Jonathanerrils/tail-dependence-ghana

# Extreme Commodity Tail Risk and Exchange-Rate Exposure in Ghana **A copula–EVT analysis of cocoa, gold and crude oil, with the Bank of Ghana's official interbank rate as the exchange-rate reference series.** Full manuscript: `paper/latex/paper.tex` (compiled PDF), target journal *Risks*. `paper/paper.md` is an earlier, superseded draft, retained for history only — do not cite it. ## Headline findings **Economic exposure.** Across ten commodity/cedi pairs, three quantile levels, two tails, and three stress-window definitions (COVID-2020, the 2024 cocoa supply shock, and their combination) — ninety simultaneous hypothesis tests — **zero cells survive correction for multiple testing**, under both a full family-wise Bonferroni correction and an independent Benjamini–Hochberg FDR procedure. No robust evidence that tail dependence among cocoa, gold and crude oil intensifies during stress, and no detectable commodity-to-cedi transmission channel at daily-to-monthly frequency. **Model risk.** Expanding the candidate copula set from the conventional four families (Gaussian, Student-$t$, Clayton, Gumbel) to eight (adding Frank, Joe, and the survival/rotated Clayton and Gumbel) does not rescue the six pairs — all four commodity-commodity pairs plus Brent–WTI — whose dependence structure the narrower set already could not fit. Every one of eight families is rejected by formal goodness-of-fit testing for all six. This is evidence of a genuine model-specification gap, not of absent dependence. **The cedi's own risk.** Independently of any commodity link, the cedi's extreme-value shape parameter ($\hat\xi = 0.67$) is roughly four times the heaviest commodity tail in the panel, and resists six independent attempts at adequate marginal modelling — including an explicit zero-inflated hurdle model and a two-state Markov-switching extension — though the paper's substantive tail-dependence conclusions are stable across all six. | | | |---|---| | | | ## Repository structure …