The identities we work from
Every dollar of government deficit is, to the dollar, a surplus somewhere in the non-government sector. That is not a theory — it is accounting. The framework starts from sectoral balances and stays stock-flow consistent the whole way down: every flow lands on a balance sheet, every balance sheet position has to be financed, and nothing nets to zero unless it actually nets to zero. The lens comes out of the Modern Monetary Theory tradition, applied strictly as a description of how the monetary system operates — never as a policy agenda.
The daily read
The Daily Treasury Statement lands at 4:00 PM ET every business day, reporting the previous day's actual settlements. We read it the same evening: the Treasury General Account (TGA) balance, withheld income and employment taxes (the real-time wage bill), gross outlays, and the settlement calendar. Withheld receipts tell you about nominal growth weeks before the survey data; the TGA path and auction settlements tell you what is about to happen to bank reserves — which is what actually moves risk appetite at the margin.
Around the daily statement sit the slower prints: the Monthly Treasury Statement for the full fiscal decomposition, the quarterly refunding documents for supply, and FINRA margin statistics for leverage. All of it is public. The edge is not the feed — it is the reading, done every day, inside one consistent framework.
The data we maintain
Everything on this site — every chart, every model, every figure in a research note — reads from one database of series we collect and maintain ourselves:
- Daily Treasury Statement — every line item, every business day since 2013: more than 350,000 entries across 3,400 trading days. TGA balances, withheld receipts, outlays, issuance and redemptions.
- Monthly Treasury Statement — the full fiscal decomposition, monthly back to 1974.
- Margin debt — FINRA member-firm statistics, monthly back to 1996.
- Investor sentiment — the AAII survey, weekly back to 1987.
- Market series — S&P 500 daily closes, volatility-surface features (the variance risk premium, dealer-positioning proxies, futures term structure), and policy rates back to 2001.
Publication schedules belong to the data, not to us, and we keep them honest: the DTS reports on a one-day lag, AAII sentiment prints on Thursdays, FINRA margin figures arrive weeks after the month they measure. The dashboards state, series by series, exactly how fresh each one is — and when an input is late, they say so rather than interpolating quietly.
From flow to position
A reading becomes a position through the same sequence every time: the flow prints, the balance-sheet effect follows mechanically, and then someone — dealers, banks, corporates, leveraged investors — is forced to act. Heavy issuance into a thin reserve backdrop is a different market than the same issuance into an ample one. We write the implication only when the mechanism is explicit end to end. If we cannot draw the chain from the print to the position, we do not publish the trade.
The model layer
Four models sit on top of the data, each deliberately narrow:
- Flow / Sentiment — measures divergences between Treasury cash flows and investor sentiment, read as three signals: deviation, momentum, and trend. Weekly by construction — each signal pins to an AAII release and is stamped for the week it applies to, never backfilled.
- Flow Phase — a three-phase classifier in the Minsky tradition, mapping debt and deficit dynamics onto equity-market regimes: distress, neutral, expansion. Fitted walk-forward, so the model never sees the period it is scored on.
- Vol Shift — classifies the magnitude of the next session's S&P move from the variance risk premium, dealer-positioning proxies, contango, and the VIX term structure.
- DeepMinsky — our simulation platform: a multi-sector system-dynamics model of households, firms, banks, the Treasury, the central bank, and the foreign sector, linked through explicit stock-flow accounting. It does not fit regressions to history; it models the causal mechanisms and lets credit cycles emerge endogenously, in the tradition of Hyman Minsky's financing taxonomy — hedge, speculative, Ponzi.
The models retrain every morning, after the day's inputs land — the dateline at the top of every page says which Treasury statement the current signals were trained on. We use them to run scenarios, not to sell point forecasts. Institutional readers can request the DeepMinsky whitepaper.
What we do not claim
No equilibrium assumptions, no certainty sold as such, and no performance claims — we publish research, not a track record. Where a figure on this site is live data it carries its source and date; where it is model output it says so. A weekly signal is presented as weekly, not dressed up as daily. When we are wrong, the framework is the tool we use to find out why.