What works in demand forecasting for wood products?

The best recent review says it depends on the horizon. A December 2024 review of 75 studies found "time series forecasts are likely to be more reliable over the short term where historical patterns are more likely to persist", while structural models and scenarios suit long-run questions. Economic drivers help only if they are easier to project than demand itself. Adding one that isn't "could lead to a less reliable projection than simply allowing a time trend". And long-range projections spread wide: six published projections of global industrial roundwood demand are 44% apart by 2050.

What the usual answer says

The usual answer is to combine historical sales with leading indicators such as housing starts, interest rates and construction spending, using statistical or machine learning models. The review agrees with the ingredients and adds a test for each indicator: is it easier to know than the demand it is meant to explain?

What the 2024 review found

The Centre for International Economics prepared the review for a government agency's outlook model. It screened the literature on wood product demand, kept studies published from 2014, and included 75. Most are national or global projections, not firm-level forecasts.

Its central warning is about drivers. "There is little value in putting in drivers" that are "subject to as much or greater projection difficulties as demand", it says, "as these drivers will also have to be projected". Where a driver's own forecast is highly uncertain, using it can do worse than a simple time trend. The review does not count how often that happens. It is a caution about method.

On horizon, it separates time series methods for the short term from structural models, built on economic relationships, for long-run questions and scenario testing.

On uncertainty, it compares six published projections of global industrial roundwood demand, most built from the same historical data. The summary says the upper projection is "13 per cent greater than the lower projection in 2030 and 44 per cent greater in 2050". A later section puts the 2030 gap at "around 12 per cent". The 2050 figure is the same in both.

On price, it finds demand "generally inelastic with most elasticity estimates less than one in absolute value", and "more responsive to changes in income than changes in own price". It also warns of structural breaks: "the development of new building products could change the relationship between dwelling construction and wood demand".

What recent forecasting studies add

Two 2025 studies from outside timber touch the short term. Saarinen, Huttunen and Rehman, studying consumer goods, found that feeding retailers' point of sale data into four manufacturers' forecasts "generally does not yield" a measurable gain in accuracy. The value came instead from "planning alignment". A July 2025 preprint by Yang, Cao and Liu, tested on retail sales data, got part of its gains by training separate models for different levels of the product and store hierarchy, "(e.g., store, category, department) to capture localized patterns".

Reading it through a data lens

The review's driver test translates into something a planner can run. In our reading, a driver passes most easily when it is already published before the demand it leads, such as building approvals issued last quarter. A driver that has to be forecast first, such as next year's housing starts, brings its own error with it.

The check is a back-test. Rebuild the last two years of forecasts twice, using only data available at each point: once from order history alone, once with the candidate driver. Keep the driver if it lowered the error, and drop it if it didn't, however sensible it sounds. Beyond a year or two, the review's spread of projections argues for ranges rather than single numbers.

The strongest objection

For many mills and distributors, the hard part is price and market cycles, not the choice of method. Order history is itself shaped by price expectations and dealer stocking, so a model trained on it can carry the last cycle's swings into the next forecast. The review's finding that demand responds more to income than to price describes national volumes over years. It says less about a distributor's monthly orders in a volatile market.

When it doesn't apply

Made-to-order businesses whose order book covers the planning horizon need little forecasting for that horizon. And the evidence is indirect. The 2024 review deals mostly with national and global projections, and the two 2025 studies come from consumer goods and retail.

Quarri for wood products is built around how a wood products plant runs, where the order book meets real capacity.

Quarri is an AI-native data platform for the timber supply chain. It connects buying, production, sales and inventory for forest management, sawmill, wood products and pulp, paper and packaging operators.