How MacroScope Works
A plain-English guide to everything on the dashboard — no finance or machine learning background required.
What is a macro regime?
The economy doesn't move in a straight line. It cycles through distinct phases — sometimes growing fast, sometimes slowing, sometimes contracting, sometimes recovering. Each phase is called a macro regime.
The regime matters because different investments behave very differently depending on which phase the economy is in. Stocks that soar in Expansion can collapse in Contraction. Bonds that look boring in Recovery can be the safest asset in a downturn.
MacroScope classifies the current economy into one of four regimes every month, using five publicly available economic signals.
The four regimes
The economy is growing. Companies are hiring, GDP is rising, and investors feel confident.
Signals: Low unemployment, tight credit spreads, moderate VIX, positive yield curve.
For stocks: Historically the best period for equities. Growth stocks outperform.
Growth is still positive but slowing. The expansion is aging — cracks are starting to appear.
Signals: Yield curve flattening, early credit stress, VIX starting to rise.
For stocks: Reduce risk. Defensive sectors (utilities, healthcare) start to outperform.
The economy is healing after a downturn. Growth is resuming from a low base.
Signals: Credit spreads compressing, unemployment falling from peak, VIX declining.
For stocks: Early opportunity — beaten-down cyclicals and financials tend to lead.
The economy is shrinking or stagnant. Fear is elevated and credit is stressed.
Signals: High VIX, wide credit spreads, rising unemployment, inverted or flat yield curve.
For stocks: Defensive positioning. Cash, bonds, and defensive equities. Avoid cyclicals.
The five macro indicators
The VIX measures how much volatility the options market expects in the S&P 500 over the next 30 days. Think of it as a fear gauge — when investors are calm it's low (under 15), when they're panicking it spikes (above 40 in 2008 and 2020).
High VIX = bad for stocks. Low VIX = calm markets.
The difference between what the US government pays to borrow money for 10 years vs. 2 years. Normally 10-year rates are higher (you want more compensation for locking up money longer). When 2-year rates exceed 10-year rates (the spread goes negative), it's called an "inverted yield curve" — and it has predicted every US recession since the 1970s.
Negative spread = recession warning. Positive spread = healthy.
The number of people employed in manufacturing in the US (in thousands). When factories are hiring, the economy is producing goods and expanding. When they're laying off, it signals contraction. This is used as a proxy for the ISM Manufacturing PMI (a classic business activity survey) because the original PMI isn't freely available from FRED.
Rising = expansion signal. Falling = contraction signal.
The extra interest rate that medium-quality ("Baa" rated) companies must pay to borrow money, compared to the US government. When this spread is wide, it means investors distrust corporate borrowers and want compensation for the risk of default. Wide spreads are a classic sign of financial stress — they spiked sharply in 2008, 2016, and 2020.
Wide spread = market stress. Tight spread = confidence.
The percentage of people in the US labor force who are actively looking for work but can't find it. It's the most widely-watched economic indicator in the world. The Federal Reserve's primary job is keeping unemployment low while controlling inflation. High unemployment = recession. Low unemployment = healthy economy — but also potential inflation pressure.
Rising = economic stress. Falling = recovery/expansion.
What is a z-score?
A z-score measures how unusual a reading is compared to its historical average.
- z = 0 — exactly average
- z = +1.0 — one standard deviation above average (unusual but not extreme)
- z = +2.0 — two standard deviations above average (very unusual — happens ~5% of the time)
- z = −1.5 — below average
For example: an Unemployment z-score of +1.77 means unemployment is significantly higher than its 35-year average — a contraction signal. A VIX z-score of −0.62 means VIX is slightly below average — markets are calm.
Z-scores let the model compare apples to oranges: a VIX reading of 35 and an unemployment rate of 7% are on completely different scales, but z = +2.5 and z = +1.8 are directly comparable.
What do the trend arrows mean?
Each indicator card shows an arrow indicating recent momentum — whether the indicator is moving up, down, or sideways over the last month.
Context matters: rising unemployment (↑) is bad; rising manufacturing employment (↑) is good. The z-score and description tell you the full picture.
Layer 1 — Hidden Markov Model (the regime classifier)
The Hidden Markov Model (HMM)is the engine that classifies the current regime. Here's the intuition:
Imagine you're blindfolded and trying to figure out what season it is just by feeling the temperature, humidity, and wind each day. You can't see the season directly — it's "hidden" — but you can observe clues. Over time you learn that certain combinations of clues (cold + dry + low wind) reliably signal winter.
The HMM does the same thing with the economy. The "season" is the regime (Expansion / Contraction / etc.). The "clues" are the five z-scored indicators. The model was trained on 35 years of monthly FRED data and learned — entirely without being told in advance — which combinations of signals cluster together into distinct economic environments.
It uses 5 hidden states internally (because the 2022 rate-shock bear market looks different from the 2008 unemployment-driven recession), but maps them to 4 named regimes for display.
The confidence percentageon the dashboard (e.g., "100% conf.") is the HMM's posterior probability — how certain it is about the current classification given the observed data. 100% means all five signals are unambiguously pointing to Contraction.
Layer 2 — AutoGluon (the transition forecaster)
AutoGluon is a machine learning toolkit from Amazon that automatically trains and combines many different models (decision trees, gradient boosting, etc.) and picks the best combination.
While the HMM tells you "what regime are we in right now," AutoGluon answers "what regime are we most likely to be in next month?"
It was trained using walk-forward validation: at each step, it was only allowed to train on data from the past and was tested on the future — exactly like a real investor who can only use historical information. This prevents the model from "cheating" by learning from future data it wouldn't have had access to.
The Transition Forecastpanel shows the probability distribution over next month's regime. A reading of "Contraction 62%" means the model assigns a 62% probability to staying in Contraction — it's not certain, but it's the most likely outcome.
What each section of the dashboard shows
Current Macro Regime (top panel)
The HMM's current classification with a confidence dial. The probability bars on the right show how the confidence is distributed across all four regimes. 100% Contraction means the model sees no ambiguity.
Macro Indicators (5 cards)
The actual values of the five FRED signals as of the most recent fully-published month (data releases lag by 2–4 weeks). Each card shows the raw value, z-score, trend arrow, and a plain-English description of what the indicator measures.
Regime History — S&P 500 Cumulative Return (chart)
The blue line shows what $1 invested in the S&P 500 in 1990 would be worth today (about 26x). The colored background bands show what regime MacroScope classified each historical month as. Notice how the red Contraction bands (2001, 2008, 2020, 2022) align with the biggest dips in the blue line.
Transition Forecast (right panel)
AutoGluon's probability distribution for next month's regime. These are model probabilities — not certainties. The model was validated on out-of-sample data with 90% accuracy at regime identification.
Historical Returns by Regime (table)
The actual median and mean monthly S&P 500 returns across all months in each regime, from 1990–2026. This is empirical — it shows what historically happened in each regime, not what a model predicts. 'Observations' is how many months of data underlie each figure.
Methodology (accordion)
Detailed technical explanations of every modeling decision: why 5 HMM states instead of 4, how the z-score weighting was calibrated, what walk-forward validation means, and an honest account of the backtesting results (including why the strategy underperforms buy-and-hold).
How often does the data update?
MacroScope updates monthly, because the underlying FRED data is monthly. Unemployment, credit spreads, and manufacturing employment are all released with a 2–4 week lag, so the "current" reading always reflects the most recently completed month.
The "Updated" date on the dashboard shows when the data pipeline was last run and the predictions were last regenerated. This is not a real-time feed — it is a monthly snapshot that gets refreshed when new FRED data is available.
Is this investment advice?
No. MacroScope is a research and education project. It demonstrates machine learning techniques applied to public macroeconomic data.
The backtesting results show that a naive long/short strategy based on regime classification underperforms simple buy-and-hold over 35 years. Regime detection is most useful as a risk awareness tool — not as a trading signal.
Past regime classifications do not guarantee future accuracy. Do not make investment decisions based on this site.