Hedge Fund Manager's Top Investments with Fred Liu

ELI5/TLDR

Fred Liu runs Hayden Capital, a 5-to-15-name concentrated fund focused on emerging tech compounders in the $1-10B market cap range — mostly consumer internet in the US and Asia. His edge is doing ground-level work (web scraping merchant density, calling Lazada employees in KL, tracking order frequency) before the street catches on, then sizing up as KPIs print. He walks through three positions in detail: Sea Limited (largest holding, held since 2018), PDD/Pinduoduo (exited after a clean double), and AppLovin (bought in the 2022 drawdown at ~3x free cash flow). The common thread is buying growth at a value price inside a “core business subsidizing a new business” structure, where the core alone protects the downside.

The Full Story

The philosophy, in plain form

Hayden runs 5 to 15 names. Holding periods are five to ten years. The target is earnings power that 3x-5x over a decade, bought at fair-to-cheap prices. Fred is explicit that portfolio returns follow a power law — top 20% of ideas generate all the alpha, the other 80% nets to roughly zero. Losers don’t blow up the book because new positions enter at tracker size (<2%) or small core (5%), and only earn a full weight (10%+) once specific KPIs print. Core compounders end up 15-20%.

The structural inefficiency he’s hunting is well-defined. These are $1-10B businesses, founded in the last five to ten years, not yet in indexes, so no natural buyer. Earnings are volatile because a profitable core business is being used to fund a new line — so the headline multiple looks like 100x and the screener skips right past. The work is figuring out what the unit economics of the hidden second business actually are, before the street does.

“The top 20% of ideas generates all of our alpha and the other 80% kind of just nets out… we don’t tend to lose much money on our losers.”

One interesting tweak he flags: markets are more volatile than they used to be. Names like Netflix swinging 40% in a couple months at several hundred billion in cap suggests long-duration capital has thinned out. Historically he’d ride the volatility. Now he’s thinking about trimming at peaks (maybe a 15% forward IRR) and reloading at troughs (30% IRR) — raising the hurdle rate to squeeze a bit more alpha out of the same names.

Sea Limited — the 2018-to-now story

Entered in 2018 as a small 2-3% position. Now the largest at ~20%.

The original thesis was Shopee, not gaming. Shopee had been live for about two years. Fred had studied e-commerce exhaustively and had a sharp view on what a winning model in emerging markets had to look like: mobile-first, long-tail categories (fashion, beauty, homegoods — hard to price-compare), and engagement-driven rather than utility-driven. The reasoning was specific to the terrain: $100 Android phones don’t have memory for many heavyweight apps, so super-apps and multi-vertical apps win, and tab-switching for price comparison barely happens.

Shopee was female-skewed. This mattered more than it sounds:

“At a Shopee seller day, they literally gave a slide of why you should go after female shoppers because they’ll write like two paragraphs of why this product is great… compared to the male review might be like A+ great product.”

Review density compounds. Lazada and Tokopedia had started in a desktop world, leading with electronics, attracting male shoppers and sparser reviews. Shopee started mobile-first with low-AOV fashion and beauty, users spent 30-60 minutes a day in the app, and shopping functioned as entertainment in markets with few alternatives.

The conviction-builders were all ground-level. Web-scraping merchant density per vertical vs. Lazada and Tokopedia. Watching order frequency climb from 2x/month toward China levels (Shopee went from 2 to 4x within two years, pre-COVID). And a specific moment in a Kuala Lumpur cafe when Lazada mid-level employees told him their largest personal position was Sea — because whenever Lazada shipped a feature, Shopee copied it in two weeks and executed three times better. Culturally Lazada was a mess post-Alibaba acquisition; the CEO was sending emails in Mandarin and employees were Google-translating them.

On today’s valuation: Fred frames Sea as trading at replacement value. Roughly $12B cash and investments, $8B in SeaMoney (lending), and Garena (PUBG Mobile-adjacent, 1.8B EBITDA) worth ~5x comparable to Krafton. Strip those, and Shopee is implied at $18-20B against $127B of 2024 GMV growing 25%+ to north of $160B this year — about 1/8 of GMV.

The debate is terminal margin. Management has whipsawed investors — negative margins into 2022, pushed to 1.2% of GMV EBITDA, backed off to reinvest, hit profitability again, now reinvesting again (guiding 0.5% of GMV EBITDA near-term, 2-3% medium-term). Fred’s anchor: China is the most competitive e-commerce market in the world, four large players all fighting, and every one of them runs at 2%+ EBITDA/GMV. You don’t need monopoly to squeeze margin — you need rational competitors staying in their lanes. The 2023 playbook repeats: stock fell 60% during the Tokopedia/TikTok fight, then 5x’d when they stopped reinvesting.

Pinduoduo — the fast-follower template, then exit

Bought in 2022 when “the world hated anything China tech” and Western funds were forced divestors. Company was $35-40B market cap, growing 65% YoY, already cash-flow positive, trading around 10x P/FCF. Sell-side was modeling $7B in profit flatlining. Fred’s view: whatever that number is, it isn’t flat. Today they sit on $50B+ of cash against what was the original market cap.

The more interesting bit is the model itself. PDD never pioneered anything — they fast-followed everywhere. The Costco-like C2M (consumer-to-manufacturer) model aggregated rural demand (600M people in tier-3+ China, GDP/capita ~$5K vs Shanghai’s $20K) and presented bulk orders to factories running at 50% capacity. Factories accepted prices above variable cost but below total cost — PDD cut out the middleman and middle-margin simultaneously. Infinite-scroll sushi-conveyor UI: an item is there for a few days or a week, then gone forever, which forces purchase urgency. Group buying required 100 buyers before the order locked.

Two idiosyncratic tailwinds that Fred thinks are non-repeatable: (1) 600M people hitting mobile internet simultaneously and sharing links on WeChat — pure virality, (2) Alibaba couldn’t retaliate on WeChat because Tencent blew up any Taobao link shared there. Groupon failed in the US because none of that was present.

Exit was simple — thesis played out, stock went from ~$40 to $100-150 range, found better risk/reward elsewhere.

AppLovin — the adtech thesis

Bought in spring 2022 starting around $27, doubled down near $13. Stock had IPO’d around $120 and was in a brutal 2022 drawdown. At the bottom it traded at ~3x free cash flow. Current price is closer to a different thesis entirely.

The 2022 mispricing was categorical. Market saw a mobile gaming studio. Reality was a mobile gaming studio plus a rapidly growing adtech network doing 100%+ growth and already 30% of revenue. Fred had studied the mobile gaming stack end-to-end in 2018 and knew the two good businesses in the ecosystem were (a) the gaming engine (Unity) and (b) the adtech network — because that network takes a 25-30% rake every time the money wheel spins. Mobile gaming is $60B+ of the $120B app economy and behaves like a casino economy: studios earn $40 of profit on $100 revenue, plow it straight back into user acquisition ads. AppLovin sits in the middle and takes the house rake.

The AT&T/ATT wedge was specific. When Apple turned off IDFA sharing, roughly 51% of users opted out of third-party data. Every ad network suddenly went blind — except AppLovin, who had bought $1B of gaming studios in 2018 and had 200M first-party players. That data trained a better targeting model. They also owned MoPub (~70% share on the sell-side of mobile ad slots) alongside ~30% demand-side share, giving full-stack pricing visibility no one else had.

The current question — this is the meaty one:

“Adam at one of the sellside conferences a couple months ago gave the stat that their conversion rate is about 1.3%… they think just through self-improvement of the model they can get to 5%.”

So the bull case on a stock now trading at ~$160B cap with ~$3.3B earnings is: gaming is saturated and the incremental growth has to come from diversity of ads — specifically e-commerce. If you’re playing Candy Crush, another gaming ad is noise. A Shein-style e-commerce ad might actually convert. Fred says the data is showing e-commerce accelerating and CPMs rising enough to price lower-tier games out of slots. Gaming alone is worth $300-400/share status quo. Anything above is a call option on the e-commerce expansion.

His analogy for why he’s comfortable holding a black-box algorithm is worth sitting with:

“It’s like Citadel Securities or Two Sigma or Renaissance Technologies… these are automated market makers with a black-box algorithm that takes a spread on making a market… these other players make a market in the financial markets and AppLovin makes a market in the ad market. But the difference is that instead of taking a couple basis points on each trade, you’re taking 30%.”

Tail risk he acknowledges: Facebook has been making noises about getting back into mobile ad mediation. Apple has the first-party data to take it if they wanted — though Fred thinks they won’t, because 60% of App Store revenue is gaming and destroying the ecosystem to capture ad revenue is a bad trade.

Shopee vs. MercadoLibre in Brazil — the two-speed LATAM

A useful side thread. MercadoLibre grew up during the funding drought post-2001, had to self-fund, so by necessity served the top 1-5% of Brazilian households — people who could spend $50 per order. That forced them to build their own logistics, which forced them to keep serving the high-end to justify the spend. It became their moat and their cage.

Shopee’s observation: the bottom 95% is a chicken-and-egg problem. Lower AOV, longer shipping tolerance, sparser geography — only solvable by spending $4B upfront to jumpstart logistics density in rural areas. Very few players can write that check. Amazon could but won’t. Shopee did. Five years in, they’re the second-largest warehouse operator in Brazil, ship more packages per day than MELI, and fulfill 70% internally through Shopee Express.

Fred’s read: both can coexist. MELI keeps the top 20%, Shopee takes the other 80%. Different business models for different income segments.

Key Takeaways

Power law sizing: 2-5% tracker positions, scaling to 10-20% only after specific KPIs print. Losers get trimmed at small sizes before they hurt.
Core-subsidizing-new-business is the structural alpha. Sea in 2018 ($400M gaming EBITDA at $4B mkt cap) and AppLovin in 2022 (gaming paying for the adtech buildout) both fit this mold. The core alone protects the downside if the new bet fails — and rational management can always shut it down.
Emerging-markets e-commerce rule: mobile-first, long-tail non-commoditized categories, female-skewed for review density, engagement-as-entertainment. GDP-per-capita dictates model — Amazon fails outside the US because logistics speed doesn’t beat price sensitivity below ~$10K GDP/capita.
Fast-follower can be the better bet than first-mover when market penetration is <5% and others are burning VC cash to find the model.
Terminal margin as the key unlock for Sea: China’s four-way e-commerce war runs at 2%+ EBITDA/GMV. Sea is guiding near-term 0.5%, medium-term 2-3%. When they stop the reinvestment cycle, the re-rate should follow — 2023 playbook was a 5x move after the pivot.
AppLovin’s 1.3% to 5% conversion path is the next leg of the thesis. Gaming market share is saturated; e-commerce ad diversity is the lever. At ~$160B cap, you’re paying for that optionality explicitly.
Manager swinging investors around is its own risk — Sea’s start-stop reinvestment has cost them credibility even when the long-term trajectory is fine.

Claude’s Take

Fred is one of the more honest interviewees you’ll hear on a podcast. He’s specific about what went right (the KL cafe moment, the web-scraping, the ATT arbitrage on AppLovin’s data) and clear about the non-repeatable parts (PDD’s one-time WeChat virality, Sea’s 2018 Shopee timing). The framework — buy the second business disguised as the first business, protected by the first business — is useful because it’s mechanical rather than hand-wavy. You can run the screen.

The Sea argument here is tighter than most you’ll see. Replacement value as a floor, China margin structure as a ceiling, manager volatility as the timing variable. The AppLovin pitch at current prices is more speculative and he admits it — you’re paying for the e-commerce conversion lift, and if that tops out at 2% instead of 5%, the math changes. The Renaissance Technologies analogy is clever marketing for what is, at the end of the day, an ad network betting on continued Apple restraint. Facebook re-entering would be genuinely disruptive.

The trimming-at-peaks shift is the most interesting operational tell. A concentrated long-only manager raising his hurdle rate because “long-duration capital has thinned out” is essentially admitting that buy-and-hold gets arbed harder today than it did a decade ago. Whether he can actually execute that without bleeding alpha — most managers can’t — is an open question.

Score: 8/10. Dense, specific, minimal fluff. Drops a quarter-point for the AppLovin section leaning a bit on analogy where more disclosure on the e-commerce ad unit economics would have helped.