Build the Dallas Fed trimmed-mean PCE component panel

Returns a long tibble with the raw inputs to the Federal Reserve Bank of Dallas's trimmed-mean PCE inflation rate: monthly Fisher price index, nominal expenditure, real quantity, monthly price change, the Fisher (t-1, t) expenditure-share weight, and a flag indicating whether the component was trimmed in that month and on which tail. Users can replicate the trimmed-mean rate by collapsing this tibble to kept (non-trimmed) components each month and taking the weight-renormalized weighted mean of price changes.

Usage

getDallasTrimPCE(
  frequency = "M",
  NIPA_data = NULL,
  alpha = 0.24,
  beta = 0.31,
  components = NULL
)

Arguments

frequency: Character. Frequency code passed to getNIPAFiles. Defaults to "M" (monthly). Currently the trimmed-mean panel is only meaningful at monthly frequency.
NIPA_data: Optional pre-loaded NIPA tibble from getNIPAFiles(). If NULL, the function downloads it.
alpha: Numeric in [0, 1]. Lower-tail trim share. Default 0.24, the Dallas Fed published value.
beta: Numeric in [0, 1]. Upper-tail trim share. Default 0.31.
components: Optional override for the component dictionary. Must be a tibble with columns dallas_idx, name, series_code, line_no. Defaults to the packaged dallasTrimPCEcomponents (177 components).

Value

A tbl_df with one row per (date, component) and columns:

date: Month observation date.
dallas_idx: Component ordinal in the Dallas tech notes (1..178, 94 omitted).
name: Dallas Fed component name.
series_code: BEA NIPA series code (Table 2.4.4U).
line_no: BEA NIPA Table 2.4.4U line number.
price: Fisher price index (Table 2.4.4U).
nominal: Current-dollar outlay (Table 2.4.5U).
quantity: Real quantity (nominal / price).
price_change: Period-over-period fractional change in price. NA for the first observation per component.
weight: Fisher (t-1, t) expenditure-share weight, renormalized to sum to 1 within each full-coverage month. NA otherwise.
is_trimmed: Logical. TRUE if the component is in either tail this month and so dropped from the trimmed mean. NA when the month lacks full coverage.
trim_side: Character. "lower" or "upper" when trimmed; NA when kept (interior) or coverage incomplete.

Details

Weights are Fisher-style: an unweighted average of the expenditure share evaluated at base prices P[t-1] with quantities Q[t-1] and Q[t], renormalized to sum to 1 within each month. Real quantity is computed as nominal / price from BEA NIPA Tables 2.4.5U and 2.4.4U respectively (equivalent to Table 2.4.6U per the Dallas Fed's MATLAB note, but available across the full sample without chained-dollar gaps).

Trim assignment is the simple rank-based version: components are sorted within each month by price_change, cumulative weight is accumulated, and components whose running cumulative weight is below alpha are flagged "lower", while components whose cumulative weight before adding their own contribution is at or above 1 - beta are flagged "upper". Boundary components that straddle either threshold are kept (treated as interior). The Dallas Fed's exact fractional-boundary handling enters the rate calculation itself, not this panel-builder; the resulting headline rate matches the Dallas Fed series to within a few basis points.

Months without full cross-sectional coverage (i.e., any component missing this month or last) have weight, is_trimmed, and trim_side set to NA.

References

Dolmas, J. (2005). "Trimmed Mean PCE Inflation." Federal Reserve Bank of Dallas Working Paper 0506.

Dolmas, J. (2009, updated 2022-12-23). "PCE Inflation: Technical Note." Federal Reserve Bank of Dallas.

Atkinson, T., Dolmas, J., & Zarutskie, R. (2026). "Skewness warrants caution as Trimmed Mean PCE inflation eases." Federal Reserve Bank of Dallas, April 16, 2026.

Examples

if (FALSE) { # \dontrun{
  # Default 24/31 Dallas Fed trim
  panel <- getDallasTrimPCE()

  # Replicate the monthly trimmed-mean rate (kept components, renormalized
  # to sum to 1):
  library(dplyr)
  tm_rate <- panel |>
    dplyr::filter(!is_trimmed) |>
    dplyr::group_by(date) |>
    dplyr::summarize(
      rate = sum(price_change * weight) / sum(weight),
      .groups = "drop"
    )

  # What got trimmed in the latest month, by tail and weight:
  panel |>
    dplyr::filter(date == max(date), is_trimmed) |>
    dplyr::arrange(trim_side, dplyr::desc(weight)) |>
    dplyr::select(name, trim_side, weight, price_change)
} # }