Macro Regime Radar

A live macro terminal that reads the US economy into one of four regimes every month, then shows the evidence, the models and the data trail underneath.

Open the live site → · Analyst Desk · Methodology · Latest daily memo

Macro Regime Radar is a one-person project: a React and TypeScript front end, a FastAPI service, and a data pipeline that runs itself on GitHub Actions. It pulls macro data from FRED, market prices and headlines; classifies the economy as Goldilocks, Overheating, Stagflation or Recession Risk; scores recession odds with a separate model; and publishes daily and weekly briefings. Every number on the site carries the date of the data behind it.

It is built for three kinds of visitors: someone who wants the macro picture in ten seconds, an analyst who wants to test an idea against history, and an engineer who wants to see how a small, fully automated data product is put together.

---

What's on the site

The site has two parts: the terminal, with seven tabs plus Methodology, and the Analyst Desk, a research and trading workspace.

The terminal

Tab	What you can do there
Dashboard	See the current regime with its odds, the five monitored signals, key market levels, what markets are pricing, and the main macro charts.
Regime Lab	Read the regime's playbook, where it sits in the cycle, the odds of moving to another regime, historical analogues, a scenario builder, the full regime history, and backtests with factor attribution.
Markets	Research any US stock or ETF, watch the macro tape, scan the sector heatmap, and see what markets are pricing and the biggest data surprises.
Credit	Track investment-grade and high-yield spreads by rating, the quality ladder, and financing conditions.
Recession	See the recession model's score, the yield-curve monitor, and sliders that show how each input moves the odds.
News & Calendar	Read priority headlines, each scored for significance and the top ones with a Claude interpretation and cited research, beside the macro release calendar.
Tools	Run an LBO calculator, or an asset allocator with seven optimizers (mean-variance, minimum variance, risk parity, Black-Litterman, HRP, HERC and CVaR) built on how each asset class did in each regime.
Methodology	How every model works, what it can and cannot claim, and where each number comes from.

Around the tabs: a market strip, an "Ask the analyst" assistant that answers questions from the stored data, a ⌘K palette to jump to any section, an alerts drawer, and a Data status drawer that shows how current each feed is.

The Analyst Desk

The Desk turns a macro view into a testable trade idea.

• Market: Overview (what fired, and what is closest to being wrong), Technicals for the S&P 500 or any US stock, sector leadership and breadth, Macro & Correlations, and the regime with the data prints that would change it.
• Research: an Event Study engine that answers "what did markets do after this kind of shock?" with a scored verdict (Reliable, Suggestive, No edge or Too few), and a Signal Ledger of every scored signal.
• Trade: Basket & Hedge prices a basket from real prices and sizes an ETF hedge; Position Monitor tracks how far each position is from being wrong.
• About this build: a Data Pipeline page that traces every panel to its source, and Build Notes.

A few Desk cards (options hedging, positioning, an event study on your basket, the volatility column on Technicals, and a Snowflake sync) are prototypes: they are drawn finished with illustrative values and say so in a footnote. They will be wired to live data over the coming weeks.

---

How the regime call works

1. Growth and inflation trends. Each month the pipeline fits the three-month slope of industrial production and of the CPI level.
2. Four quadrants. Rising growth with falling inflation is Goldilocks; both rising is Overheating; falling growth with rising inflation is Stagflation; both falling is Recession Risk.
3. Odds, not just a label. A temperature-scaled softmax over how far each trend sits from its own history turns that into odds for all four regimes, so the site can say how narrow the lead is.
4. A separate recession model. A logistic regression trained on NBER recession dates scores recession odds from the yield curve, unemployment, the high-yield credit spread, industrial production and the inflation-breakeven spread, read three months earlier. It is labelled as its own model and never mixed with the classifier's odds.

The Methodology page covers the signals, the backtests and the limits of each model.

---

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗