Nowcasting

Estimating something that has already happened but has not been measured yet. Headline statistics arrive weeks or months after the period they describe, so a model is used to infer the answer in the meantime from indicators that are published sooner. The word is borrowed from meteorology, where it describes forecasting the weather that is happening now, and the borrowing is exact: the target is the present rather than the future, and the uncertainty comes from measurement rather than from prediction.

The commonest target is gross domestic product, which is published quarterly and late. A nowcasting model maps the monthly and weekly series that feed into it, such as retail sales, industrial production, trade and survey indices, onto a running estimate for the quarter in progress, and revises that estimate every time one of those inputs is published. The Federal Reserve Bank of Atlanta runs a public model of this kind, other reserve banks and many private forecasters run comparable ones, and the same machinery is applied to inflation and to employment.

The tripwire is reading the output as data. It is the product of a model, and it moves for two quite different reasons: because the economy changed, or because a single volatile input landed and the model reweighted. Early in a quarter very few inputs have arrived, so the estimate rests on assumptions and swings widely; late in the quarter it usually settles, because most of what the official statistic will eventually count has already been published. A jump in a nowcast is therefore a fact about the model's inputs before it is a fact about the economy.

Whether nowcasts improve on simply reading the survey of professional forecasters is unsettled. Studies find the mechanical models competitive or better close to the release date, when most inputs are in, and weaker earlier, when a human forecaster's judgment about an unusual month is worth more than a model that has not seen one before. Central banks publish them with that caveat attached, and describe them as a summary of the data received rather than a forecast the institution stands behind.