Risk assessment — Critical Materials Atlas

Three established ways to score supply risk — each letting the data or the criticality literature, not the weights I chose, decide.

The three methods

1 · Entropy-weighted TOPSIS (data-driven weights). Shannon entropy gives each criterion a weight by how much it varies across the 32 materials, then TOPSIS ranks each by closeness to the worst case (Hwang & Yoon 1981; Achzet & Helbig 2013). The weights it produces: origin gap 0.44China share 0.34trade HHI 0.10mine HHI 0.06refining conc. 0.03hard to substitute 0.02governance risk 0.01no recycling 0.00. The ranking diverges moderately from my fixed-weight index (Spearman ρ 0.69) — the data load most weight on China-share and origin-gap, pushing tungsten, magnesium and magnets up.

2 · GeoPolRisk (Gemechu et al. 2016) — stage concentration × the governance risk of the producers, for the mine and trade stages.

3 · Monte-Carlo supply-at-risk — each producer fails with a governance-derived probability and a random severity; over 20,000 draws I report the mean loss (ESaR) and the 95% tail (VaR/CVaR) — the shock scenarios made probabilistic.

Caveat: entropy rewards dispersion, not importance, so it nearly zeroes near-uniform criteria like recycling — read TOPSIS as a data-driven complement to the fixed index, not a verdict. All on public data, deterministic (fixed seed); disruption probabilities are governance-derived assumptions, not forecasts.

Material	TOPSIS	fixed	GPR mine	GPR trade	ESaR	CVaR95
Beryllium, unwrought	100	74	32	77	21%	65%
Strontium carbonate	87	57	33	27	17%	39%
Magnesium, unwrought	77	60	78	50	14%	42%
Rare-earth permanent magnets	72	52	52	38	26%	70%
Lithium carbonate	71	64	19	40	16%	52%
Tungsten, unwrought	70	46	67	32	16%	45%
Aluminium ores / bauxite	68	45	18	60	17%	35%
Phosphorus	66	57	21	34	27%	56%
Hafnium, unwrought	51	32	16	10	21%	51%
Silicon, < 99.99%	51	47	52	16	26%	71%
Natural graphite	48	51	63	20	28%	74%
Fluorspar, >97% CaF2	44	51	46	25	27%	68%
Arsenic	43	31	38	12	17%	41%
Tantalum, unwrought	43	33	35	11	33%	64%
Ferro-vanadium	42	45	51	8	17%	42%
Gallium	42	57	100	10	27%	88%
Germanium	41	44	81	10	27%	80%
Feldspar	30	25	15	21	8%	16%
Phosphate rock	29	31	21	16	22%	46%
Nickel, unwrought	29	32	34	6	13%	32%
Titanium, unwrought	28	43	45	12	12%	33%
Ferro-niobium	24	64	79	52	16%	49%
Manganese ore	24	46	20	32	23%	50%
Baryte	24	15	21	11	17%	35%
Cobalt oxides & hydroxides	22	43	85	16	30%	60%
Antimony, unwrought	21	35	34	19	13%	28%
Natural borates	21	42	40	41	15%	39%
Coking coal	19	24	22	15	20%	54%
Refined copper cathodes	18	23	12	10	12%	24%
Platinum, unwrought	12	39	52	12	24%	63%
Helium	9	22	25	12	19%	58%
Palladium, unwrought	6	25	35	11	15%	33%

Computed by build_riskmethods.py from data.json + flows_2024.json + wgi.json → riskmethods.json. Tail risk highest: Gallium, Germanium, Natural graphite, Silicon, < 99.99%. A screening panel — the Monte-Carlo probabilities are transparent governance-based assumptions, not predictions.

Three ways to score the risk — letting the data, not me, decide