We are excited to share the US version of our News Inflation Pressure Index, ahead of the general release for all regions and themes on 25 November 2020.
The US NIPI relies on the news selected by our NewsBot: a unique model to track inflation news in real-time. The results are very encouraging, even better than our expectations.
As the chart below illustrates, the NIPI has tracked pretty much all turning points in headline CPI inflation in recent years. As inflation forecasters, we can tell this result is as good as alternative macro data get to help forecast official data.
You will find more information about the NIPI below the chart. The language models we have built to get the data are described here.
Further technical details
Since the CPI is a monthly series and our NIPI is available daily, we report both series at the time of release at a daily frequency in the above chart.
The NIPI is reported without lag (it is available almost instantaneously, with just a few hours of delay following the news publication). The CPI is shown as of release date (typically 10 to 14 days after the end of the month for which the data is reported).
Headline CPI inflation is the 3-month-over-3-month rate of seasonally adjusted headline CPI inflation. For illustration purposes, we have found this metric to offer a good balance between near-term dynamic and trend.
The News Inflation Pressure Index
The NIPI is a balance of positive and negative news as detected by our models. It is normalised based on historical news volumes so that 50 corresponds to an equal volume of positive and negative news, while a value close to 100 (or 0) would indicate a one-way maximum historical volume of positive (negative) news. The NIPI on a given date aggregates the news of the previous 30 days up to that date.
The NIPI covers all inflation news but wages. Indeed, our analysis suggested that minimum wage headlines in particular were not necessarily relevant for the near-term inflation forecast. We let our models detect those wage news and remove them.