bootstrap_irf_bands: Recursive residual bootstrap confidence bands for IRFs

In pvarife: Panel VAR Models with Interactive Fixed Effects

Recursive residual bootstrap confidence bands for IRFs

Description

Constructs confidence bands for structural impulse responses via a recursive residual bootstrap. For each bootstrap replicate, the idiosyncratic residuals are resampled (by time index, preserving the contemporaneous correlation across variables), a new panel is generated recursively from the estimated VAR dynamics and the (fixed) estimated common component, the full pvarife model is re-estimated, and IRFs are computed. This is computationally expensive but does not rely on asymptotic approximations.

Usage

bootstrap_irf_bands(
  fit,
  n_periods,
  shock = 1L,
  diff_vars = integer(0),
  identification = c("short_run", "long_run"),
  n_boot = 200L,
  level = 0.95,
  seed = NULL
)

Arguments

fit	An object of class "pvarife_result".
n_periods	Positive integer. Number of IRF horizons.
shock	Positive integer. Index of the structural shock (default 1).
diff_vars	Integer vector. Variables to cumulate (default none).
identification	Character. "short_run" (default) or "long_run". Passed to compute_irf for each bootstrap replicate.
n_boot	Positive integer. Number of bootstrap replicates (default 200).
level	Numeric in (0, 1). Confidence level (default 0.95).
seed	Optional integer seed for reproducibility.

Details

Each bootstrap panel is generated as

y^*_{i,t} = c + \sum_{l=1}^{\ell} \Theta_l y^*_{i,t-l} + \hat F_t \hat\lambda_i + e^*_{i,t},

where (c, \Theta_l) are the estimated coefficients, the common component \hat F_t \hat\lambda_i is held fixed at its estimate, the first \ell periods are initialised at the observed data y^*_{i,t} = y_{i,t}, and e^*_{i,t} are residuals resampled with replacement (whole time rows, so the cross-variable correlation is kept). Because the path is generated recursively, the bootstrap correctly propagates the VAR dynamics — unlike a fixed-design scheme that reuses the original lags.

Scope of the bands. Only the idiosyncratic errors are resampled; the common component \hat F_t \hat\lambda_i and the reduced-form covariance structure are held at their estimates. The resulting bands therefore capture idiosyncratic-error uncertainty only and under-cover when the common factors account for a large share of the variation (as they do in the simulation design of Tugan 2021). This routine is intended as a robustness check; for coefficient inference use asymptotic_var or summary.pvarife_result, and for the paper's IRF bands use irf_bands.

Value

An object of class "pvarife_bands" with components irf, lower, upper, level, and method = "bootstrap".

See Also

irf_bands, compute_irf

Examples

sim   <- sim_pvarife(n_units = 20, n_time = 15, n_vars = 2,
                     n_lags = 1, n_factors = 1, seed = 1)
fit   <- pvarife(sim$y, n_lags = 1, n_factors = 1, n_out = 5, n_in = 3)
bands <- bootstrap_irf_bands(fit, n_periods = 6, n_boot = 20, seed = 42)
plot(bands)

pvarife documentation built on June 13, 2026, 5:06 p.m.