What drove the 2020-2022 lumber price spike? Event regressions point to COVID and the 2021 British Columbia floods

Introduction

Between 2020 and 2022 the price of softwood lumber in North America rose by more than 300 percent, according to the paper's abstract, and then fell back. For a sawmill the swing was a windfall followed by a reversal, and it raised the question of what actually drove it. Four economists from the Department of Economics at the University of Victoria in British Columbia (Rebecca Zanello, Yin Shi, Atefeh Zeinolebadi and G. Cornelis van Kooten) tackle that question in a short article published on 13 January 2023 in the journal Forests, in the special issue "Forest Product Markets, Sustainability, and Societal Impacts".

The paper sets the scene with Canada's stake in the market: it cites van Kooten and Schmitz for a 7.2 percent share of national exports, CAD 1.5 billion of government revenue and a CAD 25 billion contribution (1.6 percent of GDP) from the softwood sector, and notes that more than half of Canadian softwood output is exported to the United States. It also recalls that the United States imposed anti-dumping (AD) and countervailing (CVD) duties on Canadian lumber repeatedly in the four decades after 1981, and that the Softwood Lumber Agreements (SLAs) regulated trade for parts of that period.

The study asks two things. First, can a calibrated random-walk style price model reproduce the volatility seen after COVID-19? Second, if not, which identifiable events (the pandemic, a flood that disrupted rail and road links from British Columbia, trade duties, wages, the exchange rate, housing starts) move the lumber price in a regression? The authors work with a composite lumber price index and a U.S. price series, measured in Canadian dollars.

Research methodology

Stochastic process. Lumber prices were modelled as a mean-reverting Ornstein-Uhlenbeck process, estimated by regressing monthly price on the previous month's price. The sample covered January 2016 to April 2022 for the composite index (75 observations) and January 2016 to March 2022 for the U.S. price (74 observations). Fit statistics were R-squared 0.8039 for the composite and 0.7317 for the U.S. series, with monthly standard deviations of 13.41 and 16.00. The authors then simulated 100-month price paths to see whether such a process could look like the observed series.

Price impact (event) models. Four reduced-form regressions explain the logarithm of the monthly composite price index. Models 1 and 2 use the full sample, January 1981 to October 2022 (501 observations); Models 3 and 4 use January 2001 to October 2022 (255 observations) and add the average hourly wage in forestry and logging. Models 2 and 4 add the combined AD/CVD rate compiled by the authors.

Variables. Composite lumber price (from Random Lengths), the Canadian dollar per U.S. dollar exchange rate, U.S. housing starts, AD/CVD rates, wages (Statistics Canada), and dummy variables for periods of the softwood dispute and trade regimes (for example the Memorandum of Understanding period 1987 to 1991, the tariff-rate-quota period, the 2006 SLA period and the post-SLA period starting January 2017), for the Great Financial Crisis (December 2007 to June 2009), for Covid (January 2020 to January 2022) and for the Pacific Northwest floods of November 2021 (PNF).

Appendix. A repeat of the regressions using a spruce-pine-fir (SPF) price index appears in Table A1.

Key findings

The price movements are not just random noise. The calibrated stochastic process gave a mean-reverting estimate, but the simulated paths did not by themselves explain the volatility, and the authors say they cannot rule out pure randomness. They then conclude from the regressions that price movements are not purely random.

Events matter. In the log-price regressions, the Covid dummy was positive and significant in all four models, and the flood dummy was positive and larger. The paper translates the coefficients into percentage effects:

• Covid: about 42.1 percent (1981 sample without AD/CVD), 38.5 percent (with AD/CVD), 32.7 percent (2001 sample without) and 33.1 percent (with); the text gives a range of roughly 32 to 42 percent.
• November 2021 Pacific Northwest floods: about 87.0 percent (1981 sample), 81.1 percent (with AD/CVD), 55.7 and 60.3 percent (2001 samples).
• The Great Financial Crisis had a negative coefficient in the tables, but the text reports its effect as 30.1 and 18.5 percent.

The abstract states that COVID-19 produced a price jump, but a smaller one than the rainfall event that restricted imports from Canada.

Trade duties show little effect on U.S. prices. Although the AD/CVD coefficient was statistically significant and small and negative in the tables (minus 0.013 in the 1981 sample and minus 0.008 in the 2001 sample), the abstract states that import duties and other trade actions had no discernible effect on U.S. prices. The portal notes the difference between a significant but tiny coefficient and the authors' headline wording, and has not reconciled them.

Other drivers. The exchange rate and housing starts were positive and significant in all models, and the wage variable was positive and significant in the 2001 models. Housing starts fell from 2005, but the authors say this had little price impact relative to the pandemic.

Explanation offered. The flood is described as an inward supply shift through transport and supply-chain disruption; the Council of Forest Industries is cited for supply chains remaining constrained until June 2022. For the pandemic the authors say the price effect is less clear and mention possible labour-supply reductions, demand increases from remodelling and home working, and rail and sawmill capacity constraints. They suggest the flood may explain why prices did not return to pre-pandemic levels.

Practical applications for sawmills

This section is the portal's interpretation and not a statement from the paper.

• Logistics are a price risk. The finding that a flood affecting British Columbia transport links moved prices more than the pandemic suggests that mills in export-dependent regions should treat rail and road resilience as part of commercial planning, for example by holding alternative shipping options or flexible inventory.
• Do not plan on a single price path. The authors could not show that a mean-reverting model alone explains the spike, which suggests that budgets and investment cases for a mill should use scenario ranges rather than a single forecast.
• Treat trade duty headlines with care. The paper reports that duties had no discernible effect on U.S. prices, implying that the cost of duties may have been borne by producers rather than passed on; this is the portal's reading, not stated in the text I read.
• Know your exposure. Exchange-rate sensitivity was significant in every model; Canadian mills selling in U.S. dollars and paying costs in Canadian dollars should model currency alongside lumber price.
• Wage costs. The positive wage coefficient reflects the price series rather than mill-level costs, so it should not be used as a cost estimate for a particular mill.

Limitations

• The authors cite data availability and the need for more granular regressors as limitations, and propose further research on export supply chains.
• The page contains several inconsistencies in dates and signs: sample dates run to October 2022 in the methods but the figures end in July 2022; the flood dummy period is listed in Table 3 in a way that conflicts with the November 2021 event; the Covid dummy ends in January 2022 while the discussion refers to the period since March 2020; the exchange-rate coefficient is positive although the authors expected it to be negative; and the Great Financial Crisis coefficients are negative in the table but presented as positive percentages in the text.
• The slope of 0.9072 in the stochastic model is positive, which the portal flagged as unusual against the stated mean-reversion formula; this was not resolved.
• The text states that the flood effect was "roughly 2.5 times" the Covid effect, but the ratios computed from the reported percentages are about 2.1 (1981 sample) and 2.5 (2001 sample with AD/CVD).
• A table note appears to give a significance level that may be a typographical error.
• Figures, diagnostic tests and the construction of the AD/CVD series were not visible in the page text used.
• This is a market-economics paper: it does not use mill-level data.
• The portal read the article through a summarised fetch, not the PDF.

Conclusions

The paper argues that the 2020 to 2022 lumber price surge cannot be dismissed as random volatility. Event regressions covering back to 1981 attribute large positive price effects to Covid (roughly 32 to 42 percent in the authors' estimates) and an even larger effect to the November 2021 floods that disrupted transport from British Columbia, while finding no discernible effect from import duties on U.S. prices. The study is concise, relies on a composite price index and event dummies, and contains several date and sign inconsistencies that a careful reader should check against the original. For sawmill managers its lasting message is that supply-chain and transport shocks, not only demand, can move lumber prices sharply.

Original research

• Authors: Rebecca Zanello, Yin Shi, Atefeh Zeinolebadi, G. Cornelis van Kooten
• Title: COVID-19 and the Mystery of Lumber Price Movements
• Journal: Forests, Volume 14, Issue 1, Article 152
• Year: 2023 (published 13 January 2023)
• DOI: 10.3390/f14010152
• Link: https://www.mdpi.com/1999-4907/14/1/152
• Affiliation: Department of Economics, University of Victoria, Victoria, British Columbia, Canada (all authors)
• Funding: none declared (no external funding)
• Licence: Open access, CC BY; footer links to CC BY 4.0 while page metadata lists CC BY 3.0

This article is an independent summary prepared by the portal; the original paper should be consulted for full methods, tables and data.