📈 economics

Carry Incentives and Net Warrant Accumulation: Evidence from SHFE Aluminum

This paper analyzes daily SHFE aluminum data from 2015 to 2025 to determine whether futures slopes exceeding financing-and-storage benchmarks are associated with increased net warrant accumulation, finding a positive but statistically imprecise link that clarifies the relationship between carry conditions and inventory dynamics while acknowledging limitations in establishing causality.

Original authors: Yuchen Han

Published 2026-10-20

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Yuchen Han

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the world of commodities, metals like aluminum do not just sit in warehouses; they move through a complex financial system where physical goods and paper contracts are deeply intertwined. When a trader buys a futures contract, they are essentially making a promise to take delivery of a specific amount of metal at a future date. To make this system work, exchanges maintain a registry of "warrants," which are legal certificates proving that a certain amount of metal is sitting in an approved warehouse and is ready to be delivered. The decision to register metal for delivery, or to cancel that registration, depends heavily on the cost of holding that metal over time. This cost, known as the "carry," includes the price of borrowing money to buy the metal and the fees for storing it in a warehouse. If the market price for future delivery is high enough to cover these costs and still leave a profit, traders have a financial incentive to keep the metal registered and waiting for delivery. This creates a natural question: when the market offers a strong financial reward for holding metal, does the total number of registered certificates actually go up?

A researcher at the City University of Macau set out to answer this question by looking at the aluminum market in China, specifically on the Shanghai Futures Exchange. The study examined daily data spanning a decade, from 2015 to 2025, to see if there was a clear link between the profitability of holding aluminum and the subsequent changes in the number of registered warrants. The researcher compared days when the market conditions suggested a profit from holding metal against days when they did not. The core of the investigation was to see if, on the days following a profitable signal, the net number of warrants—meaning new registrations minus cancellations—increased more than usual.

The analysis revealed a distinct pattern. On average, the net number of warrants increased by approximately 1,380 tonnes in the trading sessions that followed days when the market conditions were favorable for carrying aluminum. This suggests that when the financial incentives align, the total stock of registered metal tends to grow. However, the picture becomes less sharp when other factors are taken into account. When the researcher adjusted the data to account for recent trends in warrant changes, the specific year, and the timing of delivery schedules, the statistical certainty of this result weakened. While the direction of the relationship remained positive, the data no longer supported a strong, definitive conclusion that the market conditions alone were driving the change with high precision.

The study also looked for a specific "tipping point" where traders might suddenly rush to register metal the moment the market became profitable. It searched for a sudden jump in activity right at the moment the cost of holding metal crossed the line into profitability. The evidence for such a sharp, immediate reaction was inconclusive. In some tests, the data showed a large jump, but in others, the signal was too faint to be sure. This indicates that the market's response is likely a gradual adjustment rather than a sudden, discrete event triggered by a single price threshold. Furthermore, the data used in the study only showed the net result of all activity—the difference between new certificates created and old ones cancelled. It could not reveal the total volume of metal moving in or out of warehouses, nor could it prove that the price conditions caused the change, only that they were associated with it.

Ultimately, the research clarifies how the broad state of the market relates to the accumulation of registered aluminum, while acknowledging the limits of what can be seen from public records. The findings confirm that favorable carrying conditions are associated with a higher net accumulation of warrants, but they stop short of proving that these conditions directly cause traders to register more metal. The study highlights that while the financial logic of holding metal is clear, the actual behavior of the market is influenced by many overlapping factors, including recent trends and seasonal timing, making it difficult to isolate a single cause-and-effect relationship. The work serves as a careful map of the connection between market incentives and inventory levels, distinguishing between a general trend and a specific, predictable reaction.

Technical Summary: Carry Incentives and Net Warrant Accumulation in SHFE Aluminum

Problem and Research Question
This study investigates the relationship between carrying conditions in commodity futures markets and the subsequent net accumulation of registered warrants in the Shanghai Futures Exchange (SHFE) aluminum market. While exchange-registered inventories link physical stocks to the delivery system, the decision to maintain a legal claim on approved metal (a warrant) depends on the return to carrying commodities, existing delivery obligations, and inventory management. The central question is whether stronger futures curve carrying incentives are associated with a faster subsequent accumulation of net registered warrants. The paper distinguishes this inquiry from existing literature that focuses on stock-price relationships or gross inventory flows, aiming instead to clarify how daily net warrant changes respond to carry conditions.

Methodology and Data
The analysis utilizes daily data from January 5, 2015, to April 25, 2025, covering 2,506 trading sessions.

Data Sources: The study combines SHFE contract-level settlements, daily aluminum warrant totals (retrieved via AKShare), and weekly warehouse reports. Financing costs are proxied using the one-month Shanghai Interbank Offered Rate (SHIBOR), and storage costs are assumed at 0.45 CNY per tonne per day.
Key Variables:
- Outcome: The net change in registered warrants ( $\Delta W_t$ ), defined as new registrations minus cancellations. The data does not separate gross registrations from cancellations.
- Carry Incentive: A fitted futures curve slope ( $b_t$ ) derived from an OLS regression of contract settlements against remaining maturity.
- Benchmark: A financing-and-storage cost proxy ( $c_t$ ) calculated as the sum of financing costs (based on SHIBOR) and storage fees.
- State Variable: A binary indicator ( $D_{t-1}$ ) equal to 1 if the fitted slope exceeds the benchmark ( $b_t > c_t$ ) on the previous session, and 0 otherwise.
Empirical Strategy:
- State Comparison: The primary specification compares the mean net warrant change following "above-benchmark" sessions against those following "below-benchmark" sessions using a lagged state variable.
- Controls: The "primary adjusted specification" includes lagged net warrant changes, year fixed effects, calendar-month fixed effects, and fixed effects for the number of calendar days remaining until the nearby contract's last trading date.
- Inference: Given the limited number of year clusters (11), the study employs cluster-robust standard errors (CRV1) and a two-sided restricted wild cluster bootstrap-t test (Webb's six-point weights) to ensure valid inference.
- Local Diagnostics: To test for discrete responses at the benchmark threshold, the study estimates piecewise regressions and analyzes "crossing events" (the first signal after five consecutive sessions below the benchmark).

Key Results

Unadjusted Association: In the unadjusted comparison, net warrant changes are approximately 1,380 tonnes higher following sessions where the futures slope exceeds the financing-and-storage benchmark compared to sessions where it does not. This result is statistically significant (Webb p-value = 0.008).
Impact of Controls: When the model includes lagged net changes, year fixed effects, and delivery-calendar controls simultaneously, the estimated coefficient decreases to 818 tonnes (primary adjusted specification). While the sign remains positive, statistical precision weakens (Webb p-value = 0.091).
Local and Event Analysis:
- Local Regressions: Piecewise regressions around the zero-gap threshold yield imprecise estimates of a discrete "jump" in warrant accumulation. For instance, a full-sample piecewise regression suggests a jump of 1,479 tonnes (p-value = 0.080), but wider windows show varying magnitudes and lower significance.
- Crossing Events: An analysis of 19 specific crossing events (where the curve moves from below to above the benchmark) shows a five-session cumulative difference of 3,737 tonnes, but this estimate is highly imprecise (p-value = 0.418).
Sensitivity: Excluding the 2017–2018 period (which contains the majority of above-benchmark observations) yields a larger positive difference (1,950 tonnes), suggesting the results are sensitive to the concentration of observations in specific years.

Limitations and Scope of Claims
The paper explicitly limits its causal and arbitrage interpretations due to several factors:

Data Aggregation: Daily warrant totals measure the net balance of registrations and cancellations. They do not reveal gross flows, physical warehouse inflows/outflows, or whether metal physically leaves storage upon delivery.
Cost Omissions: The benchmark is a proxy that omits specific financing tenors, margin funding, delivery/inspection charges, tax conditions, and transaction costs. Consequently, the study does not claim to identify an executable arbitrage bound or a specific holder's profit.
Causality: The design does not establish a causal mechanism. Common market conditions (e.g., demand shocks, logistical constraints) may influence both the futures curve and warrant accumulation simultaneously. Reverse regressions relating subsequent slope changes to lagged warrant levels were found to be imprecise.
Threshold Validity: The benchmark is researcher-constructed; crossing it is not an externally assigned treatment. Therefore, the study does not claim to identify a sharp, discrete registration trigger at the specific cost estimate.

Significance and Contribution
The paper's primary contribution is to distinguish a broad market-state association from claims about a discrete registration threshold. It provides empirical evidence that, in the SHFE aluminum delivery system, periods of favorable carry conditions (where the futures slope exceeds a financing-and-storage benchmark) are associated with higher subsequent net warrant accumulation. However, the study clarifies that this association is a conditional inventory-adjustment relationship rather than proof of a specific arbitrage mechanism or a discrete behavioral response at a precise cost level. The findings highlight the utility of state comparisons in understanding inventory dynamics while acknowledging the limitations of public aggregate data in isolating gross flows or causal arbitrage returns.

Technical Summary: Carry Incentives and Net Warrant Accumulation in SHFE Aluminum

More like this