Illegal mining · Market design
Why can restrictions on mineral buyers fail to reduce illegal mining? This paper studies how control over mineral sales and essential imports can keep households mining, even when their returns fall. It examines when verifying origin at extraction can change these incentives.
Abstract
I develop a theory of how the organization of local markets shapes the effects of supply-chain regulation on illegal mining. An armed intermediary buys gold from households and sells them essential imports. What it pays for gold, it recovers through import margins. Households must cover their subsistence needs. Falling returns can make them mine more, not less. Downstream, laundering, unregulated buyers, and retention give the intermediary ways around restrictions. Restrictions can therefore change sales and receipts without changing extraction. Imperfect certification instead separates certified from unverified gold. This raises the profitability and fiscal value of legal mining. Once funded, clean innovation expands legal opportunities and the prospective tax base. Authorities who profit from illegal mining then find intervention optimal once the fiscal value of legal mining is high enough. This holds even though intervention is costly, may fail, and can be postponed. With livelihoods financed and production safe, illegal mining can end
Citation: Elies Lucio. For a Fistful of Gold. 2026.
What changes when mineral markets face tighter rules?
A more reliable certificate, a mine that produces less and a better livelihood are different outcomes. This research follows the connections between them: where minerals can still be sold, how buyers and miners share the costs, and what makes investment and public action possible.
The figures are comparisons inside models anchored to observed data. They show what follows from stated assumptions, rather than measured effects of an existing regulation. The five mineral markets are gold, cobalt, tantalum, tin and tungsten.
Download the six-page brief PDF · 6 pages ↗How we found these results
Start with observed evidence
Production and price records set the scale of each mineral market, using 2024 as the reference year. Trade records describe reported routes. Mine visits and fiscal reports help document institutions.
Describe how people respond
A model connects production, sales, payments and investment. Where observations do not tell us how people respond, the research states assumptions and explores alternatives. A country name identifies the setting; some responses still need to be measured.
Change one policy or a named package
Researchers solve a baseline, change specified rules, then compare outcomes within that same model. They check quantities and financial accounts separately. Unresolved calculations stay visible as unavailable results.
Does cleaner certification mean less mining?
What we compare
The model raises the assumed chance of finding a false origin claim after the initial checks from 25% to 50%, keeping other buyers’ spending fixed. The main comparison uses the same production-cost assumption across models.
Main finding
In this static comparison, extraction from the potentially exposed group falls by less than 0.51% in every mineral. In the original model, material from that group remaining on the certification register falls by roughly 56–58%.
The register changes much more than extraction
Each starting level = 100%. The solid bar shows what remains. The hatched portion shows the decrease from 100%, moving left.
Exposed-group extraction
Full scale: 0–100%
Magnified: 99–100%
The outlined last 1% of the extraction scale is enlarged here. Each outcome uses its own starting level.
Exposed material still in the certified channel
Full scale: 0–100%
Original model.
View the data
| Mineral | Original model | Sell or retain (Muir) | Buyer competition (VCG) | Exposed material still in the certified channel · Original model |
|---|---|---|---|---|
| Gold | -0.00888% | -0.00666% | -0.00666% | -57.764% |
| Cobalt | -0.50559% | -0.37916% | -0.37915% | -55.755% |
| Tantalum | -0.13541% | -0.10155% | -0.10155% | -56.959% |
| Tin | -0.13504% | -0.10128% | -0.10128% | -56.961% |
| Tungsten | -0.13207% | -0.09905% | -0.09905% | -56.979% |
Why this happens
Imagine a checkpoint becoming better at rejecting ineligible deliveries. Its register becomes cleaner. But material rejected there can still find another buyer. A large change at the checkpoint can therefore accompany a small change at the mine.
How to read it
The potentially exposed group means producers assigned an origin-compliance concern in the model. Its size is partly assumed. Its extraction is not total mining, measured illegal production or environmental damage. Each percentage uses its own model’s baseline and the outcome named in the figure.
Technical details
This change concerns the policy-set probability of late discovery, not the overall detection rate. Reporting and verification effort also affect whether a false claim is detected.
In the rebuilt examples, Muir means an intermediary chooses what to sell or retain, including resale. VCG means buyers compete for packages and their payments depend on the opportunities they leave to other participants.
The matched cost assumption is supply exponent 1: the marginal cost of extra production rises proportionally with output.
The rebuilt static models direct 25% of both source groups through a separate retail channel. Stronger detection changes the bulk contract while retail acquisition charges stay fixed; this partly explains their smaller extraction response.
The close Muir and VCG results in this example do not establish that the trading rules always give the same result.
Who bears the cost of tighter rules?
What we compare
A buyer faces a new charge, while the village’s selling rules, access to household goods and dependence on mining for basic needs stay fixed. The figure compares the same hypothetical village with and without this charge, using each model’s own starting point.
Main finding
In the two active sell-or-retain examples, extraction is unchanged while the intermediary’s profit falls by about 0.89%. In the rebuilt buyer-competition example, there is no production either before or after. The same headline of “no change” can therefore describe very different situations.
Follow both production and the intermediary’s profit
Each starting level = 100%. The solid bar shows what remains. The hatched portion shows the decrease from 100%, moving left.
Original model
Sell or retain (Muir)
Both active examples keep producing while the intermediary’s profit falls. The buyer-competition example starts and ends with no production; it cannot demonstrate resilience.
View the data
| What we compare | Starting situation | Production | Intermediary profit |
|---|---|---|---|
| Original model | Producing in both cases | 0% | -0.88516% |
| Sell or retain (Muir) | Producing in both cases | 0% | -0.88509% |
| Buyer competition (VCG) | No production in either case | Not defined: zero starting level | Not defined: zero starting level |
| What we compare | Production · Before | Production · After | Intermediary profit · Before | Intermediary profit · After |
|---|---|---|---|---|
| Original model | 0.318625 | 0.318625 | 0.055357 | 0.054867 |
| Sell or retain (Muir) | 0.3186 | 0.3186 | 0.055362 | 0.054872 |
| Buyer competition (VCG) | 0 | 0 | 0 | 0 |
Why this happens
A buyer who also provides food or credit can adjust several terms of a relationship. If households have few alternatives, they may keep mining after the terms worsen. An intermediary can also change sales channels or hold material instead of immediately reducing purchases.
How to read it
These are constructed examples, not surveyed villages or country estimates. In one rebuilt VCG example, production is zero before and after the change: zero contraction there means continued inactivity, not a working mine protected from the restriction.
Technical details
The displayed charge rises from 0 to 0.1 illustrative monetary unit per mineral unit; it is not a 10% tax. The selected full configuration combines selling arrangements, an intermediary supplying household goods and dependence on mining for basic needs. The original example uses zero retention value.
The accounts distinguish payments to households, procurement costs, profit, retained mineral stock and processed output. A payment is not automatically a household welfare measure.
Across the broader institutional experiments, the rebuilt VCG selling-rule comparison switches between an auction and a posted-price arrangement; neither simply contains all the options of the other. In the figure above, trading rules stay fixed and only the charge changes. Small Muir changes require the paper’s numerical-refinement qualification.
What makes change last, and who pays?
What we compare
The dynamic models follow investment and government decisions over successive periods. They compare stronger detection alone with a package that also gradually lowers other buyers’ spending by about 25%, with research support and public intervention available.
Main finding
Among the displayed available pairs, the combined package reduces the exposed group’s extraction by roughly 13% at period 100. Detection alone has much smaller effects. The larger reduction includes the assumed fall in outside spending; it is not a result of certification alone.
The combined package changes the production incentive
Each starting level = 100%. The solid bar shows what remains. The hatched portion shows the decrease from 100%, moving left.
Stronger detection
Combined package
Figure: rebuilt sell-or-retain model (Muir), model period 100. The combined package also reduces outside spending; its larger effect cannot be attributed to certification alone.
View the data
| Mineral | Policy comparison | Original model | Sell or retain (Muir) | Buyer competition (VCG) |
|---|---|---|---|---|
| Gold | Stronger detection | -0.00888% | -0.00666% | -0.00666% |
| Gold | Combined package | -13.38416% | -13.35324% | -13.35336% |
| Cobalt | Stronger detection | -0.52592% | -0.40082% | -0.4023% |
| Cobalt | Combined package | -12.5687% | -12.7333% | -12.73061% |
| Tantalum | Stronger detection | Unavailable | ≈ 0% | ≈ 0% |
| Tantalum | Combined package | Unavailable | -13.27705% | -13.27276% |
| Tin | Stronger detection | -0.00022% | ≈ 0% | ≈ 0% |
| Tin | Combined package | -13.411% | -13.27626% | -13.27199% |
| Tungsten | Stronger detection | 0.00038% | ≈ 0% | ≈ 0% |
| Tungsten | Combined package | -13.3157% | -13.27013% | Unavailable |
Why this happens
Investment depends on who earns the return from an improvement. Public action depends on whether future collected taxes justify costs today. Even a programme with later receipts may need cash upfront, while refundable deposits remain money owed back.
How to read it
These are model periods, not years or forecasts. Ownership and technology differ across models. Weaker tax collection prevents intervention in some original examples but delays it in some rebuilt examples; it does not universally stop action.
Technical details
Very small late-period screening effects can occur after nearly all reachable producers have already changed status. They do not establish that the earlier path was unchanged.
Keep country budgets, programme spending, private cash needs and refundable deposits separate. A surplus for one actor does not automatically finance another.
The calculations cover finite experiments. They do not establish indefinite continuation or which policy delivers the greatest social benefit.
The trading rules behind the results
The models share evidence but organize transactions differently. Their names identify these rules; knowing them helps explain why similar production totals can hide different payments or incentives. In the rebuilt models, decisions need not change proportionally: a new package, different buyer or decision to retain stock can cause a jump. “Nonlinear” describes this structure; it does not mean a more accurate country estimate.
The original model, extended to more minerals
Keeps the original production, certification and investment framework, then adds mineral markets and country scales. This provides continuity with the earlier gold and cobalt research.
An intermediary chooses sales and retention — Muir
The seller chooses what to deliver and what to retain, taking resale into account. The model follows how these choices affect receipts and actual physical output.
Buyers compete for packages — VCG
Packages go to the allocation with the highest modeled value, including the seller’s option to retain stock. A buyer’s payment reflects the value other participants forgo because that buyer participates.
What the reconstruction makes explicit
| Account | Original framework | Rebuilt framework |
|---|---|---|
| Buyer payments | Preserves the original transaction rules. | Recalculates Muir receipts with resale and retention; VCG payments remove the whole buyer in the payment comparison. |
| Retention and resale | Keeps the original selling framework and separately tested stock values. | Makes retaining material part of the seller’s choice; Muir also follows resale. |
| Processing | Uses the original production accounts. | Tracks physical conversion separately from sales revenue; money received is not output. |
| Ownership | Preserves the original technology and research incentives. | Specifies who owns improvements and earns access payments; public adopters do not generate private ownership returns. |
| Payment timing | Keeps separate programme, deposit and government accounts. | Adds explicit installation payments, dated refunds and funded service and custody obligations. |
Stronger detection
Makes a false origin claim more likely to be discovered. It changes the reward from entering the certified channel.
Other buyers’ spending
Changes the money available outside that channel. The combined package assumes a reduction; the research does not establish how to achieve it.
Household alternatives
Changes access to other income, subsistence needs and goods supplied by an intermediary, which can affect households’ ability to leave mining.
Research support
Changes the reward for developing and adopting improvements. Grants can encourage investment but also create a financing obligation.
Actual tax collection
Changes how much of a nominal tax liability reaches the government. A stated tax rate alone does not show what is available to finance intervention.
What the evidence supports
Production sources include USGS, BGS and the World Gold Council; price records retain their stated product stage. UN Comtrade describes reported trade routes. IPIS mine visits, EITI reports and the Rwanda mining study add institutional context. These sources answer different questions and are not one uniform sample.
Gold, cobalt, tantalum, tin and tungsten are modeled separately. Coltan is an ore or concentrate; the tantalum market uses its contained tantalum. Rough diamonds provide a certification comparison, while copper and nickel provide industrial context.
The production archive contains 6,277 historical records; its separate 2024 BGS panel has 309 country–resource combinations. Trade records cover 2005–2024, with a narrower six-country panel for detailed routes. Separate 2025 updates do not replace the 2024 calibration.
The papers and downloadable data identify the source, assumptions and saved result behind each comparison. Earlier gold and cobalt benchmarks are archived separately from the global extension.
What still needs measurement
Matched evidence on miners’ payments, production costs, buyers, household alternatives and actual enforcement is needed to estimate policy responses.
Reported routes do not reveal every buyer or re-export. Visited mine sites do not form a representative national sample. Some trade categories combine several minerals.
Scenario ranges are not confidence intervals. Failed or unavailable model comparisons remain identified and are never filled with zero.
The results do not measure changes in conflict, environmental damage or household welfare, and do not establish programme affordability.