How PFX is built

1. What PFX measures

PFX is a family of monthly price indexes for the U.S. 48x40 wood pallet market. Delivered indexes measure what buyers pay per pallet with freight included. Yard indexes measure what suppliers receive at their yard (free on board), the price suppliers set their business by. The headline, the PFX National Pallet Index, is the delivered price of recycled 48x40 pallets by the truckload.

2. Standard specification

• Size: 48x40 stringer pallets (block pallets have their own index).
• Condition: recycled (new is its own spec).
• Lot size: a full truckload; observations under 400 pallets are excluded.
• Delivered basis: delivered to the buyer's dock, freight included, U.S.. Yard basis: free on board (FOB) at the supplier's yard, no freight.
• Heat treatment: observations for heat-treated (ISPM 15) pallets are normalized by removing a $1.25 premium until PFX measures its own heat-treatment adder.
• Grades are mapped from common trade names (#1, A, premium; #2, B, standard; AAA; block). Observations without a clear grade are excluded.

3. Inputs and hierarchy

PFX uses the best available evidence for each value, in this order:

1. Completed transactions: delivered, invoiced or paid loads with price, quantity, grade, date and delivery location, dated to the month the price was agreed (weight 1).
2. Contributor submissions: completed-trade prices reported by buyers and suppliers in the contributor program (weight 0.6; treated as transactions once verified).
3. Firm quotes: supplier quotes for a specific load (weight 0.4).
4. Published offers: publicly posted price ranges (weight 0.25), used only to seed a value before enough data exists.
5. Model carry-forward: when none of the above meets the minimums, the last value is carried forward by the change in the BLS wood pallet price index and flagged Tier C.

Never used: loads that are agreed but not yet delivered, Chesapeake Pallets' own unaccepted quotes, opinions or estimates, forward-looking prices or intentions, and deals whose price was itself set off PFX or another index. Contributor submissions enter a value only when at least 5 distinct contributors report in that cell.

Today the largest source of transactions is Chesapeake Pallets, a national pallet broker that shares owners with Pallet Forecast. Its loads enter PFX through a read-only export with buyer and supplier identities replaced by one-way hashes. The contributor program exists to widen the sample so no single source dominates.

4. Calculation

• Each observation is normalized to the standard specification (section 2).
• Outliers are screened within each month, side and grade: an observation more than 3.5 robust standard deviations from the median (1.4826 × median absolute deviation) is excluded and logged. When more than half the prices are identical, the median absolute deviation is zero, so the screen uses 1.2533 × the mean absolute deviation instead.
• The assessed price is the volume-weighted median of the remaining observations. The published range is the volume-weighted 25th to 75th percentile, the middle half of trades.
• When sources are blended, each observation's weight is its quantity (capped at 2,000 pallets) times its input-quality weight from section 3.
• If the current month is too thin, the window widens to the trailing two, then three, months. The window used is published with every value.

5. Publication thresholds and quality tiers

Counterparties are the distinct buyers (delivered indexes) or supplier yards (yard indexes) on the other side of the trades. The concentration rules stop any single trading relationship from setting the index: Tier A refuses a value when one counterparty is above 25% of volume, and Tier B caps that counterparty's weight at 35%. The share each value's largest counterparty held before capping is recorded in the release file.

Indicative and Benchmark

Every PFX series launches as PFX Indicative. A series becomes a PFX Benchmark only after 6 consecutive months at Tier B or better, with at least 3 data sources independent of Chesapeake Pallets supplying at least 30% of the weight.

Freight

Delivered values are measured as transacted, freight included. A planned refinement (version 1.2) will normalize each load to a standard delivery distance using a fitted freight-versus-miles curve, once enough loads carry the distance data it needs. Until then, regional mix can move the national delivered value, which is one reason Tier A needs several counterparties.

6. Composite, index points and history

• The composite is a fixed-weight blend: Grade A 60% and Grade B 40%. Weights are reviewed each January against the year's traded volume.
• Index points are set so the 2025 = 100 (the 2025 calendar-year average of the composite).
• Assessed months chain from the prior month, so a change in the mix of contributors cannot jump the index.
• Backcast. Months before the first assessed value are linked to the BLS wood pallet producer price index (series WPU084101), scaled to the first PFX level. Backcast history is labeled as such on every chart and in every data file.

7. Regional indexes

Regional delivered indexes publish for Northeast, Southeast, Midwest, South Central, West once a region reaches Tier B on its own data. Until then the region shows “building coverage.”

8. Releases, revisions and the audit trail

• PFX: Monthly at 10:00 a.m. ET on the 2nd business day after BLS publishes that month's Producer Price Index. Every value, free and paid, is released at the same moment.
• Forecast: Updated in the same release, using the BLS data published two business days earlier.
• Every release goes through a QA report, then is approved by a named person, sealed with a SHA-256 hash of its contents and archived.
• Published values are final. Errors are corrected in a dated revision note; the original value stays in the archive.
• Material methodology changes are published for 30 days of public comment before they take effect.
• Any series will get at least 6 months' notice before it is discontinued.
• Before the first approved release, the site shows a provisional value, clearly labeled.

9. Governance, conflicts and confidentiality

Conflict of interest. Pallet Forecast is published by the team behind Chesapeake Pallets, which buys and sells pallets. We manage that conflict with published, code-enforced rules; concentration caps; a disclosed count of contributors and counterparties on every value; a named approver; and an immutable archive. We are working toward an oversight committee with at least one member independent of Chesapeake Pallets, and toward independent review of the methodology.

Confidentiality. No buyer, supplier or contributor is ever identified. Raw submissions are stored privately, hashed and used only in aggregate.

Competition law guardrails. PFX collects completed-trade and firm-quote prices only, never future pricing plans. It publishes aggregates with minimum counterparty and concentration rules and never shares one participant's data with another.

Corrections and complaints. Email editor@palletforecast.com. We acknowledge within two business days and publish any correction as a dated revision note.

10. Forecast methodology

The forecast targets the BLS producer price index for wood pallets and pallet containers (WPU084101), forecasting the log change over each horizon from 1 to 12 months. PFX dollar forecasts apply that percentage path to the current PFX level until PFX has 24 months of assessed history, after which its own pass-through will be estimated.

Data

Models

• Random walk. No change from the latest value. The benchmark every model has to beat.
• Random walk with drift. Extends the average monthly change of the last five years.
• Autoregressive AR(p). Monthly changes explained by their own recent history; lag order chosen by AIC (up to 6).
• Damped-trend exponential smoothing. Holt's linear method with a damped trend on log prices; parameters fit by one-step error.
• Driver regression. Direct multi-horizon ridge regression on pallet momentum, hardwood and softwood lumber, sawmill prices, diesel, truckload freight, manufacturing and warehouse jobs, wood-products wages, and an error-correction term on the pallet-to-inputs price ratio.

Combination and ranges

• The published forecast combines the random walk, AR, exponential smoothing and driver models. Weights are the inverse of each model's mean squared error over its last 60 out-of-sample forecasts at that horizon (Bates-Granger), shrunk halfway toward equal weights, which forecasting competitions show is more robust.
• Prediction ranges are empirical: past ensemble errors at each horizon, scaled by the market's volatility when each forecast was made (24-month standard deviation of monthly changes), then re-scaled to today's volatility. A calibration step then widens a horizon's bands whenever its 80% range covered too few outcomes over the last 36 forecasts, so coverage holds near target.
• The driver regression's ridge penalty is chosen per horizon on 2012 to 2015 only.

Testing

Rolling-origin backtest: from mid-2010, every month the models are refit using only data available then and forecast 1 to 12 months ahead. Accuracy is reported on 2016-01 to 2026-08 (128 origins), a window not used to tune anything. The ensemble is clearly better than a no-change forecast at 1 to 6 months (Theil's U 0.73 at 1 month, 0.94 at 6) and modestly better beyond (0.92 at 12 months). It calls the 3-month direction right 80% of the time.

Limits, stated plainly. In the test window the 80% ranges contained the outcome 75% of the time at 1 month and 85% at 12 months. Lumber increases have historically passed into pallet prices much more than lumber declines, which a linear model cannot fully capture. No model foresees a shock before it reaches the data. These numbers are a backtest; live forecasts are archived monthly from October 2026 so the live record can be judged on its own.

Full scorecard and current forecast: the forecast page.

11. Data sources and notices

• U.S. Bureau of Labor Statistics, Producer Price Index and Current Employment Statistics, via the BLS public API. Pallet Forecast is not affiliated with or endorsed by BLS. BLS.gov cannot vouch for the data or analyses derived from these data after the data have been retrieved from BLS.gov.
• Transaction and quote data from Chesapeake Pallets (disclosed above) and from contributors under the contributor terms.
• PFX values are information, not financial advice, and not an offer to buy or sell pallets.

12. Version history

• 1.2 (2026-10-21): adopted before first release, so no consultation period applied. Tier B now caps any one counterparty at 35% of the weight instead of refusing the value; Tier A is unchanged. The outlier screen falls back to the mean absolute deviation when the median absolute deviation is zero, so ordinary prices next to a cluster of identical prices are no longer excluded. Published offer ranges are used only to seed a value, as stated in section 3, and never enter an assessed value.
• 1.1 (2026-10-21): adopted before first release after an external review of benchmark practice: no Tier A on a single data source; counterparty caps of 25% (Tier A) and 35% (Tier B); contributor cells need 5 contributors; completed loads only; Indicative and Benchmark designations; one monthly release keyed to the BLS calendar; forecast ensemble weights shrunk toward equal and adaptive interval calibration.
• 1.0 (2026-10-21): first draft methodology (never used for a release).