Columbia University · Workshop · 2026–2027

Reading the economy in real time.

Most measures of economic activity are published after the period they describe and revised later. This workshop asks what can be learned before then from weekly and daily indicators. We develop the methods step by step, beginning with state-space models and ending with a real-time estimate of GDP.

observations, as released filtered estimate 95% band — the honest answer

Course schedule

01The Nowcasting Problem · State Space &Amp; the Kalman FilterP0 — the sealed envelope

02Dynamic Factor Models · Weekly Economic TrackersP1 — a weekly tracker from FRED

03Seasonal Adjustment at High FrequencyP2 — hourly electricity load

04Text as Data · LLM Sentiment IndicatorsP3 — a daily sentiment index

05ML for Time Series · Foundation ModelsP4 — the foundation-model horse race

06Nonlinear Tools · Markov Switching, MIDAS, Growth-at-Risk—

07Capstone Sprint · Assembling the Dashboardcapstone build

08Demo Day · Nowcasts vs. the Proscapstone ships

The capstone

Your nowcast vs. the Federal Reserve's.

Every student ships a live dashboard — fresh data on a schedule, a current-quarter GDP nowcast with honest bands, every revision attributed to the release that caused it. Scored against GDPNow on the same axes.

+1.8%

reference build · ±1.1 · vs GDPNow +2.0

About

I put this together for other students who want to work on forecasting and nowcasting. These methods usually get picked up in fragments, a paper here and a replication file there, which makes joint projects slow to start and hard to trust.

Eight sessions, each pairing one method with the code that runs it on real data. By the end you should be able to build a nowcast, defend its error bars, and read someone else’s model without a long preamble. Python and a first course in econometrics are enough to begin.