Monetary policy shock and stance series for the United States, United Kingdom, and Australia, bundled in R.

When the Fed, Bank of England, or ECB announces a rate decision, part of the move is expected by markets and part is not. The unexpected part is what economists call a monetary policy shock: an exogenous change in policy that lets empirical macroeconomists separate the causal effect of policy from policy's response to the economy. Every paper running an impulse response function (IRF), local projection (LP), vector autoregression (VAR), or event study needs a shock series as its identifying input. Shadow-rate measures play the same role during zero-lower-bound episodes, when the policy rate itself is pinned at its floor.

Those series live today as XLS and CSV files scattered across individual authors' personal pages, Federal Reserve research data pages, and GitHub mirrors, with inconsistent column names, date formats, and update cadences. Over three decades the methodology for measuring these shocks has evolved: from narrative reading of FOMC minutes (Romer and Romer 2004), to high-frequency identification around announcement windows (Kuttner 2001; Gurkaynak, Sack, and Swanson 2005; Nakamura and Steinsson 2018), to sign-restricted decompositions of joint rate-and-stock moves (Jarocinski and Karadi 2020), to orthogonalisation against pre-announcement public data or the Fed's Greenbook forecasts (Bauer and Swanson 2023; Miranda-Agrippino and Ricco 2021). Shadow-rate term-structure models (Wu and Xia 2016) address the zero-lower-bound regime separately. Each technique produces a subtly different series, and modern empirical papers increasingly report impulse responses under several as a robustness check.

mpshock centralises these series as tidy data frames with provenance, loaded in one call, no network required.

mpshock is for empirical macroeconomists, central-bank research staff, rates strategists, and PhD students. Typical uses are supplying the identifying input to a local projection, VAR, proxy-SVAR, or event study of asset-price responses to FOMC, MPC, or RBA announcements. If your specification needs an exogenous monetary policy shock, this package supplies it.

install.packages("mpshock")

# Or install the development version from GitHub
# install.packages("devtools")
devtools::install_github("charlescoverdale/mpshock")

Every call into mpshock follows the same pattern: ask for a series by name, then work with it as a tidy data frame. Here's a guided tour through the main functions. Output shown with #> prefix is what you will see in your console.

# Attach the package
library(mpshock)

# mp_list() returns a metadata table with one row per bundled series
# (we select a subset of columns here for a readable display)
mp_list()[, c("series", "author", "type", "start", "end", "n")]
#>                    series                                           author        type      start        end   n
#> 1      nakamura_steinsson                    Nakamura and Steinsson (2018)       shock 2000-02-01 2014-03-01 170
#> 2           bauer_swanson                         Bauer and Swanson (2023)       shock 1988-02-01 2023-12-01 431
#> 3              gss_target Gurkaynak, Sack, Swanson (2005) / Swanson (2021)       shock 1991-07-01 2015-10-01 292
#> 4                gss_path Gurkaynak, Sack, Swanson (2005) / Swanson (2021)       shock 1991-07-01 2015-10-01 292
#> 5    jarocinski_karadi_mp                     Jarocinski and Karadi (2020)       shock 1990-02-01 2024-01-01 404
#> 6   jarocinski_karadi_cbi                     Jarocinski and Karadi (2020) information 1990-02-01 2024-01-01 404
#> 7 miranda_agrippino_ricco               Miranda-Agrippino and Ricco (2021)       shock 1991-01-01 2019-06-01 342
#> 8                  wu_xia                                Wu and Xia (2016) shadow_rate 1960-01-01 2022-02-01 385

# Load the Nakamura-Steinsson policy news shock into the object `ns`
ns <- mp_shock("nakamura_steinsson")

# head() shows the first six rows plus a provenance header
head(ns)
#> <mp_shock> nakamura_steinsson [monthly, 6 obs, 2000-02-01 to 2000-07-01]
#>        date      shock
#>  2000-02-01  0.0215744
#>  2000-03-01 -0.0028683
#>  2000-04-01  0.0000000
#>  2000-05-01  0.0335943
#>  2000-06-01 -0.0061341
#>  2000-07-01  0.0000000

ns is an mp_shock object (inherits from data.frame). It has a date column set to the first day of each month and a shock column in percentage points. Months with no FOMC meeting are coded zero, matching the convention in Gertler-Karadi (2015) and the authors' maintained releases.

# mp_source() prints the paper, DOI, source URL, and method description
mp_source("nakamura_steinsson")
#> nakamura_steinsson
#>   Nakamura and Steinsson (2018)
#>   DOI: 10.1093/qje/qjy004
#>   URL: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/HZOXKN
#>   Policy news shock: first principal component of five interest-rate futures
#>   surprises in a 30-minute window around FOMC announcements, aggregated to
#>   monthly.

Always cite the underlying paper when you use a series in academic work. See Citation for the full list.

# Subset to the 2008 financial-crisis months using start and end arguments
crisis <- mp_shock("nakamura_steinsson",
                   start = "2008-01-01",
                   end   = "2008-06-30")
head(crisis)
#> <mp_shock> nakamura_steinsson [monthly, 6 obs, 2008-01-01 to 2008-06-01]
#>        date      shock
#>  2008-01-01 -0.0692382
#>  2008-02-01  0.0000000
#>  2008-03-01  0.2043770
#>  2008-04-01 -0.0755810
#>  2008-05-01  0.0000000
#>  2008-06-01 -0.0252113

The January and March 2008 rows capture the inter-meeting 75-basis-point cuts that stand out as the largest surprises of the GFC window.

# Aggregate the monthly series to quarterly by summing within each quarter
ns_q <- mp_to_quarterly(ns, method = "sum")

head(ns_q)
#> <mp_shock> nakamura_steinsson [quarterly, 6 obs, 2000-01-01 to 2001-04-01]
#>        date      shock
#>  2000-01-01  0.0187061
#>  2000-04-01  0.0274602
#>  2000-07-01  0.0126884
#>  2000-10-01  0.0537947
#>  2001-01-01 -0.1058595
#>  2001-04-01  0.0328807

method = "mean" (arithmetic mean within the quarter) and method = "end" (last month's value, used for stock-like series such as the Wu-Xia shadow rate) are also supported.

# mp_cumulate() adds a `shock_cum` column with the running cumulative sum
ns_cum <- mp_cumulate(ns)
head(ns_cum)
#> <mp_shock> nakamura_steinsson [monthly, 6 obs, 2000-02-01 to 2000-07-01]
#>        date      shock shock_cum
#>  2000-02-01  0.0215744 0.0215744
#>  2000-03-01 -0.0028683 0.0187061
#>  2000-04-01  0.0000000 0.0187061
#>  2000-05-01  0.0335943 0.0523004
#>  2000-06-01 -0.0061341 0.0461663
#>  2000-07-01  0.0000000 0.0461663

Pass window = 12L for a rolling 12-month sum instead.

Not every bundled series has a single shock column. The Wu-Xia shadow rate carries three data columns: the first-differenced shock for pipeline compatibility, the level shadow_rate, and the effective federal funds rate effr for reference.

# Load the Wu-Xia shadow rate over the first GFC zero-lower-bound episode
wx <- mp_shock("wu_xia", start = "2008-01-01", end = "2008-06-30")

head(wx)
#> <mp_shock> wu_xia [monthly, 6 obs, 2008-01-01 to 2008-06-01]
#>        date       shock shadow_rate effr
#>  2008-01-01 -1.04279824    2.655856 3.22
#>  2008-02-01 -0.51799019    2.137866 3.01
#>  2008-03-01 -0.26333728    1.874529 2.51
#>  2008-04-01 -0.06210523    1.812423 2.37
#>  2008-05-01  0.14401369    1.956437 1.98
#>  2008-06-01  0.09757498    2.054012 2.47

A common robustness check: run the same regression with different identification strategies and see whether the impulse responses line up. Here's how to line up two shock series on a common monthly grid.

# Load two alternative US monetary policy shocks over the same window
ns <- mp_shock("nakamura_steinsson", start = "2005-01-01", end = "2013-12-31")
bs <- mp_shock("bauer_swanson",      start = "2005-01-01", end = "2013-12-31")

# Merge by date; attach suffixes so the two shock columns can be told apart
panel <- merge(
  ns[, c("date", "shock")],       # NS shock series
  bs[, c("date", "shock")],       # BS orthogonalised surprise
  by = "date",
  suffixes = c("_ns", "_bs")
)

head(panel)

# Pearson correlation of the two shock series over the common sample
cor(panel$shock_ns, panel$shock_bs, use = "pairwise.complete.obs")

A correlation in the 0.4 to 0.7 range is typical: the series agree on the biggest FOMC surprises but diverge on the harder-to-classify meetings. Papers should report IRF results under at least two shock series for robustness (Brennan, Jacobson, Matthes, and Walker 2024, FEDS 2024-011).

The Wu-Xia series ends in February 2022 because the Atlanta Fed paused the update once policy rates normalised. The series remains the standard reference for the two zero-lower-bound episodes (2008-12 to 2015-12 and 2020-03 to 2022-02). The UKMPD is live-maintained by the Bank of England and is the flagship UK source for post-1997 work.

Users should be aware of a few caveats. Each per-series help file (?nakamura_steinsson, ?bauer_swanson, etc.) discusses these in more depth.

• Shock versus surprise. "Surprise" refers to the raw market reaction to an announcement (e.g. NS first principal component, raw Bauer-Swanson MPS); "shock" refers to the exogenous structural component after removing predictable public information (e.g. MPS_ORTH, Miranda-Agrippino-Ricco residual, Jarocinski-Karadi MP). mpshock uses shock as a column name for pipeline uniformity; consult each series' docs for what the underlying object is.
• Information-effect debate. The Nakamura-Steinsson (2018) "Fed information effect" interpretation is challenged by Bauer and Swanson (2023, AER) and Miranda-Agrippino and Ricco (2021). When estimating macro IRFs, MPS_ORTH (bauer_swanson) and the informationally-robust shock (miranda_agrippino_ricco) are the current methodological leaders; NS and JK-MP remain widely cited.
• Monthly aggregation. All event-study series are aggregated to calendar-month by summation; months with no FOMC meeting are coded zero. Bu, Rogers, and Wu (2021, JME) recommend coding no-meeting months as NA for proxy-SVAR/LP-IV to avoid downward-biased variance. mpshock does not apply this; apply post hoc if needed.
• Wu-Xia sensitivity. Shadow-rate estimates depend on the lower-bound choice, number of factors, and yield maturities (Krippner 2020, JMCB). Robustness to Krippner's alternative SSR is recommended for any result that relies on the shadow-rate level.

• Euro-area shocks: Jarocinski-Karadi ECB version, Altavilla et al. (2019) Euro-Area Monetary Policy Database.
• Romer and Romer (2004) US narrative shocks.
• Bolhuis, Das, and Yao (2024 IMF WP) cross-country panel of 29 central banks' HFI shocks.
• A vignette reproducing a Gertler-Karadi (2015) style impulse response using mpshock plus lpirfs.
• Brennan, Jacobson, Matthes, and Walker (2024) cross-series comparison helper.

Cite both the package and the underlying paper(s) for the series you use. Package citation:

citation("mpshock")

Series citations:

Original two-factor GSS: Gurkaynak, Sack, and Swanson (2005), International Journal of Central Banking 1(1): 55-93.

Report bugs or request additional series at GitHub Issues. Pull requests for new series are welcome; please include a licensing check in the PR description.

monetary policy, impulse response, local projection, high-frequency identification, FOMC, shocks, shadow rate, empirical macroeconomics.