A plain-English guide to nowcasting.

The cleanest way to know what tomorrow's data print will show, before it shows, is to read a nowcast. The Cleveland Fed publishes a daily inflation nowcast; the Atlanta Fed publishes a daily GDP nowcast (GDPNow); the New York Fed publishes a weekly nowcast as part of its Weekly Economic Index. Each model takes the universe of already-released indicators and translates them into a real-time forecast of the headline series. The models routinely beat the Bloomberg consensus by 30 to 50 basis points on release day. With Friday's core PCE the next major test, this is the explainer on how the nowcasts are built, what their track record actually is, and how to read them against the upcoming print.

What a nowcast actually is

A nowcast is a statistical forecast of a current-period economic indicator constructed from already-released data. The technique was developed at the European Central Bank in the mid-2000s and has become a standard tool at most G7 central banks. The architecture is the same across most published nowcasts:

1. Identify the target series. Headline CPI year-on-year. Core PCE month-on-month. GDP annualised quarterly. The specific series whose next print the nowcast is forecasting.
2. Identify the input panel. The set of higher-frequency indicators that contain information about the target. For inflation nowcasts, this includes weekly retail gasoline prices, daily crude futures, weekly chain-store sales, and monthly ADP wages. For GDP nowcasts, the panel is much wider: ISM, durable goods, payrolls, hours worked, inventory data, etc.
3. Fit a state-space or factor model. The relationship between the input panel and the target series is estimated econometrically. The model typically combines a state-space (Kalman filter) or dynamic-factor structure with bridge equations that translate the high-frequency inputs into lower-frequency targets.
4. Update daily as new inputs arrive. The nowcast is republished whenever a new indicator is released. The convergence path from initial estimate to release day is monotonic in expectation; the variance narrows as more inputs land.

The three main US nowcasts

• Cleveland Fed inflation nowcast. Published daily on the Cleveland Fed website. Forecasts headline CPI, core CPI, headline PCE, and core PCE for the current and next quarters. The model is published with full documentation and replicable code. The one-month-ahead median forecast error in the post-2010 sample is roughly 10 basis points.
• Atlanta Fed GDPNow. Published twice weekly. Forecasts GDP annualised quarterly growth. Started in 2014; the median forecast error on release day in the post-2015 sample is roughly 30 basis points. GDPNow tends to start volatile (large standard error 8 weeks ahead) and converge as data accumulates.
• NY Fed Weekly Economic Index (WEI). Published weekly. Combines 10 indicators into a single "year-over-year growth" estimate that proxies for real-time GDP growth. WEI is the slowest-moving of the three; it captures cyclical regime changes well but misses single-month inflection points.

How the nowcast compares to consensus

Why nowcasts beat consensus

Three structural reasons:

1. Information processing speed. Nowcasts update mechanically whenever new data lands. Consensus surveys ask economists for a point forecast; survey respondents typically update slowly between releases because the cost of producing a fresh forecast every week is high. The nowcast captures the new information faster.
2. No anchoring bias. Survey respondents anchor to their own previous forecast (a behavioural tendency well documented in the FOMC SEP). Nowcasts have no such anchoring; the model output is the model output.
3. Wider input panel. The Cleveland Fed inflation nowcast uses roughly 12 separate data series; GDPNow uses 17. A median consensus economist might track 3 or 4. The information advantage compounds.

The track record is consistent. Cleveland nowcasts beat consensus on inflation prints by 10 basis points on average in the post-2010 sample. GDPNow beats consensus on GDP prints by 30 basis points on release day. Both margins are statistically significant.

What nowcasts get wrong

Three failure modes worth knowing:

1. Regime breaks. Nowcasts assume the historical input-to-target relationship holds. When the economy shifts regime (post-COVID labour-supply changes, post-Iran energy-shock pricing), the model lags. The 2022 inflation surge was poorly nowcast by published models for roughly six months.
2. Tail prints. The model averages over the input panel. When a single input has a tail value (a one-off ISM print), the model dampens it. A real tail print in the target series is systematically under-nowcast.
3. Revision interactions. Nowcasts use the published data series, which are themselves revised. The nowcast for the same target can change not because the target moved but because a prior input was revised. The Cleveland Fed publishes revision history; reading it matters when the nowcast moves sharply on no apparent news.

How to read the nowcast into tomorrow's PCE

The Cleveland Fed nowcast as of yesterday's update sits at 0.18 percent month-on-month for May core PCE, against the Bloomberg consensus of 0.20 percent. The nowcast is suggesting a marginally softer print than consensus is pricing.

Three reads on what that implies:

• The base case favours a soft surprise. The nowcast model has been within 5 basis points of realised on the last six PCE prints. A 0.18 print against a 0.20 consensus would not be a regime change, but it would be the kind of marginal undershoot that lets Warsh's look-through framing get a clean data tailwind.
• The tail risk is on the upside. Nowcasts under-state tail prints by construction. If May core PCE prints 0.30 or above, the nowcast under-shot, and the financial-conditions reaction on the upside will be sharper because no one was positioned for it.
• The 5y5y breakeven will move first. Real-money reaction to the print runs through the breakeven compression first, the curve second, FX third. Watch the breakeven tape in the first 15 minutes after the 8:30 release.

What nowcasts do not tell you

1. The market reaction. A nowcast that forecasts the print correctly does not tell you how the market will react. The market reacts to the surprise relative to consensus, not to the realised print directly.
2. The framework implication. A 0.18 core PCE print supports both the dovish and the hawkish framings depending on the rest of the composition. The supercore decomposition is more informative than the headline number.
3. The next nowcast. Today's reading forecasts tomorrow's print. The next quarter's forecast restarts from scratch as the input panel refreshes; the convergence path begins again.

Related reading

• Today's analysis applies this directly: DXY consolidates. Claims confirmed labour-market firm.
• The inflation companion: A plain-English guide to PCE vs CPI
• The CPI components: A plain-English guide to CPI components
• The surprise index that translates nowcast errors into FX moves: A plain-English guide to economic surprise indices