The Margin Debt Manual

Research note. Every figure on this page is computed at render from FINRA's margin statistics as we maintain them (monthly, 1997–present, latest report August 2026) — the same file that powers the live margin debt page — plus the Fed's Z.1 accounts for the pre-1997 history. Methods in How we checked it; editorial standards in How we use AI.

What is margin debt?

Money borrowed against brokerage accounts to hold securities. When an investor buys stock in a margin account and doesn't pay full cash, the broker lends the difference; FINRA collects those debit balances from every member firm and publishes one monthly total. That total stood at $1.45T in the August 2026 report — up 37% from a year earlier. One boundary to keep in mind from the start: this is brokerage margin only. Securities-backed loans at banks, futures leverage, and the borrowing inside options positions never appear here, so the true amount of money riding on stock prices is larger than any number on this page.

The latest report — what changed

The August 2026 statistics put margin debt at $1.45T, up $37B (+2.6%) from July 2026 and 37% above a year earlier. Free credit balances total $0.43T, leaving net free credit at −$1.03T. Excess leverage — margin-debt growth minus SPY's return — reads +18.3 points, the 89th percentile of the record. The next report, covering the following month, should land around the third week of October 2026; every number on this page recomputes when it does.

How the number is built

FINRA aggregates three balances from member firms each month: debit balances in margin accounts (the headline "margin debt"), free credit balances in cash accounts, and free credit balances in margin accounts — the last two being money sitting uninvested. The report lands in the third week of the following month, which matters more than it sounds: whatever the current print says, the borrowing it describes happened five to seven weeks ago. We date every point by its publication date instead of the month it covers, and our Bubble Tracker applies the same lag before letting margin data vote. The series in this form starts in January 1997; FINRA's predecessor series from the NYSE covers earlier decades, and the Fed's Z.1 accounts reach back further still on a quarterly basis. The predecessor series are related but built differently, which is why our headline chart stops at 1997 instead of splicing quietly — the long-history section below shows the older data with the junction marked.

How to read it — never as a raw level

Margin debt sets all-time highs constantly, for the same reason the market does — both grow with prices and with the economy. A record level, on its own, is a headline and not a finding. The readings that carry information are relative, and the live page charts all three: the level divided by SPY (is borrowing outrunning the thing it finances?), the level as a share of M2 (is it outrunning the money supply?), and what we call excess leverage — the year-over-year growth of margin debt minus the year-over-year return of SPY. That last one is the speed camera. Right now it reads +18.3 points — the 89th percentile of the full history. Borrowing is growing much faster than the market it's buying.

There's a fourth read most coverage skips: net free credit, the cash sitting in accounts minus the debt borrowed against them. Investors currently hold $0.43T of free credit against $1.45T of margin debt — a net position of −$1.03T (the record was set June 2026). When the cushion is thin and prices fall, margin calls sell into weakness — that's the mechanism people actually worry about when they worry about this series.

Margin debt vs GDP and market cap — the denominators disagree

The two most-quoted scalings of this series currently tell opposite stories, and both are worth seeing. Against nominal GDP, margin debt stands at 4.48% — above the 3.0% peak of March 2000 and the 3.8% peak of August 2021, a record for the FINRA era. Against total equity market value, it is 1.55% — below the 2.3% high of February 2008, because market cap itself has grown faster than the economy. Which denominator is right? Neither, alone. Scaling by GDP says borrowing is extreme relative to the economy that ultimately backs it; scaling by market cap quietly assumes today's prices are the right yardstick — circular, when margin borrowing is part of what holds those prices up. We treat both as context and lean on the scale-free reads above (excess leverage, net free credit), which need no denominator at all.

The long history — margin leverage since 1947

The FINRA series starts in 1997, but the Fed's Z.1 financial accounts track margin loans quarterly back to the 1940s — and at the 1997 junction the two agree within about 2% ($101B vs $103B), close enough to read as one story if the seam stays visible. Scaled by GDP, the pattern is stark: for its first fifty years, margin borrowing lived below 1.3% of GDP. The go-go market of the late 1960s peaked at 1.2% (1968); the run-up to the 1987 crash reached 1.2%. The modern era broke the scale — 3.0% at the dot-com peak, 3.8% in 2021, and 4.5% today. Before all of that sits the one episode our data cannot reach: 1929, when brokers' loans are estimated by financial historians to have approached 8% of GDP before the crash unwound them. That number comes from historical estimates, not our dataset, and we label it accordingly — but it is the reason margin debt has been a watched series for nearly a century.

What happened at past margin debt peaks?

Three episodes dominate the modern history, and they teach two different lessons. Here is each one from our data — the peak, the market top near it, and where the debt went afterward.

Lesson one: the peak itself warned nobody. Margin debt topped the same month as the market in 2000 and within three months of it in 2007 and 2021 — by the time you could see the peak in a report published weeks later, the top was behind you. Lesson two: the surge was different. Year-over-year growth above today's 37% has appeared in exactly 38 months of the 344 on record, and they sort into four clusters: the run-ups to those three peaks — and the spring of 2026. Growth that fast has never yet ended quietly. Small samples deserve suspicion, ours included; three prior episodes are a pattern. They are not a law.

What happens when margin calls hit

The mechanics are worth spelling out, because they are the entire reason a borrowing statistic doubles as a risk gauge. Under Regulation T an investor can borrow up to half of a purchase; after that, maintenance rules require the account's equity to stay above a floor — 25% by rule, and typically 30–40% at the broker's discretion. When prices fall far enough to breach the floor, the broker issues a margin call: deposit cash, or positions get sold. And there is the feedback loop — the selling is not optional and not price-sensitive. Forced sales push prices lower, which breaches more accounts, which forces more selling. Deleveraging that starts as one account's problem propagates into everyone's prices; the episode table above shows the aggregate result, with margin debt shrinking 40–50% from each peak. This is why net free credit matters more than the debt level alone: the cash cushion is what stands between a routine decline and a cascade. When it is thin — and it is currently negative by $1.03T — the market's shock absorber is, in aggregate, already spent.

Where this gauge will mislead you

• The lag. Each print describes borrowing from roughly two months ago. Anyone claiming margin data "called" a move that happened inside that window is reading a report that didn't exist yet.
• Peaks are coincident, not leading. The table above is the evidence. This series confirms tops in hindsight; it does not schedule them.
• You can't see who is borrowing. One total blends retail accounts, hedge funds and family offices. The 2021 Archegos losses sat outside this series entirely — prime-broker swap leverage never touches FINRA's number.
• Denominators are a choice. As the ratio section shows, %-of-GDP says record and %-of-market-cap says below 2007 — in the same month. Anyone quoting only the scarier one is selling something.
• The long history is a splice. Z.1 and FINRA measure related but different things; cross-era comparisons are directionally useful and precise to nobody.

The last 12 reports

Full series from 1997 as CSV: margin-debt.csv.

How we checked it

The core series is FINRA's monthly margin statistics as we maintain them, 1997 to the latest report, dated by publication rather than reference month. Peaks and troughs in the episode table are the highest and lowest monthly readings inside each window; "peak YoY" is the fastest year-over-year growth in the episode's run-up; the market-top dates are the S&P 500's closing highs of each cycle. Percentiles rank the latest value against every month in the history. The GDP and market-value ratios join each month's debt to the latest quarterly Z.1/BEA figure published on or before it — an as-of join, never interpolation — and the pre-1997 history uses the Fed's Z.1 margin-loan series only up to the 1997 splice, where it agrees with FINRA within about 2%. Everything recomputes when a new report lands, so the numbers in this Manual always match the live page.

Frequently asked questions