Measuring Economic Resilience and Recovery with ERRI
In ERRI: Economic Resilience and Recovery Index

knitr::opts_chunk$set(collapse = TRUE, comment = "#>", fig.width = 7,
                      fig.height = 4.5)

Purpose

Economic resilience is not a single observed variable. ERRI represents it as six complementary dimensions calculated relative to an estimated no-shock counterfactual. Let the observed outcome for unit $i$ at time $t$ be $Y_{it}$ and its counterfactual be $Y^{(0)}_{it}$. The scaled adverse gap is

$$ G_{it}=\frac{Y^{(0)}{it}-Y{it}}{s_i}, $$

where $s_i$ is the pre-shock standard deviation, absolute mean, or one.

The package measures maximum adverse gap (depth), summed adverse gap (cumulative loss), time required to remain within a tolerance, strength of recovery, post-shock residual volatility relative to pre-shock volatility, and positive performance beyond the counterfactual after recovery.

Example

library(ERRI)
dat <- erri_example_data()
head(dat)

The example contains three fictional regional income series. The shock begins in 2020.

fit <- erri(dat, time = "year", outcome = "income", unit = "region",
            shock_time = 2020, method = "trend", scale = "sd",
            epsilon = 0.25, consecutive = 2)
fit

plot(fit, type = "trajectory", unit = "North")

plot(fit, type = "index")

Uncertainty

Residual bootstrap intervals propagate uncertainty in the pre-shock counterfactual. At least several hundred replications are recommended for an empirical study.

boot <- erri_bootstrap(fit, R = 99, seed = 2026)
subset(boot$intervals, measure == "ERRI")
rank_probability(boot)

Weight sensitivity

The default weights are equal. The following analysis draws random weights from the simplex and recalculates scores and rankings.

sens <- erri_sensitivity(fit, R = 250, seed = 2026)
aggregate(ERRI ~ unit, sens, function(x) c(mean = mean(x), sd = sd(x)))

Interpretation and limitations

A higher score denotes stronger measured resilience under the selected model, scale, tolerance, and weights. The score is not automatically causal. A shock date must be substantively justified, and a trend, mean, or AR(1) counterfactual may be inadequate when other events affect the outcome. Report component estimates, bootstrap intervals, and weight sensitivity rather than only the composite index.