Package {reviser}

Type:	Package
Title:	Analyzing Revisions in Real-Time Time Series Vintages
Version:	0.3.0
Description:	Analyzes revisions in real-time time series vintages. The package converts between wide revision triangles and tidy long vintages, extracts selected releases, computes revision series, visualizes vintage paths, and summarizes revision properties such as bias, dispersion, autocorrelation, and news-noise diagnostics. It also identifies efficient releases and estimates state-space models for revision nowcasting. Methods are based on Howrey (1978) <doi:10.2307/1924972>, Jacobs and Van Norden (2011) <doi:10.1016/j.jeconom.2010.04.010>, and Kishor and Koenig (2012) <doi:10.1198/jbes.2010.08169>.
License:	MIT + file LICENSE
Encoding:	UTF-8
LazyData:	true
Depends:	R (≥ 4.1.0)
RoxygenNote:	7.3.3
Imports:	dplyr, stats, tidyr, pillar, ggplot2, car, sandwich, systemfit, calculus, rlang, scales, tibble, lubridate, KFAS, numDeriv, withr
Suggests:	tsbox, testthat (≥ 3.0.0), knitr, rmarkdown, purrr
Config/testthat/edition:	3
VignetteBuilder:	knitr
URL:	https://docs.ropensci.org/reviser/, https://github.com/ropensci/reviser
BugReports:	https://github.com/ropensci/reviser/issues
NeedsCompilation:	no
Packaged:	2026-09-02 08:05:16 UTC; marcburri
Author:	Marc Burri [aut, cre, cph], Philipp Wegmueller [aut, cph]
Maintainer:	Marc Burri <marc.burri91@gmail.com>
Repository:	CRAN
Date/Publication:	2026-09-02 08:30:02 UTC

reviser: Analyzing Revisions in Real-Time Time Series Vintages

Description

reviser is an R package designed for working with time-series vintages data. The package provides tools to clean, visualize, and analyze time-series revisions. To learn more about reviser, start with the vignettes: browseVignettes(package = "reviser") The package will be actively developed and maintained.

Useful links:

• https://docs.ropensci.org/reviser/

• Report bugs at https://github.com/ropensci/reviser/issues

Author(s)

Maintainer: Marc Burri marc.burri91@gmail.com (ORCID) [copyright holder]

Authors:

• Philipp Wegmueller philipp.wegmueller@seco.admin.ch [copyright holder]

See Also

Useful links:

• https://docs.ropensci.org/reviser/

• https://github.com/ropensci/reviser

• Report bugs at https://github.com/ropensci/reviser/issues

Extract Parameter Estimates from a Revision Model

Description

Extract Parameter Estimates from a Revision Model

Usage

## S3 method for class 'revision_model'
coef(object, ...)

Arguments

Value

A named numeric vector of parameter estimates.

See Also

Other revision nowcasting: fitted.revision_model(), jvn_nowcast(), kk_nowcast(), logLik.revision_model(), nobs.revision_model(), plot.revision_model(), predict.revision_model(), print.revision_model(), residuals.revision_model(), revision_model, states(), summary.revision_model(), vcov.revision_model()

Examples

df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- na.omit(dplyr::select(df, -c("id", "pub_date")))
fit <- kk_nowcast(df, e = 1, model = "KK", method = "OLS")
coef(fit)

Diagnose Revision Quality

Description

Generic function to provide diagnostic summaries for revision analysis objects.

Usage

diagnose(object, ...)

Arguments

Value

Method-specific diagnostic output.

See Also

Other revision analysis: diagnose.revision_summary(), get_first_efficient_release(), get_revision_analysis(), print.lst_efficient(), print.revision_summary(), summary.lst_efficient(), summary.revision_summary()

Examples

# Example usage with revision analysis results
df <- dplyr::select(
  get_nth_release(
    na.omit(
      tsbox::ts_pc(
        dplyr::filter(reviser::gdp, id == "US")
      )
    ),
    n = 0:3
  ),
  -"pub_date"
)

final_release <- dplyr::select(
  get_nth_release(
    na.omit(
      tsbox::ts_pc(
        dplyr::filter(reviser::gdp, id == "US")
      )
    ),
    n = "latest"
  ),
  -"pub_date"
)

# Get revision analysis results
results <- get_revision_analysis(df, final_release, degree = 5)

# Diagnose revision quality
diagnose(results)

Diagnose Method for Revision Summary

Description

Provides a quick diagnostic summary of revision quality with color-coded pass/fail indicators for key metrics.

Usage

## S3 method for class 'revision_summary'
diagnose(object, alpha = 0.05, ...)

Arguments

Value

A tibble with diagnostic results.

See Also

Other revision analysis: diagnose(), get_first_efficient_release(), get_revision_analysis(), print.lst_efficient(), print.revision_summary(), summary.lst_efficient(), summary.revision_summary()

Examples

# Example usage with revision analysis results
df <- dplyr::select(
  get_nth_release(
    na.omit(
      tsbox::ts_pc(
        dplyr::filter(reviser::gdp, id == "US")
      )
    ),
    n = 0:3
  ),
  -"pub_date"
)

final_release <- dplyr::select(
  get_nth_release(
    na.omit(
      tsbox::ts_pc(
        dplyr::filter(reviser::gdp, id == "US")
      )
    ),
    n = "latest"
  ),
  -"pub_date"
)

# Get revision analysis results
results <- get_revision_analysis(df, final_release, degree = 5)

# Diagnose revision quality
diagnose(results)

Fitted Latent Values from a Revision Model

Description

Returns the smoothed estimate of the model's latent signal for the in-sample periods, i.e. the revision-adjusted series. The signal is the latent efficient value for a kk_model and the latent true value for a jvn_model.

Usage

## S3 method for class 'revision_model'
fitted(object, ...)

Arguments

Value

A tibble with columns time, estimate, lower and upper.

See Also

Other revision nowcasting: coef.revision_model(), jvn_nowcast(), kk_nowcast(), logLik.revision_model(), nobs.revision_model(), plot.revision_model(), predict.revision_model(), print.revision_model(), residuals.revision_model(), revision_model, states(), summary.revision_model(), vcov.revision_model()

Examples

df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- na.omit(dplyr::select(df, -c("id", "pub_date")))
fit <- kk_nowcast(df, e = 1, model = "KK", method = "MLE")
head(fitted(fit))

Vintages Data

Description

A collection of real-time datasets.

Usage

gdp

Format

A tbl_pubdate vintages object in the long layout: a tibble carrying the class attribute c("tbl_pubdate", "tbl_vintage", "tbl_df", "tbl", "data.frame"), with quarterly observations and 4 variables:

time

Date of the observation

pub_date

Publication date of the vintage

value

Numeric, real GDP (seasonally adjusted)

id

Country code

Because it is a vintages object rather than a plain tibble, the generics print(), summary() and plot() work on it directly. See validate_vintages() for the data contract and tbl_vintage for the class hierarchy.

Details

• GDP: Quarterly Vintages (Billions of real dollars, seasonally adjusted)

• Timeframe: Q1 1980 - Q4 2024

• Real-Time Vintages: Q4 2002 - Q4 2024

Sources

• All the data is from the realtime database of Indergand and Leist (2014). Countries:

• CHE:

  • Switzerland

  • Source: SECO

• US:

  • United States

  • Sources: FRED, OECD

• EA:

  • Euro Area

  • Sources: Eurostat, OECD

• JP:

  • Japan

  • Sources: Cabinet Office (Japan), OECD

References

Indergand, R., Leist, S. A Real-Time Data Set for Switzerland. Swiss J Economics Statistics 150, 331–352 (2014). doi:10.1007/BF03399410

See Also

validate_vintages(), tbl_vintage

Examples

# Load gdp dataset
data(gdp)
head(gdp)

# It is a vintages object, so the generics dispatch on it directly.
class(gdp)
summary(gdp)

Calculate the Number of Days Between Period End and First Release

Description

Computes the number of days between the publication date (pub_date) of a release and the time period (time) end date for each record in the dataset.

Usage

get_days_to_release(df)

Arguments

Details

The function calculates the difference between pub_date and time for each row in the dataset. The result is expressed as the number of days between the release publication date and the corresponding time period end. If the dataset is in wide format, it will first be transformed into long format using vintages_long.

Value

A data frame with an additional column days_to_release representing the number of days between the publication date (pub_date) and the time period (time) for each release.

See Also

Other revision utilities: get_first_release(), get_fixed_release(), get_latest_release(), get_nth_release(), get_releases_by_date(), get_revisions()

Examples

# Example data
df <- dplyr::filter(reviser::gdp, id == "US")

# Calculate days to release
df_with_days <- get_days_to_release(df)

Identify the First Efficient Release in Vintage Data

Description

Identifies the first release in a sequence of vintages that is "efficient" relative to the final release. A release is deemed efficient if it satisfies specific conditions of unbiasedness and efficiency, tested using a Mincer-Zarnowitz type linear regression and hypothesis testing.

Usage

get_first_efficient_release(
  df,
  final_release,
  significance = 0.05,
  test_all = FALSE,
  robust = TRUE
)

Arguments

Details

The function performs the following steps:

1. Validates inputs and ensures both df and final_release are in the correct format.

2. Iteratively tests each release for efficiency using a linear regression model of the form:

   final = \beta_0 + \beta_1 \cdot release_i + \epsilon

   The null hypothesis for efficiency is:

   • \beta_0 = 0 (no bias)

   • \beta_1 = 1 (efficiency) Uses heteroskedasticity and autocorrelation consistent (HAC) standard errors for robust hypothesis testing.

3. Stops testing when the first efficient release is found (unless test_all = TRUE).

If no efficient release is found, a warning is issued.

Value

A list of class lst_efficient with the following elements:

• e: The index of the first efficient release. (0 indexed)

• data: A long-format data frame containing the vintage data with the final release appended.

• models: A list of linear regression models fitted for each release.

• tests: A list of hypothesis test results for each release.

References

Aruoba, S. Boragan, "Revisions Are Not Well Behaved", Journal of Money, Credit and Banking, 40(2-3), 319-340, 2008.

See Also

Other revision analysis: diagnose(), diagnose.revision_summary(), get_revision_analysis(), print.lst_efficient(), print.revision_summary(), summary.lst_efficient(), summary.revision_summary()

Examples

# Example data
df <- get_nth_release(
  tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
  n = 0:3
)

final_release <- get_nth_release(
  tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
  n = 10
)

# Identify the first efficient release
result <- get_first_efficient_release(
  df,
  final_release,
  significance = 0.05,
  robust = FALSE
)

result <- get_first_efficient_release(df, final_release, significance = 0.05)

# Access the index of the first efficient release
result$e

Extract the First Data Release (Vintage)

Description

Filters the input dataset to return the earliest release (or vintage) for each time period.

Usage

get_first_release(df, diagonal = FALSE)

Arguments

Details

For each time period, the function identifies the release with the earliest publication date (pub_date). A new column release is added and labels all rows in the resulting data frame as release_0. If diagonal is set to TRUE, the function only returns the real first releases. That is historic values for which no vintages exist are not returned.

Value

A filtered data frame containing only the first release(s). The resulting data frame is assigned the class tbl_release to indicate its structure.

See Also

Other revision utilities: get_days_to_release(), get_fixed_release(), get_latest_release(), get_nth_release(), get_releases_by_date(), get_revisions()

Examples

# Example data
df <- dplyr::filter(reviser::gdp, id == "US")

# Get the first release for each time period
first_release <- get_first_release(df)

Extract Vintage Values from a Data Frame

Description

Some statistical agencies make a final revision of their data after a certain period of time in a give month in the year. This function extracts values from a given month or quarter a specified number of years after the initial release.

Usage

get_fixed_release(df, years, month = NULL, quarter = NULL)

Arguments

Value

A filtered data frame containing values matching the specified criteria.

See Also

Other revision utilities: get_days_to_release(), get_first_release(), get_latest_release(), get_nth_release(), get_releases_by_date(), get_revisions()

Examples

df <- dplyr::filter(reviser::gdp, id == "US")
dta <- get_fixed_release(df, month = "July", years = 3)
dta <- get_fixed_release(df, month = 7, years = 3)
dta <- get_fixed_release(df, quarter = 3, years = 3)

Extract the Latest Data Release (Vintage)

Description

Filters the input dataset to return the most recent release (or vintage) for each time period.

Usage

get_latest_release(df)

Arguments

Details

For each time period, the function identifies the release with the latest publication date (pub_date) and adds a column release that labels the release as release_N, where N is the release index (zero indexed).

Value

A filtered data frame containing only the most recent release(s). The resulting data frame is assigned the class tbl_release to indicate its structure.

See Also

Other revision utilities: get_days_to_release(), get_first_release(), get_fixed_release(), get_nth_release(), get_releases_by_date(), get_revisions()

Examples

# Example data
df <- dplyr::filter(reviser::gdp, id == "US")

# Get the latest release for each time period
latest_release <- get_latest_release(df)

Extract the Nth Data Release (Vintage)

Description

Filters the input dataset to return the Nth release (or vintage) of data for each time period. The function supports selecting the first, latest, or a specific numbered release.

Usage

get_nth_release(df, n = 0, diagonal = FALSE)

Arguments

Details

The behavior depends on the value of n:

• Non-negative integer: The function retrieves the Nth release for each time period (e.g., 0 = first release, 1 = second release, etc.).

• "first": Retrieves the first release for each time period (via get_first_release).

• "latest": Retrieves the most recent release for each time period (via get_latest_release).

Value

A filtered data frame containing only the specified release(s). The resulting data frame is assigned the class tbl_release to indicate its structure. If diagonal is set to TRUE, the function only returns the real first releases. That is historic values for which no vintages exist are not returned.

See Also

Other revision utilities: get_days_to_release(), get_first_release(), get_fixed_release(), get_latest_release(), get_releases_by_date(), get_revisions()

Examples

# Example data
df <- dplyr::filter(reviser::gdp, id == "US")

# Get the first release (n = 0)
first_release <- get_nth_release(df, n = 0)

# Get the latest release
latest_release <- get_nth_release(df, n = "latest")

# Get the second release (n = 1)
second_release <- get_nth_release(df, n = 1)

# Get the first and second release (n = 0:1)
releases <- get_nth_release(df, n = 0:1)

Get Data Releases for a Specific Date

Description

Filters the input dataset to return the releases corresponding to a specific time period (date).

Usage

get_releases_by_date(df, date)

Arguments

Details

This function filters the input data based on the specified date in the time column. The input dataset must have the pub_date and time columns, with time being the period to match against the given date. If the dataset is in wide format, it will first be transformed into long format using the helper function vintages_long.

Value

A data frame containing the releases for the specified date. The returned data frame will include the same structure as the input, filtered to only include rows matching the date in the time column.

See Also

Other revision utilities: get_days_to_release(), get_first_release(), get_fixed_release(), get_latest_release(), get_nth_release(), get_revisions()

Examples

# Example data
df <- dplyr::filter(reviser::gdp, id == "US")

# Get releases for a specific date
date <- as.Date("2020-04-01")
releases_on_date <- get_releases_by_date(df, date)

Revision Analysis Summary Statistics

Description

Calculates a comprehensive set of summary statistics and hypothesis tests for revisions between initial and final data releases.

Usage

get_revision_analysis(df, final_release, degree = 1, grouping_var = NULL)

Arguments

Details

This function performs a variety of statistical analyses to understand the nature of revisions between the initial and final data releases. The function:

• Checks the input data for consistency and transforms it as necessary.

• Merges the initial and final release datasets by their time variable and optional grouping variables (id or release).

• Computes summary statistics such as the mean, standard deviation, and range of the revisions.

• Performs hypothesis tests for bias, efficiency, and correlation using robust methods (e.g., Newey-West standard errors).

• Includes tests for seasonality, noise, and news effects.

Key tests include:

• Bias Tests: Tests for the presence of mean bias and regression bias.

• Autocorrelation and Seasonality: Tests for serial correlation and seasonal patterns in revisions.

• Theil's U Statistics: Measures predictive accuracy of the initial releases relative to the final values.

• Noise vs. News: Differentiates between unpredictable errors (noise) and systematic adjustments (news).

The function supports grouped calculations based on the presence of id or release columns in the input.

The following statistics and tests are calculated (See the vignette vignette("revision-analysis") for more details):

• N: The number of observations in the group.

• Frequency: The inferred data frequency (e.g., 12 for monthly or 4 for quarterly data).

• Bias (mean): The mean revision, testing whether revisions are systematically biased.

• Bias (p-value): p-value from a t-test evaluating the significance of the mean revision.

• Bias (robust p-value): Newey-West HAC robust p-value for the mean revision test.

• Minimum: The minimum revision in the group.

• Maximum: The maximum revision in the group.

• 10Q: The 10th percentile revision.

• Median: The median revision.

• 90Q: The 90th percentile revision.

• MAR: The mean absolute revision.

• Std. Dev.: The standard deviation of revisions, indicating their variability.

• Noise/Signal: The ratio of the standard deviation of revisions to the standard deviation of final values.

• Correlation: The Pearson correlation between revisions and initial values, testing the relationship.

• Correlation (p-value): p-value for the significance of the correlation.

• Autocorrelation (1st): The first-order autocorrelation of revisions, measuring persistence.

• Autocorrelation (1st p-value): p-value for the first-order autocorrelation test.

• Autocorrelation up to 1yr (Ljung-Box p-value): p-value for the Ljung-Box test for higher-order autocorrelation.

• Theil's U1: A normalized measure of forecast accuracy, comparing the root mean squared error (RMSE) of revisions to the RMSE of final and initial values.

• Theil's U2: Compares forecast changes to actual changes.

• Seasonality (Friedman p-value): p-value from the Friedman test for seasonality in revisions.

• News joint test (p-value): p-value for the joint news test.

• News test Intercept: The estimated intercept from the news test regression.

• News test Intercept (std.err): The standard error of the intercept in the news test regression.

• News test Intercept (p-value): p-value for the intercept in the news test regression.

• News test Coefficient: The estimated coefficient for the value in the news test regression.

• News test Coefficient (std.err): The standard error of the coefficient in the news test regression.

• News test Coefficient (p-value): p-value for the coefficient in the news test regression.

• Noise joint test (p-value): p-value for the joint noise test.

• Noise test Intercept: The estimated intercept from the noise test regression.

• Noise test Intercept (std.err): The standard error of the intercept in the noise test regression.

• Noise test Intercept (p-value): p-value for the intercept in the noise test regression.

• Noise test Coefficient: The estimated coefficient for the final_value in the noise test regression.

• Noise test Coefficient (std.err): The standard error of the coefficient in the noise test regression.

• Noise test Coefficient (p-value): p-value for the coefficient in the noise test regression.

• Fraction of correct sign: The fraction of correct sign changes in revisions.

• Fraction of correct growth rate change: The fraction of correct sign changes of growth rates in revisions.

Value

A data frame with one row per grouping (if applicable) and columns for summary statistics and test results. The resulting data frame is of class revision_summary.

See Also

Other revision analysis: diagnose(), diagnose.revision_summary(), get_first_efficient_release(), print.lst_efficient(), print.revision_summary(), summary.lst_efficient(), summary.revision_summary()

Examples

# Example usage:
df_small <- dplyr::filter(
  reviser::gdp,
  id == "US",
  time >= as.Date("2018-01-01")
)

df <- dplyr::select(
  get_nth_release(df_small, n = 0:2),
  -"pub_date"
)

final_release <- dplyr::select(
  get_nth_release(df_small, n = "latest"),
  -"pub_date"
)

results <- get_revision_analysis(
  df,
  final_release
)

Calculate Revisions in Vintage Data

Description

Computes revisions in vintage data based on specified reference points: a fixed reference date, the nth release, or a specified interval. This function allows users to analyze differences between data vintages across time.

Usage

get_revisions(df, interval = NULL, nth_release = NULL, ref_date = NULL)

Arguments

Details

The function supports three mutually exclusive methods for calculating revisions:

• Reference date (ref_date): Computes revisions relative to a fixed publication date.

• Interval (interval): Computes revisions relative to vintages published interval periods earlier.

• Nth release (nth_release): Computes revisions relative to the nth vintage release for each reference period.

If no method is explicitly specified, interval = 1 is used by default.

Input validation ensures that only one of ref_date, nth_release, or interval is specified.

Value

A data frame (tibble) of class tbl_revision, with the following columns:

• pub_date: The publication date of the vintage.

• time: The reference period (e.g., quarter or month).

• value: The calculated revision, i.e., the difference between the observed value and the reference value.

See Also

Other revision utilities: get_days_to_release(), get_first_release(), get_fixed_release(), get_latest_release(), get_nth_release(), get_releases_by_date()

Examples

# Example data
df <- dplyr::filter(reviser::gdp, id == "US")

# Calculate revisions using an interval of 1
revisions_interval <- get_revisions(df, interval = 1)

# Calculate revisions using a fixed reference date
revisions_date <- get_revisions(df, ref_date = as.Date("2023-02-01"))

# Calculate revisions relative to the nth release (2nd release)
revisions_nth <- get_revisions(df, nth_release = 1)

Jacobs-Van Norden Model for Data Revisions

Description

Estimate the Jacobs and Van Norden (2011) state-space model for real-time data revisions, allowing for news and noise components and optional spillovers.

Usage

jvn_nowcast(
  df,
  e,
  ar_order = 1,
  h = 0,
  include_news = TRUE,
  include_noise = TRUE,
  include_spillovers = FALSE,
  spillover_news = TRUE,
  spillover_noise = TRUE,
  method = "MLE",
  alpha = 0.05,
  standardize = FALSE,
  solver_options = list()
)

Arguments

Value

An object of class "jvn_model" with components:

states

A tibble of filtered and smoothed state estimates. If solver_options$return_states = FALSE, this is NULL.

jvn_model_mat

A list containing the state-space matrices Z, Tmat, R, H, and Q.

params

A data frame of parameter estimates and standard errors.

fit

The raw optimizer output returned by the selected numerical optimizer.

loglik

The maximized log-likelihood.

aic

Akaike information criterion.

bic

Bayesian information criterion.

convergence

Optimizer convergence code.

data

The input data after preprocessing.

scale

Scaling metadata returned by jvn_standardize() when standardize = TRUE; otherwise NULL.

se_method

The standard-error method used.

cov

Estimated covariance matrix of the parameter estimates, if available; otherwise NULL.

p0_regularized

Logical; whether computing the Kalman filter's stationary initial-state covariance at the converged estimate required numerical regularization, which can happen for a fit close to a non-stationary boundary. NULL when solver_options$return_states = FALSE, since the initial-state covariance is then not computed. summary() reports this when TRUE.

References

Jacobs, Jan P. A. M. and Van Norden, Simon (2011). Modeling data revisions: Measurement error and dynamics of "true" values. Journal of Econometrics.

See Also

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), kk_nowcast(), logLik.revision_model(), nobs.revision_model(), plot.revision_model(), predict.revision_model(), print.revision_model(), residuals.revision_model(), revision_model, states(), summary.revision_model(), vcov.revision_model()

Examples

gdp_growth <- dplyr::filter(
  tsbox::ts_pc(reviser::gdp),
  id == "EA",
  time >= min(pub_date),
  time <= as.Date("2020-01-01")
)
gdp_growth <- tidyr::drop_na(gdp_growth)
df <- get_nth_release(gdp_growth, n = 0:3)

result <- jvn_nowcast(
  df = df,
  e = 4,
  ar_order = 2,
  h = 0,
  include_news = TRUE,
  include_noise = TRUE
)
summary(result)

Generalized Kishor-Koenig Model for Nowcasting

Description

Implements a generalized Kishor-Koenig (KK) model for nowcasting and forecasting with state-space models, allowing for multiple vintages of data, efficient estimation, and Kalman filtering and smoothing.

Usage

kk_nowcast(
  df,
  e,
  h = 0,
  model = "Kishor-Koenig",
  method = "MLE",
  alpha = 0.05,
  solver_options = list()
)

Arguments

Details

The function supports multiple models, including the full Kishor-Koenig framework, Howrey's model, and a classical approach. It handles data preprocessing, estimation of system equations using Seemingly Unrelated Regressions (SUR), and application of the Kalman filter. This is the first openly available implementation of the Kishor-Koenig model (See the vignette vignette("nowcasting_revisions") for more details).

Value

A list with the following components:

states

A tibble containing filtered and smoothed state estimates. If solver_options$return_states = FALSE, this is NULL.

kk_model_mat

A list of KK model matrices, such as transition and observation matrices.

ss_model_mat

A list of state-space model matrices derived from the KK model.

model

The fitted KFAS::SSModel object. If solver_options$return_states = FALSE, this is NULL.

params

Estimated model parameters with standard errors.

fit

The raw fit object returned by the selected estimator.

loglik

Log-likelihood value for MLE fits; otherwise NULL.

aic

Akaike information criterion for MLE fits; otherwise NULL.

bic

Bayesian information criterion for MLE fits; otherwise NULL.

convergence

Convergence status.

e

The number of the efficient release (0-indexed).

data

The input data after preprocessing to wide format.

se_method

The standard-error method used by MLE; otherwise NULL.

cov

Estimated covariance matrix of the parameter estimates, if available; otherwise NULL.

References

Kishor, N. Kundan and Koenig, Evan F., "VAR Estimation and Forecasting When Data Are Subject to Revision", Journal of Business and Economic Statistics, 2012.

See Also

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), jvn_nowcast(), logLik.revision_model(), nobs.revision_model(), plot.revision_model(), predict.revision_model(), print.revision_model(), residuals.revision_model(), revision_model, states(), summary.revision_model(), vcov.revision_model()

Examples

# Example usage:
df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(
      dplyr::filter(reviser::gdp, id == "US")
    ),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- dplyr::select(df, -c("id", "pub_date"))
df <- na.omit(df)

e <- 1 # Number of efficient release
h <- 2 # Forecast horizon
result <- kk_nowcast(df, e, h = h, model = "Kishor-Koenig")

# Fitted models are reached through the extractor generics inherited from
# the parent class, not by indexing into the object.
coef(result)
logLik(result)

Extract the Log-Likelihood of a Revision Model

Description

The returned object carries the degrees of freedom and effective number of observations used by the model, so stats::AIC() and stats::BIC() reproduce the values reported by summary().

Usage

## S3 method for class 'revision_model'
logLik(object, ...)

Arguments

Value

An object of class logLik.

See Also

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), jvn_nowcast(), kk_nowcast(), nobs.revision_model(), plot.revision_model(), predict.revision_model(), print.revision_model(), residuals.revision_model(), revision_model, states(), summary.revision_model(), vcov.revision_model()

Examples

df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- na.omit(dplyr::select(df, -c("id", "pub_date")))
fit <- kk_nowcast(df, e = 1, model = "KK", method = "MLE")
logLik(fit)
AIC(fit)
BIC(fit)

Number of Observations Used to Fit a Revision Model

Description

Returns the effective number of observations behind the reported information criteria. Under the default ic_n = "Tp" this is the number of time periods times the number of releases (KK) or vintages (JVN) modeled.

Usage

## S3 method for class 'revision_model'
nobs(object, ...)

Arguments

Value

A single integer.

See Also

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), jvn_nowcast(), kk_nowcast(), logLik.revision_model(), plot.revision_model(), predict.revision_model(), print.revision_model(), residuals.revision_model(), revision_model, states(), summary.revision_model(), vcov.revision_model()

Examples

df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- na.omit(dplyr::select(df, -c("id", "pub_date")))
fit <- kk_nowcast(df, e = 1, model = "KK", method = "MLE")
nobs(fit)

Plot Revision Model Results

Description

Plot filtered or smoothed estimates for a selected state from a fitted revision model. Defined once for the parent class revision_model and inherited by kk_model and jvn_model objects alike; the state shown when state is not given is chosen by the concrete class.

Usage

## S3 method for class 'revision_model'
plot(x, state = NULL, type = "filtered", ...)

Arguments

Details

This method requires the state estimates to be available. A model fitted with solver_options$return_states = FALSE did not retain them, and plotting it fails with a message naming that option, in the same way states(), fitted(), residuals() and predict() do.

Value

A ggplot2 object.

See Also

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), jvn_nowcast(), kk_nowcast(), logLik.revision_model(), nobs.revision_model(), predict.revision_model(), print.revision_model(), residuals.revision_model(), revision_model, states(), summary.revision_model(), vcov.revision_model()

Examples

df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(
      dplyr::filter(reviser::gdp, id == "US")
    ),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- dplyr::select(df, -c("id", "pub_date"))
df <- na.omit(df)

e <- 1 # Number of efficient release
h <- 2 # Forecast horizon
result <- kk_nowcast(df, e, h = h, model = "Kishor-Koenig")

plot(result)

Plot Method for Vintages Data

Description

Plots a vintages object along the dimension implied by its class: publication date for a tbl_pubdate, release number for a tbl_release. Defined once for the parent class tbl_vintage and inherited by both. plot_vintages() remains the entry point when the dimension, title or plot type need to be set explicitly.

Usage

## S3 method for class 'tbl_vintage'
plot(x, ...)

Arguments

Value

A ggplot2 object.

See Also

Other revision graphs: plot_vintages(), theme_reviser()

Examples

df <- dplyr::filter(reviser::gdp, id == "US")
plot(df)

release_data <- get_nth_release(df, n = 0:5)
plot(release_data)

Plot Vintages Data

Description

A flexible function to visualize vintage data using various plot types such as line plots, point plots, bar plots, or boxplots. The function ensures that input data is validated and appropriately transformed before plotting.

Usage

plot_vintages(
  df,
  type = "line",
  dim_col = "pub_date",
  time_col = "time",
  title = "",
  subtitle = "",
  ylab = ""
)

Arguments

Details

The plot_vintages function is designed to handle data frames in both wide and long formats. It ensures that the provided data frame includes the necessary columns for plotting. If the dim_col column contains more than 30 unique values, only the most recent 30 are plotted. Additionally, the function supports custom themes and color scales using scale_color_reviser, scale_fill_reviser, and theme_reviser.

The function raises an error if:

• The type argument is not one of "line", "point", "bar", or "boxplot".

• The specified dim_col is not a column in df.

• title, subtitle, or ylab are not character strings.

Value

A ggplot2 plot object representing the specified vintage data visualization.

See Also

theme_reviser(), scale_color_reviser(), scale_fill_reviser()

Other revision graphs: plot.tbl_vintage(), theme_reviser()

Examples

# Example data
df <- data.frame(
  time = rep(seq.Date(from = as.Date("2022-01-01"),
  by = "month", length.out = 12), 3),
  value = runif(36, 50, 100),
  pub_date = rep(c("2022-01-05", "2022-02-07", "2022-03-03"), each = 12)
)

# Line plot
plot_vintages(
  df,
  type = "line",
  dim_col = "pub_date",
  title = "Line plot",
  subtitle = "Randomly generated data"
  )

# Point plot
plot_vintages(
  df,
  type = "point",
  dim_col = "pub_date",
  title = "Scatter plot",
  subtitle = "Randomly generated data"
  )

# Bar plot
plot_vintages(
  df,
  type = "bar",
  dim_col = "pub_date",
  title = "Bar plot",
  subtitle = "Randomly generated data"
  )

# Boxplot
plot_vintages(
  df,
  type = "boxplot",
  dim_col = "pub_date",
  title = "Boxplot",
  subtitle = "Randomly generated data"
  )

Forecasts from a Revision Model

Description

Returns the out-of-sample estimates of the model's latent signal produced by the forecast horizon h supplied to kk_nowcast() or jvn_nowcast(). The horizon is fixed at estimation time, so refit with a different h to change it.

Usage

## S3 method for class 'revision_model'
predict(object, ...)

Arguments

Value

A tibble with columns time, estimate, lower and upper. Has zero rows when the model was fitted with h = 0.

See Also

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), jvn_nowcast(), kk_nowcast(), logLik.revision_model(), nobs.revision_model(), plot.revision_model(), print.revision_model(), residuals.revision_model(), revision_model, states(), summary.revision_model(), vcov.revision_model()

Examples

df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- na.omit(dplyr::select(df, -c("id", "pub_date")))
fit <- kk_nowcast(df, e = 1, h = 2, model = "KK", method = "MLE")
predict(fit)

Print Method for Efficient Release Results

Description

Print Method for Efficient Release Results

Usage

## S3 method for class 'lst_efficient'
print(x, ...)

Arguments

Value

The function returns the input x invisibly.

See Also

Other revision analysis: diagnose(), diagnose.revision_summary(), get_first_efficient_release(), get_revision_analysis(), print.revision_summary(), summary.lst_efficient(), summary.revision_summary()

Examples

df <- get_nth_release(
  tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
  n = 0:3
)

final_release <- get_nth_release(
  tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
  n = 10
)

result <- get_first_efficient_release(df, final_release, significance = 0.05)
print(result)

Print Method for Revision Models

Description

Default print method for objects inheriting from revision_model. Dispatches to summary.revision_model() for a consistent console display.

Usage

## S3 method for class 'revision_model'
print(x, ...)

Arguments

Value

The input x, invisibly.

See Also

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), jvn_nowcast(), kk_nowcast(), logLik.revision_model(), nobs.revision_model(), plot.revision_model(), predict.revision_model(), residuals.revision_model(), revision_model, states(), summary.revision_model(), vcov.revision_model()

Examples

df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- na.omit(dplyr::select(df, -c("id", "pub_date")))
fit <- kk_nowcast(df, e = 1, h = 2, model = "Kishor-Koenig", method = "MLE")
fit

Print Method for Revision Summary

Description

Print Method for Revision Summary

Usage

## S3 method for class 'revision_summary'
print(x, interpretation = TRUE, digits = 3, ...)

Arguments

Value

The function returns the input x invisibly.

See Also

Other revision analysis: diagnose(), diagnose.revision_summary(), get_first_efficient_release(), get_revision_analysis(), print.lst_efficient(), summary.lst_efficient(), summary.revision_summary()

Examples

df <- dplyr::select(
  get_nth_release(
    na.omit(
      tsbox::ts_pc(
        dplyr::filter(reviser::gdp, id == "US")
      )
    ),
    n = 0:3
  ),
  -"pub_date"
)

final_release <- dplyr::select(
  get_nth_release(
    na.omit(
      tsbox::ts_pc(
        dplyr::filter(reviser::gdp, id == "US")
      )
    ),
    n = "latest"
  ),
  -"pub_date"
)

results <- get_revision_analysis(df, final_release, degree = 1)
print(results)

# Print without interpretation
print(results, interpretation = FALSE)

Print Method for Vintages Data

Description

Print method for objects inheriting from tbl_vintage. Delegates to the tibble print method, which calls tbl_sum.tbl_vintage() to generate the custom header.

Usage

## S3 method for class 'tbl_vintage'
print(x, ...)

Arguments

Value

The input x is returned invisibly.

See Also

Other helpers: summary.tbl_vintage(), tbl_sum.tbl_vintage(), tbl_vintage, validate_vintages(), vintages_long(), vintages_wide()

Examples

df <- dplyr::filter(reviser::gdp, id == "US")
release_data <- get_nth_release(df, n = 0:3)
print(release_data)

Residuals of a Revision Model

Description

Difference between the observed target release and the smoothed estimate of the model's latent signal. These are measurement residuals of the release the model treats as its target – the efficient release for a kk_model, the most mature release included in the estimation for a jvn_model – not one-step-ahead prediction errors.

Usage

## S3 method for class 'revision_model'
residuals(object, ...)

Arguments

Value

A tibble with columns time and residual.

See Also

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), jvn_nowcast(), kk_nowcast(), logLik.revision_model(), nobs.revision_model(), plot.revision_model(), predict.revision_model(), print.revision_model(), revision_model, states(), summary.revision_model(), vcov.revision_model()

Examples

df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- na.omit(dplyr::select(df, -c("id", "pub_date")))
fit <- kk_nowcast(df, e = 1, model = "KK", method = "MLE")
head(residuals(fit))

Fitted Revision Models

Description

reviser represents every fitted revision-nowcasting model as an S3 object that inherits from the common parent class revision_model. The two concrete classes are kk_nowcast(), which returns a kk_model, and jvn_nowcast(), which returns a jvn_model; both carry c("<family>_model", "revision_model", "list") as their class attribute.

The parent class holds everything the two families share. The standard extractor generics coef(), vcov(), logLik(), nobs(), fitted(), residuals(), predict() and the reviser generic states(), together with print(), summary() and plot(), are defined once for revision_model and inherited by both families. Only the handful of behaviors that genuinely differ between the families are dispatched separately, through the internal generics model_family(), spec_lines(), signal_state(), target_column() and default_plot_state().

A fitted object is a list with at least the components params (a data frame with columns Parameter, Estimate and Std.Error), states (a long tibble of state estimates, or NULL when the model was fitted with return_states = FALSE), loglik, n_param, n_ic, cov and data. A new model family becomes a full citizen of this system by returning an object with those components, prepending "revision_model" to its class attribute, and supplying methods for the five internal generics above.

Value

This topic documents a class rather than a function. kk_nowcast() returns a list of the components described above with class attribute c("kk_model", "revision_model", "list"), and jvn_nowcast() returns the same with "jvn_model" in place of "kk_model".

See Also

kk_nowcast(), jvn_nowcast(), states()

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), jvn_nowcast(), kk_nowcast(), logLik.revision_model(), nobs.revision_model(), plot.revision_model(), predict.revision_model(), print.revision_model(), residuals.revision_model(), states(), summary.revision_model(), vcov.revision_model()

Examples

df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- na.omit(dplyr::select(df, -c("id", "pub_date")))
fit <- kk_nowcast(df, e = 1, model = "KK", method = "OLS")

# The fitted object carries the shared parent class.
class(fit)
inherits(fit, "revision_model")

# The extractor generics are inherited from that parent.
head(coef(fit))
head(states(fit))

Extract the Latent State Estimates of a Revision Model

Description

Accessor for the state paths of a fitted revision-nowcasting model. Provides programmatic access to the estimated states instead of reaching into the object with fit$states. The method is defined once for the parent class revision_model and is inherited by kk_model and jvn_model objects alike.

Usage

states(object, ...)

## S3 method for class 'revision_model'
states(object, filter = c("smoothed", "filtered", "all"), state = NULL, ...)

Arguments

Value

A tibble with columns time, state, estimate, lower, upper, filter and sample.

See Also

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), jvn_nowcast(), kk_nowcast(), logLik.revision_model(), nobs.revision_model(), plot.revision_model(), predict.revision_model(), print.revision_model(), residuals.revision_model(), revision_model, summary.revision_model(), vcov.revision_model()

Examples

gdp_growth <- dplyr::filter(
  tsbox::ts_pc(reviser::gdp),
  id == "EA",
  time >= min(pub_date),
  time <= as.Date("2020-01-01")
)
gdp_growth <- tidyr::drop_na(gdp_growth)
df <- get_nth_release(gdp_growth, n = 0:3)

fit <- jvn_nowcast(df = df, e = 4, ar_order = 2, include_noise = FALSE)
head(states(fit))
head(states(fit, filter = "filtered", state = "true_lag_0"))

Summary of Efficient Release Models

Description

Provides a detailed summary of the regression model and hypothesis test for the first efficient release identified by the get_first_efficient_release() function.

Usage

## S3 method for class 'lst_efficient'
summary(object, ...)

Arguments

Details

This function prints the following information:

• The index of the first efficient release.

• A summary of the regression model fitted for the efficient release, which includes coefficients, R-squared values, and other relevant statistics.

• The hypothesis test results for the efficient release, showing the test statistic and p-value for the null hypothesis of unbiasedness and efficiency.

The function assumes the object includes:

• e: The index of the first efficient release (0-based).

• models: A list of linear regression models for each release.

• tests: A list of hypothesis test results corresponding to each release.

Value

Returns a tibble with the following columns:

• id: The identifier of the time series (if present in input data).

• e: The index of the first efficient release.

• alpha: The intercept coefficient of the regression model.

• beta: The coefficient of the slope.

• p-value: The p-value for the joint hypothesis (alpha = 0 and beta = 1).

• n_tested: The number of releases tested.

See Also

Other revision analysis: diagnose(), diagnose.revision_summary(), get_first_efficient_release(), get_revision_analysis(), print.lst_efficient(), print.revision_summary(), summary.revision_summary()

Examples

# Example usage
df <- get_nth_release(
  tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
  n = 1:4
)

final_release <- get_nth_release(
  tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
  n = 10
)

# Identify the first efficient release
result <- get_first_efficient_release(df, final_release, significance = 0.05)
summary(result)

Summary Method for Revision Models

Description

Computes and displays a summary of a fitted revision model, including the estimated specification, convergence status, information criteria, and parameter estimates. Defined once for the parent class revision_model; the family-specific header and specification block are supplied by the concrete class.

Usage

## S3 method for class 'revision_model'
summary(object, ...)

Arguments

Value

The input object, invisibly.

See Also

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), jvn_nowcast(), kk_nowcast(), logLik.revision_model(), nobs.revision_model(), plot.revision_model(), predict.revision_model(), print.revision_model(), residuals.revision_model(), revision_model, states(), vcov.revision_model()

Examples

df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- na.omit(dplyr::select(df, -c("id", "pub_date")))
fit <- kk_nowcast(df, e = 1, h = 2, model = "Kishor-Koenig", method = "MLE")
summary(fit)

Summary Method for Revision Summary

Description

Summary Method for Revision Summary

Usage

## S3 method for class 'revision_summary'
summary(object, interpretation = TRUE, ...)

Arguments

Value

The function returns the input object invisibly.

See Also

Other revision analysis: diagnose(), diagnose.revision_summary(), get_first_efficient_release(), get_revision_analysis(), print.lst_efficient(), print.revision_summary(), summary.lst_efficient()

Examples

# Example usage with revision analysis results
df <- dplyr::select(
  get_nth_release(
    na.omit(
      tsbox::ts_pc(
        dplyr::filter(reviser::gdp, id == "US")
      )
    ),
    n = 0:3
  ),
  -"pub_date"
)

final_release <- dplyr::select(
  get_nth_release(
    na.omit(
      tsbox::ts_pc(
        dplyr::filter(reviser::gdp, id == "US")
      )
    ),
    n = "latest"
  ),
  -"pub_date"
)

# Get revision analysis results
results <- get_revision_analysis(df, final_release, degree = 5)

# Summarize revision quality
summary(results)

Summary Method for Vintages Data

Description

Reports the layout, time coverage, number of vintages and missing-value count of a vintages object. Defined once for the parent class tbl_vintage; the block describing the vintages themselves is supplied by the concrete class.

Usage

## S3 method for class 'tbl_vintage'
summary(object, ...)

Arguments

Value

The input object, invisibly.

See Also

Other helpers: print.tbl_vintage(), tbl_sum.tbl_vintage(), tbl_vintage, validate_vintages(), vintages_long(), vintages_wide()

Examples

df <- dplyr::filter(reviser::gdp, id == "US")

# Long format
release_data <- get_nth_release(df, n = 0:3)
summary(release_data)

# Wide format
wide_release <- vintages_wide(release_data, names_from = "release")
summary(wide_release$US)

# Publication-date vintages
summary(vintages_wide(df)$US)

Tibble Summary for Vintages Data

Description

Provides the custom header shown when a vintages object is printed. This method is called automatically by pillar. It is defined once for the parent class tbl_vintage and inherited by tbl_pubdate and tbl_release objects alike, so that both report the same fields.

Usage

## S3 method for class 'tbl_vintage'
tbl_sum(x, ...)

Arguments

Value

A named character vector where names are labels and values are the corresponding information. The vector is used by pillar to format the tibble header. An object whose columns no longer match either documented layout falls back to the plain tibble header, so that a broken object can still be inspected; summary() reports the problem.

See Also

Other helpers: print.tbl_vintage(), summary.tbl_vintage(), tbl_vintage, validate_vintages(), vintages_long(), vintages_wide()

Examples

df <- dplyr::filter(reviser::gdp, id == "US")
release_data <- get_nth_release(df, n = 0:3)
pillar::tbl_sum(release_data)

Vintages Data Objects

Description

reviser represents real-time vintages as tibbles that carry an additional class recording which dimension indexes the vintages. Data indexed by publication date are of class tbl_pubdate, data indexed by release number are of class tbl_release, and both inherit from the common parent class tbl_vintage, ahead of the usual tibble classes. A fitted class attribute therefore reads c("tbl_pubdate", "tbl_vintage", "tbl_df", "tbl", "data.frame"). An object that carries both a pub_date and a release column is classed as both, with tbl_release taking precedence.

The parent class holds everything the two representations share. The print(), summary(), plot() and pillar::tbl_sum() methods are defined once for tbl_vintage and inherited by both, so that the two stay consistent with one another. Only the parts that genuinely depend on the indexing dimension are dispatched on the child classes, through the internal generics vintage_labels(), vintage_value_cols() and vintage_detail().

Both representations may be stored in either a long or a wide layout, and the methods below detect which and report accordingly. The layouts, and the columns each one requires, are specified under "Data contract" in validate_vintages(), which also checks an object against them. Use vintages_long() and vintages_wide() to convert between layouts.

Value

This topic documents a class rather than a function. The functions that build vintages data, such as get_nth_release(), vintages_long() and vintages_wide(), return tibbles whose class attribute is c("tbl_pubdate", "tbl_vintage", "tbl_df", "tbl", "data.frame") or the same with "tbl_release" in place of "tbl_pubdate".

See Also

validate_vintages(), vintages_long(), vintages_wide()

Other helpers: print.tbl_vintage(), summary.tbl_vintage(), tbl_sum.tbl_vintage(), validate_vintages(), vintages_long(), vintages_wide()

Examples

df <- dplyr::filter(reviser::gdp, id == "US")

# Release vintages carry the shared parent class.
releases <- get_nth_release(df, n = 0:3)
class(releases)
inherits(releases, "tbl_vintage")

# So do publication-date vintages, in either layout.
class(vintages_wide(df)$US)

# The print, summary and plot methods are inherited from the parent.
summary(releases)

Custom Visualization Theme and Color Scales for reviser

Description

These functions provide a custom visualization theme and color scales for use with ggplot2, inspired by the tsbox package.

Usage

theme_reviser(
  base_size = 12,
  legend.position = "bottom",
  legend.direction = "horizontal"
)

colors_reviser()

scale_color_reviser(...)

scale_fill_reviser(...)

Arguments

Details

• theme_reviser: Defines a minimal theme with custom adjustments for axis titles, plot titles, subtitles, captions, and legend positioning.

• colors_reviser: Provides a predefined set of colors, including a soft black, a palette suitable for colorblind readers, and additional colors for extended use.

• scale_color_reviser: A ggplot2 color scale that uses the custom colors_reviser palette.

• scale_fill_reviser: A ggplot2 fill scale that uses the custom colors_reviser palette.

Value

A customized ggplot2 theme, color palette, or scale.

See Also

Other revision graphs: plot.tbl_vintage(), plot_vintages()

Other revision graphs: plot.tbl_vintage(), plot_vintages()

Other revision graphs: plot.tbl_vintage(), plot_vintages()

Other revision graphs: plot.tbl_vintage(), plot_vintages()

Examples

library(ggplot2)
ggplot(mtcars, aes(x = wt, y = mpg, color = factor(cyl))) +
  geom_point(size = 3) +
  theme_reviser() +
  scale_color_reviser()

Vintages Data Classes and Their Validation

Description

reviser stores vintages in two S3 classes that sit on top of a tibble and record what the columns mean. Both are produced by the package's own constructors; you normally do not create them by hand.

tbl_pubdate

Vintages identified by publication date, as returned by vintages_wide(), vintages_long() and get_revisions().

tbl_release

Vintages identified by release number, as returned by get_first_release(), get_nth_release(), get_latest_release(), get_fixed_release() and get_releases_by_date().

validate_vintages() checks that an object conforms to the contract below. This is useful after manipulating a vintages object with external tools, which can leave the class attribute in place while breaking the assumptions the methods rely on.

Usage

validate_vintages(x)

Arguments

Value

x, invisibly, if it is valid. Otherwise an error describing the first problem found.

Data contract

Every object of either class has a time column of dates and may carry an optional id column identifying the series. Beyond that, each class has two permitted layouts:

long

A key column (pub_date for tbl_pubdate, release for tbl_release) together with a value column.

wide

One column per vintage. For tbl_pubdate the column names are publication dates in ⁠%Y-%m-%d⁠ form; for tbl_release they are release labels matching release or final.

Columns must be atomic and scalar-valued; list columns are not permitted. The two classes are not mutually exclusive: a long release table carries both a release and a pub_date column and holds both classes, with tbl_release taking precedence for method dispatch.

A valid object also inherits from the shared parent class tbl_vintage, which is where its methods live; validate_vintages() checks for that too. See tbl_vintage for the class hierarchy and the methods it provides.

Operations that drop the class

The vintages classes sit on top of a tibble, so the dplyr verbs (filter(), mutate(), select(), arrange(), slice()) and [ preserve them. A few functions rebuild the object from scratch and return a plain tibble instead; tidyr::drop_na() is the one most likely to be met in a vintages workflow. The data are unaffected, but plot(), summary() and the vintages print header no longer dispatch. Pass the result back through vintages_long() or vintages_wide(), or apply the operation before the release-extraction step, to get the class back.

See Also

Other helpers: print.tbl_vintage(), summary.tbl_vintage(), tbl_sum.tbl_vintage(), tbl_vintage, vintages_long(), vintages_wide()

Examples

df <- dplyr::filter(reviser::gdp, id == "US")

releases <- get_nth_release(df, n = 0:3)
validate_vintages(releases)

# A malformed time column is rejected
broken <- releases
broken$time <- as.character(broken$time)
broken$time[1] <- "not a date"
try(validate_vintages(broken))

# So is a class attribute that contradicts the columns
mislabelled <- vintages_wide(df)$US
class(mislabelled) <- c("tbl_release", class(mislabelled))
try(validate_vintages(mislabelled))

Extract the Parameter Covariance Matrix of a Revision Model

Description

Extract the Parameter Covariance Matrix of a Revision Model

Usage

## S3 method for class 'revision_model'
vcov(object, ...)

Arguments

Value

The estimated parameter covariance matrix.

See Also

Other revision nowcasting: coef.revision_model(), fitted.revision_model(), jvn_nowcast(), kk_nowcast(), logLik.revision_model(), nobs.revision_model(), plot.revision_model(), predict.revision_model(), print.revision_model(), residuals.revision_model(), revision_model, states(), summary.revision_model()

Examples

df <- get_nth_release(
  tsbox::ts_span(
    tsbox::ts_pc(dplyr::filter(reviser::gdp, id == "US")),
    start = "1980-01-01"
  ),
  n = 0:1
)
df <- na.omit(dplyr::select(df, -c("id", "pub_date")))
fit <- kk_nowcast(df, e = 1, model = "KK", method = "MLE")
vcov(fit)

Convert Vintages Data to Long Format

Description

Converts a vintages dataset from wide format to long format, optionally adding id if the input is a list of data frames. The long format contains one row per combination of time and names_to (e.g., pub_date or release), with values stored in a single value column.

Usage

vintages_long(df, names_to = "pub_date", keep_na = FALSE)

Arguments

Value

A long-format vintages object: a tibble carrying the tbl_pubdate or tbl_release class and their shared parent tbl_vintage. If the input is a list of wide-format data frames, the output is a single combined long-format object.

Long-format input is returned with the vintages class attached, which is also how to recover the class after an operation that dropped it (see the "Operations that drop the class" section of validate_vintages()). Input that is already a long-format vintages object warns, because the call is then a no-op.

See Also

Other helpers: print.tbl_vintage(), summary.tbl_vintage(), tbl_sum.tbl_vintage(), tbl_vintage, validate_vintages(), vintages_wide()

Examples

# Example wide-format data
long_data <- dplyr::filter(reviser::gdp, id == "US")

# Convert to wide format
wide_data <- vintages_wide(long_data)

# Example list of wide-format data frames
wide_list <- list(
  A = wide_data$US,
  B = wide_data$US
)

# Convert list to long format
long_data <- vintages_long(wide_list, names_to = "pub_date")

Convert Vintages Data to Wide Format

Description

Converts a vintages dataset from long format to wide format, optionally grouping by id if present. The wide format uses one column per unique value of the names_from parameter, with observation dates (time) as rows and values (value) as cell contents.

Usage

vintages_wide(df, names_from = "pub_date")

Arguments

Value

If an id column is present, the function returns a named list of wide-format data frames, one for each unique id. Otherwise, it returns a single wide-format data frame.

See Also

Other helpers: print.tbl_vintage(), summary.tbl_vintage(), tbl_sum.tbl_vintage(), tbl_vintage, validate_vintages(), vintages_long()

Examples

# Example wide-format data
long_data <- dplyr::filter(reviser::gdp, id == "US")

# Convert to wide format
wide_data <- vintages_wide(long_data)

# Example list of wide-format data frames
wide_list <- list(
  A = wide_data$US,
  B = wide_data$US
)

# Convert list to long format
long_data1 <- vintages_long(wide_data, names_to = "pub_date")
long_data2 <- vintages_long(wide_list, names_to = "pub_date")