Empirical claim #3 — Indonesian hilirisasi reflex window
Generated 2026-05-23 16:59 UTC by `scripts/py/intelligence/event_study_hilirisasi.py`. Re-run anytime; pure-Python, no network.
Pre-registered cohort
All IPTM actions with issuer_country = ID, action_type ∈ {export-control, industrial-policy, regulatory}, severity >= 4, target_sectors intersects the minerals / batteries cluster (ev-batteries, critical-minerals, critical-minerals-processing, nickel-downstream, nickel-processing, battery-supply-chain, copper-refining, downstreaming, bauxite-alumina), announced_date in [2018-01-01, 2026-12-31], with a full T+120 trading-day window available in the ETF dataset.
Matched: 10 actions in cohort; 0 excluded for insufficient post-event ETF data (announced after 2025-12-01). Final n = 10.
The claim
Severity-4+ Indonesian hilirisasi-regime actions (export-controls, downstream-processing mandates, royalty escalations, RKAB quota cuts, REE-management designations) are historically followed, over the 20- to 120-trading-day windows, by:
- EWA (Australia) / EWC (Canada) / EPU (Peru) / ECH (Chile) / ARGT (Argentina) — the five largest ex-Indonesia alternative suppliers of nickel / copper / cobalt / REE / lithium — outperforming the ACWI global benchmark; AND
- SPY (US, the primary downstream-consumer market for restricted Indonesian intermediate metals) trailing ACWI.
EIDO + MCHI are reported alongside for transparency. EIDO direction is ambiguous (issuer-country export-revenue compression vs. domestic-processing-premium uplift). MCHI direction is the subsidiary-capture hypothesis — Tsingshan / Huayou / GEM affiliates listed in HK/Shanghai would benefit from hilirisasi if they hold the in-Indonesia JV equity the policy creates; this is reported but is NOT the headline claim.
This is a historical association, not a forecast.
Results — abnormal returns vs. ACWI
Median log-return on each ETF minus log-return on ACWI over the same window. Two-sided sign test against H0: median = 0 (binomial exact). Wilcoxon signed-rank p is the more powerful rank-based test (normal approx; reported when n >= 6 non-zero observations).
Horizon: T+20d
| ETF | n | median (%) | mean (%) | IQR (%) | % neg | sign-test p | Wilcoxon p |
|---|---|---|---|---|---|---|---|
| EWA | 10 | -0.79 | -0.64 | 2.60 | 60% | 0.754 | 0.241 |
| EWC | 10 | -0.50 | -0.23 | 3.05 | 50% | 1.000 | 0.878 |
| EPU | 10 | +0.49 | +1.72 | 11.02 | 50% | 1.000 | 0.508 |
| ECH | 10 | +2.39 | +1.43 | 8.76 | 30% | 0.344 | 0.575 |
| ARGT | 10 | -0.19 | +0.25 | 11.05 | 50% | 1.000 | 0.959 |
| EIDO | 10 | -2.69 | -4.10 | 6.49 | 90% | 0.021 | 0.013 |
| MCHI | 10 | -3.17 | -0.25 | 9.23 | 60% | 0.754 | 0.799 |
| SPY | 10 | -0.14 | -0.37 | 2.25 | 50% | 1.000 | 0.508 |
Horizon: T+60d
| ETF | n | median (%) | mean (%) | IQR (%) | % neg | sign-test p | Wilcoxon p |
|---|---|---|---|---|---|---|---|
| EWA | 10 | +0.34 | -0.11 | 4.94 | 40% | 0.754 | 0.799 |
| EWC | 10 | +1.52 | +1.42 | 4.23 | 30% | 0.344 | 0.114 |
| EPU | 10 | +3.93 | +4.24 | 12.14 | 30% | 0.344 | 0.333 |
| ECH | 10 | -0.95 | +1.50 | 21.32 | 50% | 1.000 | 0.721 |
| ARGT | 10 | -3.56 | -0.24 | 15.58 | 60% | 0.754 | 0.878 |
| EIDO | 10 | -1.19 | -1.31 | 8.82 | 50% | 1.000 | 0.575 |
| MCHI | 10 | -3.99 | -4.27 | 10.81 | 70% | 0.344 | 0.114 |
| SPY | 10 | -0.11 | -0.51 | 2.91 | 60% | 0.754 | 0.575 |
Horizon: T+120d
| ETF | n | median (%) | mean (%) | IQR (%) | % neg | sign-test p | Wilcoxon p |
|---|---|---|---|---|---|---|---|
| EWA | 10 | -0.25 | -2.43 | 5.33 | 80% | 0.109 | 0.114 |
| EWC | 10 | +3.19 | +2.20 | 2.04 | 20% | 0.109 | 0.074 |
| EPU | 10 | +14.03 | +8.37 | 21.32 | 40% | 0.754 | 0.074 |
| ECH | 10 | +1.24 | -3.81 | 18.89 | 40% | 0.754 | 0.575 |
| ARGT | 10 | -1.87 | -0.65 | 26.88 | 50% | 1.000 | 0.721 |
| EIDO | 10 | -7.99 | -10.71 | 8.76 | 100% | 0.002 | 0.005 |
| MCHI | 10 | -0.25 | -3.58 | 10.19 | 50% | 1.000 | 0.386 |
| SPY | 10 | +0.44 | +0.50 | 3.86 | 50% | 1.000 | 0.508 |
Results — raw log-returns (for reference, not abnormal)
| Horizon | EWA | EWC | EPU | ECH | ARGT | EIDO | MCHI | SPY | ACWI |
|---|---|---|---|---|---|---|---|---|---|
| T+20d | +1.91% | +2.09% | +4.39% | +5.05% | -1.26% | +0.39% | -0.15% | +3.37% | +2.70% |
| T+60d | +4.11% | +7.71% | +8.64% | +7.77% | +3.93% | +2.16% | +1.72% | +5.62% | +5.52% |
| T+120d | +7.46% | +12.40% | +22.22% | +13.23% | +6.31% | +4.10% | +7.46% | +11.76% | +9.50% |
Events used
| # | Date | Type | Severity | Action |
|---|---|---|---|---|
| 1 | 2019-08-30 | export-control | 5 | 2020-01-01-indonesia-nickel-ore-export-ban |
| 2 | 2022-12-21 | export-control | 4 | 2023-06-10-indonesia-bauxite-ore-export-ban |
| 3 | 2024-05-30 | regulatory | 4 | 2024-05-30-indonesia-permen-esdm-6-2024-smelter-completion |
| 4 | 2024-05-30 | export-control | 4 | 2024-05-30-indonesia-permendag-10-copper-concentrate-export-ban |
| 5 | 2025-02-18 | industrial-policy | 4 | 2025-03-19-indonesia-uu-2-2025-fourth-amendment-minerba |
| 6 | 2025-02-24 | industrial-policy | 4 | 2025-02-24-indonesia-danantara-sovereign-wealth-fund |
| 7 | 2025-04-11 | industrial-policy | 4 | 2025-04-11-indonesia-pp-19-tiered-minerba-royalty |
| 8 | 2025-06-05 | industrial-policy | 4 | 2025-06-05-indonesia-pp-28-2025-risk-based-licensing-nickel-smelter-moratorium |
| 9 | 2025-09-30 | regulatory | 4 | 2025-09-30-indonesia-permen-esdm-17-rkab-annual-quota |
| 10 | 2025-11-14 | industrial-policy | 4 | 2025-11-14-indonesia-permen-esdm-18-rare-earth-management |
Honest caveats
1. Very small sample (n=10). Smallest cohort in the pipeline. Effect-size intervals are wide; sub-cohort analysis is entirely under-powered; small permutations of the cohort criteria would materially move the headline numbers.
2. Exact-day overlap + clustered overlapping windows. The two 2024-05-30 events (Permendag 10 + Permen ESDM 6) share post-event window — observations 3 and 4 produce identical abnormal returns. PP 19 → PP 28 → Permen ESDM 17 → Permen ESDM 18 cluster within 60-120 trading days of each other. Trade-independence is violated; p-values are anti-conservative upper-bounds.
3. No factor-return adjustment. Most of the 2024-2026 EWA / EWC / EPU / ECH / ARGT outperformance loads on the commodity factor; a factor-purged alpha would attenuate.
4. Cohort-period confounding is severe. 2020 COVID nickel-ore-ban shock + 2022 commodity super-cycle + 2024 Tsingshan nickel oversupply collapse + 2025 copper rally + 2025-Q4 / 2026-Q1 REE/cobalt re-pricing (multi-driver, including the MOFCOM architecture from claim #2) — the hilirisasi mechanism cannot be cleanly separated from these joint macro forces.
5. No out-of-sample validation. Future hilirisasi packages may differ if the regime saturates, if ex-Indonesia capacity absorbs the price-tightness pop, or if the 2026 §232 regime re-prices the alt-supplier ETFs through a tariff channel orthogonal to hilirisasi.
6. Cohort criteria pre-registered; test-ticker list is post-hoc. issuer_country = ID, action_type in {export-control, industrial-policy, regulatory}, severity >= 4, and the minerals/batteries target_sectors filter were pre-specified before the analysis. The test ETFs (EWA / EWC / EPU / ECH / ARGT) were chosen post-hoc from dossier-identified alt-suppliers; a random draw of other ex-Indonesia commodity exporters would not necessarily reproduce the same magnitudes.
7. EWZ deliberately excluded from the headline. Vale carries both Indonesia exposure (PT Vale Indonesia Sorowako + Pomalaa) and Brazil exposure (Carajás iron ore + Salobo copper). Which side dominates EWZ's response is the empirical question this study cannot answer cleanly — EWZ would be both alt-supplier and issuer-country exposure in one ticker. The 2026-05-23 Vale dossier (docs/intelligence/dossiers/vale.md) lays out the structural mechanism.
8. Joint evidence vs. claims #1 + #2 is structural, not mechanical. Claims #1 + #2 tested trade-restriction architectures (US demand-side, CN supply-side). Claim #3 tests a value-added-capture architecture — forced domestic value-add, not export restriction. If the alt-supplier responses pattern across all three claims, the reflex holds across three distinct architectures. If claim #3 fails where #1+#2 held, the value-added-capture channel is structurally weaker — a useful finding in its own right.