The Chart That Keeps Me Awake at Night (Margin Debt)

Margin debt just hit a record. Here’s what it actually means — and why the number everyone watches is the wrong one.

A free companion to tomorrow’s finale, «When Should You Sell Semiconductors?» No paywall — because this is the piece I most want every reader to understand before the next drawdown, not after it.

---

Tomorrow we publish the last part of our semiconductor trilogy: when to sell a parabola you cannot time. The answer there is a process, not a prediction — twelve rules that work whether the chips top next month or melt up into 2028.

Today is the companion piece. Because there is one gauge sitting underneath this whole market that tells you why the endgame, when it comes, tends to come fast. It is old, it is unglamorous, and right now it is flashing.

It is called margin debt. Let me show you how it works, why the number everyone quotes is the wrong one to watch, and how it fits the semiconductor setup we’ve spent three parts mapping.

What It Is, in One Sentence

Margin debt is the total amount investors have borrowed against their portfolios to buy more stock. You own $100,000 of shares, you borrow another $50,000 against them to buy more — that $50,000 is margin debt. Multiply across every brokerage account in America and FINRA reports the total once a month.

It is the single most reliable indicator of speculative excess in US equities, for one simple reason: it is not a survey, not a vibe, not sentiment. It is a claim. Real money, borrowed against real shares, that must be paid back regardless of what the market does next.

The current reading (May 2026): a record $1.42 trillion, up 53.7% from a year ago. Investors added roughly half a trillion dollars in borrowed money to their stock bets in just twelve months — a pace of borrowing not seen in nearly three decades.

The Mistake Almost Everyone Makes

Here is the part that matters, and it is the part most people get wrong.

They watch the dollar level. «$1.42 trillion, all-time high!» But the dollar level is almost useless, because it rises with the market — of course margin debt hits a record when the S&P hits a record. The level trending up with prices tells you nothing.

The signal lives in the rate of change — and, more precisely, in the direction of that rate of change.

Chart 1: US margin debt, year-over-year change — 67 years. Every major top was marked by the rollover, not the peak

FINRA margin-debt %YoY since 1960 — the 55% zone and the 2000 / 2007 / 2021 turns, Fundstrat, Finra

Look at the year-over-year growth rate across seventy years of history, and a pattern jumps off the page. Before every major top — 1972, 2000, 2007, 2021 — the growth rate climbed into a danger zone, rolled over, and started to fall. And it was the rollover, not the peak, that marked the edge.

Read that again, because it is the whole point: the top is not the highest reading. The top is the turn. The market doesn’t ring a bell at a level. It rings it when the borrowing that drove the rally starts, quietly, to reverse.

You don’t even need to wait for the extreme to see this bite. Drop the threshold to a more common trigger — every time the year-over-year growth rate breaks above 45% — and you get a far larger, more usable sample stretching back to 1960. Mark each of those breakouts on the S&P 500 and the clustering is hard to unsee: they bunch around the frothy, late-cycle moments, not the quiet middles of bull markets.

Chart 1a: The S&P 500, with every instance margin-debt %YoY broke above 45% marked — since 1960

S&P 500 (log) with 45%-breakout markers — the late-cycle clustering, Fundstrat, Finra

And here is the part that turns a pattern into evidence. Take all of those 45% breakouts and average what the S&P 500 did next — one month, three months, six months, a year out. This is the honest, full-sample version of the claim, and it avoids the small-numbers problem of the 55% line: enough instances to actually mean something.

Chart 1b: Every instance margin-debt %YoY broke above 45%, and the S&P 500’s forward returns — since 1960

Forward S&P 500 returns after 45% breakouts — 1m / 3m / 6m / 12m, all instances, Fundstrat, Finra

Read the forward-return table as weather, not a clock. It does not say «sell on the breakout» — plenty of these signals fired and the market kept climbing for months. What it says is subtler and more useful: once leverage is accelerating this fast, the distribution of outcomes gets worse — forward returns thinner, drawdowns deeper, the ride rougher. You are being paid less to take more risk. That is the definition of a late-cycle tape.

There’s a rougher, higher line that has flagged outright euphoria for two generations — around 55% year-over-year growth. Where 45% is a common late-cycle warning, 55% is the rare, red-hot extreme: in roughly seventy years it has only been breached a handful of times, so treat it as a small-sample tell, not a law of physics. Today we sit at 53.7%, up from 53.3% the prior reading — through the 45% warning already, and now climbing toward that extreme line. We are not through 55% yet. And the day that matters isn’t the day we cross it — it’s the day the line stops climbing and turns down. This month of July, we will get updated numbers and it very likely will cross the 55% threshold.

Why Leverage Is the Part That Matters Most

The dollar amount isn’t really the danger. The mechanism is. And the mechanism is human.

On the way up, borrowed money feels like genius. You borrow, you buy, the position rises. You were right — and you were right with more money than you had. The market hands you a reward and calls it proof. So you do it again, bigger. This is the FOMO engine: every gain amplified, every bull looking like a prophet, the greed feeding on itself.

Here’s the psychological trap underneath it. Nobody adds leverage at the bottom, when it’s actually safe. They add it at the top, when it feels safe. By the time the borrowing is spiking, it isn’t the cautious getting greedy — it’s people who have been right for two years straight and have quietly stopped believing they can be wrong. Confidence is highest exactly where the ground is thinnest.

Then it turns. And the same psychology runs in reverse. A normal 10% dip triggers the first margin calls. The first call doesn’t feel like a signal — it feels like an insult, so people meet it, double down, defend the position. The market takes the cushion. The next call comes, and now it isn’t pride, it’s survival — and everyone sells at once. Forced selling begets forced selling. FUD — fear, uncertainty, doubt — floods in exactly where FOMO ruled a month earlier.

Chart 2: Leveraged-ETF trading activity — every dip bought harder than the last, until it isn’t

«BTFD» — S&P 500 vs aggregated leveraged long-ETF value traded, Topdown Charts, LSEG

That is how a correction becomes a crash. Not because the news got worse — but because the borrowed money all gets called back at the same time. Up is the party. Down is the bill.

Where This Meets the Semiconductor Story

Now connect it to what we’ve been writing all week, because the two stories are the same story.

Remember from Part 1: semis at a historic extreme, record inflows, narrowing leadership. Here’s the leverage underneath that. Roughly 85% of all leveraged-ETF assets now sit in just three sectors: technology, AI, and semiconductors. In a single recent month, leveraged-ETF rebalancing drove over $100 billion of net buying — $38 billion of it into semiconductors alone.

Chart 3: Leveraged-ETF assets under management — and how much of it now sits in semis

Leveraged ETF AUM since 2020 — Semis $66B (30%), Tech ex-semis $81B (37%), Other, Citadel Securities

Sit with what that means. These products are mechanically forced to buy more of what’s rising and sell more of what’s falling, every single day. On the way up, that’s a self-reinforcing loop that makes the semi rally look far more organic than it is. But the loop runs both ways. On June 5, a 3x leveraged semiconductor ETF fell 31% in a single session — a preview, in miniature, of what forced deleveraging looks like when the most crowded, most levered corner of the market gets a scare.

Chart 4: The 3x leveraged semiconductor ETF (SOXL) — assets exploding into the climb

SOXL assets under management, 2022–2026 — the vertical late-stage spike, Koyfin, Kobessi

This is the machinery beneath Part 2’s warning. The semiconductor «story» is already weaker than the price. Layer record leverage — disproportionately concentrated in those exact names — on top of a cracked story, and you have the setup where a routine pullback can become something much faster and deeper. Not because the businesses changed overnight, but because the borrowed money didn’t leave gently.

So What Do You Actually Do With This?

Here is where today’s piece hands off to tomorrow’s — and where I want to be very careful, because froth is not a timing tool.

Margin debt cannot tell you when. The market can stay irrational, and leverage can push well past 55%, for far longer than feels possible. People made fortunes in 1999 buying with both hands. Anyone who tells you this number means «sell now» is selling you a forecast, and we don’t trade forecasts.

What it can do is tell you the weather is dangerous — that the fuel for a violent unwind is stacked high, so when a pullback comes, it may be sharper than the news alone would justify. That’s context, not a trigger.

And here is the reframe that connects it to everything we believe. A hot, over-leveraged market with a cracked leader is not only a risk. Handled with a cool head, the shakeout it eventually produces is one of the great buying opportunities. This is exactly the tension in tomorrow’s finale: two hand-drawn roadmaps, a century apart, suggest we look more like the middle of this cycle than the end — a shakeout ahead, then a final melt-up, with a path that could plausibly carry the S&P 500 and Nasdaq 100 toward 10,000 and 40’000+, respectively, into the back half of the decade. The margin-debt rollover, if and when it comes, is likely to be that shakeout — weak hands liquidated, leverage flushed, the capitulation that hands stock from the late money back to the patient.

Which is why the discipline is the same discipline we preach everywhere: Keep a cool head. Don’t chase the hot tape. Wait for the pitch instead of swinging at everything.

The investor who survives the end of a great run isn’t the one who called the top — nobody calls the top. It’s the one who watched the metrics, kept some dry powder, and had rules ready in advance so that when the forced selling comes, they are the buyer, not the liquidated.

So watch the line. Not for a magic number — for the turn. While it climbs, the party continues and leverage keeps building. The day it rolls over is the day to pay full attention — because that is historically when the deleveraging begins, and with it, both the danger and, for the prepared, the opportunity.

Up is the party. Down is the bill. And the bill, when it arrives, is where fortunes are made by whoever kept their head.

Tomorrow: the finale — When Should You Sell Semiconductors? The twelve rules that let you act on all of this without ever having to guess the date. It’s a process, not a prediction. Read this today; use the rules tomorrow.

Read the rest of the trilogy: Part 1 — The Story Is Weaker Than It Looks · Part 2 — The Climax Run

Love,

Thierry

---

This is educational content, not investment advice, and not a recommendation to buy or sell any security or to use margin. Leverage magnifies losses as well as gains. Figures are drawn from FINRA and public reporting as of the May 2026 release. Past performance is not a reliable indicator of future results. arvy AG is authorised by FINMA as a manager of collective assets under CISA Art. 24.