Overview

The Federal Reserve evidence treats futures performance as conditional, not universal. Against a random-walk forecast, futures generally outperform when more than five percent from spot, with median relative RMSE of 0.70 far below spot and 0.74 far above, versus 0.97 and 0.96 for moderate deviations. Futures materially outperform the specified momentum model across reported commodities and maturities, while gains over mean reversion are smaller and less uniform. Wheat and many metal comparisons are exceptions, and sparse cells, benchmark choice, and full-sample mean-reversion estimates limit the conclusion.

Learning outcomes

Large futures-to-spot deviations often carry more relative information than modest ones against a random-walk baseline, but they are not directional promises.
Futures broadly outperform the specified momentum forecast, while their advantage over mean reversion is smaller and more heterogeneous.
Wheat exceptions, sparse metal states, benchmark assumptions, and medians prevent a universal ranking of futures.

Keywords

RMSEdeviationmomentumreversionwheatmetals

When the Curve Helps—and When It Does Not

The empirical question is now specific: when a commodity futures price is compared with a later spot outcome, does it improve on a named backward-looking forecast? The Federal Reserve evidence answers that question conditionally, across commodities, maturities, and different distances between futures and spot. The useful judgment is not to crown futures a universal winner, but to identify the benchmark, the curve state, and the exceptions attached to each result.

A relative RMSE below one means that futures had a lower root-mean-squared error than the selected benchmark in the stated comparison. It does not mean that the futures price was close to the later spot price in every observation. The same contract can beat one benchmark and lose to another, and a group-level median can conceal individual episodes. Reading the result requires keeping its denominator in view.

flowchart LR F["Futures forecast"] --> R["Random-walk comparison"] F --> M["Momentum comparison"] F --> V["Mean-reversion comparison"] R --> E["Relative RMSE"] M --> E V --> E E --> C["Below 1 favors futures"] C --> H["Conditional on commodity, maturity, and curve state"]

The three branches are separate comparisons, not three votes on one universal ranking. Each branch has its own assumption about how recent prices behave. The final box is deliberately conditional: even when futures are below one against several baselines, the result belongs to the stated commodity, horizon, and futures-to-spot state. It is evidence of incremental performance under a defined test, not a certificate that the quote will equal the later spot price.

The advantage over a no-change forecast grows at the edges

The comparison with a random-walk forecast provides the clearest first pattern. A random walk uses today’s spot price as the forecast of the future spot price. It is not trying to interpret the curve, infer a supply shock, or extrapolate a recent trend. For futures to beat it, the contract price must contain information that improves on simply carrying the current spot price forward.

The Federal Reserve note reports that when futures differ from spot by more than five percent, futures outperform the random-walk forecast for most commodity maturities. The reported advantage generally becomes larger as the deviation grows. In the table’s four futures-to-spot groups, the median relative RMSE is 0.70 in the far-below-spot group and 0.74 in the far-above-spot group. In the two moderate-deviation groups, the corresponding medians are 0.97 and 0.96. Because values below one favor futures, the pattern says that the largest departures from spot often carry more incremental forecasting information than modest departures.

That is an important result, but it is not a directional rule. The far-below group combines observations in which futures stand substantially below spot; it does not establish that spot will fall until it reaches the futures price. The far-above group similarly indicates a relative forecasting advantage in the reported comparison, not that spot must rise to meet the contract. The table is about error relative to a no-change alternative, not about the sign of a future price movement.

The pattern also varies by commodity and maturity.

Published by DeepMox

FAQ

Why does futures performance improve at larger deviations from spot?

In the reported sample, larger deviations often coincide with more incremental information relative to a random-walk baseline. The relationship is conditional and does not show that the deviation causes better forecasts or predicts a direction.

Why is momentum a weak comparator when shocks reverse?

Momentum can fail when a recent shock does not persist. Temporary disruptions, structural changes, or offsetting forces can interrupt the trend that a momentum forecast extends.

What do the wheat and metals exceptions show?

They show that the cross-commodity pattern is not universal: wheat often favors a random walk or mean reversion in reported cells, while metals have few large-deviation observations and often favor mean reversion.

Previous9 minHow to Tell Whether Futures Add Information

Next9 minBetter Than the Baseline Does Not Mean Accurate