Albacore — An Indicator of Core Inflation

Methodology

Timeliness is key. For monetary policy, this means having a measure that is as indicative as possible of future inflation conditions. Goulet Coulombe et al. (2024) introduce a simple machine learning algorithm, Assemblage Regression, which optimizes weights of inflation subcomponents such that the aggregate is a new core inflation series that succeeds in that goal. This is intentionally low-tech machine learning-wise (a generalized nonnegative ridge regression) so that the output can be interpreted as a well-defined macroeconomic aggregate.

Replacing the subcomponents matrix by the corresponding one of empirical order statistics, switches the algorithm to learn supervised trimmed inflation. This results from the regression now shrinking and selecting parts of the distribution of realized price changes at any point in time. Thus, we have two variants of adaptive learning-based core inflation (Albacore): one in the original components space and the other in ranks space.

Main Findings

First, if you shall trim, do so asymmetrically. Our supervised trimmed indicator puts a much higher weight than traditional indicators on the upper region of the price distribution.

Second, we find more harmony with traditional core definitions in components space, with, e.g., the exclusion of energy and low weight on food goods at longer horizons.

Third, we consider building core inflation measures specialized for upside and downside risks via a quantile regression extension and find strong deviations from classic indicators for high inflation risk.

The paper also considers small side ventures applying the novel assemblage regression in rank space, like (i) supervised temporal aggregation of headline inflation for the US and (ii) a geographical assembling of inflation from euro area member states.

Time-Series Estimates

Below we show monthly inflation estimates for the United States, the Euro Area, and Canada. The estimates are available for three types of time-aggregation: (i) a 3-Month average, (ii) a 6-Month average, (iii) and a Year-over-Year reading.

If you want to get an idea of what the Albacore Models project for average inflation over the next 12 months, select “3-Month Average” in the dropdown menu.