Commodity Pricing Framework Report
AI-synthesized walkthrough of the GS "10 Lessons with Jeff Currie" note — reads commodity price as structural + cyclical components.
Commodity Pricing Framework Report
AI-synthesized note. This report is the reconciled output of a multi-agent analysis of the source PDF (see the Subagent Reconciliation section below), not a hand-authored note — treat the quantitative claims as source-dependent.
Source: Goldman Sachs Global Investment Research, "Commodity Insights: 10 Lessons Learned with Jeff Currie," September 26, 2023.
Concept focus: commodity pricing as a two-part framework: Price = Structural Component + Cyclical Component.
Note on evidence: this PDF is a sell-side commodity research note and retrospective framework essay, not an academic paper with fully reproducible datasets. The notes below separate what the paper claims from my interpretation and flag places where the PDF's evidence is suggestive rather than conclusive.
Note on method: this page and the RAG transit planning report were produced the same way — a multi-agent read of a source PDF, reconciled into one note. The transit paper is worth reading alongside this one for a different reason than its subject: it argues for the technique these notes are an instance of, and is candid about where grounding in a source document stops protecting you.
Executive Summary
The central concept is that commodity prices, especially oil prices, should be read as the sum of a structural component and a cyclical component. The structural component is tied to long-dated prices, marginal costs, project economics, capital scarcity, and investment uncertainty. The cyclical component is tied to spot market tightness, inventories, storage costs, interest rates, and the shape of the futures curve. The paper introduces this directly in Lesson 1 and Exhibit 1 (p. 3), then builds the remaining lessons around it.
The paper's strongest framework claim is that timespreads contain useful information about near-term physical tightness. Lesson 2 argues that spot-futures arbitrage links timespreads to convenience yield, storage costs, and interest rates (p. 5). Exhibits 3 and 4 show that OECD commercial stocks and an inventory/rates model line up with Brent 1m/36m timespreads reasonably well (pp. 6-7). This is useful evidence, but it is model and correlation evidence, not proof that timespreads never mislead.
The most important learning move is to stop asking for "the" oil price. The paper argues that the relevant object is the whole term structure: spot prices, long-dated prices, timespreads, quality/location differences, and how these interact with physical constraints. The closing appendix letter makes this point explicitly by saying Jeff Currie's insight was that there are "Oil Prices," not one oil price (p. 25).
My interpretation: the concept is best learned as a diagnostic map. First ask whether a price move is structural or cyclical. Then ask whether the evidence comes from inventories/timespreads, marginal project costs, producer behavior, investment uncertainty, or substitution across fuels. The framework is powerful because it links finance to physical bottlenecks. Its weakness is that many of the exact magnitudes depend on proprietary datasets, historical classifications, and models not fully reproduced in the PDF.
Subagent Reconciliation
All three subagents agreed on the main structure:
- The paper is a 10-lesson commodity research note, not a conventional academic paper.
- The core method is a framework synthesis supported by charts, prior Goldman research, economic intuition, and selected empirical models.
- The most important concept is the structural/cyclical decomposition of commodity prices.
- The highest-risk claims are quantitative claims that rely on proprietary data, small samples, or model choices.
No substantive disagreement emerged. The main reconciliation choice in this report is terminology: where the prompt asks for "experiments," I use "evidence and tests" because the PDF mostly presents observational charts, models, and historical case studies rather than controlled experiments.
Glossary
| Term | Meaning for this paper | Paper citation |
|---|---|---|
| Structural component | The long-dated price component, usually tied to marginal production cost, project economics, geology, technology, politics, and cost of capital. | Lesson 1, pp. 3-4 |
| Cyclical component | The spot-minus-long-dated component, tied to inventory tightness, convenience yield, storage/insurance costs, and interest rates. | Lesson 1-2, pp. 4-6 |
| Forward curve / term structure | The set of commodity prices for different delivery dates. The paper treats the curve shape as information about physical tightness and future supply incentives. | Exhibit 1, p. 3; p. 25 |
| Spot price | Price for near/immediate delivery. In this framework, spot prices react strongly to current physical shortages or surpluses. | Lessons 2-3, pp. 5-8 |
| Timespread | Difference between nearby and longer-dated futures prices, such as Brent 1m/36m. The paper treats this as a core signal of near-term tightness. | Lessons 1-3, pp. 4-8 |
| Backwardation | Nearby prices above later-dated prices. The paper associates this with tight markets and high convenience yield. | p. 5; p. 17 |
| Contango | Later-dated prices above spot/nearby prices. The paper associates this with ample inventories and the cost of carrying inventory. | p. 5; p. 17 |
| Convenience yield | The non-cash benefit of holding physical inventory, such as avoiding stockouts or smoothing production. | p. 5; p. 8 |
| Cost of carry | Storage, insurance, and financing costs of holding a commodity. | p. 5 |
| Marginal cost / marginal producer | Cost of the last unit needed to satisfy demand. Long-dated oil prices are described as being set by the higher-cost marginal producer. | pp. 3, 10, 17 |
| GS Top Projects | Goldman Sachs' upstream project dataset used to estimate project breakevens, capex, reserves, production, and marginal incentive prices. | Lesson 4, p. 10 |
| Hurdle rate | Minimum return needed to justify investment. The paper argues uncertainty can raise hurdle rates and therefore long-dated prices. | p. 10; p. 14 |
| Old economy | Physical sectors such as energy, mining, and materials. The paper argues underinvestment in these sectors can set up later commodity tightness. | Lesson 5, pp. 11-13 |
| Exploitation phase | A phase where existing spare capacity can meet demand, limiting the need for high long-dated prices. | p. 11 |
| Investment phase | A phase where new, more expensive capacity is needed, so long-dated prices must rise enough to fund projects. | p. 11 |
| Real options | Investment logic where uncertainty can make waiting valuable. The paper uses this to argue that uncertainty can delay supply investment. | Lesson 6, p. 14 |
| Geological decline rate | Natural production decline from existing oil fields after peak output, requiring investment just to maintain supply. | p. 14 |
| OPEC / OPEC+ | Producer groups that can influence production, inventories, and curve shape. The paper argues they do not control the long-run price level. | Lessons 8-9, pp. 17-19 |
| Supply elasticity | How much supply responds to price changes. The paper argues non-OPEC supply elasticity drives OPEC pricing power. | Lessons 9, pp. 18-19 |
| BTU / mmBtu | Energy-content units used to compare fuels on a common heat basis. | Lesson 10, pp. 20-21 |
| Fuel substitution | Switching from one fuel to another when relative prices move far enough, such as coal-to-gas or gas-to-oil switching. | Lesson 10, pp. 20-21 |
Paper Walkthrough
Framing
The paper frames Jeff Currie's contribution as showing how physical constraints force commodity prices to extremes to change behavior: incentivizing investment, discouraging demand, releasing inventory, or encouraging substitution. This framing appears in the opening quote and introduction (pp. 2-3).
The paper's stated contribution is a retrospective synthesis of 10 commodity-market lessons from Goldman's research history, especially in energy (p. 2). It is not trying to introduce a single new econometric result. It is trying to teach a durable way to read commodity markets.
Lesson 1: Price = Structural Component + Cyclical Component
What the paper claims: commodity prices can be decomposed into a long-dated structural component and a short-term cyclical component. The structural component is linked to marginal cost, while the cyclical component is linked to inventories and delivery tightness (pp. 3-4; Exhibit 1).
My interpretation: this is the master key for the report. When oil prices move, the first diagnostic question is whether the move is about near-term barrels or long-run investment incentives.
Evidence: Exhibit 1 is conceptual. Exhibit 2 shows 36-month forward WTI prices were relatively stable in the 1990s and later moved sharply, especially during 2003-2008 (p. 4). This supports the idea that long-dated prices can shift when the supply-investment regime changes.
Lesson 2: Timespreads Do Not Lie
What the paper claims: timespreads are disciplined by spot-futures arbitrage and therefore reveal near-term physical tightness, at least over the supply-cycle horizon (p. 5). It formalizes the relation as:
S = psi(stocks) - k + F - rS
Timespreads = S - F = psi(stocks) - k - rSHere S is spot price, F is futures price, psi(stocks) is convenience yield, k is storage/insurance cost, and rS is financing cost (p. 5).
My interpretation: "timespreads do not lie" is a useful slogan but too strong literally. A better learning version is: timespreads are hard to ignore because arbitrage and inventory economics constrain them.
Evidence: Exhibit 3 shows a negative relationship between OECD commercial stock deviations and Brent 1m/36m timespreads, with reported R^2 values of 0.62 and 0.66 before and after interest-rate adjustments (p. 6). Exhibit 4 shows a realized-vs-predicted timespread model from roughly 2005-2024 (p. 7).
Lesson 3: A Barrel in the Hand is Worth Two in the Bush
What the paper claims: oil is more of a spot asset than an anticipatory asset. Current supply shocks have a larger price impact than future supply shocks because storage is costly and physical availability matters now (p. 8).
My interpretation: oil does price the future, but it discounts distant future tightness more heavily than equities or some metals because holding physical barrels is costly.
Evidence: Exhibit 5 says oil prices appear about four times more responsive to a current supply shock than to a future OPEC shock (p. 8). Exhibit 6 says metals and equities respond significantly to revisions in more distant GDP growth while oil does not show the same significant response (p. 9).
Verification caution: Exhibit 5 averages 7 current-supply shocks and 128 future-supply shocks (p. 8). That uneven sample makes the exact "four times" number worth treating cautiously.
Lesson 4: Track the Top Projects
What the paper claims: long-dated oil prices should reflect the cost and hurdle rate required to bring the last needed supply project to market. Goldman uses its Top Projects cost curve to track this marginal incentive price (p. 10).
My interpretation: if the cyclical side asks "how full are tanks?", the structural side asks "what project has to be financed next?"
Evidence: The paper says the 2023 GS Top Projects report includes 477 oil projects representing over 50% of global oil and gas supply over the next seven years, with data on capex, costs, production, reserves, technology, returns, and breakeven price (p. 10). Exhibit 7 shows the cost curve steepening and shrinking in 2023 (p. 10).
Verification caution: the dataset is proprietary and not reproduced in the PDF.
Lesson 5: Revenge of the Old Economy
What the paper claims: long periods of low returns and underinvestment in commodity sectors sow the conditions for later tightness, higher marginal costs, and higher prices (pp. 11-13).
My interpretation: commodity cycles are partly capital-stock cycles. Underinvestment can look harmless until demand meets depleted or aging capacity.
Evidence: Exhibit 8 shows the average age of the US mining-sector capital stock rising into the early 2000s and again after the mid-2010s (p. 12). Exhibit 9 shows energy firms' share of top-20 US corporate profitability at 20% in 2012, 5% in 2019, 0% in 2020, and 25% in 2022 (p. 13).
Verification caution: the "revenge" framing is narrative and cyclical. The figures are consistent with it but do not by themselves prove a full causal cycle model.
Lesson 6: Investing Under Uncertainty Boosts Long-Dated Prices
What the paper claims: uncertainty can raise long-dated commodity prices by increasing hurdle rates and delaying investment. The paper applies this first to 2000s supply technology uncertainty and then to decarbonization-driven demand uncertainty (pp. 14-15).
My interpretation: uncertainty can be bullish for long-dated prices if it discourages investment faster than it lowers expected future demand.
Evidence: The paper uses real-options logic on p. 14 and shows in Exhibit 10 that long-run oil-demand forecasts have shifted substantially and now span a wide range across scenarios (p. 15).
Verification caution: the paper acknowledges the direct effect of falling oil demand would be negative, then argues uncertainty can offset it (p. 14). That is plausible but depends on investor behavior, policy, capital access, decline rates, and project lead times.
Lesson 7: Long-Term Shortages Create Near-Term Surpluses
What the paper claims: expectations of long-term shortage can raise long-dated prices and drag spot prices higher. Those higher prices can weaken demand and stimulate near-term supply, creating a near-term surplus (p. 16).
My interpretation: the framework allows the market to look paradoxical: high prices can coexist with inventory builds if the price is being pulled up by long-run investment needs rather than immediate shortage.
Evidence: The paper cites the mid-2000s and 2022 as examples, and estimates that about 80% of the 2022H2-2023H1 selloff reflected policy and market responses to the 2022H1 oil/inflation spike (p. 16).
Verification caution: there is no dedicated exhibit for the 80% attribution on p. 16.
Lesson 8: Cartels Can Control Curve Shape, But Not the Long-Term Price Level
What the paper claims: OPEC can influence inventories and curve shape, but long-dated prices are set by the marginal high-cost producer, which OPEC is not (p. 17).
My interpretation: OPEC has more power over the short end of the curve than over the long-run incentive price.
Evidence: Exhibit 11 presents model estimates for OPEC production-boost decisions. A 10 percentage-point rise in Brent 1m/36m timespreads is associated with a 7 percentage-point higher probability of an OPEC production boost (p. 17).
Verification caution: the table reports ordered multinomial logit estimates but not full model diagnostics, standard errors, or out-of-sample validation.
Lesson 9: Non-OPEC Supply Elasticity Drives the Oil World Order
What the paper claims: OPEC pricing power rises when OPEC market share is high, non-OPEC supply elasticity is low, and global oil demand is inelastic (p. 18). The shale boom increased non-OPEC elasticity and weakened OPEC power; later financial discipline reduced elasticity and restored some OPEC+ power (pp. 18-19).
My interpretation: the "oil order" changes when the competitive fringe changes. Shale mattered not only because it added barrels, but because it changed how quickly supply could respond to price.
Evidence: Exhibit 12 shows US public oil producer reinvestment rates falling to around the 40-60% range since 2021 (p. 19). Exhibit 13 uses a two-stage least squares model to estimate non-OPEC supply elasticity after a 10% real Brent price increase (p. 19).
Verification caution: Exhibit 13 is model-sensitive. Results depend on the demand-shock construction, controls, sample splits, and significance treatment.
Lesson 10: Never Be Afraid to Go Where the BTUs Take You
What the paper claims: commodity markets can balance through substitution across fuels when prices converge on an energy-content basis (pp. 20-21).
My interpretation: substitution is the demand-side counterpart to supply response. When one fuel becomes expensive enough relative to another, users with the right equipment and contracts may switch.
Evidence: The paper discusses feed/fuel substitution in the 2007-2008 food inflation crisis and coal-to-gas switching after US shale gas growth (p. 20). Exhibit 14 charts price ranges for switching between US natural gas, coal, and oil (p. 21).
Verification caution: Exhibit 14 shows price incentives, not actual switching volumes. Volume estimates such as more than 7 Bcf/d of incremental gas demand require separate data (p. 20).
Concept Map
Method Diagram
Evidence Table
| Evidence | What the paper uses it to support | Evidence strength | Caveat |
|---|---|---|---|
| Exhibit 1: forward curve decomposition (p. 3) | Price can be read as marginal cost plus delivery premium/discount. | Conceptually useful. | Schematic only; not empirical magnitude. |
| Exhibit 2: 36-month forward WTI, 1990-2022 (p. 4) | Long-dated oil prices were stable in the 1990s, then shifted in later investment regimes. | Visually supportive. | Does not by itself identify causes. |
| Page 5 equations | Timespreads link to convenience yield, storage/insurance, and rates. | Strong as economic intuition. | psi(stocks) is unobserved and maturity/compounding details are simplified. |
| Exhibit 3: stocks vs Brent 1m/36m timespread (p. 6) | Lower inventories are associated with stronger timespreads. | Quantitative correlation, reported R^2 0.62-0.66. | Correlation/model evidence, not causal proof. |
| Exhibit 4: inventory model vs realized timespreads (p. 7) | Inventory/rates model captures many ups and downs in timespreads. | Useful visual model check. | Footnote says recent residual may reflect Russia risk premium or forward-looking behavior. |
| Exhibit 5: current vs future supply shock (p. 8) | Oil responds more to current supply shocks than future shocks. | Directionally persuasive. | Uneven sample: 7 current shocks vs 128 future shocks; no confidence bands shown. |
| Exhibit 6: GDP revisions and asset moves (p. 9) | Metals/equities are more anticipatory than oil. | Suggestive cross-asset regression evidence. | Missing bars mean insignificant/omitted, not necessarily zero. |
| Exhibit 7: GS Top Projects cost curve (p. 10) | Marginal project economics anchor long-dated price. | Important but proprietary. | Underlying project dataset is not reproduced. |
| Exhibit 8: US mining capital-stock age (p. 12) | Underinvestment/aging capacity preceded old-economy tightness. | Historical support. | "Oldest since late 40s" is hard to verify from chart alone. |
| Exhibit 9: Fortune 500 profit rankings (p. 13) | Energy profitability cycles show old-economy relevance rising/falling. | Clear descriptive evidence. | Depends on ranking methodology and classification choices. |
| Exhibit 10: oil-demand forecast revisions and scenarios (p. 15) | Demand uncertainty is large and can affect investment. | Strong for showing uncertainty. | Scenario selection and vintages matter. |
| Page 16 historical narrative | Long-term shortages can create near-term surpluses. | Conceptually coherent. | The 80% selloff attribution lacks a dedicated exhibit. |
| Exhibit 11: OPEC decision model (p. 17) | Timespreads and stocks help predict OPEC production decisions. | Model-based support. | In-sample table only; no standard errors or out-of-sample validation shown. |
| Exhibits 12-13: reinvestment and supply elasticity (p. 19) | Lower reinvestment and lower non-OPEC elasticity restore OPEC+ pricing power. | Plausible model and descriptive support. | Sensitive to company universe, estimates, model design, and sample periods. |
| Exhibit 14: switching price ranges (p. 21) | Fuel substitution can balance markets when energy-equivalent prices diverge. | Good evidence of price incentives. | Does not show actual switching volumes. |
Claims vs Interpretation
| Topic | What the paper claims | My interpretation |
|---|---|---|
| Timespreads | Timespreads do not lie about fundamentals over the supply-cycle horizon (p. 5). | Timespreads are disciplined by arbitrage and inventory economics, but risk premia, rate changes, and forward-looking behavior can complicate the signal. |
| Spot vs future shocks | Oil is much more responsive to current supply shocks than future supply shocks (p. 8). | This is a useful intuition, but the exact four-times magnitude needs source validation. |
| Marginal costs | Long-dated prices should reflect the cost of bringing on the last needed unit (pp. 3, 10). | This is the best structural anchor, but measuring the marginal project is hard and dataset-dependent. |
| Old economy cycles | Underinvestment in commodities creates later tightness and higher prices (pp. 11-13). | The cycle logic is compelling, but "revenge" is an interpretive narrative rather than a standalone model. |
| Decarbonization | Demand uncertainty can support long-dated oil prices by delaying investment (pp. 14-15). | Possible, but it competes with bearish demand, policy, stranded-asset, and financing effects. |
| OPEC | OPEC controls curve shape more than long-run price level (p. 17). | Useful distinction: OPEC can manage inventory and short-end scarcity better than it can repeal marginal-cost economics. |
| BTU substitution | Fuel substitution can resolve near-term imbalances (pp. 20-21). | Good lens, but actual substitution depends on equipment, contracts, regulation, infrastructure, and dispatch constraints. |
Caveats
- The PDF is a Goldman Sachs research note and tribute to Jeff Currie's tenure, not a neutral academic literature review. It summarizes and celebrates a research tradition.
- The disclosure appendix says the research is based on information Goldman considers reliable but does not represent as accurate or complete, and that opinions, estimates, and forecasts are as of the report date and may change (p. 23).
- The disclosure appendix also states that Goldman has investment banking, trading, and other business relationships across markets, and that firm personnel may hold positions or express views inconsistent with the research (p. 23).
- The report includes investment-risk disclaimers, including that past performance is not a guide to future performance and futures/options can involve substantial risk (p. 24).
- Several exhibits depend on proprietary Goldman data or models: GS Top Projects (p. 10), OPEC decision probabilities (p. 17), producer reinvestment estimates and non-OPEC elasticity (p. 19).
- Some claims are time-sensitive. For example, the note forecasts Brent at $100/bbl in 2024 (p. 19). That should be checked against realized market data before being reused.
- The strongest empirical evidence is observational. The report generally supports "this framework helps explain patterns," not "this variable alone causes the outcome."
Missing Background and Assumptions
To understand the paper well, fill in these background areas:
- Futures curves: contango, backwardation, delivery months, benchmark grades, and why storage links spot and futures prices.
- Inventory economics: convenience yield, storage limits, insurance costs, financing costs, and the role of interest rates.
- Oil benchmarks: Brent, WTI, quality/location differences, and why "oil price" is not one object.
- Project economics: capex, breakeven price, reserves, decline curves, cost of capital, and long-cycle vs short-cycle supply.
- OPEC economics: dominant-producer models, spare capacity, quotas, market share, and competitive fringe supply.
- Shale economics: short-cycle drilling, decline rates, financing conditions, and why shale changed supply elasticity.
- Energy units and substitution: BTU, mmBtu, Bcf/d, mb/d, coal-to-gas switching, and dispatch constraints in power generation.
- Real options: why uncertainty can delay investment even when expected returns look positive.
Follow-Up Reading
Start with these topics, not necessarily these exact sources only:
- A futures-market primer on contango, backwardation, cost of carry, and convenience yield.
- A commodity term-structure or storage-theory treatment, such as the classic theory of storage literature.
- A basic oil-market reference covering Brent, WTI, OPEC, spare capacity, and the Strategic Petroleum Reserve.
- A project-economics primer covering capex, breakeven prices, decline rates, reserves, and cost of capital.
- Dixit and Pindyck, Investment Under Uncertainty, for the real-options logic behind delayed investment.
- IEA and OPEC long-term oil-demand scenario reports, read comparatively because scenario assumptions drive conclusions.
- CFTC Commitments of Traders documentation if you want to understand the positioning variables used in OPEC-decision models.
Open Questions
- Can the inventory/timespread relationship in Exhibits 3-4 be replicated using public data, including the interest-rate adjustment?
- How robust is the "four times more responsive" claim in Exhibit 5 if the supply-shock classification, event windows, or sample periods change?
- What is inside the GS Top Projects dataset, and how sensitive is the marginal-cost curve to project inclusion/exclusion?
- How did the paper attribute 80% of the 2022H2-2023H1 oil selloff to policy and market responses, and what was the counterfactual?
- Does the OPEC production-decision model in Exhibit 11 predict out-of-sample decisions, or mostly fit historical behavior?
- How stable is the non-OPEC supply elasticity estimate in Exhibit 13 after controlling for shale financing conditions, service costs, and geology?
- How much actual fuel switching occurred in the episodes behind Exhibit 14, as opposed to price incentives that were technically available?
- How should the framework update after 2023 given realized 2024-2026 oil prices, OPEC+ behavior, US shale discipline, and energy-transition policy changes?
One-Sentence Takeaway
The concept to learn is not a price forecast; it is a disciplined way to ask whether a commodity price is being driven by physical tightness now, marginal supply investment later, or the messy feedback loop between the two.