Determinants of CETA preferential rate utilization (2018-2024): A France-Canada comparative analysis

ISBN: 978-0-662-74062-9

July 2026

Abstract

Eight years after its entry into force, the utilization of tariff elimination offered under the Comprehensive Economic and Trade Agreement (CETA) to French and Canadian exporters is satisfactory, but there is still room for improvement. This study aims to move beyond analyzing trends in the preference utilization rate of CETA provisional since its provisional entry into force in September 2017 by identifying and quantifying the incentives and operational obstacles involved in its utilization by French and Canadian exporters.

Drawing on a novel bilateral panel econometric analysis (2018–2024) with high granularity (HS8 level), this study models the CETA preference utilization rate by incorporating variables related to the complexity of rules of origin and the structure of supply chains. The analysis shows that a single factor cannot explain the failure to claim CETA’s tariff elimination.

A key common barrier to the utilization of tariff preferences is the transit of goods through third countries outside the European Union (notably, the United States), which complicates compliance with the agreement’s direct transport rule. This indirect trade results from a two-fold dynamic: (1) operators acting on cost–benefit considerations and trade-offs between logistical gains and tariff advantages, which favours transit hubs outside the EU; and (2) lack of coordination stemming from insufficient information about the direct transport rule among exporters, importers, freight forwarders and customs brokers, which makes it more difficult for French and Canadian operators to obtain the preferential rate.

Beyond this shared barrier, the study highlights additional determinants specific to each exporter ecosystem. French exporters are less likely to claim CETA’s preferential rate when their exports are more concentrated in Canada’s Western provinces and when the goods involved are consumer goods. Meanwhile, Canadian exporters are more constrained by the complexity of the rules of origin and by the need to reach a critical mass of exports in value in order to benefit from preferential tariffs. Air freight is also associated with the under-utilization of tariff preferences.

This variation in findings suggests that policies to support the utilization of CETA preferential tariffs would benefit from moving beyond a generic promotion of the agreement and instead adopting a differentiated approach tailored to exporters’ profiles. While logistical compliance and coordination within value chains represent shared challenges, origin traceability for French exporters of consumer goods and the clarity of the rules of origin for Canadian exporters call for distinct policy responses.

Table of contents

1. Introduction and literature review
   1. 1.1 The key determinants of preferential tariff utilization
   2. 1.2 An important obstacle: the degree of restrictiveness of preferential rules of origin
   3. 1.3 Beyond the traditional determinants: logistics and global value chains
2. Data, variables, econometric strategy and descriptive statistics
   1. 2.1 Assembling the analysis panels: a granular bilateral France-Canada analytical framework
   2. 2.2 Variables and hypotheses tested in the models: quantifying the incentives and barriers to utilization of CETA’s preferential tariff
   3. 2.3 Economic modelling strategy
   4. 2.4 Descriptive statistics and stylized facts: ex-ante heterogeneous constraints
3. Results of the econometric analysis
   1. 3.1 Determinants of CETA preferential rate utilization for French exports to Canada: the predominant role of logistical constraints
   2. 3.2 A parallel analysis of the determinants of CETA preferential rate utilization for Canadian exports to France
   3. 3.3 Comparative overview: the shared logistical challenge of indirect trade and distinct trade-off strategies
4. Conclusion
5. References
6. Technical appendix
   1. Methodological appendix A – construction of the panel variables
   2. Appendix B – multicollinearity analysis
   3. Appendix C: robustness tests

1. Introduction and literature review

The cornerstone of free trade agreements (FTA) is the granting of tariff preferences, designed to stimulate trade in products between member countries by reducing or even eliminating tariffs. However, a corpus of economic studies has identified a persistent issue: businesses make only partial use of these preferences. This phenomenon, known as the under‑utilization of tariff preferences, means that substantial volumes of eligible trade continue to be subject to Most‑Favoured‑Nation (MFN) tariffs, thereby reducing the tariff advantages expected from FTAs.

An exporting business’s decision to claim or forgo preferential treatment is partly the result of an economic trade‑off: the monetary gains from tariff exemption must outweigh the associated compliance and administrative costs (Cadot et al., 2006). Understanding the factors that shape this trade‑off is crucial for the design and evaluation of trade policies. As a first step, this analysis begins with a review of the literature synthesizing academic work on the determinants of preference utilization rate (PUR) to situate this research within the existing literature. The research is structured in three parts: an examination of the established key determinants; an analysis of the major barrier posed by rules of origin (ROO); and finally, an exploration of more recent research avenues that motivate our agreement‑specific approach to the EU–Canada trade agreement (CETA).

1.1 The key determinants of preferential tariff utilization

The economic literature broadly agrees on a set of determinants that form the foundation of PUR analyses.

• Preferential margin (PM): the most direct determinant is the preferential margin, defined as the difference between the MFN tariff and the preferential tariff (which is zero under CETA). A high PM represents a larger potential saving and thus a stronger financial incentive to incur compliance costs. Empirical studies almost unanimously confirm a positive and statistically significant relationship between the size of the PM and the utilization rate (Keck & Lendle, 2012). When the margin is small, utilization rates are also low.
• Transaction and business characteristics: compliance costs (documentation, certification, auditing) include a substantial fixed component, meaning their relative weight decreases as export value rises. The literature consistently shows that PURs are positively correlated with export value. Larger businesses—benefiting from dedicated resources and exporting higher volumes—are better able to absorb these fixed costs than SMEs, which exhibit significantly lower PURs (Hayakawa, Kim, & Lee, 2014).

1.2 An important obstacle: the degree of restrictiveness of preferential rules of origin

If the preferential margin represents the benefit of using a trade agreement, the rules of origin (ROO) represent its principal cost. ROO are the criteria used to determine whether a product can be considered as “originating” from a country and thus eligible for the preferential tariff. Their purpose is to prevent simple trade diversion; however, excessively complex or restrictive ROO can function as a non‑tariff barrier, offsetting the gains from tariff liberalization (Krishna & Krueger, 1995).

Researchers have developed composite indices to measure the restrictiveness of preferential rules of origin. The Australian Productivity Commission’s approach (Gretton & Gali, 2005) and the work of Estevadeordal and Suominen (2003) involve coding each type of rule for each product and assigning it a restrictiveness score on an ordered scale. For example, a change in tariff classification (CTC) at the chapter level (two-digit HS code) is more restrictive than a CTC at the subheading level (six-digit code). More recent studies, such as those by Ayele (2024) and Ayele et al. (2023), have refined these indices to account for the flexibilities offered by modern agreements (choice between multiple rules, tolerance clauses, cumulation), which reduce restrictiveness. Studies using such indices consistently find a negative and significant effect of the restrictiveness of ROO on PUR.

1.3 Beyond the traditional determinants: logistics and global value chains

CETA, which has been provisionally applied since 2017, is a “new‑generation” agreement that includes ambitious provisions on ROO (self‑certification, cumulation). While official reports show rising utilization rates since 2018, they remain suboptimal^{Footnote 1}, and no study appears to have yet provided an exhaustive, product‑level (HS8) analysis of the determinants of CETA’s PUR—particularly one that incorporates variables capturing the complexity of modern trade.

Recent literature suggests that traditional models overlook crucial dimensions. The overlapping of businesses within global value chains (GVCs) makes tracing the origin of multiple inputs particularly difficult, and this may hinder PUR. Logistical constraints are also largely absent from existing analyses. The plausible economic hypothesis is that for urgent or high‑value goods, the need for rapid delivery (air freight) or the complexity of reaching distant final destinations within a large territory (intra‑national geography) can make the opportunity cost of a potential customs delay exceed the tariff saving. Lastly, the direct transport rule, a logistical constraint often overlooked in analyses, can prove to be a major barrier in a world of global logistics hubs.

This research addresses precisely what has been overlooked. While the traditional determinants are well understood, their interaction with the operational realities of modern supply chains under CETA remains unknown. The aim is to identify which determinants account for the under‑utilization of CETA tariff preferences by French and Canadian exporters.

To address this question, two product‑level econometric models (HS8) were constructed for the 2018–2024 period—one panel for French exports and one panel for Canadian exports—explaining PUR as a function of the preferential margin, a CETA‑specific restrictiveness index for rules of origin inspired by the methodologies of Ayele (2024), Ayele et al. (2023), and Gretton & Gali (2005), and variables capturing these logistical dimensions. The comparative analysis of French and Canadian trade flows will help distinguish the operational challenges inherent to CETA from the specific features of each export ecosystem.

2. Data, variables, econometric strategy and descriptive statistics

2.1 Assembling the analysis panels: a granular bilateral France-Canada analytical framework

The analysis utilizes two annual, symmetric panel datasets covering the 2018–2024 period: one for French exports to Canada and the other for Canadian exports to France. The unit of observation is the product at the 8‑digit level of the tariff nomenclature (HS8), which provides a finer level of granularity than at the HS6 level.

The construction of this database required substantial work in collecting, matching, and harmonizing information from multiple sources (Global Affairs Canada, Statistics Canada, Eurostat (Comext), the International Trade Centre, UN Comtrade, and the UN Statistics Division). One key methodological challenge was ensuring temporal consistency in the unit of observation. Because tariff classifications evolve over time, it was necessary to build a concordance table to convert all data into the 2024 HS8 nomenclature, thereby ensuring comparability across years. To avoid bias from non‑commercial flows or small‑number effects, the analytical sample was cleaned by removing observations in which the annual value of eligible exports is below Can$1,500 (approximately €1,000). The sample was therefore restricted to cases with a strictly positive preferential margin (PM). This ensures that the analysis captures economically meaningful arbitrage behaviour and excludes situations in which CETA may be claimed due to administrative excess or for reasons unrelated to tariff savings.

2.2 Variables and hypotheses tested in the models: quantifying the incentives and barriers to utilization of CETA’s preferential tariff

The modelling aims to explain variation in the PUR using three categories of variables.

(i) Dependent variable

• Preference utilization rates (PURs): these are defined for each HS8 product and each year as the share of exports (in value terms) that actually benefited from the CETA preferential tariff, relative to the total value of exports eligible for that rate. The data sources differ across panels: Global Affairs Canada for Canadian imports from France, and Eurostat (Comext) for French imports from Canada. The computation of the PUR is detailed in Methodological Appendix A.1.

(ii) Variables of interest

1. Preferential margin (PM): the difference (in percentage points) between the applicable MFN tariff and the CETA preferential tariff (often zero). It captures the direct financial incentive to use the agreement. This variable is log-linearized when performing the regressions. Source: ITC. Hypothesis: an expected positive effect (+). “The larger the tariff advantage, the stronger the incentive to claim the CETA preferential rate, as it provides a meaningful price‑competitiveness advantage in export markets.”
2. Preferential rules of origin (ROO-RI) restrictiveness index: a variable constructed specifically for this study, given the absence of an existing indicator for CETA. Drawing on the methodologies of Ayele (2024), Ayele et al. (2023), and Gretton & Gali (2005), this index—ranging from 0 (least restrictive) to 100 (most restrictive)—quantifies the textual complexity of the rule of origin applicable to each product. It incorporates the type of rule (e.g., change in tariff classification, minimum regional value content), its level of stringency (e.g., change in tariff classification at the 4‑digit versus 6‑digit level), and the required thresholds. The detailed coding protocol and computation method are presented in Methodological Appendix A.2.Source: authors’ calculations based on Annex 5 of the CETA Protocol on Rules of Origin. Hypothesis: an expected negative effect (-). “The more complex it is to demonstrate that a product qualifies as “originating” under CETA (documentation requirements, audit risk), the weaker the incentive to claim the CETA preferential rate.”
3. Share of indirect trade: a variable that measures, for each product, the share of exports (in value terms) that transit through at least one third country outside the European Union before reaching their final destination. It is designed to capture the challenges associated with complying with the agreement’s direct transport rule. Due to limitations in the source data, the variable is constructed at the HS6 level. The underlying data exhibit temporal discontinuities, particularly during the pandemic period. To obtain a complete time series, a rigorous and economically grounded imputation strategy—tailored separately for each panel—was implemented. Recognizing that this imputation represents a strong assumption, the robustness of our findings was assessed using subsamples restricted to years with original, non‑imputed data for each panel. The detailed imputation procedure is presented in Methodological Appendix A.3. Source: UN Comtrade. Hypothesis: an expected negative effect (-).“The more a product transits through third countries, the more difficult it becomes to demonstrate compliance with CETA’s direct transport rule.”

(iii) Control variables

1. Export value: the logarithm of the product’s total export value in the previous year (t–1). Using data lagged by one year helps control scale effects and mitigates endogeneity biases in the estimates. Source: Global Affairs Canada/Eurostat (Comext). Hypothesis: an expected positive effect (+) (economies of scale). “The larger the export volumes, the greater the tariff savings from claiming the CETA preferential rate.”
2. Product type (binary): binary variables classify each product as a consumer product, a capital product, or an intermediate product (reference category), based on the BEC classification. They are constructed at the HS6 level and then assigned to the HS8 level. Source: UN Statistics Division. Hypothesis: an expected negative effect (-) for consumer goods, whose value chains are often more complex.
3. Air freight (binary): a binary variable equal to 1 when more than 50% of the product’s trade (in value terms) is transported by air. It is constructed at the HS6 level. Source: Eurostat (Comext) / UN Comtrade. Hypothesis: a negative effect (-).“A good predominantly traded by air may indicate a preference for delivery speed over other considerations.”
4. Weighted export distance (available only in the panel of French exports to Canada): a variable that measures the weighted average distance between France and the Canadian provinces that serve as the final destinations of the exports. Source: Authors’ calculations based on Statistics Canada and distances computed via Google Maps (between Paris and the Canadian provincial capitals). The detailed coding and computation protocol is presented in Methodological Appendix A.4.Hypothesis: an expected negative effect (-).“The further away the destination, the higher the transport and information costs associated with CETA, which means that the decision to make use of the preferential CETA rate becomes a lower priority.”

2.3 Economic modelling strategy

The model selected for this study is a fractional logit panel model. This approach is the most appropriate because the dependent variable (PUR) is a proportion bounded between 0 and 1, for which a standard linear model could produce inconsistent predictions. This type of model is designed for this type of data (Papke & Wooldridge, 1996) and ensures reliable estimates.

The model controls for fixed effects at the product (HS8) level as well as by year. This technique helps limit unobserved heterogeneity that is constant for a given product (e.g., its intrinsic complexity, the structure of its market) and controls for macroeconomic shocks common to all products in a given year (e.g., health crisis, exchange‑rate fluctuations). The coefficients capture the effect of changes in the explanatory variables within a given product over time.

Three specification challenges are rigorously addressed:

• Endogeneity is mitigated through the use of lagged export values (t–1).
• Multicollinearity is assessed ex ante using correlation matrices and variance inflation factors (VIF). No multicollinearity issues were detected in either panel.
• Standard errors are systematically clustered at the HS6 level to account for the fact that some variables are constructed at a more aggregated level. This approach corrects potential within‑product-group error correlation and yields more reliable significance tests.

The resulting model equation is:

$E({\mathrm{PUR}}_{it}\mid X_{it},{\alpha }_i,{\delta }_t)=G({\alpha }_i+{\delta }_t+{\beta }_1{\mathrm{MP}}_{it}+{\beta }_2{\mathrm{ROORI}}_{it}+{\beta }_3{\mathrm{IndirectTrade}}_{it}+{X_{it}}^{\prime }\Gamma )$

In which:

• $E({\mathrm{PUR}}_{it}\mid X_{it},{\alpha }_i,{\delta }_t)$ is\; the\; conditional\; expectation\; of\; the\; preference\; utilization\; rate\; for\; the\; product$$i in the year t.
• G(.) is the logistic distribution function, which ensures that predicted values remain within the [0,1] interval.
• ${\alpha }_{i}and{\delta }_t$ are product (HS8) and year fixed effects, respectively.
• ${\mathrm{MP}}_{it}$ , ${\mathrm{ROORI}}_{it}$ and ${\text{IndirectTrade}}_{it}$ are the three main variables of interest: the preferential margin, the rules‑of‑origin restrictiveness index, and the share of indirect trade.
• ${X^{\prime }}_{it}$ is a vector of control variables, including the lagged logarithm of export value (t–1), binary indicators for product type (consumer goods, intermediates, capital goods), the air‑freight binary variable, and the logarithm of weighted export distance.
• $\Gamma$ is the vector of coefficients associated with these control variables (lagged export value, product type, air freight, export distance).

The interpretation strategy compares two specifications for each panel: an unweighted model, which reflects the dynamics of a “typical” product line, and a model weighted by export values, which captures the dynamics of the economically dominant flows in the bilateral trade relationship. Together, these two perspectives provide a comprehensive view of the underlying economic mechanisms.

2.4 Descriptive statistics and stylized facts: ex-ante heterogeneous constraints

Before presenting the econometric modelling results, the descriptive statistics offer an initial overview of the dynamics at play and reveal asymmetries between French and Canadian exports. Diagnostic tests were conducted to ensure the absence of multicollinearity among the variables. The correlation matrices and variance inflation factors (VIF), presented in Appendix B, confirm that the model specifications are correct.

A description of the number of observations for each of the modelled samples is provided in the summary table. The table details the structure of the final sample used for the estimations. After excluding observations with missing data and applying the time lag required to address endogeneity (which removes 2018 from the regressions), the analysis covers a total of 6,797 observations for the French panel and 9,690 for the Canadian panel.

Summary table: structure of the final analysis sample, 2019-2024 (after lag adjustments)

Source: Global Affairs Canada; Eurostat (Comext database). Author's calculations.
Note: Observations retained have a strictly positive preferential margin and correspond to goods with an average annual trade value of at least Can$1,500  (approximately €1,000) between 2018 and 2024.

Observation n°1: diverging trends in French and Canadian PURs since 2022

Figure 1 shows the evolution of the PUR for French and Canadian exports, revealing a clear “scissor‑like” pattern. For France, after an initial learning phase, the PUR stabilizes above 60%—the threshold for “good utilization”—but remains below the empirical performance benchmark of 70%. For Canada, the trend moves in the opposite direction: after peaking in 2021, the PUR drops sharply, falling below the 50% warning threshold. This decline is largely driven by imports of HS8 product 27101967 (bituminous oils/petroleum oils). France is the main EU importer of this product, with a CETA preference utilization rate close to 0% between 2022 and 2024. UN Comtrade data indicate that most of these Canadian exports transit through the island of Bonaire (Netherlands), which lies outside the EU customs territory. This routing may raise questions about compliance with CETA’s direct transport rule. Without CETA utilization, the applicable EU tariff on product 27101967 is 3.5%. However, the EU grants a tariff exemption when these products are used for the construction, repair, or maintenance of ships or offshore platforms. The available data do not allow us to determine whether these imports benefit from this exemption or enter the EU subject to customs duties. Excluding product 27101967, the PUR for Canadian exports is higher, but the downward trend observed between 2021 and 2024 persists.

Figure 1 – Trends in Preference Utilization Rates for France and Canada, 2018 to 2024

Source: Global Affairs Canada; Eurostat (Comext database). Author's calculations.
Note: Observations retained have a strictly positive preferential margin and correspond to goods with an average annual trade value of at least Can$1,500  (approximately €1,000) between 2018 and 2024

Observation n°2: contrasting trajectories by type of exported good

Beyond overall averages, analyzing how the PUR evolves by product category (Figures 2.1 and 2.2) reveals sharply contrasting trajectories for France and Canada.

For France, there is a strong increase in the PUR for capital goods. Starting from very low levels in 2018 (25%) and 2019 (10.3%), the PUR rose to reach the “performance” threshold of 70% between 2021 and 2023, suggesting a successful learning phase among French manufacturers. Consumer goods—the main pillar of French exports to Canada—remain stable around 63–66%, above the “good utilization” threshold, although they do not manage to cross the symbolic 70% performance threshold. For Canada, the trajectory is marked by greater volatility. The most striking development is the collapse of the PUR for consumer goods, which fell from over 70% in 2019 to a low of 36.7% in 2023. This decline pulls down Canada’s overall performance. Intermediate goods show a bell‑shaped pattern: a rise until 2021 (72.8%), followed by a steady erosion through 2024 (54.9%).

This comparison suggests that the challenges related to the use of CETA’s preferential tariff may differ in nature between French and Canadian exporters. It should be noted that the data used for figures 2.1 and 2.2 cover only the products included in the analysis sample—that is, products eligible for preferences (MFN duties > 0), whether or not they actually received the preferential rate.

Figure 2.1 – Trends in PUR by product category – French exports to Canada, 2018 to 2024

Source: Global Affairs Canada; Eurostat (Comext database). Author's calculations.
Note: Observations retained have a strictly positive preferential margin and correspond to goods with an average annual trade value of at least Can$1,500  (approximately €1,000) between 2018 and 2024.

Figure 2.2 – Trends in PUR by product category – Canadian exports to France, 2018 to 2024

Source: Global Affairs Canada; Eurostat (Comext database). Author's calculations.
Note: Observations retained have a strictly positive preferential margin and correspond to goods with an average annual trade value of at least Can$1,500  (approximately €1,000) between 2018 and 2024.

Observation n°3: a non-linear response to the preferential margins

The descriptive analysis of utilization rates by the size of the preferential margin (figures 3.1 and 3.2) challenges the intuitive idea of a simple linear relationship in which “the greater the gain, the higher the utilization.”

For France, products with a medium preferential margin (2.5%–5%) show the strongest performance, with CETA utilization reaching 85.4% in 2024. Paradoxically, utilization for the segments offering the largest tariff advantages (5%–10% and >10%) is lower, though still satisfactory, ranging between 60% and 67% in 2024. This suggests that for products with high tariff stakes, the financial incentive may run up against operational barriers—supply‑chain constraints, logistical challenges, rules of origin—that limit tariff optimization.

For Canada, products with a high preferential margin (5%–10%) are associated with very strong CETA utilization, with rates exceeding 80% since 2019 (81.9% in 2024). By contrast, there is a sharp and concerning drop for products with a 2.5%–5% margin, whose utilization falls to 26.5% in 2024, as well as weak performance for products with margins above 10% (51%). This points to a concentration of Canadian effectiveness on a narrow core segment, while other categories appear either neglected or constrained by operational barriers to fully leveraging CETA.

Figure 3.1 – PUR by preferential‑margin level – French exports to Canada, 2018 to 2024

Source: Global Affairs Canada; ITC Market Access Map. Authors' calculations.

Figure 3.2 – PUR by preferential margin level – Canadian exports to France, 2018 to 2024

Source: Eurostat (Comext database); ITC Market Access Map. Authors' calculations.

Observation n°4: more complex preferential rules of origins for French exporters

French and Canadian exporters do not face the same level of tariff complexity when complying with CETA’s preferential rules of origin. Figure 4 presents the distribution of the rules of origin restrictiveness index (ROO-RI) from 2018 to 2024. The key question is whether this regulatory “wall” constitutes a real barrier for exporters in both countries—and whether it is particularly burdensome for French exporters.

Figure 4 – Average CETA rules of origin restrictiveness index, calculated on the basis of exported products

Source: Author’s calculations based on Annex 5 of the CETA Protocol on Rules of Origin.

Observation n°5: a growing constraint linked to logistical issues

Figures 5.1 and 5.2 illustrate how the share of indirect trade has evolved for exported goods with a positive preferential margin—that is, goods for which CETA has eliminated tariffs. For France, there is a clear increase in the volume of dutiable exports to Canada that transit through non‑EU countries before reaching their final destination. Although the data are more limited on the Canadian side, we still observe that slightly more than one‑quarter of Canadian export volumes to France that remain subject to MFN duties (with no tariff elimination under CETA) transit through third countries before reaching France. The 6‑point increase between 2019 and 2024 is noteworthy. This trend suggests that managing the direct transport rule is becoming an increasingly significant challenge, particularly for French exporters.

Figure 5.1 – Trend in the share of indirect trade – French exports to Canada

Source: UN Comtrade. Authors' calculations. HS6 level.

Figure 5.2 – Trend in the share of indirect trade – Canadian exports to France

Source: UN Comtrade. Authors' calculations. HS6 level.
Note: As data for the years 2019–2021 (France) and 2020–2023 (Canada) are not available, this graph shows only the years for which original data are available to illustrate the observed trend.

These stylized facts sketch the portrait of two export ecosystems facing heterogeneous challenges. The econometric analysis that follows aims to disentangle these correlations, quantify the impact of each factor, and explain why, under the same agreement, their performance trajectories may follow divergent paths.

3. Results of the econometric analysis

This section presents the results of the panel models, which rely on product‑level (HS8) and year fixed-effects applied to the analytical sample (2018–2024). The aim is to quantify the impact of financial incentives, barriers, and constraints on the preference utilization rate (PUR).

The analysis begins with French exports to Canada (3.1), followed by a mirror analysis of Canadian exports to France (3.2), and concludes with a comparative synthesis (3.3). Robustness tests for these models are presented in Appendix C.

3.1 Determinants of CETA preferential rate utilization for French exports to Canada: the predominant role of logistical constraints

3.1.1. Sequential analysis: unpacking the economic mechanisms that shape access to tariff elimination under CETA

To isolate the effect of each determinant, a sequential approach is adopted in which variables are introduced in successive blocks. This method makes it possible to assess the stability of the coefficients and to highlight omitted‑variable bias.

a) Unweighted modelling of export values: an accumulation of constraints

The analysis of the unweighted models (Table 1), which reflect the “typical” product line, shows that the decision to use CETA stems from an accumulation of operational constraints rather than simple tariff arbitrage.

• Low sensitivity to traditional incentives (models 1–3): the preferential margin (pref_margin) is not statistically significant, even though its coefficient is large. The rules‑of‑origin complexity index (roo_ri) paradoxically shows a positive and marginally significant effect, suggesting a selection mechanism in which products subject to the most complex rules of origin are associated with utilization of the preferential tariff.
• The importance of the intrinsic characteristics of the exported product (models 4 and 5): the inclusion of product categories points to a structural determinant. Whether a good is classified as a consumer good (d_cons) or a capital good (d_cap), relative to an input, appears to be a significant barrier to the utilization of CETA tariff preferences, a factor that persists right through to the final model.
• The predominance of logistical constraints (models 6 and 7): the inclusion of geographical and logistical variables alters the hierarchy of effects. Distance (log_avg_dist) and, above all, the share of exports routed indirectly (share_indirect) emerge as very powerful and statistically significant barriers (p<0.001), confirming the absence of a decisive effect of financial incentives on the unweighted flows of export value.

In sum, the PUR for a typical French tariff line results from a combination of constraints operating simultaneously. The strongest driver is logistics (transit and distance), followed closely by the intrinsic nature of the product (consumer and capital goods).

b) Weighted modelling of export values: the problem of indirect trade

An analysis of the weighted models (Table 2), which reflect the key economic challenges, reveals a tension between cost advantages and logistical constraints.

• The preferential margin emerges as a clear driver for CETA utilization… (models 1–6): in the intermediate models, the preferential margin (pref_margin) has a positive and significant effect. For large flows, financial incentives act as a driver of CETA utilization. Alongside this, the constraint associated with being a consumer good (d_cons) remains a substantial barrier.
• …highly constrained by certain logistical choices (model 7): the inclusion of the variable share_indirect accounts for the model’s entire explanatory power. Its coefficient is negative and highly significant (-3.560; p<0.001). Once this factor is taken into account, the financial incentive, the scale effect and the intrinsic characteristics of the products lose their statistical significance.

The sequential analysis highlights how operational barriers to the utilization of CETA unfold for French exporters. The effects observed in the intermediate models (financial incentive, product type) were masking a more dominant constraint. Once the structure of the logistics chain is introduced, it accounts for the entire explanatory power and proves to be the most important determinant. For large flows, there is indeed a financial incentive, but even so, transit via third countries outside the European Union remains a major operational barrier to the utilization of CETA.

Table 1 – Sequential regression (value unweighted)

Source: Global Affairs Canada. Authors' calculations.
Data: Global Affairs Canada.
Note: Fractional logit panel model with fixed effects for product (HS8) and year. Standard errors clustered at HS6 level.
Statistical significance: *** p<0.001; ** p<0.01; * p<0.05; . p<0.10

Table 2 – Sequential regression (value-weighted)

Source: Global Affairs Canada. Authors' calculations.
Data: Global Affairs Canada.
Note: Fractional logit panel model with fixed effects for product (HS8) and year. Standard errors clustered at HS6 level.
Statistical significance: *** p<0.001; ** p<0.01; * p<0.05; . p<0.10

3.1.2. Interpretation of the two final models – comparison between the value-unweighted and weighted-value model

The analysis of the final models (Table 3) confirms the dichotomy between the two types of flows. The interpretation of the average marginal effects (AME) enables the economic magnitude of each determinant to be quantified.

a) Unweighted model: three types of operational barriers to the utilization of CETA’s preferential tariffs

For a typical product line (Table 3 – (1)), the decision to use CETA does not appear to be a simple cost–benefit trade‑off, but rather the outcome of a cascade of operational barriers.

Three factors stand out for their negative impact. Their AME indicates their magnitude:

1. Product characteristics: on average, a French consumer good exported to Canada (d_cons) is associated with a PUR that is 20.3 percentage points lower than that of an intermediate good. For capital goods (d_cap), the gap is 12.8 points. This reflects the greater operational difficulty of tracing origin within the value chains of final goods.
2. The logistical constraint: a 10‑percentage‑point increase in the share of transit through a third country (share_indirect) is associated with a 4.8‑point decrease in the PUR. Transit through a non‑EU hub (for example, the United States) complicates the proof of direct transport, either by breaking documentary traceability or through an intermediate customs clearance that disrupts customs control.
3. The geographical constraint: a 10% increase in weighted distance (log_avg_dist) is associated with a 4.2‑point decrease in the PUR. A greater distance from the Canadian destination province appears to dilute information or encourage logistical choices that are unfavourable to CETA compliance, while the financial incentive and volume effect have no significant impact in this case.

The financial incentive (pref_margin) and the volume effect (log_v_total_lag1) are statistically insignificant, with AME close to zero.

For an average product line, exporters are penalized in their utilization of CETA because of the characteristics of the exported product, its logistics chain, and its final destination within Canada. Policies aimed at supporting the utilization of CETA must take these barriers into account, in addition to the more traditional barriers already identified (lack of awareness of the agreement, and SMEs’ difficulty in accessing practical information on how to benefit from it).

b) Value-weighted: indirect trade as a primary constraint

For the products that account for the largest share of French exports to Canada (Table 3 – (2)), exporters appear to have overcome the challenges linked to product characteristics and distance. Only one barrier remains: indirect trade.

The coefficient associated with share_indirect is the only statistically significant one. Its AME shows that, for a given product, a 10‑percentage‑point increase in the share of indirect transit is associated with a 5.1‑point decrease in the PUR.

This finding can be explained by two economic mechanisms:

1. A strategic trade-off: for the main products exported to Canada, exporters prioritize the efficiency of integrated North American continental supply chains rather than strictly bilateral ones, even at the expense of tariff optimization. Indeed, large businesses that, since the early 2000s, have favoured foreign investment and overseas establishment over exports (Direction générale du Trésor, 2020) tend to view North America as a single integrated market, with major logistics hubs located in the United States for reasons of scale, connectivity, and cost. A French business may therefore choose to optimize logistics costs rather than claim tariff elimination under CETA—especially if it has negotiated advantageous pricing for managing its flows to the United States. For some exporters, the decision to utilize CETA ultimately comes down to the question: “Is it worth bearing the cost of reorganizing an efficient North American logistics chain in order to obtain the tariff savings under CETA?”
2. A lack of coordination between exporters, importers, and customs brokers: assuming that (1) exporters maximize profitability, (2) make decisions independently, (3) have imperfect information about how CETA’s direct transport rule operates, and (4) base their decisions on this partial information, then exporters primarily make logistical choices according to shipping costs. Importers have no recourse but to accept or reject these choices. Indirect trade is therefore a constraint imposed on importers. Operators who were better informed of the direct transport rule would have recognized the benefits of CETA and would likely have favoured direct transport between France and Canada.

Although the preferential margin and volume have positive coefficients, their statistical uncertainty indicates that they are not sufficient to offset this dominant logistical constraint.

For the largest flows, the under‑utilization of CETA is not simply a matter of lacking information about the agreement, insufficient expertise, or the complexity of the rules of origin. It is the result of a decision (or inaction) whereby businesses, faced with a trade-off between local optimization (tariff savings on the Canadian market) and global efficiency (North American logistics structure), prioritize the latter. The issue is less about training businesses about how CETA works, than about helping them reassess the profitability of this trade‑off by reducing the perceived cost of logistical adaptation. Strengthening coordination between exporters, importers, and freight forwarders is therefore crucial to ensure that all trade stakeholders are fully aware of the advantages under CETA.

Table 3 – Final models (value-weighted and unweighted)

Source: Global Affairs Canada. Authors' calculations.
Data: Global Affairs Canada.
Note: Fractional logit panel model with fixed effects for product (HS8) and year. Standard errors clustered at HS6 level.
Statistical significance: *** p<0.001; ** p<0.01; * p<0.05; . p<0.10

c) Do the “reciprocal tariffs” introduced by the U.S. in 2025 mark a turning point?

This analysis, based on data from 2018 to 2024, shows that the strategic trade‑off shaping the largest French export flows to Canada has so far favoured the efficiency of integrated North American logistics chains over the tariff optimization under CETA. Is this situation likely to persist? U.S. protectionist trade policy suggests that this trade-off is in the process of being redefined.

The introduction of U.S. tariffs on all countries in April 2025 (including the European Union) has altered exporters’ calculations when weighing logistical gains against tariff advantages. Canadian import data from UN Comtrade for the first half of 2025, compared with the first half of 2024, reveal a significant decline in transit through the United States for French exports to Canada. In early 2025, the share of such transit fell, marking a break with past trends. This decline has been offset by an increase in direct transport from France, concentrated for now on products with a positive MFN rate—mainly consumer goods.

Transit through the United States appears to be no longer merely an opportunity cost (the unclaimed CETA advantage), but a direct and potential double cost (U.S. tariffs upon entry, followed by the risk of losing CETA preferences and paying Canadian MFN duties upon exit). Under this new calculation, direct transport is economically more rational. The U.S. tariff shock appears to be triggering structural adjustments, prompting businesses to reassess their logistics chains. Over time, this trend is likely to naturally strengthen compliance with CETA’s direct transport rule. For public authorities, this represents an opportunity to support this logistical transition in order to promote better utilization of the preferential rate under CETA.

3.1.3. Sub‑sample analysis: highlighting heterogeneity in behaviour

To refine the study, two additional analyses were conducted to examine the specific characteristics of each product type and to assess the cumulative effects.

a) Consumer goods versus other goods: distinct determinants of CETA utilization

French export behaviour may vary depending on the type of goods exported. To test this hypothesis, two models were estimated: one for consumer goods and another covering intermediate and capital goods. The results are presented in Tables 4.1 and 4.2.

The analysis by type of good confirms that indirect trade is a constraint regardless of the type of good or the size of the flows. In the estimated models, the variable share_indirect is negative and significant (***). Its impact on non-utilization of CETA is substantial: a 10-point increase in the transit share is associated with a drop in PUR ranging from 4.3 (unweighted model) to 5.1 points (weighted model).

Beyond the barrier posed by indirect trade, the segmented analysis shows two decision patterns:

• Exporters of industrial goods (intermediates and capital goods) behave in a way consistent with most of the hypotheses set out earlier: for weighted flows, the PUR is positively influenced by the financial incentive (pref_margin) and by economies of scale (log_v_total_lag1). A 10% increase in export value is associated with a 1.2‑point rise in the PUR. For French exporters of intermediate and capital goods to Canada, reaching a critical mass and benefiting from a clear tariff advantage both influence the decision to utilize CETA. For unweighted flows, the influence of the tariff advantage disappears, and distance (log_avg_dist) is a highly significant barrier (***).
• For consumer goods, economies of scale (log_v_total_lag1) have no significant effect on CETA utilization. The tariff incentive (pref_margin)serves as a driver. French exporters of consumer goods therefore, tend to claim the preferential rate under CETA regardless of export volumes. However, they remain heavily constrained by operational barriers. Distance (log_avg_dist) is a significant barrier regardless of which model is used.

The analysis points to a paradox: the rules‑of‑origin complexity index (roo_ri) is positively and significantly correlated with the PUR (***). This could suggest that only the most sophisticated businesses—those capable of handling complex rules—have high utilization rates. Large exporters of consumer or industrial goods are mainly major groups with dedicated export‑management departments, unlike many SME‑type exporters. It may also reflect an omitted‑variable bias.

Overall, scaling up and improving the dissemination of information on tariff savings under CETA are two avenues for increasing the utilization of CETA’s tariff preferences among exporters of intermediates and capital goods. Meanwhile, for exporters of consumer goods, the management of logistical complexity (distance, transit) requires particular attention.

Table 4.1 – Sub-sample models (value weighted): consumer goods vs other

Source: Global Affairs Canada. Authors' calculations.
Data: Global Affairs Canada.
Note: Fractional logit panel model with fixed effects for product (HS8) and year. Standard errors clustered at HS6 level.
Statistical significance: *** p<0.001; ** p<0.01; * p<0.05; . p<0.10

Table 4.2 – Sub-sample models (value unweighted): consumer goods vs other goods

Source: Global Affairs Canada. Authors' calculations.
Data: Global Affairs Canada.
Note: Fractional logit panel model with fixed effects for product (HS8) and year. Standard errors clustered at HS6 level.
Statistical significance: *** p<0.001; ** p<0.01; * p<0.05; . p<0.10

b) Analysis of interactions between variables: the cumulative effect of logistical constraints

Another research question is whether these barriers interact, creating cumulative effects that further penalize French exporters. The analysis focuses on two economically plausible interactions: the trade‑off between financial gain and logistical constraint (pref_margin × share_indirect), and the synergy between two logistical constraints (share_indirect × log_avg_dist).

First, the test of the trade‑off between tariff advantage and logistical cost is not statistically conclusive, whether the model is value-weighted or not. While the coefficient of the interaction term is negative, it is not significant. Second, the other interaction model, for which results are presented in Table 5, reveals a negative and significant synergy between indirect trade and distance.

This negative interaction coefficient brings to evidence a logistical “toxic cocktail,” in which the penalizing effect of indirect trade is amplified by geographical distance (for example, shipping to Western Canada via the United States). Distance becomes a substantially greater barrier when the supply chain is indirect, heightening the risk of breaks in documentary traceability.

Overall, the analysis shows that logistical barriers do not simply add up; they compound. A value chain that passes through non‑EU countries and involves complex geographical routing is particularly detrimental to CETA utilization.

Table 5 – Model with interaction variable: indirect trade x distance

Source: Global Affairs Canada. Authors' calculations.
Data: Global Affairs Canada.
Note: Fractional logit panel model with fixed effects for product (HS8) and year. Standard errors clustered at HS6 level.
Statistical significance: *** p<0.001; ** p<0.01; * p<0.05; . p<0.10

3.2  A parallel analysis of the determinants of CETA preferential rate utilization for Canadian exports to France

To provide a comprehensive perspective, a parallel analysis was conducted on the panel of Canadian exports to France. This approach helps distinguish shared challenges (logistics) from features specific to the Canadian export ecosystem.

3.2.1. Sequential analysis: results more closely aligned with the initial hypotheses

Unlike the French panel, the step‑by‑step inclusion of variables in the Canadian models (Table 6 and Table 7) shows a high degree of stability in effects. The coefficients do not offset one another, suggesting that Canadian exporters face a set of independent and cumulative factors.

a) Unweighted model: a multifactor trade‑off in CETA utilization

For a typical Canadian product (Table 6), the decision to utilize CETA reflects a balance between drivers and barriers that goes beyond a simple opportunity‑cost calculation based on the preferential margin (models 1–6):

• Drivers: the volume effect (log_v_total_lag1) is a positive and highly significant driver (***), demonstrating that economies of scale are a major determinant of CETA utilization. Whether a good is a consumer good (d_cons) or a capital good (d_cap) is also associated with greater CETA utilization (** and *** respectively).
• Barriers: the rules of origin complexity index (roo_ri) is a significant negative constraint (**). The use of air freight (d_air_freight) and indirect trade (share_indirect) are also significant barriers (** and ***).

However, the preferential margin (pref_margin) is never statistically significant.

b) Weighted model: the predominance of logistical factors

For the largest trade flows (Table 7), the model highlights the predominance of air freight and indirect trade as barriers to the utilization of CETA. The significance of the preferential margin (pref_margin), observed in the initial models, gradually diminishes. The volume effect (log_v_total_lag1) and whether the good is a capital good (d_cap) are significant drivers (* and **), and remain so through to the final model. Further, the introduction of logistics variables shows that the use of air freight (d_air_freight) is a major and highly significant barrier, and its inclusion in the model diminishes the influence of the other variables. Indirect trade (share_indirect) also emerged as an additional barrier, though statistically less significant than in the unweighted model or the French model. For major Canadian trade flows to France, the decision to claim CETA is dominated by a trade-off in logistics where the use of air freight—which prioritizes speed—is the main barrier to tariff optimization.

Table 6 – Sequential regressions (value unweighted)

Source: Office of the Chief Economist. Authors' calculations.
Data: Eurostat (Comext)
Note: Fractional logit panel model with fixed effects for product (HS8) and year. Standard errors clustered at HS6 level.
Statistical significance: *** p<0.001; ** p<0.01; * p<0.05; . p<0.10

Table 7 – Sequential regressions (value weighted)

Source: Office of the Chief Economist. Authors' calculations.
Data: Eurostat (Comext)
Note: Fractional logit panel model with fixed effects for product (HS8) and year. Standard errors clustered at HS6 level.
Statistical significance: *** p<0.001; ** p<0.01; * p<0.05; . p<0.10

3.2.2. Interpretation of final models

The analysis of the final models (Table 8) quantifies the factors that influence the decision to utilize CETA for Canadian products exported to France.

a) Unweighted model: a mix of barriers and drivers for utilization of CETA

On average, the decision to use the CETA preferential rate is a calculation in which several factors weigh into the balance.

• Drivers of CETA utilization among Canadian exporters: unlike the French context, the scale effect (log_v_total_lag1) is significant (***)—reaching a critical mass encourages utilization. In addition, exporting finished products (consumer or capital goods) is associated with a PUR roughly 21 percentage points higher than for intermediate goods.
• Rules of origin as an operational barrier: the rules of origin restrictiveness index (roo_ri) is a significant negative factor (**). A 10‑point increase in the complexity index reduces the PUR by 7.8 points. This is a key difference: Canadian exporters appear to be discouraged by the textual complexity of the rules.
• Logistical barriers: indirect transit and air freight both weigh negatively on utilization, reducing the PUR by 1.2 and 3.4 points, respectively.

The successful utilization of CETA by a Canadian exporter hinges on reaching a critical mass and overcoming both the administrative complexity of the rules of origin and logistical constraints.

b) Value-weighted model: the importance of logistical variables

When the analysis focuses on the largest export flows, the decision to utilize CETA depends primarily on logistical factors:

• Air freight as the dominant barrier: reliance on air freight (d_air_freight) is the strongest and most significant barrier (***), associated with an 11.5‑point reduction in the PUR. For the most heavily exported products, the priority given to speed outweighs tariff optimization. Indirect trade is also clearly identified as a barrier to the utilization of CETA, although less significant than in the French context.
• Characteristics of Canadian capital goods are more conducive to utilization of CETA: classification as a capital good (d_cap) remains a strong driver (**), with average CETA utilization 16.9 points above that of intermediate goods.
• Volume and financial incentives as modest drivers of the utilization of CETA: the scale effect remains a relevant factor in the equation, although its statistical weight diminishes once the air‑freight variable is included. Financial incentive (pref_margin) shows a very large positive coefficient (+60.01), indicating a potentially strong influence, although its statistical variability calls for caution.

For major flows, financial incentives and volume encourage the utilization of CETA, but these drivers are directly hindered by the logistical constraints of air freight and indirect trade.

Table 8 – Final models (value weighted and unweighted)

Source: Office of the Chief Economist. Authors' calculations.
Data: Eurostat (Comext)
Note: Fractional logit panel model with fixed effects for product (HS8) and year. Standard errors clustered at HS6 level.
Statistical significance: *** p<0.001; ** p<0.01; * p<0.05; . p<0.10

3.2.3. Sub-sample analysis: patterns in the utilization of differentiated tariff preferences

As in the French context, additional sub‑sample analyses were conducted.

a) Consumer goods versus other goods: contrasting behaviour by product type

Estimates from the separate models for consumer goods and for “other goods” (intermediate and capital goods) reveal distinct decision‑making patterns in the utilization of CETA’s tariff preferences (tables 9.1 and 9.2).

• For industrial goods (capital goods and intermediates), the unweighted model shows that the decision to utilize CETA is dominated by operational barriers. The PUR is negatively and significantly affected by the complexity of the rules of origin (roo_ri), by air freight, and by indirect trade. A negative selection effect is also noted on the preferential margin. The weighted model qualifies this paradoxical finding, as the preferential margin once again becomes a positive incentive, balanced against logistical constraints (air freight and transit), which remain major barriers. The result is that, for a typical product, the utilization of CETA is hindered by administrative constraints, whereas logistical issues remain predominant for the largest flows.
• For consumer goods, exporters’ behaviour is characterized by a strong sensitivity to the preferential margin: it is a positive and significant driver regardless of the modelling choice (unweighted — *** — or weighted — *). Unlike industrial goods, the complexity of the rules of origin is never a significant barrier, and the scale effect plays only a limited role (positive and significant only in the unweighted model). In short, if exporters identify an attractive tariff advantage, eligibility for CETA will be considered. Only indirect trade persists as a cross‑cutting barrier.

Table 9.1 – Sub-samples (value weighted): consumer goods vs. other goods

Source: Office of the Chief Economist. Authors' calculations.
Data: Eurostat (Comext)
Note: Fractional logit panel model with fixed effects for product (HS8) and year. Standard errors clustered at HS6 level.
Statistical significance: *** p<0.001; ** p<0.01; * p<0.05; . p<0.10

Table 9.2 – Sub-samples (unweighted): consumer goods vs. other goods

Source: Office of the Chief Economist. Authors' calculations.
Data: Eurostat (Comext)
Note: Fractional logit panel model with fixed effects for product (HS8) and year. Standard errors clustered at HS6 level.
Statistical significance: *** p<0.001; ** p<0.01; * p<0.05; . p<0.10

b) Analysis of interaction between variables: a positive interaction between two constraints

The interaction analysis (Table 10) reveals a positive and statistically significant (*) interaction between the use of indirect transit and the use of air freight.

This result indicates that the cumulative negative impact of these two constraints is less than the sum of their individual effects. This suggests that logistical constraints are not homogeneous: for air‑freight flows—where the main constraint is the “speed vs. cost” trade‑off—the administrative burden of transit appears to weigh less heavily than it does for maritime or land flows.

However, this result should be interpreted with caution. Unlike the French context, this panel does not include a control variable for geographical distance. It is possible that this interaction term partly captures unobserved distance‑related effects, which may explain why the result is different from the “toxic cocktail” observed in the French context.

Table 10 – Model with interaction variable: indirect trade x air freight

Source: Office of the Chief Economist. Authors' calculations.
Data: Eurostat (Comext)
Note: Fractional logit panel model with fixed effects at the product (HS8) level as well as by year. Standard errors clustered at the HS6 level.
Statistical significance: *** p<0.001 ; ** p<0.01 ; * p<0.05 ; . p<0.10

3.3 Comparative overview: the shared logistical challenge of indirect trade and distinct trade-off strategies

A comparative analysis of French and Canadian exports provides a bilateral overview of the utilization of CETA. The findings point to a key conclusion: while both partners face specific challenges, they share a common and dominant logistical barrier.

a) Indirect trade: a shared barrier

The most robust finding of this study is the confirmation, in both panels, of the significant role played by the logistics chain. In almost all estimated models, the share of indirect trade (share_indirect) emerges as a negative, high‑magnitude, and statistically significant constraint.

This situation can be explained by two mechanisms, the second of which appears most likely for SMEs and new exporters:

• The exporter’s sequential rationality: the operator rules out the option of utilizing CETA as soon as the logistical feasibility (compliance with direct transport) is deemed too complex, even before considering the tariff‑preference trade‑off. If feasibility is confirmed, the exporter then weighs up the gains from logistical optimization (even if it means not utilizing CETA) against the potential tariff advantage under the agreement. This mechanism requires the exporter to have a thorough understanding of the information and effective coordination with the importer and freight forwarders and customs brokers – a scenario likely to apply to large groups.
• A lack of coordination between exporters, importers, and the freight forwarder: the exporter chooses the logistical route based on transport cost, relying on incomplete information about CETA’s direct transport rule. This creates an exogenous non‑compliance constraint for the importer, who is then unable to claim the preference. The negative correlation observed for the indirect‑trade variable (share_indirect) would thus be the symptom of this coordination failure and of insufficient information transmission along the value chain—a scenario that is highly likely for SMEs and new exporters.

Regardless of which interpretation is preferred, the operational conclusion remains the same: non‑compliance within the logistics chain is the common barrier to increased utilization of CETA. This is a shared challenge, although it is more detrimental to French exporters than to Canadian exporters. The negative impact of indirect trade on the utilization of CETA is three to seven times more severe for French exports than for Canadian exports. This heightened sensitivity in the French context can be explained by the “toxic cocktail,” in which the constraint of transit is compounded by the long distances to final destinations within the Canadian market.

b) Distinct decision-making rationales

Beyond logistical considerations, the decision parameters governing the utilization of CETA differ depending on the exporter’s country. These asymmetries are summarized in Table 11 and can be grouped into five points:

1. Ambiguous role of the tariff incentive: the preferential margin (pref_margin), a traditional driver in the implementation of trade agreements, plays an ambiguous role here. For French exports, it appears to be a positive but statistically fragile driver, suggesting that it is not the primary determining factor. For Canadian exports, the effect is highly segmented: paradoxical for industrial goods (where a high margin is associated with low utilization, a symptom of a selection effect), it acts as a powerful and robust driver for consumer goods.
2. Sensitivity to the administrative complexity of the rules of origin: there is a significant asymmetry in how exporters respond to CETA’s rules of origin (roo_ri). For Canadian exporters, the textual complexity of the rules represents a significant negative constraint, confirming a sensitivity to the administrative burden. For France, the effect is paradoxically positive. This suggests that French exporters face more qualitative challenges regarding the traceability of value chains (which the textual index captures only imperfectly), while Canadians struggle with the difficulty of interpreting the agreement.
3. The scale effect as driver: the volume effect (log_v_total_lag1) is a key determinant. For Canada, it is a robust and highly significant driver of performance: reaching a critical mass is essential to offset the costs of CETA compliance. For France, this effect is weak or even non-existent, suggesting that the decision to utilize CETA is more automatic and less dependent on export volume.
4. The trade-off between CETA compliance costs and delivery speed: reliance on air freight (d_air_freight) is a major and systematic constraint for Canadian exports, whereas it has no statistically clear effect for France. This strategic trade‑off is more salient for Canadian exporters, likely reflecting differences in the nature of the goods traded (which are more sensitive to delivery times—particularly capital goods) or differences in French and Canadian logistics cultures.
5. The structure of traded goods: lastly, each country’s specialization creates contrasting challenges. For France, whose strength in the Canadian market lies in consumer goods, these products are less likely to be associated with utilization of CETA because of the challenges they pose in terms of supply chain traceability. For Canada, whose strength lies in industrial goods, consumer goods, by contrast, emerge as a driver of performance, likely representing markets that are easier to manage.

Overall, Canadian exporters behave in a more traditional way, whereas French exporters face more systemic challenges linked to the very nature of their flagship products exported to Canada.

Table 11 – Comparative table of the determinants of CETA utilization

c) Conclusion of the comparative analysis

The comparative analysis of the two export panels paints a contrasting picture of CETA utilization. It shows that although French and Canadian exporters face a common logistical challenge—indirect trade—they do not approach it with the same strengths, nor with the same sensitivities.

Under‑utilization of CETA is not a monolithic problem. It results from two distinct trade-off logics. For France, the challenge is concentrated in the management of complex value chains for consumer goods, and is compounded by a “toxic cocktail” of logistical factors (indirect trade and export distance). For Canada, the challenge is more traditional, centred on understanding administrative barriers, reaching a critical export mass for a given product, and better managing the speed‑cost trade‑off in air freight.

The heterogeneity of findings illustrates that under-utilization of CETA is not a monolithic problem: it stems from distinct trade-off logics, which call for a differentiated analysis based on the profile of exporters and the nature of the trade flows concerned.

4. Conclusion

Eight years after it entered into force, the Comprehensive Economic and Trade Agreement (CETA) still offers untapped potential for optimization. This study went beyond the simple observation of under‑utilization of its tariff preferences by quantifying its determinants. It does so through a bilateral econometric analysis of unprecedented granularity.

The main conclusion of this work is that the failure to utilize preferential tariffs is not merely a matter of insufficient information; it is the outcome of specific trade-off logics and a lack of coordination between exporters, importers, and freight forwarders. While indirect trade constitutes a shared barrier to CETA utilization, the French and Canadian ecosystems follow different decision‑making logics. For French exporters, the challenge lies in tracing value chains for consumer goods and shipping distances. For Canadian exporters, the challenge is administrative complexity and scaling up.

This heterogeneity suggests that trade agreements should not be promoted in a one-size-fits-all manner, but rather with support tailored to the specific needs of economic players. Ultimately, this study demonstrates that, to unlock CETA’s full potential, the challenge is no longer so much to convince businesses of its existence as to equip them to navigate the operational and logistical complexities of its implementation.

While this HS8‑digit product‑level analysis has helped identify robust mechanisms, further research using business‑level microdata from French and Canadian customs would be particularly valuable. It would allow researchers to control for business‑specific characteristics (size, productivity, export experience) and to test the impact of Incoterm choice on utilization rates, thereby validating the hypothesis of coordination failures within the value chain. Moreover, improving the granularity and availability of data on transit flows would be crucial for refining the measurement of logistical barriers. Finally, extending this analysis to all EU Member States would help determine the extent to which the barriers identified for France are specific or shared by other economies with similar export structures.

5. References

Academic research articles

Ayele, Y. (2024). Assessing the restrictiveness of rules of origin in the AfCFTA. ODI.

Ayele, Y., Gasiorek, M., & Koecklin, M. T. (2023). Trade preference utilization post-Brexit: the role of rules of origin. World Trade Review, 22(3-4), 436-451.

Cadot, O., Carrère, C., De Melo, J., & Tumurchudur, B. (2006). Product-specific rules of origin in EU and US preferential trading arrangements: an assessment. World Trade Review, 5(2), 199-224.

Estevadeordal, A., & Suominen, K. (2003, April). Rules of origin: a world map. In seminar 'Regional Trade Agreements in Comparative Perspective: Latin America and the Caribbean and Asia Pacific (pp. 6-7).

Gretton, P., & Gali, J. (2005, September). The restrictiveness of rules of origin in preferential trade agreements. In 34th Conference of Economists, University of Melbourne, Australia (pp. 26-28).

Hayakawa, K., Kim, H., & Lee, H. H. (2014). Determinants on utilization of the Korea–ASEAN free trade agreement: margin effect, scale effect, and ROO effect. World Trade Review, 13(3), 499-515.

Keck, A., & Lendle, A. (2012). New evidence on preference utilization. World Trade Organization Staff Working Paper No. ERSD-2012-12.

Krishna, K., & Krueger, A. O. (1995). Implementing free trade areas: Rules of origin and hidden protection.

Papke, L. E., & Wooldridge, J. M. (1996). Econometric methods for fractional response variables with an application to 401 (k) plan participation rates. Journal of Applied Econometrics, 11(6), 619-632.

Institutional reports and data sources

Direction générale du Trésor (2020). French businesses’ international strategies.

Eurostat. Comext database on merchandise trade. Accessed several times in 2024–2025.

Global Affairs Canada. Data on the utilization of preferential tariffs. Accessed on several occasions in 2024 and 2025. 

International Trade Centre (ITC). Market Access Map. Accessed several times in 2024-2025.

Statistics Canada. Canadian International Merchandise Trade Data. Accessed several times in 2024-2025.

UN Comtrade. International merchandise trade database. Accessed several times in 2024-2025.

UN Statistics Division. Correspondence Tables for Broad Economic Categories (BEC). Accessed several times in 2024-2025.

6. Technical appendix

Methodological appendix A – construction of the panel variables

A.1. Calculation of the CETA preferential utilization rate (PUR)

The model’s dependent variable, the preferential tariff utilization rate (PUR), is defined for each product i (at the HS8 level) and each year t. It represents the share of eligible exports that actually benefited from the CETA preferential treatment. It is calculated as follows:

${\text{PUR}}_{it}=\frac{\text{Value of exports of product i in year t cleared under the CETA preferential tariff}}{\text{Total value of exports of product i in year t that are eligible under CETA}}$

In which:

The numerator is the value of trade flows for which the exporter claimed and obtained the CETA preferential tariff.

The denominator is the total value of trade flows for the same product that were eligible for preferential treatment under the agreement, whether or not the exporter actually claimed it.

By construction, ${\text{PUR}}_{it}$ is a proportion bounded between 0 and 1. The data for the numerator and the denominator come from detailed customs datasets provided by Global Affairs Canada (for imports into Canada) and Eurostat (for French imports).

A.2. Calculation of the preferential rules of origin restrictiveness index (ROO-RI)

The ROO‑RI index is a constructed variable ranging from 0 (least restrictive) to 100 (most restrictive), based on a four‑step protocol:

1. Systematic codification: the legal text of Annex 5 of the CETA Protocol on Rules of Origin was analysed for each tariff line (at the HS8 level). Each rule was broken down into its characteristics and entered into a structured database.
2. Assignment of baseline scores: each type of rule was assigned an initial score reflecting its intrinsic degree of restrictiveness, as summarized in the table below.

Score adjustment principles

Source: Author’s calculations based on Annex 5 of the CETA rules of origin protocol. Methodology inspired by Ayele (2024), Ayele et al. (2023), and Gretton & Gali (2005).

3. Application of adjustment principles. The baseline scores were adjusted to reflect the complexity of the rules of origin:
   • Flexibility principle (“OR” rule): when several options are available, the final score is based on the least restrictive option, from which a “flexibility bonus” is subtracted (example: CTH or RVC → lower score than CTH alone).
   • Complexity principle (“AND” rule): when several requirements are cumulative, the final score is based on the most restrictive option, to which a “complexity penalty” is added (example: CTH and RVC → higher score than CTH alone).
   • Specific clauses: additional adjustments were made for relaxation clauses (“however…”) or restrictive clauses (“except for…”).

A.3. Construction and imputation of the indirect trade variable (share_indirect)

The variable  is intended to approximate the barrier posed by the direct transport rule. It was constructed using a multi-step protocol to account for the problem posed by data discontinuities on the UN Comtrade website.

For each product (at the HS6 level) and each year for which data were available, the variable was calculated as the ratio of the value of exports reported as transiting through a non‑EU third country to the total value of exports to the final destination (Canada or France, depending on the panel).

The source data contained gaps for several consecutive years, largely coinciding with the disruption of global supply chains during the COVID-19 pandemic. Simply excluding these years would have biased the analysis and reduced the size of the analytical sample. An imputation strategy was therefore required, differentiated across panels to best reflect their respective data contexts.

1) Imputation for the panel of French exports to Canada (missing data: 2019, 2020 and 2021)

For this panel, data were available for the years 2018, 2022, 2023 and 2024; an interpolation‑based method was deemed most appropriate to capture the underlying trend.

• Linear interpolation: for products with observed data in 2018 and in 2022, 2023 and 2024, the values for 2019, 2020 and 2021 were estimated using linear interpolation, modelling a gradual adjustment of supply chains.
• Last/first observation carried forward: for products with observed data only at the start or end of the period, the last or first known value was carried forward to fill in the missing years.

2) Imputation for the panel of Canadian exports to France (missing data: 2020 to 2023)

For this panel, a non-linear approach was adopted to account for the pandemic period.

• Strong assumption: the pandemic caused a break (“before”/“after”) rather than a gradual trend.
• Imputation rules:
  • Crisis period (2020-2021): it is assumed that actors attempted to maintain their pre‑pandemic strategies. The last observed value (2019) was therefore carried forward using the Last Observation Carried Forward (LOCF) method.
  • "New normal" period (2022-2023): it is assumed that a new logistical structure, observed in the 2024 data, had been established. The first post‑crisis observed value (2024) was therefore carried backward to fill the preceding years, using the Next Observation Carried Backward (NOCB) method.
    This approach creates a sharp break between 2021 and 2022, which is economically more plausible than a linear interpolation that would have artificially smoothed the impact of the COVID‑19 pandemic shock.

After these steps, a few missing values remained for products with highly intermittent trade flows. In the context of this variable, a missing value was interpreted as the absence of any observed indirect flow. All remaining missing values were therefore imputed as zero, on the assumption that if no indirect flow was detected, its share is effectively zero.

It is important to acknowledge that any imputation strategy relies on strong assumptions. To ensure that the modelling results were not an artefact of this procedure, the robustness of the findings was systematically assessed by re‑estimating the final models on a sub-sample restricted to years with genuine (non‑imputed) data only. As shown in Appendix C, the main conclusions—particularly the negative and significant effect of indirect trade—remain unchanged, which in hindsight confirms the relevance of the imputation and assignment method.

A.4. Construction of the weighted average export distance variable (avg_dist)

To capture the impact of transport costs and logistical complexity linked to intra‑national geography, a weighted average distance variable was constructed for the panel of French exports to Canada. This approach, directly inspired by gravity models in international trade, aims to provide a more accurate measure of the geographical barrier than a simple bilateral distance. This variable is available only in the France → Canada panel, as HS8‑level data on the regional destinations of Canadian exports to the various French regions are not publicly available from the French customs website.

Distance is used as a proxy for transport and information costs. The underlying assumption is that as the average export distance increases, the relative weight of these logistical costs in the exporter’s total cost structure also rises. Consequently, the relative attractiveness of the tariff advantage under CETA diminishes, which may discourage compliance efforts. A negative effect of this variable on the PUR is expected.

The variable was calculated for each product i (at HS8 level) and each year t following a three-step process:

Step 1: Collection of bilateral distances

Distances in kilometres was calculated between France’s economic centre (Paris) and the capital or main economic metropolis of each Canadian province and territory (e.g., Paris–Montréal, Paris–Toronto, Paris–Vancouver, etc.). These distances are time‑invariant.

Step 2: Calculation of each province’s weight in exports

Using detailed trade data from Statistics Canada, the share of French exports for each product i and each year t destined for each Canadian province/territory (p) was calculated: 

Step 3: Calculation of the weighted average

The weighted average distance for each product-year was obtained by calculating the sum of the bilateral distances, weighted by the share of exports to each province:

${\text{avgdistance}}_{it}=\sum _{p=1}^N{\text{w}}_{ipt}{\text{d}}_{Fp}$

In which:

• $w_{ipt}$ is the share of the export value of product i in year t destined for Canadian province p.
• $d_{Fp}$ is the geographical distance between France (F) and Canadian province p.
• N is the total number of Canadian provinces and territories.

The logarithm of this variable (log_avg_dist) was used in the econometric regressions in order to interpret its effect in terms of elasticity and to mitigate potential influences of extreme values.

This variable is available only in the France → Canada panel because regional‑destination data for Canadian exports to the various French regions at the HS8 level were not available from the French customs website.

Appendix B – multicollinearity analysis

Before estimating the econometric models, an in‑depth analysis was conducted to check for the potential presence of multicollinearity among the explanatory variables. Multicollinearity—defined as a strong linear correlation between two or more variables—does not introduce bias into the estimated coefficients, but it can inflate their standard errors, thereby making statistical significance tests unreliable. To assess this risk, two tools were used for each panel: a correlation matrix and the calculation of the variance inflation factor (VIF).

B.1. Assessment for the panel of French exports to Canada

(a) Correlation matrix – Figure B.1 presents the Pearson correlation matrix for the model’s variables. A correlation is generally considered problematic when its absolute value exceeds 0.7 or 0.8. Visual inspection reveals a strong negative correlation (–0.86) between the binary variables d_cons (consumer goods) and d_int (intermediate goods). This is mechanical: a product classified as a consumer good cannot simultaneously be an intermediate good. It is the result of the construction of mutually exclusive binary variables, not an issue of economic multicollinearity. Aside from this case, all other correlations among explanatory variables are low to moderate, indicating no major risk.

(b) Variance inflation factor – The VIF measures how much the variance of an estimated regression coefficient is inflated due to multicollinearity. The common empirical rule is that a VIF above 5 is a threshold of concern, and a VIF above 10 indicates severe multicollinearity. Figure B.2 reports the VIF values for each variable. All VIFs are well below the threshold of 5. The highest values, for d_cons (3.65) and d_int (3.61), confirm the negative correlation observed earlier but remain at a fully acceptable level.

Table B.1. – Pearson correlation matrix – Panel of French exports to Canada

Source: Authors' calculations
Note: correlations with |r| > 0.7 are considered problematic. The strong negative correlation between d_cons and d_int is mechanical (the variables are mutually exclusive).

Table B.2. – Variance inflation factors (VIF) – Panel of French exports to Canada

Source: Authors' calculations

B.2. Assessment for the panel of Canadian exports to France

(a) Correlation matrix – Figure B.3 presents the correlation matrix for the Canadian panel. Similar to the French panel, the strongest negative correlation (–0.72) was observed between d_cons and d_int, for the same mechanical reasons. A moderate positive correlation (0.45) was found between the tariff (pref_margin) and the value of exports (log_v_total), which is economically coherent (trade flows tend to be larger on tariff lines with higher stakes). No other correlation reaches a level that would be considered problematic.

(b) Variance inflation factor – confirms the absence of risk. All VIF values are well below the threshold of 5. The highest VIFs, for d_cons (2.6) and d_int (2.15), are even lower than in the French panel and raise no concerns.

Table B.3. – Pearson correlation matrix – panel of Canadian exports to France

Source: Authors' calculations
Note: correlations with |r| > 0.7 are considered problematic. The strong negative correlation between d_cons and d_int is mechanical (the variables are mutually exclusive).

Table B.4. – Variance inflation factors (VIF) – Panel of Canadian exports to France

Source: Authors' calculations

The assessments conducted on both analytical panels confirm that the models are properly specified with respect to multicollinearity risk. The coefficient estimates and their significance tests can therefore be considered statistically reliable in this regard.

Appendix C: robustness tests

To verify the robustness of the assessment’s conclusions, a series of robustness tests was conducted. These tests aim to ensure that the results are not an artefact of the data‑imputation strategy, the chosen level of aggregation, or the inclusion of certain specific flows.

C.1. Robustness to aggregation: estimates at the HS6 level

The purpose of this robustness test is to verify whether the main results obtained at the HS8 level are sensitive to the level of aggregation, given that several of our variables were available only at the HS6 level and were assigned at that level.

To do this, the final models (value-weighted and unweighted) for the French and Canadian panels were re‑estimated at the HS6 aggregation level (tables C.1 and C.2). For the French panel, the core mechanisms are validated. The dominant role of indirect trade (share_indirect) as a barrier (*** in both models) was confirmed. Likewise, the other barriers identified in the unweighted model (negative and significant effects of share_indirect, log_avg_dist, and d_cons) are robust to this change in aggregation. For the Canadian panel, the robustness test confirms the results of the HS8‑level estimation. The barriers related to rules of origin, air freight, and transit, as well as the drivers linked to scale and finished products, are reproduced. Similarly, in the value-weighted model, the barrier identified for air freight and the positive effect identified for the variable d_cap are confirmed. Taken together, the mechanisms identified in our HS8‑level models are not statistical artefacts arising from the choice of aggregation level, but genuinely reflect underlying economic phenomena.

Table C.1. – Robustness test under HS6 aggregation (unweighted and value-weighted) – Panel of French exports to Canada

Note: Fractional logit panel model with fixed effects at the product (HS6) level as well as by year. Standard errors clustered at the HS6 level.
Statistical significance: *** p<0.001 ; ** p<0.01 ; * p<0.05 ; . p<0.10
Source: Global Affairs Canada. Authors' calculations.
Data: Eurostat (Comext)

Table C.2. – Robustness test under HS6 aggregation (unweighted and value-weighted) – Panel of Canadian exports to France

Note: Fractional logit panel model with fixed effects at the product (HS6) level as well as by year. Standard errors clustered at the HS6 level.
Statistical significance: *** p<0.001 ; ** p<0.01 ; * p<0.05 ; . p<0.10
Source: Global Affairs Canada. Authors' calculations.
Data: Eurostat (Comext).

C.2. Robustness to the imputation strategy: exclusion of imputed years for the indirect trade measurement value

This second test aims to verify the validity of the effect observed for the indirect trade variable by ensuring that it is not an artefact of the imputation strategy. To do so, the final models were re‑estimated on subsamples restricted to the years for which original (non‑imputed) data are available (tables C.3 and C.4). The results of this test not only confirm the initial conclusions regarding the indirect trade variable for both panels, but also reinforce the observation made about the complexity of rules of origin for the Canadian panel.

For the French panel, the effect of indirect trade is confirmed with stable magnitude and significance (***)—demonstrating that its role as a barrier is not an artefact. The other main effects (barriers related to capital goods and distance) are also confirmed. It should be noted that the reduced temporal variation in this restricted sample prevents reliable estimation of the coefficients for d_cons and roo_ri, due to their collinearity with the fixed effects. For the Canadian panel, two findings emerge. First, the effect of indirect trade is not only confirmed and negative, but appears stronger than in the initial model. This suggests that the imputation strategy was conservative and that the true magnitude of this logistical barrier is likely more severe. Second, the test reveals that the barrier posed by the complexity of rules of origin (roo_ri) is also stronger and robust (***) compared with the initial estimates. Overall, this robustness test validates the role of indirect transit as a barrier for both partners. Moreover, it suggests that for Canadian exporters, the importance of logistical and administrative barriers (notably the complexity of rules of origin) may have been underestimated in the main analysis.

Table C.3. – Robustness test to the imputation strategy (original years only) – French panel

Note: Fractional logit panel model with fixed effects at the product (HS6) level as well as by year. Standard errors clustered at the HS6 level.
Statistical significance: *** p<0.001 ; ** p<0.01 ; * p<0.05 ; . p<0.10
Source: Global Affairs Canada. Authors' calculations.
Data: Global Affairs Canada.

Table C.4. – Robustness test to the imputation strategy (original years only) – Canadian panel

Note: Fractional logit panel model with fixed effects at the product (HS6) level as well as by year. Standard errors clustered at the HS6 level.
Statistical significance: *** p<0.001 ; ** p<0.01 ; * p<0.05 ; . p<0.10
Source: Global Affairs Canada. Authors' calculations.
Data: Eurostat (Comext).