Inflation is a particularly difficult macroeconomic variable to forecast. Trained economists in central banks and leading financial institutions who have had a hard time in the last 18 months can testify.
In the forecasting literature, the random walk benchmark, a naive forecast which assumes inflation stays the same between two periods, is hard to beat by even the most sophisticated models.
One key reason is that idiosyncratic shocks explain most of the volatility, especially in the short term.
Said differently, in order to predict inflation, one needs to forecast a myriad of largely unrelated price moves.
It is a well-known inflation feature, even before the pandemic - see for instance this 2020 Fed paper, showing that most of the fluctuations in core PCE prices over the last decade have been idiosyncratic in nature.
Now, this should make the forecast data intensive, not necessarily impossible.
The trouble is, idiosyncratic shocks do not necessarily have determinants that nicely fit in structured time series. When they do, the forecaster would often discover the time series input afterwards.
We could give 100 examples here, from the March 2017 US telecom services price rise (Verizon), all the way to the product shortages that led to retail price increases during COVID-19: there is no easy way to monitor web data plans prices, insurance fees or maple syrup shortages in time series.
Except, through news stories.
Scanning for news relevant to the near-term inflation forecast is a resource intensive process, involving many trained economists analyzing news events.
That's where the NewsBot which detects "news relevant to near-term inflation forecasting" and the NIPI databases which aggregate that information in quantitative signals come in.
Inquiries: reach out