Literature Review: Fed Staff on Computer Software and Accessories

One of the questions Riccardo received most frequently during his trip to London was: “What do you think about ‘Computer Software and Accessories’? Have you read the Fed note?”

For this reason, we provide a brief review of the FEDS Note, Measurement of “Computer Software and Accessories” Inflation, by Alessandro Barbarino, Anthony M. Diercks, and Stephen Miran.

The TL;DR is straightforward: the argument made by Barbarino and coauthors is, in our view, deeply flawed. The idea that current U.S. inflation (or most of it) can be explained by a single item—or by a narrow set of items—does not even pass the basic test we present below. Our recommendation is therefore not to take the note seriously as an explanation of current inflation dynamics.

What the note does

The FEDS Note studies the recent surge in the PCE “Computer Software and Accessories” price index and asks whether it reflects genuine inflation or measurement problems. From November 2025 to March 2026, this category made an unusually large contribution to core PCE and especially core goods PCE inflation. The authors show that core PCE rose at a 4.4 percent annualized rate over the latest four months, with software accounting for about two-thirds of a percentage point, despite being only 1.2 percent of the core PCE basket. The effect is larger for core goods, where the category accounts for 5.1 percent of the basket and explains about 2.8 percentage points of the 5.5 percent annualized increase.

Main results

The note argues that part of the surge is likely measurement error. The first issue is a category mismatch. BEA uses the CPI “Computer Software and Accessories” index to deflate the PCE category, but the CPI category includes blank media and flash drives, while the PCE category consists mainly of software publishers and data processing services and excludes blank media. The second issue is that software pricing has shifted toward subscriptions and AI-enhanced products, making matched-model price comparisons less reliable. The third issue is the lack of explicit quality adjustment for software, which matters if AI features raise product quality while measured prices rise.

Quantitatively, the category rose at a 73.1 percent annualized pace from November 2025 to March 2026, compared with a long-run historical decline of about 5.3 percent per year. Flash drives and blank media may account for about one-tenth to one-quarter of the measured surge, depending on the assumed weight. The authors estimate that a category mismatch adjustment alone would reduce the annualized software price increase from 73.1 percent to 52.0 percent. Adding a partial quality adjustment would lower it further to 22.0 percent, reducing core goods PCE inflation from 5.47 percent to 3.68 percent and core PCE inflation from 4.38 percent to 3.96 percent. Under a more aggressive quality pass-through assumption, core goods PCE would fall to 2.73 percent.

Our critical (technical) comment

The note is useful because it identifies a very specific and potentially large measurement problem in a category that has recently become macro-relevant despite its small weight. The argument is plausible. A CPI index that partially contains flash drives and blank media is a poor deflator for a PCE category that appears to contain no physical accessories. The quality-adjustment point is also important, especially if AI-enhanced software bundles more functionality into higher subscription prices.

The weakest part is the quality adjustment. Using frontier AI benchmark improvements as a proxy for consumer software quality is highly speculative. Benchmark gains do not map cleanly into consumer utility, and even less cleanly into the quality of the average software product in the PCE basket. The category mismatch adjustment is more convincing than the AI quality adjustment. The note is therefore strongest as a warning about potential mismeasurement (which btw, it is true for other items (i.e. medical prices) and has been known for decades..), not as a definitive corrected inflation estimate.

A more basic critique

Riccardo speaking. Are they really doing this again? Since 2021, I believe this is roughly the twentieth time someone has focused on a single item and concluded that “if we exclude this item from the basket, inflation is much lower or back at target.”

Once again, we would stress two important points. First, this procedure is inherently risky because different sectors reset prices at very different times. Not only but in a world dominated by negative supply shocks, sector-specific shocks also become more frequent. Second, if it were true that inflation were currently driven by only one item (or by a very limited set of items), then the distribution of the entire basket (the unweighted distribution) would itself be consistent with the target.

Regarding the second point, we report below (Figure 1) the cumulative density function (CDF) of the month-on-month SAAR distribution of core CPI items. As we have repeated after every CPI/PCE report for the past five years, the entire distribution is still not consistent with a 2 percent target (indeed, it has never been consistent with 2 percent at any point since 2021).

Figure 1. CDF of core CPI items cross-section MoM (saar)

As can be seen, the distribution contains substantially more mass above target than the 2016–2019 distribution (when core inflation was close to 2 percent), and it also exhibits a much fatter right tail (there are many more items printing at 20 percent+ MoM SAAR).

Translated into plain language: this is not merely a story about a single item, nor is it simply a measurement issue. Even today, the entire distribution remains inconsistent with a 2 percent inflation target and continues to display significantly greater dispersion than in the pre-Covid period.

Contact

Want something more tailored?

We provide tailored consulting on ad-hoc projects.