Implements the Jacobs & Van Norden (2011) state-space model for analyzing data revisions, allowing for news and noise components in measurement errors.
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,
solver_options = list()
)Arguments
- df
A matrix or data frame where each column represents a different vintage estimate for the same time period. Rows are time periods.
- e
An integer indicating the number of data vintages to include in the model. Must be greater than 0.
- ar_order
Integer specifying the AR order for true values (default = 2).
- h
Integer specifying the forecast horizon (default = 0).
- include_news
Logical, whether to include news component in measurement error (default = TRUE).
- include_noise
Logical, whether to include noise component in measurement error (default = TRUE).
- include_spillovers
Logical, whether to include spillover effects (default = FALSE).
- spillover_news
Logical, whether spillovers apply to news component (default = TRUE).
- spillover_noise
Logical, whether spillovers apply to noise component (default = TRUE).
- method
A string specifying the estimation method to use. Options are Maximum likelihood ("MLE") for now.
- alpha
Significance level for confidence intervals (default = 0.05).
- solver_options
List of options for the optimizer:
trace: Integer controlling output level (default = 0)
maxiter: Maximum iterations (default = 1000)
startvals: Named vector of starting values (optional)
transform_se: T/F whether standard errors should be constrained to be positive in optimization. . - method: String specifying optimization method (default = "L-BFGS-B").
se_method: Method for standard error calculation (default = "hessian")
n_starts: Number of random starting points for multi-start optimization
Value
A list of class 'jvn_model' with components:
- filtered_true
Filtered estimates of true values
- smoothed_true
Smoothed estimates of true values
- forecast_true
Forecasted true values (if h > 0)
- forecast_vintages
Forecasted vintage values (if h > 0)
- jvn_model_mat
Model matrices (Z, T, R, H, Q)
- params
Estimated parameters with standard errors
- fit
Optimization results
- loglik
Log-likelihood value
- aic
Akaike Information Criterion
- bic
Bayesian Information Criterion
- data
Input data
References
Jacobs, Jan P.A.M. and Van Norden, Simon, "Modeling Data Revisions: Measurement Error and Dynamics of 'True' Values", Journal of Econometrics, 2011.
See also
Other revision nowcasting:
kk_nowcast(),
plot.jvn_model(),
plot.kk_model(),
print.jvn_model(),
summary.jvn_model()
Examples
gdp <- dplyr::filter(
tsbox::ts_pc(
reviser::gdp
), id %in% c("EA"),
time >= min(pub_date),
time <= as.Date("2020-01-01")
)
gdp <- tidyr::drop_na(gdp)
df <- get_nth_release(gdp, n = 0:4)
# Estimate model
result <- jvn_nowcast(
df = df,
e = 3,
ar_order = 2,
h = 4,
include_news = TRUE,
include_noise = TRUE
)
#> Warning: 2 parameter(s) have problematic standard errors (2 NaN)