Where the cycle is now, and how the model has done

Regime now: Goldilocks, with financial conditions tightening

Nowcast from Chicago Fed CFNAI-MA3, core PCE prices (BEA) and the Chicago Fed NFCI, scored on first-release data; stress from daily S&P 500 ETF returns.

The nowcast puts September 2026 in Goldilocks: growth at or above trend with core inflation not rising. Growth is estimated at +0.085 on the CFNAI-MA3 (zero is trend) and core PCE inflation at 2.94%, -0.10 points over three months. The last published label still reads Stagflation, because the official data run 75 days behind: that gap is what the nowcast exists to close.

Risk-off scorecard: 4 of 7 signals

One line per pillar: is it pointing risk-off right now, and since when. The count is a plain sum of the published labels, no weights and no fitting. It was 1 four weeks ago. Note that the growth signal uses the last published label (data through July 2026); the nowcast already reads above trend.

What a high count has meant: higher volatility, not lower returns. With four or more signals on, S&P 500 ETF volatility averaged 22.9% against 11.6% with one or none, and four-week periods were up 59% of the time against 70%. Average returns were not lower: risk-off weeks are when markets pay most for bearing risk. And the count is no early warning on its own: the worst four-week fall in the sample (-31.4%) began in a week with one or no signals on. Use the count to size, not to call direction.

What changed

Every label, the date its current reading was first published, and what it replaced.

Regime history: as published vs revised

The top row is what an investor saw on each date. The bottom row applies the same rule to today's revised data. Where they differ, a backtest on revised data would have traded a regime nobody could see at the time.

Data behind this section: Chicago Fed National Activity Index, 3-month average CFNAIMA3 · PCE price index excluding food and energy (core PCE) PCEPILFE · Chicago Fed National Financial Conditions Index NFCI · Chicago Fed NFCI risk subindex NFCIRISK · Federal Reserve total assets WALCL · Treasury General Account (TGA) WTREGEN · Overnight reverse repurchase agreements (Fed RRP) RRPONTSYD · Chicago Fed NFCI credit subindex NFCICREDIT · /v1/model/latest

Growth and inflation: the two axes of the quadrant

Chicago Fed National Activity Index (85 monthly indicators, 3-month average) and the BEA core PCE price index, latest vintage; labels point-in-time.

Growth is above trend when the CFNAI-MA3 is at or above zero. For the quadrant, inflation counts as rising when core PCE inflation is above its level three months earlier; the separate inflation direction label adds a ±0.05 point band, so it can read flat while the quadrant reads rising. Shading shows the quadrant label; switch between the label as published and the label on revised data.

The last twelve months

The first-published label applies the rule to the data as it was released at the time, about 75 days after the month; the revised quadrant applies it to today's data. Where they differ, the month was revised.

Data behind this section: Chicago Fed National Activity Index, 3-month average CFNAIMA3 · PCE price index excluding food and energy (core PCE) PCEPILFE · macro_quadrant · growth_level · inflation_direction

Financial conditions: tightening, easing or flat, every week

Chicago Fed National Financial Conditions Index (105 measures of risk, credit and leverage) and its risk subindex, weekly; labels point-in-time.

A move of more than 0.02 in the NFCI over four weeks labels conditions as tightening (up) or easing (down). The nowcast predicts the change for the week not yet published: for the week to 9 October 2026 it estimated +0.0389; the published figure is +0.042. Shading is the label as it was published each week.

What happened next: one week after each label

Tightening weeks have not been followed by lower average returns; they have been followed by much higher volatility (22.0% against 13.1% after easing weeks). The label is a risk measure, which is how it is best used: for sizing, not for direction. Price returns, distributions excluded, since 2011-05-27.

Rates, curve, inflation expectations and the dollar

The context a regime label sits in. Pick a series; shading is the macro quadrant as published. Weekly (Friday) values of the daily series.

Data behind this section: Chicago Fed National Financial Conditions Index NFCI · Chicago Fed NFCI risk subindex NFCIRISK · 10-year Treasury yield DGS10 · 2-year Treasury yield DGS2 · 10-year minus 2-year Treasury spread T10Y2Y · 10-year minus 3-month Treasury spread T10Y3M · 10-year TIPS real yield DFII10 · 10-year breakeven inflation T10YIE · Effective federal funds rate DFF · Broad trade-weighted US dollar index DTWEXBGS · WTI crude oil price DCOILWTICO · US economic policy uncertainty index USEPUINDXD · fincond_direction · /v1/model/nowcast · SPY · IWM

Liquidity: is the Fed adding dollars to markets or draining them?

Federal Reserve H.4.1 balance sheet and Treasury General Account, New York Fed reverse repo operations, H.6 money stock; weekly, labels point-in-time from ALFRED vintages.

Fed net liquidity is the Fed's balance sheet less the cash that sits outside markets: the Treasury's account at the Fed and money parked in the reverse repo facility. In the week to 7 October 2026 it stood at $5,865bn, -1.56% over 13 weeks, so the label is Contracting. Fed assets $6,748bn, Treasury account $880bn, reverse repos $2.3bn, bank reserves $3,030bn. M2 is up 5.66% on the year (August 2026).

What happened next: one week after each label

Unlike financial conditions, liquidity has carried a directional signal. Since 2011 the S&P 500 ETF returned +19.6% a year in the weeks after an expanding label and +4.6% after a contracting one, at similar volatility (17.1% and 16.3%). Weekly returns are noisy and these are averages over hundreds of weeks, not a forecast for next week.

Liquidity and credit together

Next-week S&P 500 ETF return, annualized, for each combination of the two weekly labels since 2012. Expanding liquidity with easing credit returned +22.1%; contracting liquidity with credit flat or easing returned -3.0% and +2.2%. The tightening column pays more on average but with far higher volatility: those are the weeks after sell-offs, when rebounds are large and so are the risks.

Data behind this section: Federal Reserve total assets WALCL · Treasury General Account (TGA) WTREGEN · Overnight reverse repurchase agreements (Fed RRP) RRPONTSYD · Reserve balances with Federal Reserve Banks WRESBAL · M2 money stock M2SL · liquidity_direction · Money, liquidity & capital flows dataset · FinzData Weekly: liquidity, flows & positioning

Credit: the price and availability of borrowing

Chicago Fed NFCI credit and leverage subindexes (weekly, ALFRED vintages), the Gilchrist-Zakrajsek excess bond premium (Federal Reserve Board, monthly), commercial paper and T-bill rates, H.8 bank credit and the Senior Loan Officer Survey.

Two labels. Credit conditions direction applies the financial-conditions rule to the NFCI credit subindex: it moved +0.025 over the 4 weeks to 2 October 2026, so Tightening. Credit stress level reads the excess bond premium, the part of corporate bond spreads not explained by default risk, which is the bond market's appetite for risk: -0.39 in September 2026, so Benign. The 3-month commercial paper spread over T-bills is +0.03 points, and the net share of banks tightening business loan standards was 0% in the latest Senior Loan Officer Survey (2026 Q3).

What happened next: credit conditions, one week later

What happened next: credit stress, one month later

Stressed credit has been the costly state: the S&P 500 ETF returned -4.2% a year in the month after a stressed reading, against +11.7% after benign ones, with volatility of 22.0% against 12.9%. One caveat in the open: the Fed re-estimates the excess bond premium every month and keeps no vintages, so this table uses today's values, not the ones published at the time. The two weekly labels are fully point-in-time.

Data behind this section: Chicago Fed NFCI credit subindex NFCICREDIT · Chicago Fed NFCI leverage subindex NFCILEVERAGE · 3-month AA financial commercial paper rate DCPF3M · 3-month Treasury bill rate DTB3 · Bank credit, all commercial banks TOTBKCR · Commercial and industrial loans, all commercial banks BUSLOANS · Net % of banks tightening C&I loan standards (SLOOS) DRTSCILM · Excess bond premium (Federal Reserve Board) · credit_direction · credit_stress · ICE BofA high-yield spread (licensed)

Recession risk: what the yield curve, the bond market and the labor market say

Treasury yields (10-year minus 3-month), the Federal Reserve Board's excess bond premium model, the real-time Sahm rule and initial jobless claims; NBER recessions shaded.

Three independent reads on the next year. The yield curve model (the New York Fed's probit on the 10-year minus 3-month spread) puts the chance of a recession within 12 months at 11.7%, from 27.9% a year ago, with the spread at +1.04 points in October 2026. The Fed's bond-market model, built on corporate credit spreads and the excess bond premium, says 9.1% (September 2026; 11.5% a year earlier). The labor market: the real-time Sahm rule reads 0.00 points against a 0.50 trigger, and initial claims average 198k over four weeks, from 228k a year ago.

These are published models, shown so you can set the regime labels against them; they are not FinzData labels and carry no FinzData track record. The yield-curve probability uses the New York Fed's published coefficients on the monthly average spread; the bond-market probability is the Federal Reserve Board's own estimate, published with the excess bond premium.

Data behind this section: 10-year minus 3-month Treasury spread T10Y3M · Real-time Sahm rule recession indicator SAHMREALTIME · Initial jobless claims ICSA · NBER recession indicator USREC · Yield curve as a leading indicator (New York Fed) · Recession risk and the excess bond premium (Federal Reserve Board)

Market stress: the probability of the high-volatility state

Two-state Gaussian Markov-switching model of daily S&P 500 ETF returns and 10-day volatility, z-scored causally, refit every 126 sessions on a rolling 750-session window. Filtered, walk-forward: each value uses only data up to that close.

The model reads 0.03% today. Across 7,666 sessions since 1996, markets spent 29.5% of days above 90% and 64.0% below 10%; it found 104 stress episodes, 36 of them lasting a month or more. The point is not direction: next-day returns were similar in both states. The point is risk: realized volatility over the following month averaged 22.6% after high-stress days against 13.7% after calm ones, and the worst next day was -10.9% against -5.7%.

What each probability level has meant

Try it: a stress-aware S&P 500 allocation

Hold the ETF, and cut exposure when yesterday's stress probability is above the threshold. Positions use the previous close's probability, cash earns the 3-month T-bill, and every change in exposure pays the cost you set. This is an illustration of how the series behaves, not a recommendation.

Recent stress episodes (probability above 50%)

Data behind this section: 3-month Treasury bill rate DTB3 · S&P 500 ETF (SPY) daily bars · /v1/prices/daily?symbol=SPY · /v1/model/stress

Regimes and returns: what each quadrant has paid, with and without hindsight

Next-month price returns of 22 ETFs, grouped by the macro quadrant known at the previous month end, since 2011-06-15.

With labels as published, the S&P 500 ETF returned +26.5% a year in Goldilocks months and +12.4% in Slowdown months. Switch to revised data and the picture moves: Goldilocks shows +8.4% and Stagflation +9.5% (against +12.5% as published). The difference is the cost of hindsight, and it is why every FinzData label carries the date it was known.

Annualized return, % over the month after each label, as published. Price returns, distributions excluded.

How long regimes last

Monthly transition probabilities on revised data since 2004 (269 months). Regimes are persistent: the chance of staying put next month runs from 54% to 74%, which is what makes a label worth trading on at all.

Data behind this section: Chicago Fed National Activity Index, 3-month average CFNAIMA3 · PCE price index excluding food and energy (core PCE) PCEPILFE · SPY · QQQ · IWM · EFA · EEM · TLT · IEF · LQD · HYG · GLD · DBC · UUP · VNQ · XLK · XLF · XLE · XLV · XLP · XLY · XLI · XLU · XLB

Track record: scored against first-release data

Point-in-time labels since 2005 (inflation) and 2011 (growth and financial conditions); nowcast scored walk-forward. Realized labels use the first release of each period.

The fair benchmark for any nowcast is the label you can already read off the last published data. For the macro quadrant that label was right 63.6% of the time for the month that mattered; the nowcast was right 75.9%. For weekly financial conditions: 71.3% against 79.5%. The liquidity and credit labels have no nowcast yet; their published labels were right 79.3% and 71.4% of the time for the following week. The forward record, scored only on dates after the model was frozen, is accumulating and will be published here as it matures.

Published label (rows) against the label realized for that period (columns), counts. The diagonal is where the published label was right; the off-diagonal cells are the turning points the nowcast is built to catch.

Data behind this section: /v1/model/labels returns label and realized_label for every date, so you can rescore any of this yourself

Methodology: every rule, written down

Labels are computed from each input as it was published on the as-of date (ALFRED vintages), so history contains no later revisions. realized_label is the same rule applied to the first release of the target period, filled once that period is published. The same rules are returned by /v1/model/catalog, free.

The nowcast

Published labels describe a period that has already ended: monthly inputs arrive about 75 days late and the NFCI a week late. Each day the nowcast estimates the label for the period not yet published, from market data available by that date, and keeps the published label beside it. Weekly, it predicts the 4-week NFCI change; monthly, it estimates CFNAI-MA3 and core PCE inflation for the target month and applies the same rules. It is scored on the first release of each period, never on later revisions.

The stress model

A two-state Gaussian Markov-switching model (Hamilton filter, EM estimation) on two causally standardized features: the daily log return of the S&P 500 ETF and its 10-day volatility. States are ordered by variance at every refit, so state 1 is always the high-volatility state. The model is refit every 126 sessions on the last 750; between refits the parameters are frozen and the filter runs forward. Values are filtered, not smoothed: no value uses data after its own close. The history on this page is the same specification run over the full price history; the live value is the one the API serves.

The risk-off scorecard

Each Friday, one point for each of: financial conditions tightening, credit conditions tightening, Fed liquidity contracting, credit stress elevated or stressed, growth below trend, core inflation rising, and a market stress probability above 50%. Every input is the label as published that day, carried forward until the next publication; no weights, no fitting. The credit stress level is the one input built on current rather than first-published data.

Recession indicators

Yield-curve probability: Φ(−0.5333 − 0.6330 × spread), where spread is the monthly average of the 10-year minus 3-month Treasury yield, the New York Fed's published specification for a recession within 12 months. Bond-market probability: the Federal Reserve Board's published estimate from the Gilchrist-Zakrajsek spread and excess bond premium. Sahm rule: the real-time series published by the St. Louis Fed. Initial claims: the Department of Labor weekly series, 4-week average.

What the model does not do

It does not forecast returns, and the tables above show why it should not be read that way: the regime and stress labels describe the risk environment. Labels can be wrong for weeks around turning points, and the walk-forward scores above include those weeks. Nothing here is investment advice.

The data behind the model

Every input, from the FinzData database. Macro values are the latest vintage; each series page shows the full history and every revision, and as_of= returns any series as it was published on a date. Licensed series are listed with their API endpoint only. Download this page's data as JSON or CSV.

Macro and market series

ETF prices

Model API

Rebuild it in Python

import finzdata                                    # pip install finzdata
c = finzdata.Client()
now    = c.model("latest")                         # every label, the nowcast and today's stress probability
quad   = c.model("labels", series="macro_quadrant")  # point-in-time history with realized_label
stress = c.model("stress", start="2007-01-01")     # daily filtered P(high-volatility state)
nfci_then = c.macro("NFCI", as_of="2020-03-20")    # the input as it was published that day
spy    = c.prices("SPY")                           # daily bars for the S&P 500 ETF

# score the published label yourself
hit = (quad.label == quad.realized_label).mean()

Get a free API key   Model access and look-back plans · Browse all macro series · Research

Sources: Federal Reserve Bank of Chicago (CFNAI, NFCI and subindexes), U.S. Bureau of Economic Analysis (PCE prices), Board of Governors of the Federal Reserve System (H.4.1, H.6, H.8, Senior Loan Officer Survey, excess bond premium), Federal Reserve Bank of New York (reverse repo), U.S. Treasury (rates, Treasury General Account), via FRED and ALFRED; ETF prices from FinzData. Built for AI, quants, data scientists and financial analysts who backtest. Clean data. Real insights. Real alpha. finzdata.com. For information only; not investment advice.