find_swans: Find outlying "black swan" jumps in trends

In bayesdfa: Bayesian Dynamic Factor Analysis (DFA) with 'Stan'

Find outlying "black swan" jumps in trends

Description

Find outlying "black swan" jumps in trends

Usage

find_swans(rotated_modelfit, threshold = 0.01, plot = FALSE)

Arguments

rotated_modelfit	Output from rotate_trends().
threshold	A probability threshold below which to flag trend events as extreme
plot	Logical: should a plot be made?

Value

Prints a ggplot2 plot if plot = TRUE; returns a data frame indicating the probability that any given point in time represents a "black swan" event invisibly.

References

Anderson, S.C., Branch, T.A., Cooper, A.B., and Dulvy, N.K. 2017. Black-swan events in animal populations. Proceedings of the National Academy of Sciences 114(12): 3252–3257. https://doi.org/10.1073/pnas.1611525114

Examples

set.seed(1)
s <- sim_dfa(num_trends = 1, num_ts = 3, num_years = 30)
s$y_sim[1, 15] <- s$y_sim[1, 15] - 6
plot(s$y_sim[1, ], type = "o")
abline(v = 15, col = "red")
# only 1 chain and 250 iterations used so example runs quickly:
m <- fit_dfa(y = s$y_sim, num_trends = 1, iter = 50, chains = 1, nu_fixed = 2)
r <- rotate_trends(m)
p <- plot_trends(r) #+ geom_vline(xintercept = 15, colour = "red")
print(p)
# a 1 in 1000 probability if was from a normal distribution:
find_swans(r, plot = TRUE, threshold = 0.001)

bayesdfa documentation built on April 3, 2025, 9:31 p.m.