Friday charts: AI investor had a situation

1. A Russian roulette equation: usually win, occasionally die.

   — Warren Buffett on leverage

In 2005, Brian Hunter generated an estimated $1 billion in trading profits in natural gas futures for the hedge fund Amaranth. 

The fund paid him a bonus of $113 million for his 2005 performance, but he hoped to make more. Reportedly targeting a $500 million bonus for 2006, he bet even bigger.

By August, he was carrying $53.5 billion in natural gas derivatives against Amaranth’s $3 billion in capital.

At that level of leverage — 18x — the firm’s capital would be wiped out with a loss of just 5.6%. In a market that sometimes moved 5% in a single day.

He had a few bad days.

“All I know is I am personally one more bad day away from stopping out,” Hunter told a colleague in September. “I can’t afford to drop below 30 for my family.”

($30 million being the minimum net worth required to live in New York City in 2006.)

On Sept. 20, Hunter lost $681 million in a single day. On Sept. 21, Amaranth faced a margin call of $4.07 billion.

Down $6 billion for the month, Amaranth liquidated the entire fund.

2. Rational people don't risk what they have and need for what they don't have and don't need.

   — Warren Buffett

In 2020, Bill Hwang borrowed billions to make his already enormous fortune much more enormous.

His family investment office, Archegos, had $20 billion of equity — all his own money. Against that, it held at least $100 billion of longs in a highly concentrated book of about eight stocks.

One of those, ViacomCBS, accounted for his firm’s entire capital — a $20 billion position. The rest of the $100 billion+ portfolio was financed by a group of investment banks.

In March, a vicious cycle of margin calls and stock sales vaporized over $100 billion of Hwang’s P&L in just a few days.

Archegos lost all of its equity and more: The banks that financed his trades lost an additional $10 billion closing them out.

“Bill Hwang wanted to be a legend of Wall Street,” a prosecuting attorney said.

He is.

3. Never forget the six-foot-tall man who drowned crossing the stream that was five feet deep on average. It’s not enough to be able to get through on average; you have to be able to survive life’s low points. 

   — Howard Marks

In 1998, Long-Term Capital Management borrowed $125 billion from investment banks against $4.8 billion of its investors’ capital.

The debt funded relative-value arbitrage trades in the Treasury market that their models said were virtually risk-free: buying a Treasury bond trading at a small discount to a nearly identical one they shorted.

So sure of their models, they amplified their returns with 25x leverage. A marked-to-market loss of just 4% would wipe out their equity.

In August, 1998, a default in Russia scrambled global bond markets, causing the relative prices of nearly identical Treasurys to diverge further than LTCM’s models suggested was possible.

By September, the fund was marked-to-market insolvent.

LTCM’s models accurately predicted where prices would end up, but not how far they might stray before getting there.

In other words, they had forgotten the six-foot-tall man.

4. A “good” investment doesn’t become a “great” investment with leverage. Leverage only magnifies gains and losses. It increases the risk of ruin.

   — Howard Marks

The latest member of the margin-call hall of fame is Leopold Aschenbrenner, the “Nostradamus of AI,” who lost an estimated $35 billion this week.

All of it was borrowed.

He had made a lot of good investments in AI-related stocks. Great ones, even. At the end of June, his fund, Situational Awareness, was up 270% on the year.

But the returns were amplified by leverage. Sure of his long-term thesis on AI, Aschenbrenner reportedly borrowed as much as $4 for every $1 of capital he had.

In the short-term, though, things went the other way: Several of his highest-conviction longs were down over 30% just this week. Several of his shorts were up. 

On Thursday, to meet the ensuing margin calls, he was forced to sell nearly his entire portfolio of publicly traded stocks to Citadel.

Leverage had increased the risk of ruin to a near certainty.

5. Great opportunities arise when buying from forced sellers. That means don't become a forced seller. Build a strategy that avoids forced selling at any price. 

   — Howard Marks

On the other side of every forced seller is an opportune buyer. In 1998, it was a consortium of banks taking over LTCM’s portfolio of Treasury bets. In 2008, it was Warren Buffett buying preferred shares from the banks.

This week, it was Citadel. 

Many of the positions it bought from Situational Awareness on Thursday morning opened more than 20% higher. SK Hynix — an $800 billion company — was up 30%.

Having reportedly bought the portfolio at a 10% discount to Wednesday’s close, it’s likely Citadel is up at least $4 billion on the trade.

Citadel had seen this movie before. It made a killing buying Amaranth’s energy portfolio in 2006, as well.

6. “Assuring survival in bad times is inconsistent with return maximization in good times.”

   — Howard Marks

Despite this week’s margin call, Situational Awareness still has $10 billion in assets under management, nearly all of it in unlisted equities, like Anthropic. 

Selling his publicly listed positions to Citadel allowed Aschenbrenner to keep the unlisted ones — and keep trading.

His fund now has no leverage at all. In part, presumably, because banks are unlikely to lend him money against stocks that don’t trade.

But perhaps also because he learned the lessons of Buffett and Marks.

“Our fund must always be structured such that we can take a loss and fight another day,” he wrote in a letter to investors this week.

Let’s check the charts.

Not helpful:

There’s a web app that tracks Aschenbrenner’s positions. It’s based on 13F filings, so not entirely up to date. But the last thing you want when you’re nearing a margin call is for everyone to know what your positions are.

Still up:

A website estimating the fund’s performance is directionally correct and ends in about the right spot. Incredibly, Situational Awareness is still up 80% on the year. The fund had gained about 270% after fees through May. At that point, it was up more than 1,000% after fees since inception.

Bargain basement:

Some of Aschenbrenner’s biggest longs were down as much as 38% over the past week. Citadel bought them all 10% below the low on the above chart.

The long-term thesis:

Callum Williams notes that for the AI industry to make an adequate return on the hundreds of billions currently being invested, it will need to generate $2.7 trillion of revenues every year in perpetuity. Seems like a lot.

Early returns:

AI has created a boom in new business formations, many of which are solo entrepreneurs. Many are doing big business: The Wall Street Journal reports that thousands of them are generating over $1 million in annual revenue.

Speaking of leverage…

Roughly 35% of hyperscalers' projected capex next year is expected to be debt-financed.

Spooky:

Bespoke calculates that this week’s Situational Awareness margin call was 918 days after introduction of ChatGPT, which marked the start of the AI boom. Incredibly, LTCM’s margin call was 918 days after the Netscape IPO, which marked the start of the dotcom boom.

7. Progress is cumulative in science and engineering, but cyclical in finance. 

   — Jim Grant 

Unlike LTCM, however, Aschenbrenner survived its margin call — and relearned the danger of leverage. 

“I will make it my mission to ensure we learn the necessary lessons from this experience,” he wrote this week.

Have a great weekend, unlevered readers.

— Byron Gilliam