to speak of. 1 new investor and 6 buys vs 1 sell in Visa and 4 buys with 1 sell in Mastercard.

That means that funds that hold V and MA like them very much and keep adding to their positions.

Both V and MA underperformed S&P500 so far, but it is clear that Q2 was a decent run. Our universe members are long-term investors so we expect them sticking to this position going forward.

Another interesting fact about Visa and Mastercard is that they often go as a pair. This duopoly is clearly competing for a market share. So together they provide a bet on payments ecosystem as a whole.

So pick your poison kind of situation. If you don’t want to pick - take both. Something like Bill Ackman did in Q2.

V and MA in Pershing now together a top 4 position.

Mag7

We already touched Alphabet above, but I want also to show just how different the story is for the Mag7.

Most of Mag7 is widely held across backrunner universe.

There is no significant changes at both 5% conviction and any-conviction level with exception of META but we will get to that.

And except for Alphabet everything else looks like it is in a state of pause in terms of bets right now.

But this state of calm is a weird one, so let’s dive deeper in.

If we look into fund by fund actions we see a picture that is very different from calm state.

My personal take is that there is no single story here for every stock in Mag7. What we must remember is that backrunner consensus first and foremost is a story about momentum factor farmed from long-term investors.

So is it profit taking on top of momentum or a conviction building cycle? Lets remove new investors and keep only those that we had in Q1 and Q2 and see how their position average size changed to get the feel on what is going on.

NVDA had a great run past 6 months and it looks like investors are piling in on money inflows. But if we look at our high-conviction subgroup with funds that hold more than 5% of their portfolio have their average position regress from 11.7% of portfolio to 10.6%. That is probably profit taking but not completely bailing on AI hype-train.

Here is the price chart of Mag7 vs SPY YTD.

You can probably tell that MSFT and META don’t fit into profit taking narrative. But that is because there is a different story with them. Both of them seem very lost in the whole AI-boom cycle and are taking a beating in both market and in our high-conviction ownership analysis.

Both of them are bleeding long term investors at alarming pace.

Here is MSFT history of >5% holders.

Most of MSFT holders are there for the long run of 5+ years. They still have profits on their position. But not much head-space left.

And here is META.

That doesn’t look good for them.

Both are in dire need of a coherent plan to justify their hefty AI CAPEX. Until they provide one - the stock will continue feeling the pressure.

Mag7 is a mixed bag right now. There are clear winners like Alphabet. And clear losers like MSFT and META waiting to be relegated to the value territory.

AMZN is holding but with signs of pressure.

We will continue tracking the Mag7 developments going forward since it’s the most widely held basket and together about 18.5% of all long positions by total $ in our universe.

Emerging stories

Space Exploration Technologies Corp

SPCX is a hot new kid on the block and I personally was wondering who from our universe known for restraint and long-term vision will jump on the hype-rocket from the IPO.

The story so far is one of pain, controversy and bold moves like acquiring Cursor and selling datacenter capacity to top AI players.

Yet when we look at most new positions initiated during Q2 - SPCX is a clear leader in both number of funds and pure $ amounts.

So lets dive in and look who our heroes are.

The highest conviction bet comes from Coatue Management’s Philippe Laffont - notorious high-growth and high-risk investor. He fits our universe perfectly with his long-term vision and little respect for traditional value-investing frameworks. A nice counter-balance and I am personally glad it surfaced that way.

The truth about that particular bet is that Laffont had SPCX before IPO as a private equity investment. A brief research on the subject in question is riveting.

Coatue bought into SpaceX during the famous $420 round (Elon in his signature style) in early 2022. After 2 stock-splits each share would roughly amount to $8.4 so with the current price at around $140 we are looking at 16x+ gain in 5 years.

I found a nice short interview of Philippe explaining his investment style and philosophy, worth a watch. He was not hyping the IPO as much as an insider should but spoke long 10 year vision on how SpaceX is a threat to traditional wired and wireless communication providers way before Elon’s August 4th announcement.

This puts this particular bet in perspective as we at backrunner will be watching closely Coatue’s SPCX position in following filings once the early investor lock drops.

Another big bet at 3.1% of portfolio comes from Altimeter Capital. This is a different beast since this one is an IPO bet. Altimeter is ran by Brad Gerstner who is another growth investor in our universe. In interviews he mentioned that this is a long-term commitment so we will see how this plays out in the following quarters.

Outflows

A quick note on net outflows. Nothing too dramatic here.

Despite the outflows TSM still holds a top 10 position in overall consensus rating. Healthy mix of long-term holders and fresh positions.

Momentum still works in TSM’s favour. It is exceptional on fundamentals screen, richly valued but possible very deserved.

ASML, again, despite the outflows gaining high-conviction holders, propelling it into the 5+ territory.

Both TSM and ASML rank highly and according to our research have a strong chance to keep gaining.

Discussion

Q2 2026 consensus picture remains stable for backrunner universe.

But detailed look reveals a mixed signal that averages out the consensus to look calm and collected. Funds rotate their holdings, thin-out their positions and restraining from deploying new capital into richly valued market.

Mag7 is getting segmented with members moving in different directions testing investors confidence.

Below is our view of how consensus momentum looks for stocks discussed in this report.

Strong

• GOOG + GOOGL with a bet of confidence from Berkshire and strong market performance recently. Without that bet probably goes into holding ground bucket with AMZN.

• V + MA keep getting new investors with a strong recent run but still below S&P500 performance.

• SPGI looks like a solid value choice getting new high-conviction investors way below 52wk highs.

Holding ground

• AMZN one of the few solid choices in AI hyperscaler race with a decent strategy and relevant experience to back it up. We did not get into AMZN but it is step above MSFT and META in terms of investor confidence. Trending down but holding steady.

• NVDA despite market worries still holding it’s high-conviction base. No signs of bailing.

• TSM did not gain any new high-conviction holders but overall looks like a strong business on fundamentals.

Stressed

• META and MSFT leaking high-conviction investors and not currently loved by markets. Core business seems solid, but AI bet lacking coherent strategy takes a toll.

Emerging

• TMO crossed into 5+ conviction holders territory after a choppy ride. Combined with decent fundamentals has a strong chance to keep gaining following.

• UBER added 1 high-conviction bet to 6 total. Sitting close to 52wk lows looks like a solid value choice. I wrote about it in June and currently own a position in my personal portfolio.

Conclusion

This wraps up the first report. We plan to continue reporting every quarter from now on.

If that report provided value, consider subscribing to our blog and sharing this piece.

Subscribe now

If you have any question on the contents of this report, methodology or disagree with anything said - please feel free to jump into comments or find us on X.

Not investment advice. Do your own research. I or backrunner team may have a position in some of the stocks discussed in this report.

Eighty disciplined managers is still eighty different opinions. In their most recent quarter, the concentrated funds I track held over two thousand distinct stocks between them. Any method that starts with “just buy what they own” fails immediately.

The whole game is compression: from a couple thousand names down to the handful actually carrying information, without trusting any single manager’s judgment, including the best one.

The compression method I landed on runs on one idea, and everything in the earlier posts was building the case for it: independent agreement between people who are expensive to convince.

Why agreement, of all things

Recall the one robust positive result in the research from post 3: the Best Ideas finding.

So the first cut is brutal: a stock only counts for a fund at all when it is at least 5% of that fund’s disclosed book. Conviction or nothing.

But post 1 taught us a single manager’s conviction, however sincere, is an anecdote. Managers are wrong constantly, and you cannot tell in advance which ones will be right this time. That is the whole reason hero-copying fails. What you can do is ask a different question: how often do several concentrated books, run by managers who mostly do not know each other, each independently decide the same stock deserves a serious slice of their capital at the same time?

The answer, empirically, is almost never. That is what makes it a signal.

Take the most recent reported quarter. The tracked funds books held 2,092 names between them. Require a name to be a conviction position for even one fund: 361 left. Require two independent conviction holders: 99. Three: 41. Five or more: 15 stocks, out of two thousand! In a good quarter this last group fits on an index card, and it changes by a name or two per quarter, not by wholesale turnover.

Deep agreement among concentrated books is genuinely rare, which is exactly what you want from a filter. A signal that fires constantly is a fire alarm with a dead battery. This one fires almost never. When it does, several disciplined investors have each put a meaningful share of their fund behind the same idea. They answer to different clients, run different processes, and cannot copy each other’s homework in advance (hopefully).

The rules

The basket from post 3 is just this idea made into a ruleset. Every rule kills one specific way the signal gets polluted:

• Conviction threshold (5% of the book). Kills furniture. A fund’s 69th-best idea is not information, per Best Ideas above.

• Two consecutive quarters to count. Kills ghosts. A position that appeared in one filing may already be gone by the time you can act (the post 2 leaks). A vote only counts once the position has sat through two quarter-ends.

• Share classes merged, one vote per issuer. Kills double counting. GOOG and GOOGL are one opinion, not two.

• Each manager’s influence capped. Kills gurus. No single fund, however famous, can drive a name into the basket on its own. Agreement means several someones, definitionally. A single vote from Berkshire is great in both credibility and dollars weight. But I believe everything is relative. $100 million from Buffet is not the same as $100 million from someone with $500 million in AUM.

• Rank by number of independent conviction holders and hold the top 10. The signal itself. Depth of agreement decides membership, nothing else.

• A name stays until the agreement dissolves. Kills churn. Entry is hard, and exit only happens when the funds themselves walk away. The basket does not fidget. In the backtest one name sat in the basket 51 of 52 quarters.

• Everything happens ~52 days after quarter close. The lag, priced honestly: the earliest date a real follower could realistically act, which post 2 showed costs a patient basket approximately nothing.

Run those rules quarter after quarter since 2013 and you get the result from post 3.

Here is the full risk picture, both legs measured identically from the same quarterly series:

Both legs measured identically, quarterly, price-only, 2013-2026.

The honest reading takes the table whole. The basket compounds 3 points faster. Behind those rows sits an asymmetry: it captures 112% of the market’s rallies and only 87% of its falls. It also runs hotter than the index, and in a decade this strong the index’s own Sharpe ends a touch higher. The caveat from every previous post is not going away either. Attribution says this edge is a harvested style, not stock-picking magic.

What agreement looks like right now

The fifteen names with five-plus independent conviction holders last quarter: Apple, Amazon, ASML, Berkshire, Alphabet, Mastercard, Moody’s, Meta, Microsoft, Nvidia, S&P Global, TransDigm, Taiwan Semiconductor, UnitedHealth, Visa.

Read that list slowly, because it is the least surprising list in finance. This is the honest texture of what concentrated smart money agrees on: wide-moat, cash-gushing, large-cap compounders. No lottery tickets, no secret microcaps. Anyone hoping consensus would reveal hidden treasure will be disappointed. Concentrated managers agree on quality so obvious everyone already knows its name. Whether being told “own the obvious compounders, hold them for years” is trivial or is the entire hard part of investing, I leave to anyone who has actually tried to hold a position through a 30% drawdown.

The basket distills that agreement down to ten names. Here is its current book, with the drift left visible:

And the names at the door:

I keep this view rebuilt every quarter, for every tracked name with the full holder detail, on backrunner’s consensus explorer. It is the screen I check first when new filings land, precisely because it moves so little. When the index card changes, something real happened.

Frequently asked questions

What is a 13F consensus basket? A rules-based portfolio of the stocks that several concentrated fund managers independently hold as high-conviction positions (5%+ of their book) at the same time, with membership reviewed quarterly after 13F filings become public. Holdings are never rebalanced back to target weights. Positions trade only when the agreement itself changes. Depth of independent agreement, not any single manager’s opinion, decides what qualifies.

Why require agreement between multiple funds? Because single-manager conviction does not persist reliably. Past winners beat the index at roughly a coin flip. Requiring several independent concentrated books to back the same name filters out both individual error and portfolio furniture. Empirically the filter is severe: of about 2,100 names held by tracked funds in Q1 2026, only 15 had five or more independent conviction holders.

Doesn’t everyone already know these consensus stocks? By name, yes. The current list is dominated by mega-cap compounders. The information is not the names but the structure: how many independent disciplined managers commit serious capital, how that agreement shifts quarter to quarter, and the discipline of holding a small, slow-changing basket through drawdowns. The tested edge, to be clear, is a style tilt harvested with discipline, not secret knowledge.

What’s next?

This post concludes the main idea that I wanted to deliver. 13F is an incredibly interesting dataset rich with insights. Over last 5 posts I’ve shown how the edge can be extracted just from aggregating the data in a reasonable way. No data-mining. Just logic.

But there is much more to 13F and I’ve built a tool to show this. It’s called backrunner as a callback to the fact that we, as retail investors run after the main money managers made their moves. Which as I showed earlier doesn’t mean the information is not useful anymore.

Check out backrunner.io

Backrunner shows the funds I consider worth following. Shows how consensus is behaving. Shows individual stocks and provides valuable insights that you can use in your own investing when analysing stocks.

There is a lot more to show. And I have couple of subjects I am currently working on:

• How basic consensus basket performance can be improved with fundamentals

• How single stock consensus is predictive of it’s performance

• How retail investors can use 13F data to improve their analysis

And most importantly - we are nearing the end of the Q2 2026 filing window, so we will see how top investors view the current market developments with all the AI and geopolitical volatility. Are we getting closer to a bubble popping or smart money choose to double-down?

If you want to support the backrunner blog - consider subscribing to this newsletter. It’s free and is a decent place for open data-driven discussion.

Not investment advice. Do your own research. Don’t let the FOMO eat you alive.

Here is the uncomfortable math from the earlier posts. The consensus basket works on a curated universe of about 80 concentrated, patient filers. The full field contains 2,170 measurable candidates. So somebody, somehow, has to pick the 80, and post 1 already proved the obvious method is broken: past performance does not identify future winners, because winning does not persist. Only losing does.

I tested the three ways people actually try to build the list. Two of them are traps.

Trap one: follow the leaderboard

The most natural idea in the world: rank all funds by their last three years against the index, follow the top 80, refresh the list every year, kick out the laggards.

I expected it to underperform. It did something more interesting.

It made more money than the curated universe: 20.0% a year against 19.0%, over 2016-2026.

But there is but.

When the market fell, as it periodically does, the leaderboard universe fell 183% as hard, nearly double the market’s own drawdowns, against 73% for the curated set.

Its worst peak-to-trough loss was 43%, in a decade where the market’s was 15%. Its Sharpe ratio, the return-per-unit-of-risk number, came out 1.01, worse than just buying the index and going outside.

One extra point of annual return, purchased with double the volatility and triple the drawdown.

The mechanism is worth spelling out.

Ranking by recent performance does not select for skill; post 1 showed skill barely persists.

It selects for whatever style is currently hot. You end up following the funds most levered to the recent regime, which means you arrive at every party exactly when the music is loudest. The leaderboard is a cursed momentum bet.

And a detail I did not expect: it does not even matter how you churn the list. I ran the annual refresh two ways, forced replacement and replace-only-if-better. They produced identical universes. Fresh winners always outscore decayed incumbents, so the leaderboard replaces its whole bottom every year like clockwork. The list never stabilizes, because “recently hot” is not a stable property of anything.

Trap two: skip the choosing, follow everyone

If picking is hard, why pick?

Take every fund that passes the basic shape bar and let the consensus mechanism sort it out. Diversification of stock pickers, the same logic that makes index funds work. I ran the basket on universes growing from the curated ~80 all the way out to 482 funds.

The return edge does not improve.

It wobbles between +3 and +4 points a year whether the basket listens to 87 funds or 482. By 80 funds the signal is saturated, and every fund after that mostly adds noise votes that degrade the risk profile. Sharpe declines monotonically as the universe grows.

More opinions past a point just means more mediocre opinions.

I also ran the reverse test. Take 300 random draws of 82 funds from the qualified field and run the basket on each. The random universes averaged about a point below the index. Some got lucky on return; 3.7% of the draws beat the curated universe’s return. But essentially none, 0.7%, matched its risk-adjusted quality. Random gets you the occasional hot streak. It almost never gets you the thing that makes the strategy holdable through a bad year.

So what actually separates the followable funds?

Not genius, and by now that should be no surprise.

When I compared what the curated universe’s basket did against the random ones, the difference was not the upside at all: it captures 109% of the market’s rallies, roughly what the random baskets do.

The difference is that it falls only 91% as hard as the market when things break, while random universes fall 116%. Same offense, better defense, quarter after quarter.

The moat is not brilliance. It is not blowing up.

Where does the defense come from?

I went looking for it in the funds’ observable paperwork, and it turns out followability has a legible signature:

1. funds that are established (a decade-plus of filings),

2. sizeable,

3. concentrated, and

4. patient.

A simple model built on those traits, trained on half the curated list, identifies the other half in years it never saw with an AUC of 0.81, which for a returns-free screen on public filings is remarkably strong.

The same structural traits that make a fund’s 13F readable (concentration, patience, the things posts 1 and 2 kept running into) also mark the funds worth reading. Discipline does the work.

Two honest limits.

First, the signature narrows the field, but it does not finish the job. “Established and concentrated” describes defensive compounders and aggressive high-beta funds alike, and the model cannot yet tell them apart. Some judgment about how a manager behaves in a drawdown still sits on top.

Second, the caveat from post 3 still applies to everything here: even the curated basket’s edge is a style the market has been rewarding, not conjuring. But we will get to that in couple of posts in this series.

The list is chosen. Now what?

You now have the whole selection argument in one place: leaderboards buy you double downside, breadth buys you nothing, and the durable markers are structural, which is why the universe I actually track is built on establishment, concentration and patience rather than on anyone’s recent scoreboard, and gets reviewed against those same markers rather than against last year’s returns.

But a chosen universe is just a reading list. Eighty disciplined managers still hold hundreds of different stocks, disagree with each other constantly, and cannot all be right.

The next question is the mechanical heart of this whole series: how do you turn eighty careful books into one signal, without trusting any single one of them? That answer has a name, it is the reason this project exists, and it gets the next post to itself.

Frequently asked questions

Should you follow the best-performing 13F funds? The data says no. A universe of trailing 3-year winners, refreshed annually, returned about one point a year more than a curated universe over 2016-2026 but fell nearly twice as hard as the market in drawdowns (183% down-capture vs 73%), with materially worse risk-adjusted returns. Recent performance selects for hot styles, not persistent skill.

Is it better to follow more funds? Past roughly 80 well-chosen funds, no. In my tests the consensus basket’s excess return stayed flat between +3 and +4 points a year as the universe grew from 80 to 482 funds, while the Sharpe ratio declined steadily. The signal saturates; extra funds add noise.

What makes a fund worth following in 13F filings? Structural discipline rather than recent results: a long filing history, meaningful size, a concentrated book, and low turnover. Those traits identified held-out curated funds with an AUC of 0.81 in our tests, and they are the same traits that make a fund’s quarterly filings an accurate picture of what it actually does.

Not investment advice. Do your own research.

Stay disciplined, resist the FOMO. Just like the best in the business.

Check out the Backrunner!

If following filings one hero at a time fails, and the lag was never the problem, an obvious question remains open: is there any way to hold this data that has actually worked? People have been studying that question in academic finance since before I had a brokerage account, and the answer has been surprisingly consistent. It just looks nothing like the way most people use 13Fs.

What twenty years of papers actually found

Four studies, told in the order that builds the argument.

Let me quickly run through them.

Hedge funds vs Mutual funds through 13F lens

Griffin and Xu looked at every hedge fund long book in the 13F record and found the average one beats mutual funds by about 1.3 points a year at best, with no sector-timing or style-picking ability at all. That is the academic version of my post 1: the field, taken whole, is unremarkable. Anyone selling you “the smart money” as a category is selling you the gray cloud from my first chart.

The 13F trade of the century that everyone missed

The strange exception that started the whole literature.

Martin and Puthenpurackal reconstructed Berkshire Hathaway’s equity portfolio from 1976 to 2006 and asked the copy-trade question directly: what if you bought what Buffett disclosed, a month after the disclosure? Double-digit abnormal returns a year, for three decades, from public documents. The market systematically under-reacted to the news. One fund is an anecdote, even a 31-year one. But it established the mechanism that matters here: disclosure is not the same thing as the market having priced the information in.

If you only had to choose one..

Anton, Cohen and Polk took every active manager and isolated each one’s “best idea”: the single position their portfolio weights say they believe in most. Those positions, and pretty much only those, beat the market: by 2.8 to 4.5 points a year depending on the benchmark, while the rest of the book adds roughly nothing. Managers are not uniformly skilled or unskilled; their skill is concentrated in the handful of names they size like they mean it. The 6% positions know something. The 40 one-percenters are diversification theater.

Free-riding is free

Verbeek and Wang built hypothetical copycat funds that hold nothing but what other funds disclosed, bought after the disclosure, and found they match or marginally beat the originals net of costs. Free-riding on disclosed holdings works, and it got easier after regulators forced more frequent disclosure. The copy is viable infrastructure. The question was only ever what to copy.

Put the four together and the shape of the answer emerges.

Don’t copy books - the average book is noise.

Copy conviction.

Because that is where the skill hides. The lag doesn’t kill it, because the market digests these documents slowly.

And don’t ride one manager. Because one manager is an anecdote with a drawdown coming.

Which suggests something none of these papers quite built: take many concentrated, patient managers, keep only their high-conviction positions, and hold the names where several of them independently agree. A consensus basket.

So I built it and ran it

The setup:

1. Roughly 80 concentrated filers.

2. Position counts as a conviction vote only when it is at least 5% of that filer’s book and has been held two consecutive quarters.

3. Merge share classes, rank names by how many independent filers are voting for them.

4. Hold the top 10 - a name stays until it drops out of the consensus.

5. The list is reviewed once a quarter, about 52 days after the filing quarter ends, and trades happen only when membership actually changes: entries are funded by exits, winners are left to run untrimmed.

6. There is no quarterly reset of weights, no trimming back to equal slices - the portfolio mostly just sits there on purpose.

Over 2013-2026 that basket compounds at 17.3% a year against the S&P 500's 14.3%, about 3 points a year over one of the strongest index decades on record.

And because a single backtest number is exactly the kind of thing previous posts taught us to distrust, the validation mattered more than the headline: the edge holds at +2.5 to +3.4 points for nearly every train/test split I tried, stays positive in 8 of 10 rolling 3-year windows (the exceptions bracket a 2021-2022 give-back you can see in the chart), survives realistic transaction costs, and survives investor's dividend taxes. Costs stay immaterial because almost nothing ever trades: across twelve years the basket swapped one name in a typical quarter and did nothing at all in roughly one quarter in five.

Boring. But works.

Check out backrunner!

The honest catch

Here is the part most write-ups of this strategy leave out, I found it while trying to break my own result.

Run the basket’s returns through a standard factor attribution, the statistical test that asks “could this performance be explained by well-known styles rather than skill”, and the answer is: mostly yes.

The basket is a large-cap quality/growth portfolio. The residual alpha the part attributable to something like stock-picking magic, is +0.8% a year with a t-statistic of 0.48. Statistically, zero.

So the honest statement of the twenty-year answer is this: following the consensus of concentrated, patient filers has been a disciplined, cheap, rules-based way to hold a style that the market rewarded, harvested from public documents on a 45-day delay.

It is not evidence anyone involved, including me, can see the future.

Whether that style keeps being rewarded, and what happens if it stops, is a real question, and it gets the full engine-room treatment later in the series. I am not going to promise you alpha. The data says I don’t have any to promise. What the data does support is narrower and, I think, more interesting: the signal is real, the discipline is replicable, and the returns leaderboard was never where it lived.

Which brings us to the next question

Every study above quietly assumes you already know which managers to listen to.

“Concentrated, patient filers” is doing enormous work in my rules, and post 1 of the series showed why it has to: in the full field, only losing persists. Pick the universe wrong and the consensus is noise.

So the next question is the one everything actually hinges on: whom, exactly, do you follow? Can you define “worth listening to” with rules, in advance, without peeking at returns? That turns out to be the hardest and most interesting problem in this whole series. Next post.

Frequently asked questions

Does following 13F filings beat the market? Following individual funds mostly does not: the average 13F filer trails the index and past winners revert to coin flips. The academic evidence supports something narrower: baskets built from many managers’ disclosed high-conviction positions, bought on the filing lag, have historically outperformed by roughly 2 to 3 points a year. Factor attribution says that edge is a style tilt, not stock-picking alpha.

What is a 13F consensus basket? A rules-based portfolio of the stocks that several concentrated managers independently hold at high conviction (large position sizes) at the same time, rebuilt each quarter after the filings become public. The idea is that agreement between independent, disciplined stock pickers is a stronger signal than any single manager’s book.

Why doesn’t the 45-day filing lag destroy the edge? Two reasons. Patient, concentrated funds barely change their books inside a quarter, so the stale copy still matches the fund (measured in post 2 of this series). And the research record shows the market digests disclosed positions slowly; even Berkshire’s disclosed buys kept outperforming when bought a month after the disclosure.

Not investment advice. Do your own research.

Every objection to 13F investing eventually arrives at the same place: the lag. By the time you see a filing, the positions are at least 45 days old. “You are buying someone’s stale homework” is the standard dismissal, and it sounds devastating.

I spent a week investigating this, and the standard dismissal turns out to be aimed at the wrong leak. The lag is nearly free. The expensive leaks are the ones nobody prices.

The lag, measured properly: almost nothing

Here is a clean way to isolate what staleness costs, with everything else held equal.

I ran that comparison for every fund in the reconstruction with at least three years of history: 2,062 funds, quarter by quarter, 2013 through 2025.

The median cost of holding the stale book is zero. Not small. Zero, to two decimal places, minus 0.00 points a year.

The average hides the real story, which is who pays:

Sort the same funds by how much of their portfolio they replace each quarter and the lag’s tax turns monotonic.

The most patient quarter of funds, turning over about 4% of the book per quarter, pays nothing at all; the copy is effectively the fund.

The highest-turnover quarter, replacing about a third of the book every ninety days, pays around 0.6 points a year for the delay.

Let me be precise about what that last number means, because it is not “fast funds are bad funds.”

A 13F is a quarter-end photograph. For a patient manager the photograph and the film are nearly the same thing. For an active trader the photograph misses the movie: everything bought and sold inside the quarter, which for a fund replacing a third of its book every ninety days may be most of what the fund actually does. Their real performance lives in trades no filing will ever show. They might be brilliant. The filing cannot tell you either way.

I can put a number on that blindness. For 54 of the reconstructed funds I have a second, independent reconstruction that does model intra-quarter trading (at assumed mid-quarter prices).

For the most patient third of those funds, the two lenses agree almost perfectly: same quarterly returns to within 0.05 points, correlation 0.998.

For the most active third the disagreement is five times wider, and at the extreme, a fund turning over 71% of its book every quarter, the two views of the same filings drift apart by about 3.4 points a year.

And both views are built from the same filings, so even that gap is a floor, not the full leak. Past a certain turnover, “this fund’s 13F performance” stops being a fact and becomes a guess with widening error bars.

So when I exclude high-turnover funds from everything that follows in this series, hear it correctly: it is not a judgment about their skill. It is an admission about the 13F limitations as an instrument.

The lag and the lens fail on exactly the same funds, the unreadable ones, and spare the patient, concentrated ones, which, after last post’s findings, are the only kind with any claim on your attention anyway.

The leaks nobody prices

So if the famous leak is a rounding error, why did the follower lens in my reconstruction still come out slightly behind the funds themselves, 24.8% versus 25.3% beating the index, a median drag of about a third of a point?

Because the copy leaks in quieter places, and those do not show up in any spreadsheet until you have tried the strategy and experienced them.

You inherit positions without theses. The filing tells you what a manager owns, never why, at what basis, or with what exit in mind. So the first 20% drawdown arrives, and you have no way to know whether the manager is adding or already gone. Borrowed conviction expires precisely when conviction is the only thing that would keep you in.

This is the largest leak in copy-trading, it is behavioral, and the individual-investor evidence (the Barber and Odean literature on how retail accounts actually trade) says it compounds: people sell the inherited winner and average down the inherited loser.

The book you see is not the book they run. No shorts, no swaps, no foreign lines, no cash. A position that looks like a bold conviction bet can be one leg of a hedged pair whose other side is invisible by law. You copy the leg, not the trade.

Some of what you see is already gone. Beyond the intra-quarter blindness we just priced, anything sold in the 45 days between quarter end and filing day is a ghost you may be buying at the very moment its seller is done with it. For patient funds this is rare, one more way the same discipline that makes a fund followable makes its filing honest.

The sizing is yours to fumble. The manager’s 6% position sits inside a book with risk controls, offsetting exposures and an investment committee. Your version of it sits inside a five-stock brokerage account next to your emotions. Same ticker, entirely different risk level.

Add it up and the arithmetic of the whole enterprise inverts.

The copy machinery, lag included, leaks a few tenths of a point a year. The funds being copied trail the index by 2.8 at the median.

Which brings us to the actual question

Following one fund fails at the selection step, before the copy even starts, and no copying technique fixes that.

But notice what this post quietly established along the way: for patient, concentrated managers, the public filing IS the portfolio, at essentially no cost to a follower. The data is a nearly lossless window into how the most disciplined stock pickers in the country are actually positioned, six weeks late.

Which raises a better question than “which manager do I copy?”

But I propose a different approach.

What happens if you stop copying anyone in particular, and instead follow what many disciplined managers agree on at the same time? That question has an academic literature, a twenty-year answer, and an honest catch.

And I will cover it in the next post.

Frequently asked questions

Does the 45-day filing lag make 13F data useless? Measured properly, no. Holding a fund’s book one quarter stale costs the median fund copier about zero points a year; it only becomes expensive (about 0.6 points) for high-turnover funds, whose filings cannot show the intra-quarter trading their strategy actually lives in.

Can I tell from a 13F when a fund actually bought or sold? No. Filings are quarter-end snapshots. Trades inside the quarter are invisible, and a position sold after quarter end still appears in the filing. Prices paid and timing are never disclosed.

So is copying individual trades from filings a good idea? The evidence says no, but not because of the lag: the funds themselves mostly trail the index, winners do not repeat reliably, and a copier inherits positions without the thesis or the exit. Filings work better as a map of where disciplined investors agree than as a trade feed.

Not investment advice. Do your own research.

Every “Warren Buffett just bought X” headline traces back to the same public document: a 13F, filed quarterly with the SEC by anyone managing over $100 million in US stocks.

The dream 13F filings sell is irresistible. Somewhere in those filings is a genius, and if you find the genius, you can just... do what they do. Free ride, minus 45 days.

I run a tool that reads these filings all day, so I had the machinery on hand to ask the question the dream depends on: if you had followed any one of these funds, would you actually make any money?

Lets just check ALL OF THEM.

How I measured it

Some mechanics first.

A 13F discloses a manager’s US stock positions at each quarter end, with a 45-day filing deadline.

The raw filing record is a zoo: banks, index shops, quant farms holding three thousand tickers, pension overlays. No “smart money” headline was ever written about those, so the first job was pruning the field down to the funds the dream is actually about.

I started with every institution that filed 13Fs for at least five consecutive years between 2013 and 2025, dead or alive: 5,830 filers. Keeping the dead ones matters; a study of survivors only flatters the field.

Then a blind shape rule, no names, no performance peeking: a fund stays if its typical filing puts at least 36% of the book in its top 10 positions and holds 90 positions or fewer. A book that looks like conviction, not an closet indexer. Although I must admit - my description is pushing it a bit, since on top 10 weight S&P500 today is around 38%.

That leaves 2,681 funds filing like actual stock pickers, and they look the part: the median book runs 33 positions with three quarters of its value in the top 10. Mostly boutiques too, median disclosed book $266M, with Berkshire sitting alone at the far end at $211B.

Each dot is one filer’s typical book shape. The gray cloud is diversification; the green corner is conviction, and it is what this series measures.

For each of those funds, each quarter, I rebuilt the portfolio from the filing, weighted by position size, and measured what that exact book returned over the following three months at dividend-adjusted prices. Chain the quarters and you get each fund’s track record as its own filings tell it. A quarter counts only when at least 85% of the book’s value has usable prices (early years are thinner), and a fund enters the results only with at least three years of measurable history: 2,170 funds clear that bar. Where an independent reconstruction exists to check against, 54 funds, the two agree almost perfectly.

One honest limitation, and it matters: a 13F shows no shorts, no foreign listings, no cash, nothing bought and sold inside a quarter. Everything below is a statement about the disclosed US long book. That is exactly the part of the portfolio the copy-trade dream is built on, which is why it is the right thing to measure.

Finding one: the typical filer loses to the index

Of those 2,170 funds, 25% beat the S&P 500 over their own window. The median fund trails it by 2.8 percentage points a year. The spread is wide in both directions, but the middle of professional stock picking, as disclosed in its own filings, underperforms a $10 index fund.

The field, one dot per fund compressed into a histogram. The median professional book trails SPY by 2.8pp a year.

If that number feels familiar, it should: it is the SPIVA result. As of
mid-2025, 87% of active large-cap funds trailed the S&P 500 over five years and 86% over ten. Mine is a different animal measured from a different document trail, concentrated 13F books instead of mutual fund returns, and it lands in the same place. Concentration and a hedge-fund letterhead do not repeal the result.

Finding two: the winning cohort doesn’t stay winning

Fine, you say, I will not follow the average fund. I will follow the winners!

So I split 2013-2025 into four 3-year eras and, instead of gawking at each era’s freak champion, followed the whole winning cohort: every fund that finished an era in the top quarter of the field. Pooled across the decade, that is 361 fund-transitions, a real sample, not an anecdote.

Here is what being a certified recent winner bought them. In the following era they beat the index 44% of the time, with a median excess of minus 0.7 points a year. A coin flip that pays slightly worse than a coin. And the coin is getting worse: the first winning cohort (2013-2016) still beat the index 62% of the time in the next era; the most recent one managed 40%.

Yesterday’s winners converge to ordinary. Yesterday’s losers stay losers.

Two honest nuances. Winners do keep their relative rank a bit more often than chance (37% repeat top-quartile against the 25% a shuffle would give), so screening on past performance is not completely information-free.

It just is not the information you wanted: staying in the top quartile of a field that mostly trails the index is a prize for the wrong contest. And the bottom of the table is far stickier than the top: last era’s losing cohort lands below the index again two to six times out of seven, era after era.

(The single best fund of each era, if you are curious, was usually not even a decision-making fund: three of the last four era champions were effectively one stock wearing a filing. Leaderboards select for accidents, which is its own reason to never read one.)

Finding three: losing persists, winning mostly doesn’t

The sharpest way I found to see it: take the 177 funds with full 12-year history, count how many of the four eras each one beat the index, and compare that to what blind luck would produce (coin flips at each era’s actual beat rate).

Both tails are fatter than chance, but not symmetrically. The bottom one is where the persistence lives.

Five funds beat the index in all four eras; luck alone predicts about two, and the gap is within noise. I checked what those five books were before anyone calls them proof of skill: one is an endowment holding a single inherited stock, three rode the mega-cap tech decade nearly wire to wire, and exactly one is a conventional diversified manager. The consistency club is mostly luck plus style exposure, not wizardry.

The other tail is different. Forty-seven funds, one in four, never beat the index in any era, where chance predicts thirty. That gap is not noise. Persistent losing is the one durable, repeatable pattern in this data. Put dryly: the market does not reliably let anyone win four eras straight, but it absolutely lets some funds lose four eras straight, and they do.

And before you ask: adding the follower’s lag, buying each book only after its filing became public, does not rescue anything. It shaves the picture slightly worse, median minus 3.1 points a year.

So the copy-trade is dead. The data isn’t.

Here is the turn, and it is where this 13F post series is headed.

Everything above measures funds one at a time, which is how everyone instinctively uses 13F filings, and it fails.

But single funds may simply be the wrong unit of analysis.

You cannot identify tomorrow’s consistent winner, and there may be no such category to identify.

But consistent losers are real, common, and detectable, which means pruning is possible even where picking is not. And the academic literature, plus my own replication of it, says something one fund at a time cannot: a basket built from the overlapping high-conviction holdings of well-chosen filers, bought on the public lag, has historically carried a real, modest edge over the index.

Not alpha, not magic, and I will show the honest attribution later in the series, but a repeatable, rules-based something where the one-genius dream is a provable nothing.

So that is the road ahead: why the copy of even a good fund leaks (next post), what the research actually found in the lagged-basket idea, how to choose whom to listen to when only losing persists, what agreement between independent stock pickers means, what the edge turns out to be made of, and what a live quarter looks like through all of it.

The Q2 filings land in August; the series ends by reading them with you.

The genius was never in the leaderboard. Whether there is anything in the crowd is a better question, and it has a more interesting answer.

Frequently asked questions

Does copying the best-performing funds from 13F filings work? Historically, no. Funds finishing a 3-year era in the field’s top quarter beat the index only 44% of the time in the following era, with a median excess of -0.7pp a year, and only about 3% of funds beat the index across all four eras since 2013, roughly what luck alone would produce.

Do 13F filings show a fund’s whole portfolio? No. They show US-listed long positions only: no short positions, no foreign-listed holdings, no cash, and nothing traded in and out inside a quarter. A hedged fund’s 13F can look nothing like its actual risk.

How out of date is 13F data? Filings are due 45 days after quarter end, and most managers file at the deadline. A January purchase can surface in mid-May. Everything in this series treats that lag as a fixed cost of the data.

Not investment advice. Do your own research.

I am not saying that S-curves don’t exist. They clearly do. Enough literature documented them in details across hundreds of years of human history.

But I suspect that no one found a robust enough way to exploit them.

S-curve is a recurring mental model that describes technology (any technology, not just Internet or AI) adoption.

New technologies diffuse and improve along an S-curve with a slow start, steep middle where adoption compounds fast and plateau at the end.

The challenge for growth investors is to correctly identify where specific company or technology situated on this curve and estimate how high the curve can go.

And this is difficult.

Take recent SaaS blow-up. Where would you put Adobe on this S-curve? Or Salesforce? Top of the curve? What if you stick AI on top of them? Bottom? Who knows?

Whale Rock Capital - clearly one of the current top-performing investment funds that I track. But struggled to outperform in 2022-2023. Why? Here is a hint.

In recent interview Alexander Sacerdote mentions that in April 2022 they had 40-50% of their AUM in software. Sold almost all of it and switched to chips and AI. Performance radically improved since.

Forgive me. But if you drive your investing strategy with S-curves and dead-sure about where to put each company, wait for it to hit 50% of the curve then cash-out, then SaaS story is the cleanest example of how darn near impossible it is to use this tool effectively.

I have a better explanation of growth investing. It’s pure vibes.

And vibes shift.

Take any SaaS name now - vibe is way down. Many of SaaS stocks are now comfortably in the value-investing territory. But be aware of falling knifes.

NVDA, TSM, SPCX - vibe is way up. But anybody trying to place it on the S-curve anywhere other than bottom part - just ignore them. Its clear FOMO-landgrab-supply-restrained bs.

What is more exciting (for me) is the gray area where vibes are not way up or way down - it’s where vibes are off.

Want to guess who is in this vibes-are-off category?

Meta and Microsoft. What a couple.

Both are still richly priced and both are candidates for SaaS-like demotion into the value-land.

But why?

Both have been dabbling in models. Both are heavily investing in building data centers for inference. Both have a strong foothold on their respective feudal lands (smartphones and desktops).

But both clinically lack any sort of vision. Or at least failing to communicate it clearly.

Meta is already paying the price.

Backrunner universe shows consistent outflows in shares held since 2024 by top investors.

Conviction is flat since 2023. That means long-term holders are taking their profits elsewhere.

Thats not how growth looks.

Microsoft is in the same place.

Shares held - trending down.

Conviction is also trending down. There are many more long-term holders of MSFT than META, so there is a certain floor which will be supporting the stock price. But overall picture looks dim.

Q2 just ended and in 45 days we will have an updated view on how META and MSFT perform in terms of convincing investors they are worth buying at current valuations. I doubt we will see any progress.

Both are still above value-land, but are trending there.

Both are still stellar businesses.

But the vibes are off.

And no vibes - no growth.

And no promise of hopping on the next S-curve can help them.

The only thing that can salvage the situation - is a clear vision. But as of right now Zuck is much more convincing delivering his vision of perfect beef, rather than how Meta can keep growing on the exponential schedule.

42:00 timestamp - pure comedy gold.

I would much prefer the same attitude towards scaling the AI-challenge.

Don’t let FOMO eat you alive!

This is not an investment advice. Do your own research.

Subscribe now

What Uber actually is

Uber is a toll booth on movement. Somebody wants a ride, a burrito, or a pallet of freight moved. Somebody else has a car, a bike, or a truck. Uber stands in the middle, matches the two, and keeps a cut of every transaction. That is the whole business. The cut, what they call the take rate, runs around 27% blended.

Three pieces. Mobility is the rides business, the profit engine, roughly $30 billion of revenue and the highest margin. Delivery is Uber Eats plus grocery, the growth engine, growing bookings faster than Mobility and now solidly profitable. Freight is digital trucking brokerage, small, still losing a little money.

The interesting bit is the flywheel sitting on top. Uber One, the membership, hit 50 million members by Q1 2026 and those members now drive about half of all gross bookings. People who use both rides and Eats spend roughly three times more and stay around far longer than single-app users. And Uber acquires those cross-app customers almost for free, by promoting Eats inside the rides app and the other way round, at roughly half the cost of buying them through ads. Layered on that is an advertising business now running past $2 billion a year and growing about 50%, which is nearly pure margin and quietly lifts the take rate.

So the engine is volume times take rate, and the volume keeps climbing while the high-margin layers (ads, membership) push the cut higher. Good engine.

Why it is cheap

Uber fell from about $102 to near $70, roughly a 29% haircut, and bottomed near its 52-week low. During all that the operations accelerated. Q4 2025 had gross bookings up 22% and segment EBITDA up 35%. Q1 2026 had bookings up 25% and EBITDA up 33%. Those are not the numbers of a business in trouble.

The market re-rated Uber down on one fear: autonomous vehicles. Tesla launched an unsupervised robotaxi across the whole Austin metro on June 3, 2026, on its own app. Waymo runs its own app in most cities it operates. The bear story writes itself. The competition is gearing up to be tough. I’ll share some data on that below.

Still if self-driving cars become the supply, and they go straight to the consumer, who needs the middleman? The toll booth gets bypassed.

That is a real question. I will come back to it, because it is the question. But notice what the de-rate is and is not. It is multiple compression on a narrative about 2028 and beyond. It is not the current cash flow breaking. The cash flow, in fact, is the best it has ever been.

Bill Ackman is in

One more data point, for what it is worth. Courtesy of my handy backrunner.io fund universe.

In Q1 2026 the consensus starts to build around UBER in the professional money manager space.

First lets look into reductions.

Pershing is a slight trim of the position 15.9% reduced to 15.7% of fund. Marathon sold 7% of shares, probably to fund new positions and additions. They still have high conviction at 10% of the fund in UBER, but their total position is relatively small. Lyrical sold 6% of their position in a porftolio-wide reduction. Still 11th largest position of the fund which is on a less concentrated side.

Everyone else on the list added/bought. Across backrunner universe I see total of 3.2 million of shares positive buying delta to a total of $241 million additional conviction in Q1 alone.

Pat Dorsey opened UBER as a 6% stake in 11 stock portfolio. And he has been on a decent run recently.

This is a rare new formed conviction signal which is alone a solid reason to look deeper into this.

Substack investors are in

There are a few substackers covering Uber.

Manu from Fundamentally Sound has a great piece on Uber. Is a shareholder and posts regular updates. He was one of the earliest among substackers I follow who spotted the potential.

Fundamentally Sound

On Uber: AVs, Take Rates, and Fragmentation

One of the primary arguments against Uber as an investment can be summarized in two words: AV disruption…

Read more

8 months ago · 170 likes · 4 comments · Manu Invests

Pete from Bulding Arks covers Uber very closely and is bullish on Uber. He mentioned that he holds it in his portfolio.

Building Arks

Review: Uber in 20 years

Summary…

Read more

5 months ago · 10 likes · 2 comments

Quality Stocks had Uber as one of 12 stock picks for 2026. Read well with later Q1 update on performance.

Quality Stocks

My 12 Top Stock Picks for 2026

Exactly 1 year ago I selected 12 stocks for 2026. This selection returned 24.4% so 6 percentage points above the S&P500 despite another year of concentration where a few stocks had most of the performance. Therefore the selection has proved to be a success. You can find the selection and the review just here…

Read more

8 months ago · 106 likes · Quality Stocks

The fundamentals: an engine that just switched on

For most of its life Uber lit money on fire. The venture years (2018 to 2022) were a subsidy machine, burning billions to buy growth.

Then it flipped.

Operating margin went from minus 29% in 2018 to plus 12% on a trailing basis. It crossed zero in 2023 and has compounded since. And the cash followed.

Free cash flow turned positive in 2022, then ramped to roughly $9.8 billion in 2025. And it is clean cash. The accruals are negative, meaning cash runs ahead of reported earnings, the opposite of the warning sign you look for.

Uber is now also retiring shares, after years of printing them to fund losses, buying back around half of its free cash flow and shrinking the count about 3.5% a year.

Now the part that trips most people up. If you pull up Uber’s P/E, it looks like a rather cheap tech stock at ~17x. Ignore it. The earnings number is junk.

Look at 2022. Operating income was minus $1.8 billion, a manageable loss. Reported net income was minus $9.1 billion. The extra $7 billion of “loss” was the paper writedown on Uber’s equity stakes (it owns chunks of Didi, Grab, Aurora and others), not anything operational. Then run it forward to 2024 and 2025, where net income is roughly double operating income. That gap is mostly a one-time tax benefit: Uber released about $6.4 billion of deferred-tax allowances in 2024 and another $5 billion in 2025, non-cash, never repeating.

Strip the tax releases and the equity marks and net income lands back near operating income, around $5 to $6 billion. Management knows this. In late 2025 it switched its headline profit metric away from net income toward adjusted operating income, which is a polite way of admitting the GAAP number overstates the real thing.

So UBER is not cheap on P/E. The real P/E is somewhere around 30. The honest lenses are free cash flow and the operating line. On those, the business is inflecting hard and generating real money.

The moat, and where it ranks

Two-sided networks are among the best business models there are. More drivers mean shorter waits, which means more riders, which means more drivers. Density compounds. Uber One adds a switching cost on the consumer side. In US rideshare, Uber holds about 76% share, with Lyft a distant, sub-scale second.

So the moat is real. But I want to be honest about its limits, because this matters for how much margin of safety I demand.

Apart from Uber One providing some sort of stickiness - switching costs are low. The same driver runs Uber and Lyft on two phones on the same dashboard. The same rider checks both apps for the cheaper fare. That caps pricing power and keeps the core service close to a commodity. Uber’s edge is liquidity and convenience, which is genuine, but it is not a toll nobody can route around.

Put it on a spectrum. A card network like Visa, or a credit-ratings business like S&P Global, is near-impossible to substitute.

Uber is a notch below that. It is a strong network, but a contestable one, and it carries a question those toll-takers simply do not have: a live technology threat that could rewire its supply side.

The right comparison set is other marketplaces, DoorDash, Airbnb, Booking, MercadoLibre, not the payment duopolies. That placement is exactly why I want a wider discount here than I would accept for a cleaner compounder I might consider with no MoS at all.

And we also covered another moat - vs AV disruption, but again, I think it is robust enough and we will cover it below.

Can the cash flow survive?

A value pick is not just cheap. It has to be durable. So: what could break this cash machine?

Ordinary competition and the cycle. Not much of a worry. The growth is volume-led, not price-led, trips up about 20% while price per trip is roughly flat. Penetration is still low, only around 10% of adults in Uber’s top markets use it monthly, so the runway is long. It is discretionary spending, so a real consumer recession would slow it, but rides and cheap meals are sticky habits, not luxury splurges. The one cost lever I keep an eye on is insurance, Uber’s single biggest variable cost, where it self-insures and carries large reserves; a bad run of claims hits the P&L directly.

The balance sheet. Fine. Net debt is around one times EBITDA, there is no dividend commitment, and the company is buying back stock rather than stretching. Nothing here forces a bad decision in a downturn.

Autonomous vehicles. This is the one that matters, and it deserves its own section.

The AV question

Everything above is favourable. Cheap, cash-generative, growing, a real network. The entire thesis comes down to one unresolved question, and the same facts read as both bull and bear.

The bull case

AV is just another kind of supply Uber plugs into its network.

Uber sold its own self-driving unit years ago and now positions as the demand-and-fleet-operations layer.

It says Waymo cars routed through Uber’s app in Austin and Atlanta run higher utilization and faster pickups than they do standalone, and autonomous trips on Uber grew about 10x year over year in Q1 2026.

I’d say that this is the most important overlooked aspect of Uber. It’s not just an app. It is everything that they’ve learned along the way to the top of the market. That is hard to replicate without burning through tons of cash. And if you are trying to build that capability while hammering CAPEX on your own fleet - that’s a steep climb.

Balancing a mixed human-and-robot fleet, and handling the unglamorous fleet ops (charging, cleaning, maintenance), is hard for a single carmaker to do alone.

In this version Uber stays indispensable and AV is not a threat - is the upside.

The bear case

Tesla and Waymo run their own apps. If self-driving supply concentrates in a few players and they go direct to riders, they compress Uber’s take rate or skip it entirely. The toll booth gets bypassed.

But lets vibe on that for a minute. Humanless taxis seem to be the obvious endgame for the industry. If you look at consumer reports - once people try them - they start to prefer them to human-operated ones. But it’s not clear-cut.

For some users privacy aspect matters. I prefer not to talk to taxi drivers for example, rather linger in my own thought. Making a private call or having an honest deep conversation with your companion is a no go.

Safety is also a concern. Drivers in the lower tiers can drive aggressively or can behave unpredictably. No driver - no risk.

So to me, personally - there is no wonder why this mode of transportation is taking off.

But. Today robotaxis are 30% more expensive. And it is not clear if they can get cheaper in the long run. Right now Waymo is in the same venture state as Uber was 5 years ago - subsidising growth. There is a case can be made that robotaxis are structurally more expensive.

There is also a symbiotic relationship between AV providers and ride-hailing service in general. Robotaxis do not cannibalise the traditional service, but expand it.

The Uber share got lower in cities where Waymo operates, true. But overall spending on transportation grew. And if the final model is Uber wiring Waymos through their app like they do in Austin and Atlanta - it’s a net win for Uber.

In Phoenix where hybrid model exists (both standalone app and Uber wiring exists) total rides for Uber increased after Waymo launch. In direct competitive markets (LA, SF and San Jose) results are mixed.

So the honest read: it’s too early to tell. Uber is now committing more than $10 billion to own AV supply directly (fleet deals, builder stakes, a 20,000-vehicle Nuro/Lucid arrangement).

This is the thing to track quarter by quarter: autonomous trips on Uber versus on first-party apps, whether the take rate holds in AV-heavy cities, and how fast the capex ramps.

UBER fair value

I built an FCFF discounted cash flow on the operating line, which routes around all that polluted net income. Revenue growing 15% fading to 7% over eight years, operating margin climbing from 13% toward 20% (still below the ~25% the segments already earn, so not a stretch), tax normalized to 22%, and a 8.94% discount rate. The terminal multiple works out to about 12x operating profit, not heroic.

That fair value is deliberately conservative, on three counts. The 22% tax rate ignores that Uber pays almost no cash tax today on old losses, so it gives up a real near-term benefit. The model holds the share count flat, crediting nothing for the ~3.5%-a-year buyback. And it excludes the $9.35 billion of equity stakes Uber owns in other companies, which are real assets, worth about $4 a share.

Fold those in and the variants look like this:

Here is what I like about that bottom row. Even if I assume Uber spends $10 billion-plus on autonomous vehicles and earns nothing back, a deliberately punishing assumption, fair value is still around $87, comfortably above the price. The downside case is not a loss, it is a smaller win. That is the kind of asymmetry I am looking for.

Against peers, the cheapness shows up clearly:

Among the genuinely profitable platform networks, Uber sits next to Booking as the cheapest on free cash, well below DoorDash and Airbnb, and it does it on the lowest beta of the group. Wall Street’s average target sits around $104 to $107, above my conservative $97. I am happy to be the more cautious number in the room.

The risks I actually worry about

AV disintermediation. Covered above. The thesis gate. If the take rate starts compressing in AV-heavy cities, I am wrong.

The capex pivot. The $10 billion-plus AV commitment is the asset-light story eroding in real time. If that spending scales faster than the cash flow can absorb, the clean FCF picture bends. Watch capex as a share of revenue.

Labour law. The US is largely settled. Europe is not: the EU Platform Work Directive must be written into national law by December 2026, and it carries a presumption that drivers are employees. Court rulings so far are split. A driver-reclassification ruling in a big market would hit the cost structure.

Earnings optics. Not a cash risk, but the equity-stake marks swing reported profit by over a billion dollars a quarter. It makes the headlines ugly and the comparison messy, and some investors will keep mistaking the GAAP number for the truth.

Multi-homing. The structural cap on pricing power. It is not going away.

Bottom line

So, is it time to buy Uber?

Almost. Not quite.

Here is the honest position: I do not own it. At around $72 the stock is about 6% above my buy line of $67.79, and roughly 35% below where I think it is worth.

That is close. It is the first genuinely cheap megacap I have looked at in a while, and it would diversify a book that leans heavily on consumer and apparel names. I want it.

But 6% above my line is still above my line. I set that 30% margin of safety precisely because the AV question is unresolved, and a network with low switching costs deserves a wider cushion than a clean toll-taker. Paying up 6% to chase it would be exactly the FOMO this newsletter is named against. So I wait. The trigger is $67.79, or a Q2 print in early August that either calms the AV fear or confirms the cash flow is holding through the capex pivot. A small dip, or one good quarter, and this moves from the watchlist into the portfolio.

Chuyka check

After all the math, my gut says the crowd is selling the wrong thing. They are pricing a 2028 robotaxi best case nightmare onto a 2026 cash machine that is compounding right in front of them. The network is the kind of thing that tends to survive transitions by absorbing them, and Uber spending $10 billion to own the threat tells me management would rather hedge than hope.

But the chuyka does not override the discipline. The AV question is genuinely open, not a fake bear narrative, and that is exactly why I want my full margin of safety rather than a partial one. The model says $97. The market says $72. My gut says the model is, if anything, conservative. And the same gut says: do not pay up six percent to feel clever. Wait for the pitch.

Position: None. Watching, with a buy trigger at $67.79 (30% below my $96.84 fair value estimate). Fair value range $87 to $101 across scenarios. Live peer data via FMP, June 2026. This is not investment advice. Think for yourself. Don’t trust online experts.

Trust your chuyka. Don’t let FOMO eat you alive.

13F is a mandatory quarterly filing for investment funds with more than $100 million in AUM. In this filing they disclose their position at end of quarter. Comparing it to the previous quarter gives a glimpse into what each fund is doing in terms of their conviction across the public long-only market.

Screening those filings for high-conviction moves, such as buying a new position at a substantial % of total holdings, adding or selling can generate a signal for own retail investors research. And that’s what I’ve been mostly doing.

Dataroma is good, but looks outdated now. Some of the facts that it provides are misleading in my opinion and might use a little fine touch.

I had to bolt on a screener that would surface decent ideas. A combination of AI agent, some python scripts and a data API.

There should be a better solution.

I went through WhaleWisdom, GuruFocus, TipRanks and other known sites that focus on consensus, yet found them too complicated. When you try to be everything for everyone - you tend to loose usability and focus.

I though of recreating a dataroma curated list of investors in WhaleWisdom but still WW did not give me the things I was personally looking for. UI was frustrating.

So I had to take a stab at this problem.

Let me introduce you to backrunner.io

I would not pretend that is it obviously better than other solutions. But it is set to improve on the old guard and add features that I was missing.

Here are those features:

• Curated list of funds - it expands core dataroma superinvestor list of investors by adding growth, quality and failed funds (yes, failed funds as well).

• Fund performance reconstruction - not just AUM growth/contraction but tracking purchases at a min/max/median quarterly price band as well as a follower performance (copycat portfolios executed at filing date deadline + 7 days) for clear picture on who is probably doing better than the broad market.

• Clean UI for easy reading and understanding of each filing, stock, who does what, who holds the stock etc.

• Robust database which allowed me to do some backtesting.

So yeah, backtesting.

After digging through some research I found that there is more to 13F than just single stock ideas.

There is a substantial body of evidence, that following great investors can generate returns that beat S&P500 index. There are different approaches to that. You can just copy the holdings of one that you fancy the most. Or you can hold the top holdings of funds. There are caveats to this but it seems to me that this strategy if implemented correctly survives the filing lag.

I will be digging into 13F Alpha deeper, so it’s probably a decent idea to subscribe to this substack.

Subscribe now

Don’t want to make this post any longer than it needs to be.

I would be glad if you check out backrunner.io and give it a feedback under this post.

I focused on making a robust and clean core screener but I have more useful features in mind.

Hope this helps.

The aim of this substack - is to cover my experience building an active portfolio after more than a decade of passive investing. That subject interested me for some time, so I’ve decided to give it a go.

Experiment is fair - when everything is transparent. I show all my positions with all ups and downs.

Currently mostly downs.

Fully transparent portfolio

Right from the start I thought I will create a live site for my portfolio which will help me track my results. So here it is.

research.chuyka.io

The site holds all my positions current and closed. Data updated daily. The detailed holdings are in the portfolio page.

Easy to view DCF models

My main issue with open portfolios on substack is that some authors do post their analysis and add an EXCEL DCF. Which is good. But downloading them and trying to get through while understanding all the assumptions relevant to evaluation - is as fun as getting wet on a cold rainy day. Not fun.

So I’ve made my tool for displaying DCF model inputs and outputs. You may have seen some of the charts that this tool produces in my Dorman thesis writeup.

Site currently holds only relevant stocks that are in my portfolio, but I will soon add others that are modelled but are on my watchlist waiting for an entry point.

Each line of the FCFF DCF is accompanied by relevant commentary and visual representation of the model output. So all calculations remain reproducible.

Models will be stored to be later compared. This approached helped me with Nike model teardown that highlighted many flows my older models had.

Now for the cool twist.

Substack ideas and chatter radar

What I personally struggle with as a solo investor - is idea generation. One of the approaches I found useful is monitoring the local substack community full of bright and determined investors. Number of substacks I’ve been reading have been growing rapidly and now is over 20. Which is a lot.

While most of the really useful stuff is behind paywalls - some of it out in the open. And what gives substack investor community it’s edge - authors really try to be genuinely helpful and thorough in their comments.

Most of the stocks on that radar are household names - NVDA, GOOGL, MSFT, META. It is fun to see how sentiment changes on them.

But most importantly sometimes a new name appears with conviction or comments from multiple authors - and that peaks my interest.

Those go into chatter picks.

I manually label them from data I pull in the background. This is where my analysis starts. A decent company goes into watchlist - and that page I will be developing next.

I hope to follow up with more useful features that I will bring from background into publicly useful tools.

Thank you for attention to this matter.

Consider subscribing.

Subscribe now

And going be rough.

Portfolio is in red this year. Total current value $54,550 against the $58,119 we started the year with. We’ve clawed back about a third of the drawdown in the past weeks. One position is up 67%. One name lost 37% from cost and I’m still holding it.

The bright spot is passive portfolio, but thats not the subject of this substack.

The portfolio right now

Positions as of Friday’s close, sorted by performance since purchase:

I will go through each position laying out my current understanding.

Garrett Motion (GTX): the one that is working

Entered March 6 at $18.49. Friday close $30.98. Up 67.6% from cost. The position is now 15.5% of the book and the largest by dollar value behind cash.

The full thesis is in the GTX deep dive post.

At the time of the analysis market priced GTX extremely attractively with a small caveat. Main business is ICE and it’s going out of fashion.

I happened to disagree. And instead of testing my conviction market suddenly repriced the stock launching it high.

I think what every one caught on is REEV. Range-extender EVs are a separate category from PHEVs. In China the pure PHEV growth has stalled, but the REEV segment grew 181% in 2023, 124% in H1 2024, and is forecast at 17% CAGR through 2030. Every REEV needs a turbocharger that runs at fixed-RPM peak efficiency. That is exactly the engineering Garrett is good at. Li Auto, BYD, Geely, Leapmotor are all REEV players. Leapmotor is being built in Europe via a Stellantis partnership starting 2026, which is Garrett’s home turf.

Q4 2025 results were operationally strong, FCF held at $403M, they announced a $250M buyback for 2026 (about 4.4% of market cap). FY2025 results in February were a clean print. The stock walked from $18 to $32 over the spring on no real news, just the bear case quietly dying as more people realized ICE wasn’t going to fall off a cliff by 2030.

My model fair value is $41 vs current price of around $30. The position is 15.5% of the book carrying +67.6% of unrealized gain. Selling now to lock the gain is tempting but I’m holding.

One thing worth mentioning. Oaktree Capital holds 6.57% of their fund in this name, going back to the Garrett bankruptcy and emergence in 2020-2021. They’ve been gradually selling. Fairfax (Watsa) holds a smaller position and has also been reducing. Big-name distressed investors taking profits is not the same as conviction breaking, but it’s a flag worth tracking.

Dorman Products (DORM): boring as a feature

Entered April 3 at $100.19. Friday $116.16. Up 15.9% from cost. About 11% of the book.

The DORM post went up April 10 a week after purchase and after a bit of growth. I know, I know. I was a bit late.

What got me sold on DORM was the ROIIC profile. And I was working to update my whole portfolio understanding based on that.

The full methodology is in the ROIIC post, but for DORM specifically: 10-year ROIIC of 11.1%, reinvestment rate 76%, intrinsic value growth around 8.5% per year. The DCF said fair value $154 against a price of $100 at the time. Both the quality lens and the price lens agreed. That doesn’t happen often. When it does, you take the trade.

Q1 came in clean. Revenue +4%, EBIT margins held in the 16-17% range, FCF generation strong, guidance reaffirmed. Capital allocation continues to be disciplined. The thesis hasn’t been tested by adverse data. The position just compounds quietly. The way it’s supposed to work.

The hope..

As with GTX I am holding DORM. No need to change anything yet.

On Holding (ONON): the bull case validated, but I bought too high

Last years purchase at $44.90. Friday $37.26. Down 17.0% from cost. About 7% of the book.

The NKE + LULU + ONON positions now drag my portfolio like never before. And the drama is in.

I don’t think I’ve ever told the story how I got into ONON. The short version is - I’ve started seeing it everywhere. Not just on runners, but every second person I’ve met at the local mall had a pair.

Held it ever since and I disappointed me every time I’ve looked at it.

On March 3 they reported Q4 and missed on guidance. Stock cratered to the high $30s. The next few weeks were ugly.

This is the only position I’ve never had a model on.

I needed a way to justify keeping it. So I built two.

A base case with conservative APAC deceleration following the Nike China growth pattern, and a bull case where APAC stays hot and tariffs get absorbed.

Base case said fair value $21.02. Bull case said $39.24. The stock was at $35. Which model was right was a $20-per-share question.

The Q1 print on May 12 sort of answered it. Revenue +26.4% in constant currency against bull-case expectation of 26.9%. EMEA +25.6% (sixth straight quarter at 25%+). APAC +61.4%, now over 20% of the business with no visible deceleration. Adj EBITDA margin guide raised to 19.5-20%, implying EBIT in the 15.3-15.8% range, ahead of even the bull case for year one. Tariffs absorbed without lowering profitability, which neither model had built in.

The bull case is now the working fair value. The ONON is the next LULU in China framing turned out to be the right read, and I had it as a side bet rather than the main case.

Pure gamble if you ask me. Could have gone the other way.

I am still down 17%. The bull-case fair value is $39.24. Even if the bull case plays out exactly, the position only recovers to about $39 against my $44.90 cost. The Q1 print closed some of the gap but the math says I paid too much.

The current question is what to do with it. Hold and see if ONON improves? Or just sell it now as it reaches my valuation?

Guess I will have to decide sooner, rather than later. It is a growth company that might run out of steam. Especially given the NKE and LULU drama that is going in the background.

One observation that matters more than the position math: ONON Q1 was the cleanest validation of any model I’ve run. APAC was the decisive factor and the framing I used in the bull case (premium athletic share-shift, runway intact) turned out to be exactly right. I’ll see if it stands.

Nike (NKE): the position I have to talk about

Entered April 14 last year $56.13. Friday $41.88. Down 25.4% from cost. About 14% of the book.

This is the awkward one. Not because of the loss. Because the loss happened in a specific way that I was transparent about.

I wrote about my mistake in a separate piece: the NKE teardown.

The post lays out the five mistakes: instant-recovery margins from year one (no turnaround drag modeled), ΔNWC set to zero across all eight years, top-line growth peaking at 10% for three consecutive years instead of building bottom-up by segment (where it tops out at 6.6%), CapEx declining while revenue grows (internally inconsistent), and using FCFE instead of FCFF which made all the other errors easier to miss.

The old model wasn’t aggressive. It was just garbage. But it’s my first. Like the first love. Utterly undefensible.

That teardown post was the most uncomfortable thing I’ve written so far. I owned a stock based on a fair value that wasn’t real. Walking through five specific errors in your own work is a different exercise than critiquing someone else’s.

It hurts.

But it had to be public because I learned a lot in the past year.

So now I’m holding a stock the new model says is overvalued. Why? Three reasons, and none of them are good enough on their own.

The brand is real. North America wholesale was up 11% in Q3 FY26. Running grew 20%. The classics cleanup is in its late innings. Elliott Hill has a coherent strategy and he’s executing it. Tariffs are a structural ~120bp permanent headwind I’ve baked into the model, but the rest of the business is recovering.

The model could still be wrong on the upside. My new fair value of $38.33 caps the terminal EBIT margin at 12.5% to reflect tariffs as permanent. If the tariff regime eases, that cap is too low. The bull scenarios for normalized margins are in the $50-70 range, but I’d need a specific things to happen to upgrade. But I am not there yet.

Third is China. Greater China comps were -10% in Q3 FY26 and guided to -20% next quarter. The competitive displacement from Anta and Li Ning isn’t just macro, it’s structural. The bull-case argument that “China comes back” is the weakest link. But NKE is investing into new experience there. Lets see how this plays out.

Together those 3 are ways for upside to show. But then again..

10-year ROIIC essentially zero. Reinvestment dominated by debt-funded buybacks rather than productive capital. The brand is still the brand, but the capital allocation pattern of the last decade was extracting value from the business, not compounding it.

The ROIIC numbers don’t tell me to sell. They tell me to be honest about what I’m holding. I am holding a brand cyclically depressed, with structural headwinds, where management is shifting from financial engineering to operational rebuilding. That’s a fine thing to own at $25. Less obvious at $42.

I’m holding a position the model says is overvalued, or I treat it as an exit candidate. The next forced data point is Q4 FY26 results in late June. Probably the right time to sit down and decide.

Nomad Foods (NOMD): the macro you can’t escape

Entered late last year at $12.68. Friday $9.80. Down 22.7% from cost. About 7% of the book.

I refreshed this model today. Fair value went from $35.84 (previous model) to $27.13. A 24% cut on the same business.

The reasons are worth talking about because they apply to almost every European consumer name with European factories.

NOMD makes Birds Eye fish sticks, Iglo frozen peas, Findus everything. 18 manufacturing plants across Europe. World’s largest buyer of MSC-certified fish. They source vegetables from European growers, chicken from South America, and they freeze everything in factories that consume meaningful electricity and natural gas. The pre-2022 EBIT margin norm was about 12.7%. The 11-year average. The 6-year average. The number I had in my head when I bought the stock.

That margin was built on a regime that no longer exists. Cheap Russian gas piped into European factories. Cheap nitrogen fertilizer from Russia and Belarus flowing into pea and spinach grower contracts. Cheap-ish diesel for cold-chain logistics. The fertilizer-driven cost of agricultural inputs sat in the low single digits as a share of revenue. Everything was cheap.

That regime is gone. The 20-F discloses something the headlines don’t make explicit. NOMD’s “hedges,” the ones the CFO references on every earnings call, are physical fixed-price supply contracts that run one to twelve months. There is no financial commodity hedge book. The Note 29 disclosure says they hedge FX and interest rates only. When the 2026 fixed-price physical contracts roll into 2027, the company is fully exposed to spot energy and spot fertilizer prices.

The CFO was explicit on the Q1 call. Coverage extends through most of 2026. The incremental inflation starts flowing through the P&L in Q4 2026 and into fiscal 2027 if current conditions hold.

I spent the morning today on a ground-up cost decomposition of NOMD’s COGS from the 20-F Note 6, cross-checked against peer disclosures. Factory energy lands at 3-4% of revenue. Logistics fuel another 1-2%. Vegetable raw materials, where fertilizer is the embedded driver, around 5%. Poultry feed (corn and soy, also fertilizer-driven) around 4-5%. Total energy plus ag-commodity-linked inputs land at 14-19% of revenue, which is roughly 19-26% of COGS. A 30% sustained increase in those inputs is a 130-290 basis point hit to EBIT margin, partly offset by the €200M three-year cost program, leaving net margin pressure around 100-150 basis points.

The terminal margin in my updated model is 10.75%, anchored to FY2025 reality, not 12.7% from previous years. The 2027 sags to 9.0% as the contracts repricing into elevated spot. Fair value falls from $35.84 to $27.13.

So am I selling? Probably not. At least not yet.

At $9.80 the stock is below the 50% margin of safety line of $13.57. The market is pricing something materially worse than the new normal model assumes.

The bull trigger to watch is macro resolution. Sustained EU TTF gas back to €20-25/MWh ($200 per thousand cubic meters), Brent in the $50-60 range, fertilizer normalized, all three together. That implies Russia-Ukraine resolution with Nord Stream restart, and Mid-East de-escalation. If any of that happens, this model is too conservative and the fair value needs to come back up toward 12% terminal margin and FV in the $30-35 range.

But it’s a strong candidate for sell. May be I am missing something. No matter how much negative factors I add to my model, it’s still valued very conservatively by the market. What am I missing?

Lululemon (LULU): the deepest cut, and the most reframed

Entered in December 2025 at $190.08. Friday $119.14. Down 37.3% from cost. About 6% of the book. The worst drawdown by percent. The most chaotic narrative.

The stock has been a study in how many ways a position can hurt you in less than a year.

Q4 FY25 earnings on March 19 missed and the stock fell from ~$170 to the low $150s. Then drift.

Then on April 22 the company named Heidi O’Neill as CEO and the stock dropped 13.33% in a single session on the news. Then more drift, down to $119 by mid-May.

The CEO event is interesting. Heidi O’Neill came from Nike. Twenty-eight years at Nike. Most recently President of Consumer, Product and Brand. The market read this as “Nike-arc CEO joining a Nike-arc company.”

The activist investor Elliott had preferred Jane Nielsen from Ralph Lauren and didn’t get her. Chip Wilson, the founder with a 4.27% stake, launched a formal GOLD-card proxy contest nominating three alternative directors including Marc Maurer from On Holding.

The board responded by appointing Chip Bergh (ex-Levi Strauss CEO) and Esi Eggleston Bracey (ex-Unilever) as independent directors. The picture is board-vs-activist-vs-founder, three-way struggle.

Most of the analysts did not like O’Neill as a CEO.

The bear case for O’Neill is that she came from the late-Nike era of DTC oversaturation and lifestyle-over-technical product. That era hurt Nike’s margins. The argument runs: she’ll do the same thing at LULU.

The reframe I now hold is that LULU doesn’t need the same thing Nike did. Nike was a wholesale-heavy business trying to transition to DTC. The over-DTC playbook hurt them because they cut wholesale and lost shelf space. LULU is already 95%+ DTC. They don’t need to dismantle wholesale. They need to defend premium positioning against Alo Yoga and Vuori without resorting to markdowns. May be some innovation and quicker product development might help.

O’Neill’s actual playbook at Nike, before the strategic drift took over, was the opposite of margin compression. She grew the NCX (NikePlus) membership program at roughly 30% per year to over 100M members. She built the SNKRS app where gamified members spent five times more than guest shoppers. She launched the House of Innovation flagships in New York and Shanghai where visitors spent 30% more with the brand in subsequent months. That is retention plus monetization-per-customer playbook. That is exactly what LULU needs to defend its premium without discounting. Whether she imports that playbook or imports the late-Nike drift is the real question. The activist contest aside, the playbook fit is better than everybody gives her credit for.

I also sourced her compensation package details from the April 22 8-K. Base salary $1.4M, target bonus 200% of base, $10M annual equity awards split 60% performance-vesting RSUs and 40% stock options, one-time signing equity $7M, and a $2M cash retention bonus. Pay at risk is about 77% of total comp. The 60/40 RSU-to-options mix is bullish-skewed for a turnaround. The specific RSU performance metrics are not yet disclosed and will land in the FY26 proxy statement, which is the next document to read.

So today I refreshed the model again. I extended the EBIT margin trough by one year to reflect what I’m calling the lost year. O’Neill doesn’t take over until September 8, 2026 because of her Nike non-compete. First strategic articulation lands around December 2026. First operating year fully under her is FY2027. The previous model had margin recovering from 18.5% in FY26 to 19.5% in FY27 to 21% in FY28. The lost-year version holds 18.5% in FY27 and pushes everything right by one year.

New fair value $185.17. 30% margin of safety line $129.62. Current price $119.14 is $10.48 inside the buy zone. Not deep in the buy zone, but inside.

I’m holding the position.

Probably adding more if it continues to drift.

CASH: 40% bracing for the impact

214 shares of an iShares T-bill ETF. About $21,500. Yields whatever the front of the Treasury curve is paying, currently around 4.3% on an annualized basis.

S&P at ATH. recently made his statement about current market valuation. And I mostly agree.

I diligently track a lot of stocks that satisfy my criteria for investing. But finding good value is impossible. And even quality is right now - all trades at a premium.

Berkshire recently disclosed that they’ve tripled their position in Google at current prices. Which can be a bullish AI signal. Yet they still sit at a substantial pile of cash. And a few fellow substack investors have been discussing what could that mean. The consensus is bullish and I will write about that in the future. I think it is different with Google in particular. Google may be one of the few companies that in my opinion will get hit the least when AI cycle will inevitably come to a close.

So this cash will go either towards ideas that I find like DORM or GTX, or will serve as a dry powder if we get a major sell-off near the end of the year.

What I’m watching

The named events in the next 30 days are short.

GTX Investor Day this Wednesday. Largest dollar position, $250M buyback announced, multi-year framework reveal. Post-event model refresh window May 25-30.

LULU proxy battle is interesting piece of drama.

NOMD is a decision. Might finalize my first loss in this experiment.

ONON decision frame still open. Q1 print validated the bull case. Waiting for Q2 and then we will see.

NKE structural call is the longest open question in the book. Held against a model that says overvalued. Will probably need to dive deep into this one after next print.

What I have learned in eight weeks

Three things, none of them surprising in retrospect.

Sitting in losing positions is harder than the books make it sound. Especially when the model fair value keeps falling with each new data point. The discipline isn’t in being right. It’s in being honest about whether the thesis is still intact while the price moves against you. I am at peace with the answer on every position right now. That does not mean I’m right.

Old models from 2025 were materially overstated. NKE’s came down 53% on rebuild. LULU’s came down 32% on first refresh. NOMD’s came down 24% today. The shape of the errors was similar across positions. Once the framework changed, the fair values changed. Learning every day.

The most useful addition to the toolkit was ROIIC. It’s not a valuation tool. It’s a conviction tool. It tells me whether the business is actually good before the DCF tells me whether it’s cheap. Two of my positions look very different after running it.

The plan stays the same. Real money, real models, real positions disclosed with entry prices and sizes. No daily updates. No signals. When I have something worth your time, I publish.

Subscribe now

Trust your chuyka. Don’t let FOMO eat you alive.

Not an investment advice. Think for yourself. Don’t trust online experts.

I guess it will be useful to compare expenditures to other investors, so feel free to jump into the comment section. But first, a little intro.

Subscribe now

A passive investing disclaimer

Overwhelming part of my investments are currently passive. I’ve been running a passive portfolio for over 10 years, and so far it compounded at 7% a year. Which is a decent result. In line with my expectations.

The passive portfolio is 4-5 ETFs (depending on the period) - IVV, IEFA, VGLT, VCLT, PCY. As you can imagine - holding this costs close to zero. Establishing this portfolio on 2015 cost me a couple of books and articles read.

Yet I could not sit still.

In 2014 before I’ve started my investment journey I’ve read the normal stuff you read before you start investing. I’ve been lucky to start with value investing stuff like Graham, Buffet with a bit of Lynch. I could have gone trading route, but God and broker commissions saved me from that choice.

Long story short - that value investing thing did not stick. I’ve figured it took too much time, which was a precious resource for a new father and a devoted oil and gas warrior. So I quickly went full passive.

Value-investing side-quest

Now, with a bit of extra cash laying around I chose to come back to my roots. I’ve set aside $55 000 and started writing online about my side-quest to invest with a value-oriented stock-picking approach.

If you’ve been following me - it’s not been smooth.

My portfolio is bleeding red and even with S&P at ATH it’s not getting any better.

Seeing my passive portfolio up in 2026 and comparing it to 8% draw down in open active portfolio made me ask myself a question. Why am I doing this?

I am spending a lot of time looking for opportunities, researching them, doing DCFs, custom data analysis tooling. Thats quiet a significant portion of my free time.

And while on this thought I’ve also started adding up my expenses to run this thing. And they go way beyond broker’s fees.

Here is the breakdown of my costs.

The total runs at $5 450 per year.

On a portfolio that is $55 000.

What have I done?

Components

And I can’t really exclude much from this table. Here is my rationale for each.

Financial Modelling Prep

I found FMP to be the best and fastest source of data for my use case. I use Ultimate tier, which is $99 per month if you pay for the whole year. It has 30 years of financial statement data, prices, earnings calls, insider actions, analyst consensus.

Everything feeds into my analysis. An overkill? Not exactly.

You need at least 10-15 years of historical data to judge the business quality through stuff like ROIIC.

Earnings calls are essential to get management sentiment. You don’t just need a single company’s earnings calls - you should look into the whole sector and cross-check how they view their business and if management you target company is delusional.

Insider actions and analyst consensus is critical to judge how market views the current state of things. If you are bullish on the stock and everyone else is selling - you should be at least cautious in your conclusions. Markets can stay irrational longer than you can stay invested.

So FMP is a bedrock that everything else is built on.

Claude AI

I use a second to highest tier of Claude Max. I don’t have a degree in CS but a relatively robust understanding of LLMs and their limitations.

The primary use is - coding all the tooling that I need for models, screeners etc., since I don’t trust LLM math.

Claude also helps with reading. LLM is ok with 10Ks and earning calls. And since I don’t have a whole day to pour through them - I use it as a shortcut and self-imposed limitation for my experiment.

The logic being - if I can’t do a research with LLM, then there is no chance for retail investor to actually make money reading other analysts digests.

The Max tier also is mandatory since I do work in short bursts. Other tiers are quickly maxed-out and no progress been made.

Gemini AI

Acts as my main research helper for specific questions through Deep Research tool.

I found that it is more robust and gives better numbers and links if you aim it at a problem. Claude is prone to hallucinations and short-cuts if you give him an open-ended task with web-search.

I also enjoy Gemini as a companion to watching series like Game of Thrones for asking lore questions and getting explanations. So cost justified in my opinion :)

Financial Times, WSJ, Bloomberg

Arguably something I can actually think of cutting.

With so much writing here on substack, I have a hard time keeping up with all three.

But there is one huge drawback to substack that I can’t deal with right now. It’s algorithmic feed. I have to sift through “this AI prompt will make your stock research for you” and “how I’ve predicted next 100 bagger” before I get to actually relevant pieces.

With news outlets it’s much easier right now. They are not polluted with self-promotion and click-bait. Only reporting on subjects that currently matter. And the reporting quality itself is very good.

So may be one of them should go. But that is a hard pill to swallow.

One of the few blogs that I am actually paying for to read. His move to substack coincided with me starting my stock-picking journey and I enjoy reading and learning from him.

I believe this to be one of the most high-value sources for both education and stock-picks. People say he predicted the last 8 market crashes that never happened. But at least he is the one with skin in the game and suffers from his public convictions for everyone to see with the rest of us. Nothing but respect and a huge shout out.

Cassandra Unchained

Cassandra Unchained is now Dr. Michael Burry’s sole focus as he gives you a front row seat to his analytical efforts and projections for stocks, markets, and bubbles, often with an eye to history and its remarkably timeless patterns.

By Michael Burry

A bit of an off beat blog for stock market dwellers. But as an oil and gas professional and an energy nerd I find it most entertaining.

The analysis is spot-on and predictions are also public and clear.

I’ve been working to add oil and gas to my portfolio, but probably being an insider and knowing all the risks I have been too hesitant to act. Now all opportunities have been priced-out for me. So until next time.

Still a worthwhile read in my opinion.

Doomberg

A lateral-thinking approach to energy, finance, and geopolitics | Subscribe below for free previews of new articles | Click through to our About page for pricing and FAQ.

Private Server Hosting

This is a part of my experiment. My aim is to have my portfolio live with every step logged for public. That means my portfolio will have to live somewhere online.

There are cheaper alternatives. But given that I am not a professional coder I have to choose options that are user-friendly.

The page is still in development and most of the screenshots you see on this substack are from there. I am aiming to bring it live sooner rather than later so stay subscribed :)

Value vs Cost

So the question remains.

Is $5 450 per year justified on a $55 000 portfolio?

Probably not.

Probably there are cheaper ways to do the exact same thing with a -2,65% from inception result.

But here at Chuyka my aim is to provide real open view into the dealings of a random retail investor that is trying to beat his own long-term passive result.

All passive-investing community is preaching - is that it is impossible.

I think it is possible. It just takes a lot of grit and perseverance to achieve this. And I am here to f-around and to find out myself.

Feel free to jump into the comments section and share how your expenses look. I bet there are different approaches to data acquisition and analysis. Will be glad to learn.

Trust your chuyka. Don’t let FOMO eat you alive.

Not an investment advice. Think for yourself. Don’t trust online experts.

If you choose to stay with me and bear my clunky English pieces I will be forever grateful. If not, I cannot blame you. Writing this is as hard as reading. But I am trying to improve. And improve I will.

The value investing cosplay circa 2026

The Graham/Buffett game in 2026 is mostly cosplay. People say “good business at a fair price” and what they actually mean is “find something cheap with a story I can write about.” The story does the work. Fundamentals show up later as decoration.

The trouble is stories are easy.

Every dropping stock has one.

Every undervalued narrative has a following of believers explaining why this one is different.

What rarely shows up is the part where someone explains, in mechanical terms, what makes a business actually good. Not “wide moat.” Not “great management.” Real numbers about real things.

I’ve been working through this question for the last few months and ROIIC is the cleanest answer I’ve found so far. Not the only answer. But the one that most directly tells you whether the next dollar this business reinvests is going to be a good investment, or just a dollar. That distinction is the whole game.

This post walks through two ways to compute it. Huber’s version and Mauboussin’s version. One is quick. One is precise. I’ll show the formulas, where the inputs live on the 10-K, and run them both through Walmart 2006-2016 to see what the numbers actually say.

This is not an original idea and I am following the footsteps of Huber and Mauboussin. But that’s the way I learn and I hope you will learn too.

Why headline ROIC lies

Return on Invested Capital (ROIC) looks simple. Earnings divided by the capital the business uses to earn them.

ROIC = Earnings / Invested Capital

A business earning 20% on capital is producing more per dollar deployed than a business earning 5%. Headline ROIC is that number for the whole company at a point in time.

The problem is that ROIC does not give the “business quality today” picture.

Capital deployed in 2010 at 25% returns sits next to capital deployed in 2024 at 8% returns and you get a respectable-looking 17% blended number. The blended number lies. As an investor buying the business today, you don’t get the historical 25%, or even 17%. You get the future returns of the capital deployed. Whatever the business does with its NEXT dollar.

Picture two buckets. First is an old capital the company deployed a decade ago when it was a different business in a different market. Second is new capital deployed over the last few years at current economics. Headline ROIC stirs both buckets together. The investor buying today is lucky to only fund on second bucket terms. If situation is improving - that’s a good thing. But I rarely see companies that are improving today with tariffs, wars and other stuff going on.

Return on Incremental Invested Capital (ROIIC) is different to ROIC in that it measures the second bucket. The marginal return on capital actually being deployed recently.

ROIIC = ΔEarnings / ΔInvested Capital

The delta is the key. We look only at incremental.

ROIIC = (Earnings_year_N − Earnings_year_0) / (InvestedCapital_year_N − InvestedCapital_year_0)

Two endpoints, one subtraction each, one division. The formula is simple. The hard part is picking the right inputs, and that is where we introduce two investing experts: Huber and Mauboussin. Both are household names.

And before we go into details - one teaser. Pair ROIIC with a reinvestment rate and you get the punchline formula for how intrinsic value of a company compounds:

Intrinsic Value Growth ≈ ROIIC × Reinvestment Rate

A business with 25% ROIIC and 80% reinvestment compounds IV at 20% per year. A business with 10% ROIIC and 25% reinvestment compounds at 2.5%. Same math. Wildly different investor outcomes over ten years.

Most of value investing debate is about the price. This formula is about the business. The price determines your entry; this formula determines whether the business is worth entering at all. And that’s exactly what we are looking for to determine if the business is actually any good.

Method 1: Huber

John Huber at Saber Capital wrote the cleanest practical guide to computing ROIIC I’ve come across. His Return on Capital series is the curriculum I wish someone had handed me when I started taking this stock-picking thing seriously. The piece that compiles the math is Calculating the Return on Incremental Capital Investments. Read it. My summary is shorter; his original is better.

Huber’s recipe is mechanically simple. Two numbers at year 0, the same two at year N.

Earnings:

Earnings = Net Income

Net income off the income statement. The number the equity holder actually pockets after everything.

Tangible Invested Capital:

Tangible IC = Total Debt + Total Equity − Goodwill − Excess Cash

Why each piece:

• Total Debt because interest-bearing capital is real capital the business has to earn a return on.

• Total Equity because book equity is the cumulative retained earnings and paid-in capital the business is operating with.

• Minus Goodwill because acquired goodwill is price paid above book value, not productive capital.

• Minus Excess Cash because cash sitting idle inflates the denominator and depresses ROIC. A rough heuristic: anything above 2% of revenue for a mature business, 5% for a growth business.

Here’s the statement-to-line-item cheat sheet:

Pair ROIIC with a reinvestment rate, computed two ways:

Method A:

Reinvestment Rate_A = ΔInvested Capital / Σ Earnings (over the period)

Of every dollar the business earned across the window, how many got plowed back into the capital base?

Method B:

Reinvestment Rate_B = (RetainedEarnings_year_N − RetainedEarnings_year_0) / Σ Earnings

Of every dollar earned, how many stayed inside the company versus walking out as dividends?

The two should match. When they don’t, the gap is telling you something.

• Method A > Method B by a lot

  • Debt or equity issuance funded the gap. Capital came from outside, not from retention. Ask: is the debt service sustainable? Was the equity issuance dilutive?

• Method A < Method B by a lot

  • Retained earnings were returned via buybacks or used for debt paydown rather than deployed into new productive capital. Ask: were the buybacks below intrinsic value, or buyback-to-nowhere?

• Method A ≈ Method B

  • Clean reinvestment via retention. Default healthy case.

The final idea stays the same either way:

IV Growth ≈ ROIIC × Reinvestment Rate

Huber uses net income because it’s simple and it’s what the shareholder actually takes home. He admits in the piece that NOPAT or owners’ earnings are more precise. Enter Mauboussin.

Method 2: Mauboussin

Michael Mauboussin and Dan Callahan published Calculating Return on Invested Capital at Credit Suisse in 2014. It is the modern industry standard, 23 pages, dense, and worth every minute. If Huber wrote the practitioner’s recipe, Mauboussin wrote the reference manual.

The core difference is the earnings input. Where Huber uses net income, Mauboussin uses NOPAT. Three changes come with that swap.

Change 1: EBITA, not EBIT.

EBITA = Operating Income + Amortization of Intangibles

Depreciation stays as an expense. It represents real physical wear on real physical assets. Amortization of acquired intangibles is just an accounting echo of past M&A and gets added back. The goal is to capture the true operating cash earning power of the business, undistorted by prior acquisitions.

Change 2: Cash taxes, not the statutory rate.

Cash Taxes = Tax Provision + ΔDeferred Tax Liability − (Interest × Tax Rate)

Three pieces. The tax provision from the income statement, adjusted for any unusual items. The change in deferred tax liability captures the gap between book and cash taxes. A growing deferred tax liability reduces the current cash burden. The third piece is the tax shield on interest, which has to be removed to put levered and unlevered companies on equal footing. Otherwise a company with lots of debt looks artificially more profitable just because its interest expense reduced its tax bill.

Putting it together:

NOPAT = EBITA − Cash Taxes

Or the modeling shortcut Mauboussin gives for projections:

NOPAT = EBITA × (1 − Cash Tax Rate)

The cash tax rate is typically a touch below the statutory rate for growing companies.

Change 3: Invested capital from both sides of the balance sheet.

Mauboussin recommends computing invested capital two ways and reconciling.

Assets side:

Invested Capital = Net Working Capital
                 + Net PP&E
                 + Goodwill
                 + Other Operating Assets

Where net working capital is current assets minus non-interest-bearing current liabilities. Accounts payable counts as non-interest-bearing. Short-term debt does not.

Liabilities + Equity side:

Invested Capital = Short-term Debt + Long-term Debt
                 + Deferred Tax Liabilities
                 + Other Long-term Liabilities
                 + Preferred Stock
                 + Shareholders' Equity

The two sides should agree. When they don’t, that is the information. Usually it is a minority interest line, a weird long-term liability, or an inconsistent excess-cash adjustment that didn’t get propagated to both sides. Hunt it down and fix it.

Mauboussin’s universal decision rule for anything borderline:

“If you are wondering whether to include an item in invested capital, simply ask if the company could generate the same level of NOPAT without the item. If not, include it. If so, exclude it.”

That single sentence resolves 80% of the edge cases.

The Mauboussin ROIIC formula itself is the same shape as Huber’s, sharper notation:

ROIIC = (NOPAT_year_N − NOPAT_year_0) / (InvestedCapital_year_N − InvestedCapital_year_0)

The two methods side by side

The two frameworks are not alternatives. They are complementary. I default to Huber for the first pass because net income is what shows up 10Ks. Mauboussin sits next to it as a divergence check. When the two answers agree, the Huber number is reliable. When they disagree by more than five percentage points, something is going on with leverage or taxes or intangibles and it is worth figuring out what.

Both are cleaner if you run them on 5-year and 10-year rolling windows rather than a single snapshot. The 5-year catches cyclicality. The 10-year smooths through a full business cycle. Reading them together tells you whether a dip is cyclical or structural. Mauboussin actually walks through a Cisco example where the 1-year ROIIC was +304% in 2012, −778% in 2013, and the 3-year rolling came out at 47%. The 1-year numbers are noise. The window matters.

Worked example: Walmart 2006-2016

This is Huber’s canonical example. The data is public, the conclusions are non-obvious, and the contrast with what headline ROIC says is exactly the teaching moment ROIIC was built for.

I’m rounding. The real numbers are in Walmart’s 10-K filings on SEC EDGAR.

Step 2: Compute Tangible Invested Capital (Huber)

Tangible IC_2006 = $42B + $53B − $13B − $6B = $76B
Tangible IC_2016 = $44B + $78B − $17B − $7B = $98B

Huber’s numbers in his worked example are $76B and $111B. Mine for 2016 came out at $98B. The difference is Huber appears to include Walmart’s accumulated other comprehensive income, minority interest, and a slightly different cash adjustment in his total capital number.

I’m going to use Huber’s numbers because that is what his published analysis uses, but you can see right here the “this calculation takes 15 minutes” is actually “this calculation takes 15 minutes once you’ve decided which version of each line item to use.” :)

Using Huber’s numbers:

Tangible IC_2006 = $76B
Tangible IC_2016 = $111B

Step 3: Compute ROIIC

ΔNet Income = $14.7B − $11.2B = $3.5B
ΔInvested Capital = $111B − $76B = $35B

ROIIC = $3.5B / $35B = 10.0%

Walmart deployed $35 billion of incremental capital over a decade and got back $3.5 billion in incremental annual earnings. That is the marginal return on everything built between 2006 and 2016. Ten cents on the dollar.

For context, Walmart’s headline ROIC in 2016 was 13.2%. The headline number is what most investor commentary was built on. The marginal number is three full percentage points lower and that is the one that actually matters for the next decade of returns.

Step 4: Compute the reinvestment rate

Method A:

Reinvestment Rate_A = $35B / $148B = 23.6%

Method B:

Reinvestment Rate_B = ($90B − $49B) / $148B = $41B / $148B = 27.7%

Pretty close to each other. Method B is a touch higher, which means retention slightly outpaced deployment, which means some of the retained capital went into buybacks rather than productive capital. For Walmart that checks out. The company was buying back shares throughout this period. Not a red flag, just a description of what they were doing.

Step 5: Compute the compounding rate

IV Growth ≈ ROIIC × Reinvestment Rate
          ≈ 10% × 24%
          ≈ 2.4% per year

The investor who bought Walmart in 2006 should have expected about 2.4% annual intrinsic value growth just from the reinvestment math. Add the dividend yield (call it 2-3%) and some multiple expansion and you’re looking at total returns in the 5-8% range. Which is roughly what Walmart did from 2006 to 2016. The math worked.

Step 6: What the numbers actually say

Walmart in 2016 looked like a quality compounder. Headline ROIC 13.2%, the largest retailer in the world, steady dividend, defensive earnings. Every generic value investing commentary could include Walmart and nobody would blink.

The ROIIC decomposition tells a different story.

Marginal capital earning 10%, reinvestment at 24%, intrinsic value compounding at 2.4%. A legacy moat in late stage. The business was still high quality, but the per-dollar returns on new investment were already fading, and the total returns story depended almost entirely on dividends and buybacks, not on the underlying business growing.

Headline ROIC said great business. ROIIC said great business that had run out of places to put capital. Both statements were true. The second one was the one that mattered for buying in 2016.

What Mauboussin would do differently

Same company, same 10-K, same period. Mauboussin’s framework changes the numbers in three places.

Operating leases. Pre-2019 Walmart had a massive off-balance-sheet operating lease book. Store leases, equipment, vehicle fleet. ASC 842 hadn’t arrived yet. Rough-cut capitalization at 6-8x annual rent expense would add somewhere around $20-30 billion to the 2006 invested capital base and $25-35 billion to the 2016 base. Both endpoints get bigger. The differential also gets bigger. Net effect: ROIIC comes down another 1-2 percentage points.

NOPAT instead of net income. Walmart is not SBC-heavy and its tax rate is stable, so the swap from NI to NOPAT is cleaner than it would be for a tech company. The EBITA add-back for amortization of intangibles is small because Walmart’s intangibles are mostly goodwill from acquisitions, which doesn’t amortize. The cash tax adjustment matters a bit. Probably moves the numerator by a low single-digit percentage.

Both-sides reconciliation of invested capital. Walmart has minority interest from its international joint ventures, which appears on the liability side but not the asset side by default. The reconciliation forces you to decide how to treat it. Mauboussin’s convention is to include it. Adds a few billion to the denominator.

Net effect on Walmart 2006-2016: Mauboussin-adjusted ROIIC probably lands somewhere around 7-8%. Huber’s number is 10%. Both point in the same direction and both tell you Walmart in 2016 was a mediocre business for incremental capital deployment.

That directional agreement is the signal. When Huber and Mauboussin agree on a name, the simpler number is reliable. When they disagree materially, you have real work to do figuring out which distortion is responsible. For Walmart they agree. I trust the Huber number. For a debt-funded roll-up like TransDigm or a SBC-heavy hyperscaler like Nvidia, I would run both and dig into any gap.

Adjustments worth knowing about

Each of these can move ROIIC by several percentage points when they apply. Links go to the primary sources.

Excess cash. Strip it. Cash above about 2% of revenue for a mature business and 5% for growth is excess. Apple historically had hundreds of billions parked on the balance sheet and its ROIC looked mediocre until you stripped the excess cash, at which point the underlying business turned into one of the highest-ROIC operations in history. Direction: strip excess cash → IC goes down → ROIC and ROIIC go up.

Goodwill, in or out. Compute both. Including goodwill answers “did past M&A create value at the prices paid.” Excluding goodwill answers “what does the underlying organic operation earn on tangible capital.” Different questions, different decisions. For evaluating future M&A specifically, ROIC is the wrong tool entirely; Mauboussin says use NPV instead.

R&D capitalization. For software, pharma, semis, anywhere R&D is above ~5% of revenue, expensing R&D as it happens understates both the capital base and the NOPAT. The fix: capitalize cumulative R&D as an intangible asset, amortize over an industry-appropriate life (3-5 years for software, 10-15 for pharma), and replace the current period’s R&D expense in NOPAT with amortization of the accumulated balance. Mauboussin’s worked example: Microsoft FY2013 R&D was $10.4B vs $4.3B in capex. R&D was 2.4x the size of capex. Without capitalization, both NOPAT and invested capital are badly understated.

Owners’ earnings as the input filter. For tech and other SBC-heavy businesses, GAAP net income overstates real economic earnings because stock-based compensation gets added back as a non-cash expense. The cash spent on buybacks to mop up the resulting dilution is real money. Burry and Clifton have the cleanest treatment if you want the full version. For Nvidia, Meta, Palantir the right earnings input for ROIIC is owners’ earnings. For non-SBC-heavy businesses the adjustment doesn’t matter.

Hidden reinvestment in the income statement. Marketing spend at consumer brands. Customer acquisition cost at subscription businesses. R&D as already covered. These run through opex but function economically as growth investments. Separating maintenance from growth spend is harder than it sounds. The realistic move is to flag where it’s happening and note that the reported capex line is undercounting true reinvestment. A consumer brand spending 10% of revenue on marketing is reinvesting heavily even if its capex says otherwise.

Wrap-up

ROIIC is a quality screen and a trajectory tool. It tells you whether a business is good and whether the “good” is getting better or worse. It does not tell you whether the price is fair. That is what DCF is for. The two work together. I now use quality filter first, valuation second.

The reason to have both Huber and Mauboussin in the toolbox is that the gap between them is itself a signal. They agree on clean operating businesses with stable tax rates and modest intangibles. They disagree on leveraged roll-ups, SBC-heavy tech, and heavy-lease retailers. Each disagreement points you at the specific thing that distorts the simpler number. That is more useful than picking one methodology and pretending you know the answer.

ROIIC is also historical. Past ROIIC does not guarantee future ROIIC. Chipotle’s 35% ROIIC from 2006 to 2015 compressed sharply after 2015 as the unit growth runway saturated. The math gives you the baseline. The qualitative judgment about whether the moat persists and the runway stays open is still on you. The math just makes the qualitative question worth asking.

What I take away from spending the last few months on this: the catchy story is almost never the answer. My investments into NKE, LULU, ONON - are the prime example. The mechanical question of whether the next dollar of capital is going to earn a real return is almost always the right question. ROIIC is the first tool I reach for now. Everything else comes after.

Trust your chuyka. Don’t let FOMO eat you alive.

Not an investment advice. Think for yourself. Don’t trust online experts.

Speaking of the recent rally - here is the current state of my portfolio. It still feels like garbage. But I stand behind most of my positions. There is still capital to deploy and that’s what I am working on right now.

The one stock that’s quietly making money sells brake hardware, exhaust manifolds, and window regulators. Not AI. Not cloud. Not athleisure. Replacement auto parts.

There’s a lesson in this and it took me a while to see it.

Subscribe for more revelations

What Dorman actually does

When the window motor on your 2014 Honda Civic dies, you have two choices. Go to the dealer and pay $400 for an OEM part. Or your mechanic orders a Dorman part for $120 that fits the same, installs the same, and often lasts longer because Dorman actually fixed the design flaw that made the original fail.

That’s the business. Dorman reverse-engineers parts that break on cars, improves them, and sells them to mechanics and DIY enthusiasts through distributors like AutoZone and O’Reilly. They have 138,000 different parts in their catalog. Last year they introduced 5,335 new ones.

The key detail: about a third of what Dorman sells has no aftermarket alternative. They’re the only option between the dealer and nothing. That’s not a brand story - it’s an engineering moat. You can’t copy a reverse-engineered transmission control module by putting a different label on a Chinese generic. You need electrical engineers who can decode proprietary software, test it, and certify it works. Dorman has been doing this for 15 years. Their competitors haven’t started.

Why aftermarket is a compounder’s game

Here’s a number I keep coming back to: the average car on American roads is 12.8 years old. That’s a record.

Cars don’t get younger. Every year a car stays on the road, it needs more parts. Window regulators fail. Brake calipers seize. Exhaust manifolds crack. Electronic modules glitch. This isn’t cyclical demand that depends on consumer confidence or interest rates. It’s entropy. Physics doesn’t care about the Fed.

The aftermarket grows when car sales are strong (more cars on the road eventually). It also grows when car sales are weak (people keep older cars longer). The only scenario where it shrinks is if everyone stops driving, which hasn’t happened in a hundred years.

And the mix is shifting in Dorman’s favor. Modern cars have more electronics, more sensors, more modules that can fail. A 2024 vehicle has roughly 3x the electronic content of a 2010 model. When a $15 mechanical part becomes a $200 electronic module, the aftermarket grows without a single additional car on the road.

Dorman’s revenue hasn’t declined in 25 years. Not during the financial crisis. Not during COVID. Not during the tariff wars. The 30-year revenue CAGR is 10.2%. I haven’t found another business with that consistency. Take that with a grain of salt, it’s been just under a year, that I’m back to active stock-picking.

Is Dorman Products actually a quality business?

Good story is not enough. “Cars get older, parts break” is a nice narrative, but narratives don’t compound. Capital does. I need to know whether Dorman earns a real return on the money it puts back into the business - or whether it’s just growing revenue while destroying value.

I’ve been working through a framework for this. It comes from John Huber at Saber Capital (check out his substack, it’s a goldmine, now archived) - a fund manager who thinks about quality in terms of what the next dollar of invested capital earns, not what the historical average looks like. The concept is called Return on Incremental Invested Capital, or ROIIC.

The ROIIC idea is simple. Take the change in net income over 10 years. Divide by the change in capital invested over the same period. If the business earned $100M more than a decade ago and deployed $500M of new capital to get there, the incremental return is 20%. Every dollar invested created 20 cents of new annual earnings.

Headline ROIC - the number most screeners show - tells you about the average return across all capital ever deployed. A business that earned 25% in 2005 and earns 10% today might show 18% headline ROIC. Looks fine. But the marginal return is collapsing. ROIIC catches that. The investor who buys today is buying the marginal future, not the historical average.

ROIIC is a great indicator of quality. We must bear in mind that it has a trailing nature and past performance does not guarantee future results. But it is still a great way to analyse a company in numbers terms.

I’m writing a full post on this framework - the math, the methodology, how I apply it across my portfolio. For now, here’s what it shows on Dorman.

What the DORM ROIIC numbers say

Three things stand out.

First, the V-shape. ROIIC compressed from 10.4% to 4.4% in 2022-2023, then recovered to 11.1%. That was in my view the SuperATV acquisition digesting. Dorman bought a powersports parts company, funded it partly with debt, and spent two years integrating. Capital went up, earnings hadn’t followed yet. By 2025, the deal is earning and ROIIC is back above pre-acquisition levels.

Second, the 5-year is higher than the 10-year. 12.3% vs 11.1%. This is the recovery signal - recent capital is earning better than the long-run average. The SuperATV acquisition drag is washing out of the shorter window. When the 5Y is below the 10Y, the business is deteriorating. When it’s above, the business is improving. Dorman is improving.

Third, the reinvestment rate is 76%. Dorman puts three quarters of its earnings back into the business. Not buybacks. Not dividends. Back into new SKUs, new distribution, new engineering capability. And it earns 11% on every reinvested dollar. That’s management vote of confidence for their business engine.

Pop the the hood

There’s a second lens on this that I find even more interesting. When I strip out the financing costs - the interest on the debt Dorman took on for acquisitions - and look at the pure operating return on tangible capital, the picture changes dramatically.

The operating return is 26.8%. Not 11.1%.

The gap - 15.7 percentage points - is entirely the cost of acquisition debt. Dorman borrowed to buy SuperATV and a new deal in 2025. The interest on that debt eats into net income, which is what the 11.1% measures. But the underlying operations are earning 27 cents on every dollar of tangible capital deployed.

Why does this matter? Because debt gets paid down. Dorman’s total debt has already dropped from $849M to $633M since the SuperATV peak. As it continues to decline, the interest cost shrinks and more of that 27% operating return flows through to shareholders. The equity-holder return is converging upward toward the operating return.

It’s like buying a rental property with a mortgage - the property earns 12% on its value, but after mortgage payments you pocket 5%. As you pay off the mortgage, your take-home rate climbs toward 12%.

The moat assessment

Where does the moat actually sit?

It’s not brand (nobody knows Dorman outside of mechanics). It’s not patents (they have 500+, but the moat isn’t a patent wall). It’s not network effects. It’s a process advantage with three layers:

Speed. Dorman identifies failure modes, reverse-engineers a fix, and launches before anyone else. 1,659 “new to aftermarket” parts last year. By the time competitors copy, Dorman has moved to the next failure mode. The moat resets constantly - it’s a treadmill, not a fortress.

Complexity. Modern cars are stuffed with electronic modules - transmission controllers, engine computers, electronic steering racks. Reverse-engineering these requires decoding proprietary software. Dorman has 15 years of accumulated capability here. A Chinese generic manufacturer can copy a brake caliper. They can’t copy a transmission control module. And every year, the electronic content per vehicle increases. The barrier gets higher over time, not lower.

Scale. 138,000 SKUs. Distributors like AutoZone want one supplier who covers the catalog, not 50 vendors. Once you’re in the system at that scale, you don’t get ripped out. And when 21% of your competitors are in financial distress (per Lazard 2025), your 14% EBIT margins let you absorb the pricing pressure that bankrupts the 5%-margin players. You get their shelf space.

None of these alone is a wide moat. Together, they create a narrow-to-moderate moat that strengthens as the vehicle fleet ages and grows more complex. That’s the quality case.

The contrast that made it click

You still can feel that I’m sour from my Nike misjudgement. I tore it down last week and I am still digesting this position.

Nike earns less today than it did in 2015. It deployed $6.9 billion of incremental capital over a decade and generated negative $54 million of incremental earnings. The entire capital increase came from debt that funded share buybacks - not a single dollar went into productive reinvestment that grew the business.

Dorman is doing the opposite. It reinvests most of its earnings, earns 11% on that reinvestment (27% on an operating basis), and the business keeps growing. The trajectory is up, not down.

One is a compounder. The other is a cash machine running on fumes, returning money it doesn’t really have.

I own both. Draw your own conclusions about which one I’m more comfortable with.

What I paid and what I think it’s worth

I bought 50 shares at $100.19 during the early April sell-off. The stock was trading below my 30% margin of safety level for the first time.

The DCF model - built segment by segment, stress-tested with two NWC scenarios, cross-checked against a March version that arrived at the same number via different assumptions - says fair value is $154 per share.

That means I bought at 35% below fair value. In my framework, that’s comfortable. Not euphoric, not skinny - comfortable.

At the current price of $110, the stock is still 28% below fair value. The margin of safety hasn’t evaporated. It’s narrowed, but it’s there.

Key model assumptions:

• Revenue growth: 7-9% (management’s 2026 guide; I use 7%)

• EBIT margin: 15.5% near-term (tariff drag), recovering to 16.5% steady state

• WACC: 8.72%

• Terminal growth: 2.5%

The model is conservative by design. It assumes margins stay below the pre-acquisition 17-20% range.

If margins recover to 18%, fair value moves to $175+. I’m not modeling that because I haven’t seen it yet.

The risks I’m watching

Customer concentration. AutoZone, O’Reilly, and Advance Auto Parts together account for a significant chunk of revenue. If Advance Auto goes bankrupt (it’s struggling), Dorman loses a customer but likely picks up the volume through other channels - mechanics still need parts. The bigger risk is pricing pressure from consolidated customers.

Tariffs. Dorman sources about 38% from China (down from 70% a few years ago). Management has been diversifying aggressively, but a Section 232 tariff at 25% on aftermarket parts would compress margins by 1-2 percentage points. This is already in my model.

The SuperATV question. Dorman bought a powersports company (UTVs, ATVs) that’s 13% of revenue and carries lower margins. This is an adjacency move - auto aftermarket company branching into recreational vehicles. Adjacency diversification is a yellow flag in the ROIIC framework. It suggests core auto runway might be narrower than management says. I’m watching whether SuperATV margins converge upward or stay as a permanent drag.

Boring doesn’t mean safe. The stock dropped 30% from its 2025 highs before I bought it. It can do that again. Dorman is illiquid by large-cap standards - a $5.5 billion market cap with moderate trading volume. It moves fast in both directions when the market gets nervous.

Why this matters beyond one stock

I started this portfolio last year, and the most important lesson so far is that my instincts about what makes a good investment were exactly backwards.

I was drawn to the names with the stories. Nike - iconic brand, turnaround, new CEO. Lululemon - premium athletic wear, loyal customers, international expansion. On Running - the hot shoe brand that everyone’s wearing. These names selling at a discount were looking like slam-dunks.

Every one of those “exciting” positions now is deeply underwater. I know it’s too early to tell, but the boring one - the one that sells exhaust manifolds and window motors - is the one actually making money.

This isn’t a coincidence. Boring businesses often compound because nobody’s paying a premium for the story. There’s no narrative to inflate the multiple. No influencer wearing Dorman brake calipers on Instagram. The stock trades on cash flow and earnings growth, not on vibes.

The ROIIC framework made this visible in a way I couldn’t see before. Nike has a famous brand and negative incremental returns. Dorman has no brand recognition outside of auto mechanics and earns 11% on every dollar it reinvests. The framework doesn’t care about your feelings. It measures capital efficiency.

I’m not selling Nike or the other positions - each has its own thesis and timeline. But if I’m being honest about where my conviction is highest, it’s in the company that makes car door handles.

Chuyka check

My gut says this is the kind of stock you buy and forget about for five years. The business doesn’t need a turnaround, doesn’t need a new CEO, doesn’t need China to cooperate, doesn’t need interest rates to drop. It needs cars to keep getting older. They will.

I am much more comfortable holding this position than many others in my portfolio. And if fundamentals stay strong I am happy holding it for years to come even if bargain price keeps persisting.

The model says it’s worth $154. My gut says the model is conservative. The fleet keeps aging. The electronics keep getting more complex. The catalog keeps growing.

Sometimes boring is the best thing you can own.

50 shares of DORM at $100.19 average cost. Current price ~$110. Fair value estimate $154. This is not investment advice. Think for yourself. Don’t trust online experts.

Trust your chuyka. Don’t let FOMO eat you alive.

Go read it if you want. It’s in Russian, but the numbers are universal.

Дело пахнет керосином

Открываем сезон активных инвестиций

Отрицать свою предвзятость в выборе первой акции для возобновления вэлью-портфеля бесполезно. Но мы помним слова классика…

Read more

a year ago · Ilya

But the story went in a wild direction.

Subscribe for wild stories

For a while, the market agreed with me. Nike ran to $74 - up 31% from my entry. I was sitting on $3,000 in profit, watching the thesis play out, feeling like a value investing god. Six months back into stock picking after a decade of index funds and already up 31% on my first high-conviction idea. Hard not to feel validated.

Then it went to $44.

I just rebuilt the model from scratch. New framework, proper assumptions, segment-level revenue, actual working capital math. New fair value came out at $38.

Not $77. Not even close. $38.

The stock currently trades around $44. I own 173 shares at a $56.13 cost basis. That’s roughly $2,000 in unrealized losses, or about -21% on the position. This mistake has a price tag, and I look at it every day in my portfolio.

This post is about the five specific mistakes that created a $43 gap between what I thought the company was worth and what it’s actually worth. Not hindsight. Every one of these errors existed when I built the original model and could have been caught then. I just didn’t know what to look for.

What the old model said

The bull case was straightforward. Nike has problems - everyone knows this. Failed DTC strategy, aging product lineup, new CEO cleaning up the mess. But the brand is iconic, free cash flow is healthy, and turnarounds at this scale tend to work eventually.

I used a simple FCF growth model. Conservative start - 5.5% growth while the turnaround plays out, then ramping to 10% as the recovery kicks in, then tapering back down to a 2.5% terminal rate. 9% WACC. Fair value: $77. At $55-56, that’s a 30% margin of safety. Textbook value entry.

I built a detailed Excel DCF.

Except it was wrong. By a lot.

What the new model says

Six months later, I rebuilt everything using an FCFF framework with segment-level revenue, explicit margin modeling, and actual working capital assumptions. Here’s the comparison.

Same company. Same public data. I was enthusiastic. After a decade of passive index funds, I was finally doing the stock picking thing again. I wanted to start buying. That energy is good for getting started. Less good for catching your own mistakes. The difference between these two models is not information - it’s questions I asked myself and answers I found.

The five mistakes

1. Real life is messy. Still try to model it.

On the revenue side, I actually did model a downturn - slow start, then ramp. I’ll show it below.

But on margins, I didn’t. I applied a flat 10% net margin from Year 1 through Year 8. No recovery curve. Just the destination.

Nike’s actual net margin was around 7% at the time. It has since fallen to 4.8%. The company was mid-CEO transition, reversing its DTC strategy, destroying $4 billion worth of classic franchise inventory, and eating tariff costs. All happening simultaneously. Every one of these drags on margins, and every one of them takes time to work through.

Look at the gap in the early years. That’s where the damage happens. The old model assumed 13% from day one while reality was sitting much lower and falling. The new model starts EBIT at 7.5% and takes six years to reach 12.5%.

The fix is to actually model the real world. Start at the current depressed level and build the margin back with specific drivers - NA wholesale returning, running category growing, classics cleanup finishing. Not the 13-14% Nike used to earn - tariffs make that ceiling unreachable now. 12.5% is the new normal in my opinion.

How did I calibrate the recovery speed? I looked at other consumer companies going through similar transitions. Lululemon’s 2013 CEO change cost 740 basis points of operating margin over three years. Gap lost 710 bps in its transition year. Under Armour lost 970 bps starting in 2022 and still hasn’t recovered. Even Starbucks, with arguably the smoothest handoff, lost 200 bps. These are all consumer brands with strong positioning. The pattern is consistent - 1 to 2 years of margin drag from decision paralysis, talent churn, strategy resets. Nike is doing CEO transition, DTC reversal, inventory cleanup, and tariff absorption all at the same time. Modelling instant recovery was fantasy.

The impact: Year 1 cash flow was overstated by roughly 60-80%. That’s $4.7 billion modelled vs $2.8 billion actual. Years 1-3 get the heaviest discounting in a DCF, but overstating them by that much still compounds through the whole model.

The lesson: If a company is changing strategy, don’t just model the final state. Try to understand how fast things actually move in the real world. Look for precedents - other companies that went through the same kind of transition. Study how their fundamentals reacted and how long the recovery took. History doesn’t repeat itself, but it sure rhymes. Use that rhythm to calibrate your model.

2. A growing business ties up more cash in inventory and receivables

I set change in net working capital to $0 for all eight projection years. So translating it to normal-people-speak - I was assuming it wouldn’t grow.

Think about what working capital actually is in a physical business. It’s sneakers sitting in warehouses waiting to ship. It’s invoices sent to Foot Locker and Dick’s that haven’t been paid yet. It’s the float between Nike paying its Vietnamese factories and collecting from its retail partners. That’s real money tied up in the physical operation of selling shoes.

When revenue grows, all of that grows with it. More shoes sold means more inventory in the pipeline, more receivables outstanding, more cash locked up in the cycle. Nike’s 11-year average is about 1% of revenue per year in incremental working capital needs. On $46-65 billion in projected revenue, that’s $500 million to $1 billion per year that the business absorbs before it becomes free cash flow.

Over eight years, ignoring this drained roughly $6-7 billion of cumulative cash flow from the model. That’s $4-5 per share in present value that simply didn’t exist.

The lesson: If the business is physical - it makes things, stores things, ships things, invoices customers - then growth costs working capital. The faster it grows, the more cash gets locked up. Model it as a percentage of revenue growth and check it against history. A $0 change only makes sense for a business that collects cash before delivering the product.

3. Build revenue from segments, not a single growth rate

As I mentioned above - I did model the downturn. The old growth curve started at 0%, ramped to 4%, and only then hit 10%. I wasn’t ignoring the downturn on the revenue side. The problem was where I assumed the recovery would peak.

Where did 10% come from?

I looked at Nike’s history and anchored to the best years. FY2012 was +16%. FY2015 was +10%. FY2023 was +9.6%. The COVID rebound in FY2021 hit +19%. If you cherry-pick the peaks, 10% looks conservative. But Nike’s 15-year median growth is 6.0%. The 10-year CAGR ending FY2025 is 4.2%. I anchored to the highlights, not the average.

And there’s a deeper issue.

The Nike that grew 10% in FY2015 may not exist anymore. Hoka, On Running, and New Balance own the cultural moment in running and lifestyle. Nike’s sportswear market share fell from 15.2% to 14.1%. China - which was 18% of revenue at peak - is structurally losing ground to Anta and Li Ning with government propping them up. Converse is in decline with no recovery timeline. The competitive landscape shifted, and a single growth number can’t tell you that.

If you try to build revenue from segments to get 10% consolidated growth - it requires miracles from Nike everywhere at once. North America is 42% of revenue and realistically does 6% in a good year. China can spike to 15% in a recovery quarter, but it’s only 13% of the total. EMEA peaks at 8%. And Converse is declining.

Blend it all together and the consolidated peak is 6.6%. Not 10%.

Look at this chart. Thirteen years of history on the left, projections on the right. You can see exactly where the 10% anchor came from - North America hit 18% in FY2014 and FY2023. China was the growth engine at 20-24% in the mid-2010s. Those peaks made 10% consolidated growth feel conservative. But those peaks came from a different Nike in a different competitive landscape. The projections on the right are deliberately calmer. Five regions, five completely different trajectories. China goes from -12% to +15% and back. North America is the steady engine but it’s slowing. No single number captures this.

The difference in terminal revenue: $75.7 billion in the old model vs $64.8 billion now. Every dollar of that excess flows through to margins and cash flow.

There’s another benefit in building by segments I didn’t expect. And this approach applies to modelling as a whole.

When you model by segment (or disaggregate other compound numbers), you’re building around the same metrics management tracks and optimizes. Earnings calls show you where their eyeballs are. Hill talks about NA wholesale recovery, China store resets, running category growth - these are specific, measurable goals. If I model NA at +6% because management is focused on wholesale rebuild, I can check that against actuals every quarter. Did they hit it? Are they ahead or behind? The model becomes a scorecard for management execution, not just a fair value calculator.

A single growth rate gives you nothing to check against. “Did Nike grow 10%?” is a useless question. “Did North America wholesale recover while China stabilized?” - that’s a proper question.

The lesson: Build revenue and other compound numbers from components. A single growth rate invites anchoring to historical peaks. Bottom-up forces you to justify where each dollar of growth comes from. It’s much harder to be accidentally optimistic when every geography has to work on its own. And it gives you a framework to judge management against their own promises.

4. CapEx must be consistent with revenue growth

I started CapEx at $919 million and declined it to $482 million by Year 8. Nike had been cutting CapEx for years - from $1.1 billion to $430 million. I just extrapolated the trend (wtf.. why?).

But those cuts are just turnaround measures. Cash conservation during restructuring. $430 million was Nike’s lowest CapEx since 2003, when the company was one-third its current size. You can’t grow revenue from $46 billion to $76 billion while spending $482 million on property, equipment, and technology. That’s 0.6% of revenue. Below maintenance for a company with $7.5 billion in PP&E.

And think about where the money needs to go. Hill’s strategy is “elevated retail presentation” - making stores feel premium again after years of underinvestment. China has 5,000+ monobrand doors that management admits “aren’t compelling”. They’re running pilot store resets in Beijing and Shanghai that show better traffic and comp growth. That’s the playbook - but rolling it out across thousands of stores costs real money. Nike Digital needs investment too. You don’t rebuild a brand experience on a declining CapEx budget.

The history tells you everything. Nike spent $900 million to $1.1 billion a year on CapEx in its prime (FY2014-2020). Post-COVID it collapsed to conservation mode - $430 million in FY2025, the lowest since 2003. The old model extrapolated that decline to $482 million by Year 8 - below FY2012 levels when Nike was half its current size. The new model ramps back to $1.3 billion, which is just getting back to where Nike was in FY2016-2019. A 3x gap by Year 8.

The lesson: Don’t extrapolate CapEx cuts during a restructuring. If your revenue is growing and your CapEx is shrinking, your model is broken. Something has to give.

5. Know what your valuation framework hides from you

The old model used FCFE - Free Cash Flow to Equity. Start with net income, add depreciation, subtract CapEx and NWC, adjust for net borrowing. Discount at cost of equity. FCFE is a perfectly valid method. For certain companies - especially financials or stable businesses with predictable capital structures - it can be the right choice.

But every framework makes trade-offs in what it shows you and what it buries. FCFE starts from net income, which blends operating performance with tax strategy, interest costs, and one-time items. When net borrowing is in the cash flow, you think less carefully about reinvestment needs. The margin assumption? Buried inside a single net income number. Working capital? Easy to forget when it’s just one more line in an already blended formula.

FCFF - Free Cash Flow to the Firm - decomposes the business differently. EBIT margin is a line item. Tax rate is a line item. D&A, CapEx, and NWC each stare at you from their own row. For a turnaround story like Nike, where the whole point is tracking margin recovery and reinvestment needs, that visibility matters.

The FCFE framework didn’t cause the other four mistakes. But it made them easier to make and harder to spot. I think didn’t deliberately skip working capital - the framework just didn’t force me to think about it.

The lesson: Understand the weaknesses of whichever framework you use. Every method hides something. If you know what’s hidden, you can check for it. If you don’t, you’ll find out when the model breaks.

The one thing I got right

In the original post, I wrote about competition. Satisfy, Bandit Running, Tracksmith. Hoka. On. I personally switched from Nike running shoes to Hoka after an injury. And I liked them more than Nike. I said so publicly.

That qualitative insight was correct. Nike’s competitive position is weaker than it was five years ago, and the running category specifically has seen real share losses. My new model reflects this - North America growth caps at 6%, not 10%, partly because brands like On and Hoka aren’t giving market share back.

But here’s the thing. In October, I wrote about the competition in one section, then ignored it in the model. The spreadsheet projected revenue growth as if Nike was still the only game in town. The gut said one thing, the model said another, and I went with the model.

If your analysis identifies a real competitive threat and your model doesn’t reflect it, one of them is lying.

The dangerous model

The most uncomfortable part of this story isn’t that the model was wrong. It’s that it felt conservative.

Revenue growth was muted in years one and two. The discount rate was above 9%. Terminal growth was a cautious 2.5%. It looked careful. It looked skeptical. If you squinted at it, it looked like the work of someone who’d done their homework.

But “conservative on the things you can see” doesn’t help if you’re wrong on the things you don’t model. Zero change in working capital, instant margin recovery, declining CapEx, no segment decomposition - these aren’t aggressive assumptions. They’re missing assumptions. The model didn’t decide to be optimistic about working capital. It just forgot to include it.

The difference between a $82 fair value and a $38 fair value is not a difference of opinion. It’s a difference of rigor.

I’ve since rebuilt every model in my portfolio using the FCFF framework. Every position gets segment revenue, explicit margin paths, modeled NWC, and CapEx tied to growth. Two of the five came back looking different. Nike was the worst - you’ve seen the numbers. Lululemon (LULU) went from $283 fair value to $192 - a 32% haircut - once I modeled the CEO transition margin sag, tariff-adjusted working capital, and regional revenue build-up instead of a single growth rate. Same pattern. Same category of mistakes.

The process upgrade was worth more than any single stock pick.

So what now

This is the part I’ve been putting off.

I own 173 shares of Nike at $56.13. The stock is at $44. My updated model says it’s worth $38. I didn’t just buy at a premium to the market - I bought at a premium to my own revised fair value. That’s a $9,700 position in a stock I now think is overvalued.

The textbook answer is simple. Sell. The model says the stock is 14% above fair value. Take the $2,000 loss, redeploy the capital into something with actual upside, and move on. Sunk cost fallacy is the enemy. Be rational.

But it’s not that clean.

Nike is still Nike. The brand isn’t broken - the execution was. North America is growing again. Wholesale is +11%. Running is up 20% for three straight quarters. The product pipeline is the strongest in years. Hill knows the business - 32 years in, he’s not guessing. The turnaround is real, it’s just slower and more expensive than the market hoped.

My model gives Nike a 12.5% EBIT ceiling because of tariffs. But tariffs are political, not structural. If they ease - and my original thesis was that they would - the margin ceiling lifts and the fair value moves up. At 13.5% terminal EBIT instead of 12.5%, fair value jumps to mid-$40s. At 14%, it’s north of $50. The model is sensitive to this assumption because it affects every year of the projection.

So the real question isn’t “is Nike overvalued today.” It’s “what’s the probability that one or more of my conservative assumptions turns out to be too conservative.”

Here’s what I’m doing. I’m holding. But with conditions.

The turnaround has specific milestones I can track. NA revenue must stay positive. Gross margins must stay above 38%. EBIT margin inflection should show up by Q2 FY2027 - that’s management’s own timeline. China needs to at least stabilize. If these hold, I keep the position and let the thesis play out. If any of them break - especially NA turning negative again or margins deteriorating further - I sell. No averaging down. No adding. This is a position I need to earn my way out of through patience and monitoring, not through throwing more capital at it.

The honest version: I’m not confident enough to add, and I’m not panicked enough to sell. The model says one thing, the qualitative picture says something slightly different, and I’m going to let the next two quarters of earnings settle the argument.

Q1 FY2027 reports in late June. That’s the next checkpoint.

Chuyka check

Here’s what the numbers don’t capture.

I liked Nike. I wear Nike. I quoted Peter Lynch in the original post - “liking a store or a product is a good reason to get interested in a company.” And it is. But somewhere between getting interested and hitting buy, the analysis has to take over from the affection.

I did the work. I built a model. I found a margin of safety. But the work had holes in it, and the margin of safety was an illusion built on missing assumptions. The price looked right because the model was shaped to make it look right - not intentionally, but through omission.

My chuyka on this one was wrong. Not because the gut read was bad - Nike is still a great brand with real competitive advantages. But because I let the gut substitute for math in places where the math mattered.

The next time I feel strongly about a company, I’ll still listen to the feeling. That’s the whole point - data gets you 90% there, chuyka gets you the rest. But I’ll also double-check that the model doesn’t have any convenient blind spots where the feeling is doing the work the spreadsheet should be doing.

Trust your chuyka. Don’t let FOMO eat you alive.

Not investment advice. Think for yourself. Don’t trust online experts.

Before we continue, I would love if you subscribed using button below. I am looking for a thoughtful conversation and push-back on my positions. I am 9% down YTD and clearly something is not working. Your feedback is very important. Never thought that transition from passive to active investing would be such a rollercoaster.

Subscribe now

Now let’s get into one of the more recent additions. I’ve bought 270 shares of GTX for $18.49 - it is now trading at around $17 amid the market sell-off. An opportunity to increase the position? Let’s test the thesis.

The bigger turbo

When I was younger I was into time attack. You take your car to a track and try to set the fastest lap in your class. I had a Mitsubishi Evolution X. Great car.

Since I couldn’t really drive, the main way to go faster was modifications. Intake, exhaust, ECU tune, cooling, tires, rip out half the interior. The basic minimum. The luxury maximum was always an engine rebuild - forged pistons, MOTEC brain, and a Garrett turbo kit.

I never got to the luxury maximum. Around the same time I got into investing. First I realized the best investment was in myself as a driver. Then I realized the best investment wasn’t in myself or the car - it was in the stock market. Sold everything. Here we are.

The dream of a bigger Garrett turbo and a louder exhaust had to go. But now I can afford to buy a piece of their company, trading under the ticker GTX.

What Garrett Motion does

Garrett Motion designs and manufactures turbochargers for passenger and commercial vehicles - gasoline and diesel. The company sells directly to OEMs: Volkswagen, BMW, Mercedes, Stellantis, plus Chinese automakers. A small share comes from aftermarket parts and automotive software.

Garrett is a former Honeywell division, spun off in 2018. In 2020 the company went through Chapter 11 bankruptcy - not from operational problems, but from asbestos liabilities inherited from Honeywell. Emerged in 2021 with a clean balance sheet, no legacy baggage.

Headquarters in Switzerland. 6,600 employees. Trades on NASDAQ. CEO Olivier Rabiller has been running the company since the spin-off.

Three major players in the turbocharger market: Garrett, BorgWarner, and what’s left of Honeywell’s turbo business. Effectively an oligopoly.

Why the stock is cheap

GTX traded at the time of original thesis at $18.23 against my DCF fair value of $41.00. P/FCF around 10x. EV/EBITDA at 8.6x. FCF yield of 9%. For a company that consistently generates $325-430 million in free cash flow, that’s not a lot.

The market is saying something like this: “Garrett is an ICE business. Electrification will kill it. Everything it earns today will slowly disappear.” That’s where the discount comes from.

52-week range: $7.01-$21.43. The stock is closer to the upper end, but still sits at the 50% margin of safety level from my valuation.

What management says

On the Q4 2025 earnings call, management explained the current situation:

• Europe is weak. Light vehicle production hasn’t recovered to pre-pandemic levels (still -21%). Diesel keeps losing share in cities.

• Commodity pass-through. Lower raw material prices mechanically reduce revenue by about 2% - margins aren’t affected.

• Currency problems. EUR/USD and JPY swings cost about $34 million for the year.

• China is the growth story. Win rate on new platforms above 50%. The bet is on PHEV/REEV - range-extender EVs where the combustion engine runs in a narrow RPM band and needs a high-performance turbo.

Fundamentals

Revenue and margins

Revenue peaked around $3.9 billion in 2023 and has been gently declining. But EBIT margin holds steady at 13-14% - that’s 3x the sector average of 4.7% per Lazard’s 2025 Global Automotive Supplier Study. Gross margin stays in the 24-27% corridor across several cycles. Not a coincidence - structural pricing resilience.

FCF swings year to year because of working capital, but averages $300-440 million annually. For a company with a $3.5 billion market cap, that’s substantial.

Capital returns

This is where it gets interesting. Since emerging from bankruptcy, Garrett has been returning cash aggressively:

• FY2023: $818 million in buybacks. More than FCF that year. Roughly 20%+ of the entire market cap repurchased in one year.

• FY2024: $304 million

• FY2025: ~$211 million + a new $250 million program announced for 2026

• Dividend: $0.28/share (1.5% yield)

Management targets converting 60% of EBITDA into capital returns. At TTM EBITDA of ~$595 million, that’s ~$357 million a year in buybacks and dividends. Promises match actions - that’s rare.

The flip side: aggressive buybacks pushed book equity into negative territory (~-$4/share). That’s not insolvency - cash generation is real - but P/B and ROE are now economically meaningless. Look at FCF yield and EV/EBITDA instead.

Balance sheet

• Net debt: ~$1.4B

• Net debt / EBITDA: 2.2x

• Interest coverage: 4.7x

• Current ratio: 0.97

Leverage is moderate. 2.2x on EBITDA isn’t aggressive for a business with stable FCF. But there’s not much buffer if volumes drop hard. A 25% fall in EBITDA pushes leverage to 3x - that gets uncomfortable.

Piotroski F-Score: 7/9

Strong. Fails on two signals: leverage ticked up slightly, current ratio sits just below 1.0. All profitability and operating efficiency signals are green.

Competitive moat

Narrow moat - engineering lock-in to the platform.

Turbocharger development is tied to a specific engine program. Certification takes 3-5 years. Once a supplier wins a platform, they stay on it for 8-12 years - the full lifecycle of that engine. OEMs don’t swap turbo suppliers mid-program.

But the moat doesn’t widen over time - it resets with each new engine generation. If ICE programs shrink, Garrett’s captive revenue pipeline shrinks with them.

One detail the market isn’t pricing in: Garrett is a consolidation beneficiary in a distressed sector. 20.6% of the auto supplier industry is already in financial distress per Lazard 2025. When weak turbo suppliers go bankrupt, their OEM content has to go somewhere. In an oligopolistic market (Garrett + BorgWarner), there aren’t many places for it to land. Garrett at 14% EBIT margin can absorb the pricing pressure that kills a competitor running at 5%.

Two catalysts that weaken the bear case

1. EU softened the ICE ban

In December 2025, the EU revised its 2035 target for banning combustion engines. Was: 100% emissions reduction (effectively a ban). Now: 90% reduction, with the remaining 10% offsettable through e-fuels, biofuels, and carbon credits.

The practical effect: OEMs that had frozen new ICE/hybrid program investments can now invest again. Every new engine program is a potential Garrett platform win for 8-12 years. The push for this change came from Germany, Italy, Hungary, and Poland - countries whose OEMs (VW, BMW, Mercedes, Stellantis) are Garrett’s core European customers.

2. China REEV is growing despite PHEV slowdown

Pure PHEVs in China have slowed (Q4 2025: -12% YoY). But REEV - range-extender EVs - is a different story: 640 thousand units in 2023 (+181% YoY), ~1.2 million in 2024 (+83%), forecast 3.2 million by 2030 (CAGR 17.1%).

Why this matters for Garrett: REEV generators run at a fixed optimal operating point - Miller/Atkinson cycle, narrow RPM band, maximum thermal efficiency (Chinese OEM engines hitting 46%). This demands more precise turbocharger engineering than a standard combustion engine. Commodity competitors lose here - plays directly into Garrett’s hands.

Li Auto tried to pivot to pure BEV - failed - and came back to REEV, targeting 550 thousand deliveries in 2026. Leapmotor is targeting 1 million units. Geely, BYD, Xiaomi, XPeng - all entering REEV. The trend is confirmed across multiple OEMs.

Peer comparison

Garrett stands out on margins - nearly double the nearest competitor. EV/EBITDA is higher, but that’s the premium for business quality. FCF yield is in line with the sector, but the quality of that cash flow is significantly better.

DCF valuation

Key inputs:

• Revenue: $3,584M base, growing at +3% per year (linear)

• EBIT margin: 14.0% (flat, at TTM level)

• Tax rate: 21%

• WACC: 6.45% (beta = 0.70, adjusted up from reported 0.376)

• Terminal growth: 2.0%

• Projection period: 8 years

At the current price, GTX trades below the 50% margin of safety level. The market is pricing in either a significantly worse revenue trajectory, or structurally lower margins, or a much higher discount rate. All three are possible - but all three at once?

Terminal value as % of EV: 72.2%. Near the 75% threshold but acceptable. Most of the value sits in the long-term franchise - turbo leadership plus potential EV products - not in near-term cash flows.

Risks

1. EV transition accelerates. The EU softened the target, but the drift toward electrification continues. If OEMs front-load their EV mix for CO2 compliance, revenue could decline 4-7% per year.

2. China REEV stalls. If the Chinese government pivots back to favouring pure BEVs, or if Li Auto and BYD start making turbochargers in-house, the thesis weakens.

3. Tariffs. Management flagged tariff risk on the Q4 2025 call. Cross-border manufacturing creates real exposure if trade wars escalate.

4. Leverage. At 2.2x net debt/EBITDA it’s manageable. A 25% drop in EBITDA and you’re at 3x - the safety cushion disappears.

5. Buybacks vs. debt. In FY2023 the company spent more on buybacks than it generated in FCF. The pace has slowed, but negative book equity is a fact. Financial flexibility is limited.

How to think about this

Garrett is not a “combustion engines forever” story. It’s a time-boxed bet on the 2025-2030 window. Over the next five years, ICE and hybrids hold roughly 60% of new vehicle sales per IEA, BloombergNEF, and S&P Global. In that window, Garrett collects cash, buys back shares, and builds new product lines - e-boosting, fuel cells, e-cooling for data centres.

The question isn’t “will ICE survive?” It will, at least until 2030. The question is: “will Garrett earn the right to compete after 2030?”

At 9% FCF yield and buybacks shrinking the share count every year, the hurdle rate for acceptable returns is very low. Even with stagnant revenue, the math works. If China REEV accelerates and the consolidation thesis plays out, the upside is significant.

I bought 270 shares of GTX at $18.49 in my active portfolio.

Chuyka Check

After all the numbers and all the models - here’s what sits with me. Garrett is the kind of stock that makes you feel slightly uncomfortable owning. The market narrative is clean and easy to understand: EVs will kill this business. And maybe they will. Eventually.

But “eventually” is doing a lot of heavy lifting in that sentence. The turbocharger industry isn’t shrinking tomorrow. It’s an oligopoly with real engineering barriers, and the China REEV angle is something most investors haven’t even looked at. The company throws off cash like it’s going out of style, and management is buying back stock at prices that my model says are half of fair value.

My chuyka says this is one of those situations where the market has priced in a death that’s 10 years away and forgotten about the 5 years of cash collection in between. I could be wrong. The EV timeline could compress. But at this price, I’m getting paid to wait and find out.

If you want to follow me and see if I am wrong or right in my assumptions - here is a field with a button for you :)

Trust your chuyka. Don’t let FOMO eat you alive.

Not an investment advice. Think for yourself. Don’t trust online experts.

I’m a different beast. Down 9% YTD, holding six losing positions and a pile of cash, watching the market grind lower while oil trades above $100 and the Fed sits on its hands.

Subscribe now

But here’s where I came from, where I am, and where this is going.

The passive decade

I started investing in 2014. My first trade was buying one share of Tesla and selling it eight minutes later. I didn’t know what I was doing, but I knew I wanted to learn.

For the first couple of years I studied value investing. Intrinsic value, margin of safety, the whole Buffett and Graham reading list. I ran a blog about it in Russian. I had opinions about stocks I barely understood. Then life happened - career got demanding, I was working 12-hour shifts on oil rigs in Iraq, and I quietly switched to passive indexing. Meb Faber’s global asset allocation. Five ETFs: IVV, IEFA, VCLT, VGLT, PCY. Set it and rebalance once a year.

I ran that portfolio for a decade. It compounded at over 7% annually - decent, not spectacular. Most of that came from the S&P 500 carrying the rest. The international diversification, the bond allocation, the carefully balanced global exposure - all of it underperformed a simple 60/40 btw for most of the years, except when it beat S&P500 in 2025. I’ve achieved my goals of financial security and now it keeps running.

The biggest lesson from ten years of passive investing wasn’t about asset allocation. It was this: sitting still is the hardest thing in investing. Not selling during drawdowns. Not chasing what’s working. Not tinkering with something that’s fine. That discipline became the foundation for everything that came after.

The itch that wouldn’t go away

Somewhere around year seven or eight, the value investing itch came back. I’d been building data science skills through Yandex’s program, applying AI to problems in oil and gas at work, and I kept thinking: the tools have changed. What used to require a Bloomberg terminal and a team of analysts - pulling financial statements, building DCF models, screening for quality factors, reading earnings calls - a single person with the right setup can now do all of that.

Not better than a professional analyst. But good enough to make an informed bet. And that’s all value investing is - an informed bet with a margin of safety.

So in early 2025 I carved out $55,000 more from oil and gas money, opened a separate brokerage account, and started building positions. One stock at a time. Full transparency - every entry price, every share count, every P&L published in real time. (Now there is a Russian version of live tracking on my blog, but I will add an English one soon.) The passive portfolio stays. This is the experiment running alongside it.

The rules of the experiment

The philosophy is simple. Simple doesn’t mean easy.

Buy businesses, not tickers. I want to understand what a company does, how it makes money, and why it will keep making money five years from now. If I can’t explain it clearly, I don’t own it.

Value is the only reason to buy. I build a DCF model for every position. Full FCFF projection, explicit assumptions, terminal value with a sanity check on the implied multiple. The model gives me a fair value. I only buy below it - ideally at a 30-50% discount. No exceptions for “great company, fair price.” If the math doesn’t work, I wait.

Cash is not a failure. Half my portfolio is in SGOV right now. That’s not indecision - valuations are rich, and I’d rather hold dollars than overpay for a narrative. The ability to act during a selloff is an asset. Most investors are fully deployed when the best opportunities show up. I’d rather miss a rally than catch a falling knife without dry powder.

Show the work. Every stock gets a published deep dive with the full model. Not just “my fair value is $X” but every assumption behind it - revenue growth, margins, capex, WACC, terminal growth rate. You can disagree with any input and rebuild the model yourself. That’s the point.

Document mistakes. When I’m wrong - and I will be wrong - the losing trades stay on the record with the original thesis next to them. No deleting posts, no quiet exits. The process matters more than any single outcome.

Where things stand right now

Let me show you what honesty looks like. Here’s the portfolio as of March 20, 2026:

Total portfolio value: $52,917. Down from $55,000 starting capital. YTD return: -8.95%.

Every single equity position is red. SGOV - the cash proxy - is the only thing in the green, and barely. The unrealized loss across the equity book is -$3,661.

This is what value investing looks like in practice. You buy stocks the market hates. Then the market keeps hating them. You sit there, check the thesis, and either add more or wait. The thesis on each of these names hasn’t changed. The prices have. That’s the difference between price and value, and being comfortable with that gap is the entire game.

Quick thesis summary on each:

• NKE - Nike at $52, trading near a 52-week low. The market is pricing in a rough transition under new CEO Elliott Hill. I think the brand is too strong to stay here. March 31 earnings will be the next data point.

• GTX - Garrett Motion, a turbocharger manufacturer spun off from Honeywell. Fair value per my DCF is $41. At $17 the market is pricing in permanent ICE decline. I disagree.

• LULU - Lululemon post-earnings. The product quality is real - I own their stuff and it lasts years. At $163 it’s priced like a broken growth story. I think margins recover.

• NOMD - Nomad Foods, European frozen food. Boring business, beaten down by a shelf filing overhang. Trading at 7x earnings for a company with stable cash flows.

• ONON - On Holding. Best-in-class growth in athletic footwear. At $38 after a guidance miss, it’s the most expensive name I own and the one I have least conviction on. Minimum position size.

Started building that set of outcasts last year. It went well, until it didn’t.

Not one of these is a crowd favourite right now. That’s by design.

The plan from here

This is what Chuyka will be:

Stock deep dives - full analysis with DCF model assumptions. When I buy something, you’ll know exactly why, with every assumption laid out for you to challenge.

Portfolio updates - real numbers, real P&L, no hiding. When positions go against me, you’ll see it happen in real time. When I add or sell, you’ll know the reasoning right as I execute.

Process posts - how I source ideas (tracking superinvestor 13F filings, sifting through 52 week lows, mostly), how I read earnings calls, how I build models. The unsexy infrastructure of doing this alone.

I publish when I have something worth your time. Not on a schedule.

I’ve been writing about investing in Russian for over a decade. But I’ve been missing something important - the conversation. Thoughtful pushback. Someone saying “your WACC is too low” or “you’re ignoring the competitive threat from X.” That kind of discussion is how investing logic sharpens. Writing in English opens that door.

If you’re a value investor, an aspiring one, or just someone who wants to see what this process actually looks like with real money on the line - stick around.

If you’re here for hot takes and trade signals, this isn’t the place. I bought Nomad Foods. I hold 50% cash. Nothing about what I do is exciting. But I think it’s right.

Trust your chuyka. Don’t let FOMO eat you alive.

Not an investment advice. Think for yourself. Don’t trust online experts.