You're Not Long Copper

At the bottom of this page Paid Subscribers will find a link with a password to access a CPI Dashboard that they can use to examine their own stocks.

Note on Returns: Returns discussed throughout this paper are returns in excess of the risk-free rate. Asset-pricing models of the nature used throughout this report are built on excess returns because their purpose is to explain risk compensation, not total return. The risk-free rate represents the return an investor can obtain without bearing market risk (at least in the world of factor models, less so in the real world), so it must be removed before analyzing how risk factors drive performance.

Introductory Note: At Massif Capital, we think of ourselves as investing principally in projects and management teams, not in commodities. What follows is evidence in numbers of that thesis.

Every investment is a wager on what might be, not what is. With that in mind, investors deploy capital in the hope, sometimes the fantasy, that a particular future payoff will materialize. That hope is frequently underpinned by a narrative, but that narrative is rarely comprehensive enough to encompass the full scope of possible outcomes, most of which are poorly understood. Rather than the nuance, it is more common to hear someone declare themselves "long energy" or "bullish on copper," while the actual instruments in their portfolio bear only a passing resemblance to the economic exposures they claim to hold. The gap in understanding is not just a matter of tracking error; it is a symptom of a deeper conceptual fuzziness.

To build an investment thesis is to attempt clarity in a fog. The investor must first decide: what world must unfold for this bet to pay off? Which variables must move, and in which direction? And by what mechanism does that movement translate into profit or loss? Within this context, the instrument chosen to implement an idea is not a mere afterthought; it is the crucible through which the thesis is tested. If the instrument is saddled with stray risks or unpredictable sensitivities, the outcome may bear little resemblance to the original expectation.

Nowhere is this confusion more glaring than in the world of commodity producer equities. To have a view on oil, copper, gold, or lithium is to have a view on supply and demand, marginal cost, inventories, or the latest geopolitical hiccup. The desired exposure is to the price of the stuff itself, perhaps with a dash of carry or convexity. But by buying commodity-producer equities, investors trade the purity of their commodity price thesis for a stake in the messy business of corporate cash flows.

At issue is the transmission mechanism linking commodity price movements to equity prices. An equity instrument is a claim on whatever remains after all other parties involved in the business have been paid. Equity value is thus hostage not just to the price of what the company sells, but to capital allocation, cost discipline, financing terms, the whims of foreign governments, currency swings, regulatory surprises, the mood of other investors, and the ever-shifting equity risk premium. The commodity price matters, but it is only one voice in a noisy chorus, and it turns out frequently not the loudest.

To say "I am long copper" while holding a copper miner is to tell only half the story. Yes, there is exposure to copper prices, but also to management competence, the fiscal health of the host country, the tides of the equity market, the state of credit, and a host of other risks that can drown out the commodity signal. Sometimes this bundle is just what the doctor ordered: operating leverage, capital-cycle optionality, a bit of balance-sheet spice, but the real question is whether these equities are the right tool through which to express a commodity view, a question that requires establishing both a link between the commodity and the equity, but also understanding the magnitude of that linkage.

The cardinal sin is thus to mistake narrative kinship for actual risk alignment. That a miner digs copper and an oil major pumps crude does not mean their shares will dance to the same tune when the music changes. Two assets may share a theme, but their performance can diverge wildly, because the machinery that turns events into returns is built differently in each case. This is not a trivial distinction; it is the heart of the matter.

What follows is an attempt to make that machinery visible. The analysis is organized around a single framework we have developed, the Commodity Purity Index (CPI), which measures, empirically and on a rolling basis, how much of a commodity producer's equity returns are actually driven by the underlying commodity and how much is attributable to factors other than the commodity itself. The findings are, to put it plainly, uncomfortable for anyone whose investment thesis rests on the assumption that owning the equity is a fair substitute for owning the commodity.

The argument here is simple enough to test. When the thesis is about the price of a commodity, the exposure you get from owning the producer's equity may be a very different animal. The point is not to say these equities are mispriced or unfit for purpose, but to ask how faithfully their returns track the commodity itself, and when they wander off the path. By teasing apart the economic plumbing from the market's shifting moods and measuring how much of equity returns comes from commodity moves versus everything else, we hope to show when owning equity is a fair stand-in for the commodity and when it is something else entirely.

The Commodity Purity Index (CPI, an unfortunate acronym) is a framework developed to measure, on a rolling basis, the extent to which the underlying commodity drives a commodity producer's stock price behavior. The score runs from 0 to 100. A score of 80 or above means the stock moves almost entirely with the commodity: it is essentially a leveraged commodity instrument wrapped in equity form. A score below 20 means that, despite being a copper miner or an oil producer, the commodity is barely registering as a driver of the stock's day-to-day behavior.

To understand what the model does, it helps to think about why a stock moves on any given day. Take a copper miner. Its share price might rise because copper prices rose, the broader stock market rallied, the US dollar weakened, or because there was news specific to that company. Each of these forces is pulling on the stock simultaneously. The CPI's job is to disentangle them. The model does this by using three rolling windows of recent history (60 Trading Days of daily returns, 120 Trading Days of daily returns, and 36 Months of monthly data)¹ and regressing daily and monthly equity returns against a global equity variable, a commodity price variable, and a USD variable. The outcome is a set of sensitivities, one for each factor included in the model. Commodity sensitivity, or how much of the stock's movement is explained by the commodity price alone after accounting for other forces, is the raw material for the CPI score.

To this factor-model setup, we add two refinements. First, commodity prices and equity markets tend to move together: when the global economy is doing well, both copper and the stock markets tend to rise. This creates a measurement problem. If copper and the market both went up on the same day, and the stock also went up, how much credit should copper get? Without a correction, the model would double-count this overlap and overstate the stock's true copper exposure. The CPI applies an adjustment that strips out the portion of a commodity's explanatory power already captured by other factors, so the final score reflects only what is genuinely unique to the commodity price.

The second refinement splits the commodity factor into two. Linear beta, the output of our multifactor model, is a serviceable approximation of sensitivity, until it isn't, and in commodity equities, it fails precisely when it matters most. Equities respond differently to dramatic moves in commodity prices than to day-to-day movements; as such, we have added a commodity convexity term that captures the nonlinear, volatility-dependent sensitivity of equities to commodity price movements that a single beta coefficient cannot capture.

For many, a working heuristic might be that a twofold increase in the commodity move produces a twofold increase in the equity move; this assumption regularly breaks down, creating convexity in equities' sensitivity to commodities. Importantly, the direction of the breakdown in linearity tells you something essential about the business underneath the stock. Positive convexity (a squared commodity coefficient greater than zero) means the equity amplifies large commodity moves, accelerating beyond what linear beta predicts; this is the signature of the high-fixed-cost producer, where operating leverage functions as a mechanical multiplier, magnifying price swings in both directions as the commodity clears or falls short of a relatively fixed cost stack.

Negative convexity inverts the picture: diminishing sensitivity to large moves, the equity failing to keep pace with the commodity as margins compress — a pattern characteristic of high all-in sustaining-cost producers and development-stage companies, where the optionality that investors are ostensibly buying erodes precisely when the underlying commodity delivers its most dramatic moves. If nothing else, the underlying company's convexity is important because it directly affects the credibility of tail-scenario return assumptions.

After these adjustments are taken into account, the CPI is computed as a combination of the commodity share, the total variance of the equity price action explained by the model, and a statistical significance penalty. The CPI is computed independently for each time period, with the 36M rolling measure being the most significant for fundamental investors, as it is the time frame over which corporate actions have the greatest impact on corporate fundamentals.

Before turning to the empirical findings, a conceptual clarification is in order; it is critical, as it is the source of both confusion regarding commodity equities and the source of long-term fundamental returns in commodity producers' equities.

There are two fundamentally different ways to ask how much of an equity's price action is "commodity" driven, and they do not give the same answer. Conflating them, it turns out, is the source of considerable confusion about what these instruments actually are.

The first measure is variance decomposition. Variance decomposition answers the question: of the total daily/weekly/monthly price volatility in this stock, what fraction is synchronized with the commodity? Critically, variance is a squared calculation and thus sign-blind. A day when copper falls two percent, and the miner falls three percent, contributes exactly as much to the variance share as a day when copper rises two percent, and the miner rises three percent. Direction is irrelevant; co-movement is everything. Variance decomposition answers the question a trader asks at two in the morning when they can't sleep: why is my position moving? For most commodity equities on most days, the answer is the commodity.

In the short term, especially day-to-day, equities can be reasonable commodity proxies, although that is not universally true, nor equally true for each producer. One of the key findings of this research is that not only does every producer respond differently to commodity movements, but even two theoretically similar companies (like Exxon and Chevron) can have wildly different responses.

The second measure is return attribution. Return attribution answers a very different question: of the total cumulative profit or loss over the holding period, how many dollars came from commodity-linked moves? This is a signed, cumulative, net-direction calculation. A day when copper adds one and a half percent to the stock's return and a day when copper subtracts one and a half percent cancel each other out. Return attribution answers the portfolio manager's year-end question: where did my returns actually come from?

While these two questions sound similar. They are not. Take, for example, Freeport McMoRan over the last 36 months; the attributable return from commodity exposure is 43.4%, yet commodities explain 60.4% of the variance during the same period.

The explanation for this oddity is not a measurement error or a model quirk. It is a mathematical property of the relationship between variance and cumulative returns, one that emerges specifically in multi-factor assets where different forces operate at different frequencies.

A continuous flow of rapidly-resolving information drives commodity prices: inventory reports, shipping data, Chinese PMI releases, OPEC statements, weather forecasts, and speculative positioning. Each piece of news moves the price; the next piece frequently counters it. Copper rises on a Chinese stimulus headline, falls on a stronger dollar print, rises on a Chilean mine strike, falls on weak manufacturing data. The volatility this produces is high because each move is large, and commodities are levered to macro surprises. But the autocorrelation is low, which means today's price movement tells us very little about tomorrow's. Put another way, each price move is driven by a different information shock, many of which point in opposing directions. The commodity factor is high-frequency: it exhibits large daily variance, low autocorrelation, and is generally mean-reverting over the short and medium term.

The model residual (price action that the model does not explain) captures something quite different. What we refer to as the residual+α component captures everything the 4-factor model cannot explain, this includes both unmodeled systematic factors/systematic under and over performance predicted by the model (a natural gas factor or a momentum factor) and fundamental company-specific drivers (capital allocation decisions, share buyback programs, dividend policy changes, hedging book mark-to-market, reserve revisions, drilling efficiency gains or misses, and management guidance credibility).

Take, for example, Diamondback Energy (FANG). Over the last 36 months (end of February), FANG has returned roughly 14.7%, of which the residual contributed a -28.9%, commodity variables contributed -11.6%, the global equity factor contributed 39.4%, and USD contributed -0.6%. The large negative residual over the last 36 months could reflect any number of things, the market's negative repricing of the stock for acquisition integration risk, or perhaps elevated leverage from the 2024 Endeavor deal, rather than indicating poor commodity exposure; the commodity factor was negative, but the residual had a larger negative impact. The residual is a low-frequency variable: small daily increments, serially persistent, compounding slowly over time.

Fundamentals are serially persistent because real-world corporate events unfold over quarters, not days. When FANG announced the Endeavor acquisition, the market did not fully reprice the stock in a single session and move on. Instead, the information percolates through multiple channels over months: the initial deal announcement reprices the stock; each subsequent earnings call updates the integration timeline; credit agencies revise leverage outlooks; analyst estimates adjust; and quarterly production data confirms or denies synergy targets. Each of these events generates a new residual return sometimes in the same direction, sometimes not. This is not because the model is broken, but because the underlying fundamental state hasn't changed. The company is still integrating Endeavor in month 3, just as it was in month 1. This is serial persistence: today's residual can predict tomorrow's residual because the root cause hasn't resolved.

Contrast this with commodity attribution: the commodity factor's daily contributions frequently flip sign, so its cumulative line is choppy. But a fundamental overhang (leverage concerns, management-credibility discount, ESG exclusion from index funds) generates a one-way daily drag that cumulates into what appears to be a relentless trend. The key insight: a slow-rising/declining residual is the model's way of telling you that something structural and persistent is repricing the stock outside its commodity exposure. It's the quantitative fingerprint of a fundamental narrative, and the steeper the slope, the stronger the market's conviction that the fundamental headwind (or tailwind) is real.

The same thing happens on the upside as the downside. Take, for example, Lundin Mining, the residual captures mine execution improvements, the re-rating associated with a major acquisition, the slow drift of analyst consensus as quarterly results confirm or deny a thesis, index rebalancing flows, and the gradual build and unwind of thematic positioning. None of these produces headlines. On any single day, the residual contribution might be +0.05% invisible beside copper's daily swings of plus or minus two percent. The crucial property is that the residual is serially correlated at the multi-week to multi-month frequency. Today's small positive residual is likely followed by tomorrow's, because the underlying drivers are persistent processes. Those imperceptible daily increments compound in one direction for extended stretches.

The variance decomposition barely notices the residual, because its daily variance contribution is tiny: squaring 0.05% gives essentially zero compared to squaring 1.5%. But the return attribution fully captures it, because 0.05% compounding in the same direction for two hundred consecutive days produces a ten percent cumulative contribution that never appeared in the daily variance decomposition at all. The commodity factor generates a large gross movement that substantially cancels; the residual factor generates a small gross movement that mostly survives.²

This frequency separation has a direct implication for how much of any equity is treated as "commodity" as a function of the investment horizon. Hold Lundin Mining (LUN) for one day: it is approximately 61% copper (variance decomposition governs). Hold it for one year: it is approximately 21% copper (return attribution governs). Hold it for five years: it is even less, as the residual and equity beta compound further while copper's contribution continues to oscillate and cancel. The commodity fraction decreases with holding period. These equities become less commodity-like the longer they are held. The equity wrapper's non-commodity components become increasingly dominant as the high-frequency commodity signal cancels out over longer horizons. This is the exact inverse of the intuition most investors bring to the category.

Financial instruments come with narratives. A gold miner is a gold company. A copper producer is a copper play. An oil major is an energy bellwether. These stories contain grains of truth, as the best fictions do. But the question that matters — the question that separates portfolio construction from marketing — is whether investors actually receive the commodity exposure those narratives promise.

In the following pages, we apply the CPI framework to eleven commodity-producing equities across three complexes: four oil and gas names (ExxonMobil and Equinor), four copper miners (Freeport-McMoRan, Lundin Mining, First Quantum Minerals, and Ivanhoe Mines), and three gold miners (Barrick Mining, Newmont, and Lundin Gold). The analysis spans a full decade of market history. Two investor archetypes frame the discussion: the commodity-first investor, who buys the equity to express a commodity view, and the fundamental-first investor, who buys the business because it is mispriced.

Begin with ExxonMobil, whose four-factor model explains only 35.9% of the return variance over the 36-month window (R² = 0.36). What is critical is that you not interpret this R² the wrong way; we are not running a statistical test seeking to create a model with high explanatory power. We are seeking to understand exactly how much commodity price movements explain equity movements. The R² is thus a ceiling variable; it represents the maximum possible contribution of commodity price action in explaining a particular equity's price action.⁴

The 36-month CPI registers at 20.1, squarely in non-commodity territory, though it rises to 48.1 at 120 days and 57.3 at 60 days, a 37-point spread. Commodity price actions account for 82.9% of the model's explained variance, yet the model explains barely a third of the stock's actual movement. Over the trailing 36 months, Exxon returned approximately 100%, of which oil contributed 5.5%, the dollar 14.5%, and the residual 110.4%. Oil-generated noise; everything else generated wealth.

Equinor introduces a structural wrinkle absent from the American peers we reviewed in our broader analysis testing this model: a negative global market beta of −0.60. Where the American names rise and fall with broad equities, Equinor moves against them. This negative market loading, combined with a commodity beta of 0.58 and a USD beta of −0.22, creates a distinctive factor profile. Over 36 months, Equinor returned -15.1%, of which the commodity variables contributed -2.1%, the residual 0.8%, USD -1.5% and the global equity factor -10.2%. For the commodity investor, Equinor's negative market beta removes the equity contamination that dilutes the signal of every other oil stock. And yet even this structural advantage does not lift the CPI above non-commodity territory.

Copper's pro-cyclical nature, its daily correlation with global equity indices runs approximately three times that of gold, sets a contamination floor that no corporate structure can engineer away.

Freeport-McMoRan, the world's largest publicly traded copper producer, records the highest 36-month CPI in the copper group at 20.9, which is classified as merely Weak. The model explains 64% of the variance with a commodity beta of 1.3. Over 36 months, Freeport returned 46.7%, with copper contributing 45.2% and the global market 55.0%. But the residual was negative: −45.0%. The company-specific factors: Grasberg's Indonesian political risk destroyed value. An investor replicating Freeport's betas in liquid instruments would have earned 100.2 points of return; the corporate vehicle delivered 46.

Lundin Mining offers a striking contrast. Its 36-month CPI of 16.0 classifies it as non-commodity, and the cross-horizon profile is unusual: the CPI declines from 16.0 at 36 months to 14.6 at 120 days and collapses to 3.2 at 60 days. Yet Lundin returned 160.0%, with copper contributing 46.3% and explaining 60% of the daily movement, the residual an extraordinary 96.3%, and if included in the variance chart (which is not standard statistical practice, only 56.8% of the daily movement). This company story was driven by the Filo Mining acquisition re-rating and operational execution across Chilean and Brazilian operations. The convexity loading of 0.15 is also favorable, the closest to symmetry of any copper miner tested.

First Quantum Minerals is the most extreme case: a 36-month CPI of 3.1, the lowest in the entire analysis. The model explains only 28.4% of the variance. The dominant driver of variance is not copper but the global equity market, which accounts for 50.9% of the explained variance, compared with copper's 40.9%. The global market beta of 1.8 is the highest among the sample's betas. Over 36 months, the stock returned a modest 21.2%, with market and copper each contributing roughly 27–28% and USD subtracting 6.5%. The Cobre Panamá suspension and attendant balance sheet risk overwhelm the copper signal entirely. Yet the convexity loading of 15.9, an extraordinary positive reading, means that when First Quantum does participate in copper rallies, it participates with ferocity. The instrument is not a copper proxy; it is a restructuring bet with an embedded call on copper prices.

Ivanhoe Mines deepens the paradox. Its 36-month CPI of 5.3 is non-commodity, with the global market as the dominant driver (58.7% of explained variance, beta of 1.8). Over 36 months, Ivanhoe returned 29.0%, while the residual was a catastrophic −42.0% — representing −144.6% of total returns. The company-specific factors destroyed more value than the entire return itself. Unlike Freeport's modest residual drag, Ivanhoe's residual drag is existential. For the fundamental investor, the question is whether Kamoa-Kakula's ramp and restart of closed parts of the mine following a 2025 mining accident, freed from the DRC's sovereign discount, can reverse the residual's direction. For the commodity investor, the question is academic.

Gold's near-independence from equity markets means that gold miners carry structurally less market contamination than copper or oil peers. What pollutes them is operational noise from geographically distributed, capital-intensive mining operations. The updated data reveal a gold complex whose members have diverged dramatically from one another.

Barrick Mining records a 36-month CPI of 36.2 with a global market beta of 0.010, making it the purest gold-factor play in the entire sample (commodity accounts for 86.0% of explained variance). Over 36 months, Barrick returned 177.0%, of which gold contributed an extraordinary 254.5%, more than the entire return, while the global market subtracted 14.8% and the residual was a 7.6%. Gold, and almost gold alone, drove Barrick. But the convexity loading of −2.0 is severely negative: Barrick amplifies gold's largest declines and dampens its largest rallies. In a 186% bull market, this asymmetry is tolerable. In a flat or declining gold environment, negative convexity becomes a confiscatory tax. Looking at the convexity beta over time is useful here as it demonstrates how much it moves around.

Newmont, the world's largest gold producer, appears to do what people want commodity producers equities. The composition of explained variance is revealing: commodity accounts for 48.1% of daily movement while the U.S. dollar accounts for 47.8%. Newmont is, in nearly equal measure, a gold bet and a dollar bet on any single day. The USD beta of −2.2 is the most extreme in the sample. Over the period, gold contributed 154.5% to Newmont's 107.5% total return, the residual contributed -28.1%, and the global equity factor -5.4%. The convexity loading of 1.7 is genuinely attractive: Newmont amplifies gold's advances and dampens when it declines. Exactly the payoff structure the narrative of leveraged mining promises, but the data rarely delivers. The firm's trailing 36-month CPI is 71.

Lundin Gold produced the highest total return in the sample: 212.85% over 36 months. Its commodity beta of 2.3 makes it the most leveraged among the gold firms in this analysis. Commodity variables contributed 149.7%, with a positive residual of 10.9% reflecting Fruta del Norte's exceptional economics. The global market beta of −0.60 is negative, adding equity market independence to a strong commodity loading. But the convexity profile has shifted dramatically: the loading of −23.3 is the most extreme negative reading in the entire sample. The 30-point spread between the 120-day CPI (32.5) and the 60-day CPI (2.3) suggests Lundin Gold's gold character is unstable at shorter horizons. The leverage is real, but the convexity cost has risen sharply.

Across all eleven equities, only one reaches the CPI Strong band at the structural rolling 36-month horizon, Newmont. The median is approximately 15.4. On a trailing three-year basis, the commodity-producing equities examined in this analysis, are not commodity instruments.

Four structural findings emerge. First, the commodity itself determines the contamination of the commodity signal. Gold miners carry near-zero global market betas (Barrick: 0.010, Newmont: 0.029, Lundin Gold: −0.60), while copper miners carry betas ranging from 0.908 (Freeport) to 1.843 (First Quantum). This hierarchy, gold cleanest, copper most contaminated, oil intermediate, appears to be a property of the commodities, not the companies.

Second, commodity contributions to equity returns cancel at extraordinary rates. Exxon's commodity cancellation ratio runs at 94.9%. Barrick's gold factor contributed 254.5% in gross terms, yet the stock returned only 177%, with the difference attributable to convexity drag. The variance decomposition reports that these equities are commodity vehicles; the return attribution reports that they delivered wealth entirely from elsewhere. Both are true. The gap between them is the cost of accessing commodity exposure through a corporate balance sheet.

Third, negative convexity is widespread but not universal. All four oil names exhibit negative convexity. Among copper miners, the pattern reverses: First Quantum (15.925), Ivanhoe (5.779), and Lundin Mining (0.147) all exhibit positive convexity. The gold complex is bifurcated: Newmont's positive 1.7 contrasts with Barrick's −2.0 and Lundin Gold's −23.3. Convexity is an attribute of the individual company's operating leverage and balance sheet, not the commodity complex.⁵

Fourth, the residual is the true swing factor. Of eleven equities, four carry meaningfully negative residuals: Freeport (−45.0%), Ivanhoe (−42.0%), and the Equinor factor structure implies similar drag. These are companies where accessing commodity exposure through the corporate form costs real money. This means a company-specific thesis is essential. On the other side, Lundin Mining (+96.3%) and Exxon (+110.4%) generated enormous positive residuals. The dispersion is 168 percentage points between best and worst. No commodity thesis, however brilliant, survives that spread without granular company analysis.

The conclusion is not that commodity equities are poor investments. Several delivered spectacular returns: Lundin Gold's 234%, Barrick's 177%, and Lundin Mining's 160% (remember these returns are net of the risk free rate). The conclusion is that commodity equities are not commodity instruments. The investor who understands this distinction will attribute outcomes correctly, size positions appropriately, and resist the category error of believing a correct commodity thesis has been validated when the source of the profit was, in fact, the residual, that opaque, undifferentiated, thoroughly corporate remainder that the Street prefers not to discuss.

The four-factor model's residual captures everything the model does not explain, and across this sample, what it does not explain is enormous. The R² values range from 0.28 (First Quantum) to 0.58 (Freeport-McMoRan), meaning that residual variance, the share of daily return movement unexplained by commodity, global market, convexity, and dollar factors, ranges from 42% to 72% of total variance. In every single equity tested, the residual exceeds the commodity factor's contribution to total variance. This is the central quantitative fact of commodity equity investing, and means better returns are not found via better commodity forecasts but by better stock picking.

The residual is not random. For First Quantum, the 72% of unexplained variance is driven by variables the CPI model was not built to capture: the legal trajectory of Cobre Panamá arbitration, the Panamanian political cycle, the credit market's evolving assessment of FM's refinancing risk, and the probability distribution of operational restart timelines. For Ivanhoe, the 63% unexplained variance embeds the DRC's sovereign risk discount, the Kamoa-Kakula ramp trajectory, and the capital market's confidence, or lack thereof, in management execution. These are specific, analyzable, and for an investor with a differentiated view, potentially the richest source of alpha in the portfolio.

But the return attribution reveals a further distinction that the variance decomposition alone cannot make. The sign of the cumulative residual, positive or negative over the measurement period, tells you whether the corporate vehicle added or destroyed value relative to what a factor-replicating portfolio would have delivered. And the dispersion is staggering. Lundin Mining's 36-month residual is +96.3%. At the other extreme, Ivanhoe's residual is −42.0%: the company-specific factors destroyed more value than the total return itself. The spread between best and worst residual is 168 percentage points across a single commodity complex. No commodity thesis, however precisely timed, survives that spread without granular company-level work.

This distinction is fundamental to how a portfolio manager might use the CPI output. A large negative residual is a warning: something about the business is currently embedding value destruction in the equity price that compounds against you regardless of commodity direction. Newmont returned 107.5% over the period, with gold contributing 154.5%. The corporate vehicle consumed roughly 18% of the commodity delivered returns via the residual. The case of FCX is even worse, with the residual (or company factors) roughly equaling the commodity factor in the opposite direction netting each other out.

A high residual variance, by contrast, is a different signal: the stock's returns are driven by factors or variables the commodity model does not capture, and the portfolio manager who understands those factors or variables has an informational edge structurally unavailable to the commodity-tracking investor. The PM's job is to separate the structural residual drag from the residual opportunity, avoid the former, and size into the latter when the analytical edge is real.

When you hold a portfolio of commodity equities, the convexity profiles aggregate rather than average out. If 7 of 11 equities exhibit negative convexity, as is the case with the companies discussed in this paper, the portfolio-level payoff function inherits that negative convexity. In a large commodity move, each negatively convex position amplifies losses more than gains. Since large commodity moves tend to affect the whole complex directionally (correlation spikes in stress), the diversification benefit that might otherwise offset individual position asymmetries is precisely absent when you need it most.

We believe we see evidence of this in the portfolio construction of Massif Capital peers. The results of a piecewise regression of Massif Capital's returns and those of a peer vs. the Bloomberg Commodity Index show returns bucketed by the performance of the Bloomberg Commodity Index, so how does Massif Capital and the peer do when the BCOM draws down 4%, or is up 4%, or is down 2%, etc? The Massif Capital portfolio is highly concentrated but constructed in a convexity aggregation-aware way, versus those of a peer who has a completely different approach to portfolio construction that is highly diversified across natural resources producers (50 or so positions vs 15), and to ensure we are not cherry picking, we have chosen a peer whose returns since our inception mirror ours. By building the portfolio in a convexity-aware way, we have dramatically reduced the sensitivity of our returns to moves in the overall commodity complex. This does not come without a cost. In up markets, this peer outperforms us, but in down markets, we outperform them. The result is our portfolio has a high but trailing upside capture ratio vs. the MSCI Commodity Producers Index (15.8% vs. peer 24.5%, Massif Capital's upside capture is 35% lower), but a downside capture ratio that is 66.8% lower (7.1% vs. 21.6%).

The CPI data make following this issue more concrete, and regularly updating readings reveals a more nuanced pattern than a blanket statement about negative convexity would suggest. Among the oil and gas names, all four exhibit negative convexity factor betas: Exxon (−1.71), Chevron (−1.47), Diamondback (−1.39), and Equinor (−1.18). The oil complex is uniformly concave. Among the gold miners, the pattern is mixed: Barrick's convexity beta of −2.0 is severely negative, and Lundin Gold's −23.25 is the most extreme negative reading in the entire sample, a reversal from prior periods when LUG exhibited favorable upside participation. Only Newmont (+1.7) now exhibits positive convexity among the gold names, amplifying gold rallies and dampening declines.

The copper complex tells a different story. Three of four copper miners exhibit positive convexity: First Quantum (+15.92), Ivanhoe (+5.77), and Lundin Mining (+0.15). Only Freeport (−1.13) is negative. The copper miners' positive convexity is difficult to explain from an economic perspective: these are, after all, companies with high fixed costs and operating leverage that amplifies gains when copper rises faster than costs adjust, and vice versa. Our only explanation is perhaps found in First Quantum's extraordinary positive convexity of 15.96, which seems high because it reflects the extreme optionality embedded in a restructuring situation rather than a stable corporate attribute. Perhaps the global consensus-positive outlook for copper is embedding similar optionality in other equities.

The instability of convexity profiles across time is itself an interesting finding. Lundin Gold's flip from positive to the sample's most extreme negative convexity, and Newmont's simultaneous reversal from negative to positive, demonstrates that convexity is not a permanent attribute of a company's operational structure. It is regime-dependent, shifting with balance sheet leverage, hedging programs, cost positions, and the market's discount rate for specific risks. The portfolio manager who sized Lundin Gold for its positive convexity properties two years ago is now holding the most negatively convex instrument in the sample. Convexity, like everything else in these corporate vehicles, must be monitored continuously.

Negative residual return drag aggregates with diversification and cannot be diversified away. When Newmont carries a −28.1% cumulative residual and Ivanhoe carries a −42.0% residual, holding both in a gold-and-copper allocation does not cancel the negative; it doubles the weight on value-destructive corporate structures. A 15-stock diversified resources portfolio that weights toward large-cap senior producers is, in expectation, capturing the organizational complexity, jurisdictional friction costs, and historical capital misallocation premium from each position simultaneously. This is not bad luck. It is the predictable consequence of selecting the largest, most widely covered commodity producers, which are exactly the positions a passive or benchmark-aware fund will overweight.

The residual opportunity, by contrast, is position-specific and diluted by diversification. First Quantum's 72% residual variance represents a specific, bounded set of risks: Cobre Panamá, the balance sheet, and the political timeline. An investor who has done the work to underwrite those risks is not exposed to 72% random variance; they are exposed to 72% variance that they believe they understand better than the market does. If FM is held at 3% weight in a 30-stock diversified resources fund, the alpha generated by that analytical edge contributes roughly 2.5% to portfolio performance if the thesis plays out. The idiosyncratic risk is largely diversified away, but so is the idiosyncratic return. The concentrated manager holding FM at 12% weight and doing the Cobre Panamá work properly generates roughly 10% portfolio-level alpha from the same analytical effort. Diversification, in this framing, is not risk management; it is edge destruction.

The CPI data make a specific case for concentration, grounded in what the residual actually represents. If the dominant source of return variance in commodity equities is idiosyncratic, and across this sample the residual ranges from 42% to 72% of total variance, exceeding the commodity contribution in every case, then the investor who correctly assesses the company-specific factors driving that residual has an edge structurally unavailable to the commodity-first or benchmark-aware investor. The commodity-first investor is paying company-specific risk without having a view on it. The fundamental-first investor is deliberately selecting into it.

The concentrated portfolio operates differently at each level of the decomposition. At the commodity factor level, it takes deliberate positions on commodity direction or regime, expressed through the stocks whose CPI and beta structure most faithfully transmit that view at the relevant horizon. At the residual level, where the real opportunity lies, it concentrates analytical resources into the specific idiosyncratic factors that dominate each stock's variance and sizes proportionally to conviction. And at the convexity level, it explicitly manages the portfolio's aggregate payoff shape, recognizing that 7 of 11 positions in this sample carry negative convexity and that the 4 positively convex names are concentrated in a single commodity complex (copper) and in one gold miner (Newmont) whose convexity profile was negative as recently as the prior measurement window.

The long fundamental holding period changes the analysis in one important respect but not in another. It helps with the residual: the idiosyncratic factors driving it (the Cobre Panamá resolution, the Kamoa-Kakula flooding issues, Fruta del Norte grade reconciliation) have specific, analyzable timelines. An investor who correctly assesses those timelines and holds through the resolution captures the re-rating that the market was too uncertain to price in advance. This is the fundamental edge in high-residual commodity equities, and it requires patience precisely because the market discounts residual risk until the uncertainty resolves. We would also hazard that the market not just prices the residual at a discount but often does not price it at all, preferring to believe it is pricing commodity price action.

It does not help with negative convexity. Convexity is not a timing problem. The asymmetry between upside and downside commodity participation is evident across both short and long holding periods. What changes at longer horizons is the analyst's ability to assess whether the operating model itself is changing, and whether the factors that generate negative convexity (financial leverage, hedging programs, fixed-cost structures, geographic diversification) are being altered by management in ways the market has not yet priced in. Lundin Gold's dramatic convexity reversal, from a positively convex optionality vehicle to the most negatively convex instrument in the sample, is a concrete illustration: the underlying business did not change, but the market's pricing of its risk did.

There is a third consideration: the differing impact of commodity prices on equity prices across different time horizons. Institutional risk infrastructure operates at the wrong frequency to measure these positions accurately. Value-at-risk models, factor risk attribution systems, and tracking-error analytics all operate at daily or weekly frequency. At that frequency, the variance decomposition governs, and the equity looks like a commodity position. The risk system correctly identifies Lundin Mining as a copper stock; 62% of its explained variance is copper-driven, its commodity beta is 1.128, and copper generates the dominant daily P&L swings. That measurement is real and useful for position sizing, margin management, and drawdown estimation.

An investor who hedges the measured copper exposure by shorting futures to neutralize the risk system's dominant factor will successfully reduce daily volatility. They will not meaningfully affect the long-run return trajectory, because they are hedging the noise, not the signal. The return attribution shows that copper contributed 37.2% of Lundin Mining's 127.1% total return, or roughly 29% of that total. The residual contributed 126.3%. The hedge neutralizes the 29% while leaving the dominant driver, the residual, fully exposed and unmanaged.

This is the institutional risk trap built into how commodity equities are classified, measured, and hedged. Risk systems that flag these positions as commodity exposures are not wrong; they are answering the right question at the wrong time horizon. The portfolio manager who takes that classification at face value will find that their commodity exposure behaved as expected on any given day and delivered something quite different over any given year. The gap between these two experiences lies in the mathematical structure of how variance and returns differ across frequencies, and it is sufficiently structural to be anticipated, measured, and managed explicitly.

The combined analysis points to a specific portfolio construction methodology that is neither the diversified commodity-exposure fund nor the standard long-only equity approach.

Diversification should occur across commodity types, not across companies within a commodity. Gold miners, copper miners, and oil equities expose the portfolio to different systematic factors with genuinely different correlation structures: gold's near-zero equity-market betas (Barrick: 0.01, Newmont: 0.02) versus copper's leveraged equity sensitivity (First Quantum: 1.84, Ivanhoe: 1.81), and oil's intermediate positioning with Equinor's unusual negative market beta (−0.37) as a structural exception. Holding two to three positions in each commodity group can provide real diversification across systematic factors. Adding the fourth, fifth, and sixth positions within each group introduces negative convexity and residual drag without adding systematic factor diversification, while diluting the idiosyncratic alpha opportunity that would have been captured by concentrating on the best fundamental ideas.

Within each commodity group, concentration should be the default. The correct number of positions is determined by how many high-quality, differentiated fundamental theses the team can underwrite, where the residual variance is high, the residual return trajectory is either positive or addressable, and the analytical edge relative to market consensus is real. For most investment teams, this is probably two to three positions per commodity group, held at meaningful weights, with portfolio construction focused on ensuring that the aggregate convexity profile is not uniformly negative.

Explicit convexity management should be part of the construction process, as it typically is not for natural resources funds. The updated data reveal that convexity is not a stable corporate attribute; it shifts across regimes, meaning the PM must monitor it continuously rather than underwrite it once. If the portfolio has 7 negatively convex positions and 4 positively convex positions, the PM should know the portfolio's aggregate convexity loading and have a view on whether the expected commodity payoff justifies that structure. Newmont's role in a gold allocation is no longer merely as a gold-exposure vehicle; it is now the only gold equity in the sample with positive commodity convexity (+1.2), providing the genuine optionality on gold upside that the narrative of leveraged mining exposure promises. Sizing it at 1–2% in a 50-stock fund eliminates this structural benefit. Sizing it at a third to half of a 10% to 15% gold allocation makes it a meaningful contributor to portfolio-level convexity improvement.

The CPI data suggest that, for most natural resources equities, the stock is primarily a vehicle for company-specific risk rather than for commodity exposure. If this is structurally true, the natural resources portfolio manager is fundamentally a stock picker who uses the commodity as part of a valuation framework and a regime filter, not a commodity investor who uses equities as an expression vehicle.

The commodity view, when the team has one, is probably better expressed separately through futures or structured instruments with known and controllable convexity properties, leaving the equity portfolio free to do what it is actually good at: capturing idiosyncratic returns from the large, analyzable residual that dominates these stocks' return variance. To do otherwise is to pay for a commodity vehicle, receive corporate equity, and then be surprised when the outcome reflects the latter rather than the former.