SV-ADF Asymmetric-Threshold Bubble Dating (Sarkar & Wells 2026)

radf_svadf implements Sarkar & Wells (2026)'s SV-ADF date-stamping procedure: radf()'s own recursive (backward) ADF t-statistic (badf), which the paper's own asymptotic theory (their Theorem 3.1) justifies under substantially weaker volatility conditions than PWY/PSY's original derivation (nearly-nonstationary stochastic volatility, not just deterministic time-varying volatility), compared against two different closed-form, sample-size-only thresholds: log(t)/10 for origination and log(t)/2 for collapse (t the current recursive window's own sample size) – both from the paper's own calibration exercise (their Section 5.1), not new simulation.

Usage

radf_svadf(data, minw = NULL, min_duration = NULL)

Arguments

data

A univariate or multivariate numeric time series object, a numeric vector or matrix, or a data.frame. A column may have leading and/or trailing NA values (an uneven/unbalanced panel where series enter or exit the sample at different times) – those periods are filled with NA in badf/bsadf and excluded from that series' adf/sadf/ gsadf. Interior NA values (a gap in the middle of a series) are not supported. When any series is padded this way, the panel statistic (bsadf_panel/gsadf_panel) is not available and is returned as NA, with a warning.

minw

A positive integer. The minimum window size (default = \((0.01 + 1.8/\sqrt(T))T\), where T denotes the sample size).

min_duration

Minimum number of consecutive periods a threshold crossing must persist to be dated (default psy_ds(n)).

Value

An object of class radf_svadf_obj: a list with the badf statistic path, the origination/collapse threshold paths, and origination/collapse date indices (NA if not detected).

Details

Origination is dated at the first run of at least min_duration consecutive points with badf above the origination threshold; collapse is dated (searching only after the origination date) at the first run of at least min_duration consecutive points with badf below the (lower) collapse threshold.

Note

Reuses radf's own badf sequence directly, compared against the two closed-form thresholds above – no simulation and no separate critical-value function, unlike radf_mc_cv/ radf_wb_cv.

Caveats

Sarkar & Wells (2026) is a non-peer-reviewed preprint, a different bar than every other source implemented in this package. The same note is emitted as a message when this function is called (see message/suppressMessages to silence it) and stored as attr(x, "caveat") on the returned object.

References

Sarkar, A., & Wells, M. T. (2026). Is there an AI bubble? Robust date-stamping for periods of exuberance. arXiv:2604.12062.

See also

datestamp for the symmetric-threshold PWY/PSY dating this complements.

Examples

# \donttest{
res <- radf_svadf(sim_data)
#> Experimental. Sarkar & Wells (2026) is a non-peer-reviewed preprint; see ?radf_svadf, Caveats section.
print(res)
#> 
#> ── radf_svadf (n = 100, minw = 19, min_duration = 5) ───────────────────────────
#> 
#> ℹ Experimental. Sarkar & Wells (2026) is a non-peer-reviewed preprint; see ?radf_svadf, Caveats section.
#> 
#>   series  origination  origination_date  collapse  collapse_date
#>     psy1           48                48        49             49
#>     psy2           23                23        24             24
#>    evans           NA              <NA>        NA           <NA>
#>      div           NA              <NA>        NA           <NA>
#>     blan           NA              <NA>        NA           <NA>
#> 
# }