Tracking narratives to forecast food prices

Narratives cannot be safely ignored in macro, especially when it comes to tracking inflation. An illustration with US food prices.

Food prices

Among inflation components, food prices are often most difficult to forecast.

First, a discrepancy can arise between raw products market prices and processed food, the bulk of food consumption in many countries.

Second, price seasonality is difficult to account for, as fresh products come and go through the year.

Finally, food distribution channels are diverse. Pricing at national supermarket chains and local stores don't follow the same patterns. Web scraping prices from the former will not always be representative of the whole sector.

The below chart illustrates the first point. The S&P GSCI Agriculture prices, a common index for (mostly US) agricultural commodities, would have missed most of the volatility seen in US food prices around the summer 2021 as prices dropped in the early part of the summer:

Food prices: US CPI vs S&P GSCI

Notes: CPI m-o-m shown as of release date, to avoid look-ahead bias. S&P GSCI Agriculture Index shown as 1-month rolling percent change. Data up to 13/10/2021, including the Sept-21 CPI release.

COVID-19 has led to all sorts of supply disruptions, from supply chains early stages to the retail level. Raw agricultural prices alone could only capture part of the story. (Note that the reasoning can be extended beyond food: raw material prices are only showing a portion of ongoing price pressures impacting the manufactured goods sector.)

Over the summer, the volume of food-related news stories has materially increased in the US. There have been reports about re-emerging shortages in grocery stores on a daily basis and anecdotal evidence showing restaurants hiking prices as food and staff costs increase, for instance. Unsurprisingly, these shocks found their way into final consumer price releases.

Our News Inflation Pressures Indices (NIPI, see below notes) summarize the inflation news flow by telling how positive or negative the news have been (as in, positive or negative for near-term inflation). Over the recent period, the specific US Food NIPI would have made a useful contribution to help track food price volatility, alongside agriculture prices and others:

Food prices: US CPI vs NIPI

Notes: CPI m-o-m shown as of release date, to avoid look-ahead bias. Data up to 13/10/2021, including the Sept-21 CPI release.

Since the 20th September 2021, the NIPI Food index has rebounded, indicating further strength in US Food inflation, which was ultimately confirmed in the Sept-21 CPI release on the 13th of October.

Natural Language Processing tools can help keep track of narratives. These language models have considerably evolved in the last three years or so. They should now be seen as complementing traditional (structured) data in a critical way.


The News Inflation Pressure Index (NIPI) measures the directionality of the news flow. When the NIPI is at 50, positive and negative news balance each other. When the index rises above 50, the news flow signals positive inflation pressures are building up, and vice versa when the index drops below 50.

The data is available daily for the following countries: US, UK, Euro area, Germany, France, Italy, Spain, Canada, Australia, India, South Africa, Mexico and Argentina.

And the following sectors: headline inflation, core inflation, energy, food, telecom services and wages.

The NIPIs are based on state of the art Natural Language Processing applied to hundreds of thousands of news sources in six languages (Chinese, English, French, German, Italian and Spanish).

The Inflation NewsBot provides a daily selection of news relevant to the near-term inflation forecast which then enter in the NIPI calculation.

For any inquiries, including trial request, you can use our contact form.