The Effect of the Gotthard Base Tunnel on Road Traffic: A Synthetic Control Approach

Ann Hesse; Hannes Wallimann; Widar von Arx

REVIEW 4 major objections 5 minor 3 references

The Effect of the Gotthard Base Tunnel on Road Traffic: A Synthetic Control Approach

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Opening the Gotthard Base Tunnel reduced car traffic on the parallel motorway by 135 to 152 vehicles per day, a statistically significant decline of under 1%.

desk verdict A transparent, methodologically interesting synthetic control study of the Gotthard tunnel's effect on road traffic, but the inference is more fragile than the paper claims. read the letter →

arxiv 2505.21129 v2 pith:72227M5U submitted 2025-05-27 econ.GN q-fin.EC

Hannes Wallimann , Widar von Arx , Ann Hesse This is my paper

classification econ.GNq-fin.EC

keywords GotthardBaseTunnelmodalshiftpolicysyntheticcontrolmethoddifference-in-differencesinduceddemandroadtraffictransportevaluationalpinecrossings

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that the opening of the Gotthard Base Tunnel caused a small but real reduction in car traffic on the parallel Gotthard motorway. Using traffic counts from other alpine crossings to build a counterfactual Gotthard, it estimates that light vehicle traffic fell by 135 cars per day under synthetic control and 152 cars per day under synthetic difference-in-differences during the April-to-October tourist season, about 0.8% to 0.9% of the pre-tunnel average. The authors then compare this road reduction with the roughly 2,100 extra daily rail passengers and conclude that most new rail demand was induced travel made possible by faster trains, not cars switching to rail. If correct, the finding matters because it suggests the Swiss modal-shift policy, built on rail investment alone, will not meaningfully reduce tourist car traffic without additional push measures.

What carries the argument

The machinery is the synthetic-control comparison applied to demeaned traffic residuals. Because the Gotthard's raw traffic, about 16,365 vehicles per day, lies far outside the donor pool range and violates the convex-hull requirement, the paper subtracts each crossing's pre-treatment monthly average from its observed counts and then averages residuals by year, leaving a series of year-to-year deviations from seasonal norms that can be matched across crossings. The synthetic Gotthard is the positive weights, 0.768 on Bernina and 0.232 on Frejus, that minimize pre-treatment squared differences in these residuals, and the treatment effect is the post-2016 gap between Gotthard's residual and the weighted donor average. A complementary synthetic difference-in-differences estimator reweights both units and time periods to allow for unit and time fixed effects.

What would settle it

Apply the same demeaning-and-synthetic-control procedure to each donor crossing as a placebo, shifting the pseudo-treatment to 2017; if the placebo distribution produces gaps as negative as -135 vehicles per day, the main estimate is not distinguishable from chance.

Watch

Extended reading notes

Core claim

The central claim is that the Gotthard Base Tunnel's opening reduced use of the parallel motorway section by a statistically significant but economically small amount: 135 or 152 fewer light vehicles per day in April through October, about 1% below the pre-treatment baseline of roughly 16,365 daily vehicles. Because the treated unit is a single motorway section whose traffic volume is far larger than any potential control, identification rests on first demeaning each crossing by its pre-treatment monthly averages and then applying synthetic control to the residual year-to-year deviations. The synthetic Gotthard is built almost entirely from the Bernina and Frejus crossings, with weights of 76.8% and 23.2%. The authors read the small road effect together with the roughly 2,000-per-day increase in rail passengers as evidence that most of the new rail demand was induced rather than shifted from cars, concluding that road and rail are not close substitutes on this tourist corridor.

Load-bearing premise

The result stands or falls on the assumption that, after removing seasonal averages, Gotthard's traffic would have followed the weighted Bernina-Frejus trend in the absence of the tunnel; this parallel-trends-in-residuals assumption is estimated from only four pre-treatment years and is not validated by placebo or permutation tests.

Editorial extensions

If this is right

If the estimate is right, the modal-shift policy achieved at most a 1% reduction in tourist car traffic on the Gotthard axis in the first three years.
Most of the roughly 2,000 additional daily rail passengers after 2016 were new trips induced by faster rail travel, not former car drivers; applying the leisure occupancy factor of 1.89, only about 270 people per day left the car.
Road and rail are not close substitutes on this tourist corridor, so travel-time savings on rail mainly generated additional mobility rather than replacing car trips.
The effect size is stable when dropping the San Bernardino crossing and when matching only on outcome lags, but it becomes larger, at -617 vehicles per day, and statistically insignificant when the donor pool is Swiss leisure routes.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the near-1% road effect is causal, then cost-benefit appraisals of the tunnel should treat most new rail ridership as induced demand with its own welfare consequences, not as a transfer from road.
The contrast between a significant effect against alpine transit crossings and an insignificant effect against Swiss leisure routes suggests the reduction may be concentrated in long-distance transit traffic rather than domestic leisure trips, a hypothesis that could be tested with licence-plate data.
A natural extension is to re-run the design on data through 2024, after the Ceneri tunnel opened and post-Covid travel patterns stabilized; if the negative gap grows, that would support the claim that travel-time savings, not the tunnel's novelty, drive the shift.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

For your radar: this is a transparent, well-written synthetic control study of a major rail tunnel's effect on parallel road traffic. The headline estimate—a drop of 135–152 light vehicles per day on the Gotthard motorway, just under 1%—is plausible, but inference is more fragile than the paper lets on. The authors do something sensible: they apply SCM and SDID to a single treated crossing, and they handle the extreme traffic volume of the Gotthard by demeaning with pre-treatment monthly means. That residual transformation is a useful practical trick, and the paper is honest that the post-treatment window is only three years. The induced demand decomposition—comparing the estimated car reduction to rail ridership growth—is clearly labeled as an interpretation rather than a direct estimate.

What's actually new: this is the first causal estimate of the Gotthard Base Tunnel's effect on the parallel road route, and it's a nice addition to the sparse ex-post literature on rail investments and modal shift. The study design is clean: a donor pool of other Alpine crossings, explicit discussion of assumptions, and sensitivity checks (excluding San Bernardino, varying predictors, and a domestic tourist-route donor pool). That last sensitivity yields a larger but insignificant effect, which the paper honestly reports.

Where it's soft: identification rests on only four pre-treatment years (2013–2016) and three post-treatment years. The synthetic control weights (76.8% Bernina, 23.2% Fréjus) are estimated from those four annual residuals. With seven donors and four periods, a close in-sample fit is not reassuring. The paper's significance evidence is a bootstrap that resamples donor units 200 times—that perturbs the weights but does not test the key parallel-trends assumption in residuals. No placebo or permutation inference is reported, no in-time placebo, no leave-one-out donors. Three donors are excluded based on visual trend inspection, which is ad hoc though not unreasonable. The stress-test note is fair: the bootstrap cannot detect a scenario where Bernina and Fréjus deviated from Gotthard for reasons unrelated to the tunnel. That said, the two methods give consistent point estimates, and the pre-period fit in Figure 4 looks respectable. The central qualitative conclusion—a modest modal shift, most new rail demand being induced—is likely directionally right, but the statistical significance is not established at the conventional level given the non-standard inference.

Bottom line: this is a worthwhile case study for transport policy, and the authors deserve credit for a transparent execution. It merits a serious peer review, but the referee should require placebo inference, a more direct test of the parallel-trends assumption, and a more cautious interpretation of the bootstrap. I'd bring it to a reading group as a cautionary tale about inference in short synthetic control panels. If the authors add those tests and the result holds, it becomes a citable data point. As it stands, I wouldn't rely on the effect size, but I might cite the residual transformation technique.

Referee Report

4 major / 5 minor

Summary. The paper estimates the short-run causal effect of the 2016 opening of the Gotthard Base Tunnel on light-vehicle traffic on the parallel Gotthard motorway section. Using monthly April–October counts for 2013–2019, the authors residualize traffic by pre-treatment monthly means, aggregate to yearly residuals, and apply synthetic control (SCM) and synthetic difference-in-differences (SDID) with a donor pool of seven other Alpine crossings. They report a reduction of 135 vehicles per day (SCM) or 152 vehicles per day (SDID), corresponding to a decline of just under 1% relative to the pre-treatment mean, and label it statistically significant based on a bootstrap that draws seven donor units with replacement 200 times. From the comparison of this reduction (about 270 persons per day using an occupancy rate of 1.89) with the roughly 2,100 additional daily rail passengers, the paper infers that most new rail demand is induced rather than shifted from cars. Sensitivity analyses include excluding San Bernardino, changing the SCM predictor specification, and using a domestic tourist-route donor pool.

Significance. The case is policy-relevant: the Gotthard Base Tunnel is the centerpiece of Switzerland's modal-shift policy, and credible ex-post evidence on whether new rail capacity reduces parallel road traffic is scarce. The paper applies standard modern causal methods (SCM and SDID) and is transparent about data sources and the short post-treatment window. If the identification held, the finding that a major rail investment reduced road traffic by less than 1% while generating substantial induced rail demand would be a useful contribution to the transport-policy literature. However, the evidentiary basis is fragile: the counterfactual is estimated from only four pre-treatment yearly observations, and the reported confidence intervals come from a nonstandard bootstrap that does not test the parallel-trends assumption. The induced-demand decomposition is also not derived transparently. The paper's strengths are its clear research question, the use of two estimators, and the explicit discussion of several identifying assumptions.

major comments (4)

[Section 6.1 and Equation (2)] The synthetic control is fitted to only four pre-treatment annual residual observations (2013–2016) while the donor pool contains seven units. With more donor units than pre-treatment periods, a close in-sample fit is almost mechanical and carries little information about whether the weighted Bernina–Fréjus average would have tracked Gotthard absent the tunnel. The paper provides no placebo or permutation inference: no in-space placebo (artificially treating a donor unit) and no in-time placebo are shown. Because the identifying assumption is exactly that the residualized counterfactual path is valid, and this is untestable with four annual observations, the reported estimate rests on an unexamined parallel-trends-in-residuals assumption. This is load-bearing for the central causal claim and needs either additional evidence (e.g., placebo tests using the excluded or alternative donor units) or a much more cautious interpretation.
[Section 6.1, bootstrap procedure] The statistical significance claim relies on drawing seven donor units with replacement 200 times from a pool of seven units. This procedure resamples the donor pool, not the underlying time series, and it cannot detect a violation of the parallel-trends assumption: if Bernina and Fréjus both deviated from Gotthard for reasons unrelated to the tunnel, the bootstrap would still produce a tight interval centered on the estimated effect. With only seven donor units, the coverage properties of this bootstrap are unknown, and the text provides no simulation or theoretical justification. The confidence intervals [-269; -127] and [-291; -84] therefore do not establish statistical significance under the model's identifying assumptions. I recommend replacing this procedure with placebo-based inference (e.g., permuting treatment assignment across donor units) or clearly reframing the reported intervals as descriptive.
[Section 7 and Figure 6] The induced-demand decomposition (80.4% of public transport users 'would have traveled by public transport anyway', leaving 17.3% as induced demand) is not derivable from the material presented. The text compares the estimated road-traffic reduction of about 270 persons/day with the observed rail ridership increase of about 2,100/day, but the calculation of the 80.4% counterfactual is not shown, and it appears to assume that Gotthard rail ridership would have grown at the national long-distance average (16% over 2013–2019) while the actual growth was 141%. No equation or reference supports this decomposition, and the conclusion that 'induced demand is the main part of the additional passengers' is therefore not reproducible from the paper. This is a major gap because the induced-demand conclusion is a headline contribution and should be supported by an explicit, falsifiable calculation.
[Section 5 and Figure B.1] Three donor units (Tauern, Brenner, Karawanken) are excluded from the analysis based on visual inspection of the residual time series, which the text describes as showing a 'distinctly different, specifically, more positive development' compared with the other crossings. This data-driven selection of the donor pool is a specification choice that can overstate the precision of the synthetic control and is not subjected to sensitivity analysis. The paper should report results with these units included (or with alternative exclusion rules) to show that the headline estimate is not an artifact of this visual selection. As written, the exclusion undermines the stated justification for Assumption 2.

minor comments (5)

[Throughout] There are numerous typographical errors that should be corrected: 'unists' (Section 6.1), 'inutes' (Section 4.2), 'an decline' (Section 7), 'hightest' (Section 5), and 'Arkhagelsky' in the reference list (should be 'Arkhangelsky').
[Section 5] The residual transformation is described in words; a short algebraic definition of the monthly demeaning and the subsequent yearly averaging is given, but the text should clarify that the synthetic control is fitted on the yearly residual series, not on the monthly series, since this has important implications for the effective number of pre-treatment periods.
[Section 6.2] In the third sensitivity analysis, the sentence 'The effect of the new Gotthard Base Tunnel on car traffic, when applying the synthetic control method, amounts to a reduction of 135 vehicles per day... which is the same as in the original result, as the San Bernardino got a zero weight' is only meaningful because the excluded route had zero weight; this should be stated as a check, not as a new finding.
[Section 4.2] Assumption 5 (no external shocks) is asserted without discussion; given that the post-treatment period includes the 2019 European heatwave and other potential demand shocks, a brief discussion of why these do not invalidate the comparison would strengthen the paper.
[Appendix A, Table 2] The table reports travel time savings for a small set of origin–destination pairs, but it is not cited in the main text when the paper claims 'travel time savings of approximately 20 to 40 minutes'; please add a cross-reference.

Circularity Check

0 steps flagged · score 1.0 of 10

Synthetic-control estimate is non-circular; score reflects only non-load-bearing self-citations, not any reduction of the result to its inputs.

full rationale

The road-traffic effect is estimated from equations (1)-(2) with donor-pool crossings (Bernina 76.8%, Fréjus 23.2%) fitted to pre-treatment (2013-2016) residual traffic; the counterfactual is the weighted average of observed donor residuals, and the post-treatment gap is the effect. No post-treatment Gotthard value enters weight fitting, so the counterfactual is not defined in terms of the effect. The residual transformation subtracts pre-treatment monthly means, making the Gotthard pre-treatment residual zero by construction; Table 1 uses this only descriptively, not as evidence. The bootstrap significance test resamples donor units and cites Wallimann, Blättler, and von Arx (2023) as an example, but the procedure is described in the text and is standard resampling; the citation is illustrative, not load-bearing. Wallimann (2024) is cited alongside Abadie (2021) for standard assumptions, and Blättler et al. (2024) for a heterogeneity discussion; neither imports a unique theorem that forces the result. The induced-demand conclusion compares observed rail ridership growth (external SBB data) with the car reduction scaled by an external occupancy rate (1.89) and a stated national-growth assumption; it is an arithmetic decomposition, not a fitted target of the synthetic control. The paper's acknowledged limitations (three-year post-treatment window, no treatment-effect heterogeneity) and the lack of placebo inference are internal-validity concerns, not circularity.

Assumptions & free parameters 4 free parameters · 8 assumptions · 0 invented entities

The central claim rests on standard synthetic control identification assumptions, plus a stronger untested parallel-trends-in-residuals assumption, a non-standard bootstrap inference choice, and an external occupancy rate used for the induced demand conversion.

free parameters (4)

Synthetic control weights (Bernina, Frejus) = 0.768, 0.232
Estimated by fitting pre-treatment residuals to minimize squared differences (Section 6.1).
SDID weights = not reported
Estimated by the synthdid package from pre-treatment data (Section 6.1).
Pre-treatment monthly means for demeaning = month-specific means over 2013-2016
Computed from each crossing's own pre-treatment data to define residuals in Section 5.
Bootstrap resampling design = seven units drawn with replacement, 200 replications
Authors' choice; no theoretical justification in Section 6.1.

assumptions (8)

domain assumption Assumption 1 (no anticipation): travelers did not alter road or rail behavior before the GBT opened in December 2016
Stated in Section 4.2, Assumption 1; needed for causal interpretation of post-2016 changes.
domain assumption Assumption 2 (comparison group): donor pool crossings are sufficiently similar to the Gotthard
Section 4.2, Assumption 2; used to justify excluding Tauern, Brenner, and Karawanken in Section 5.
domain assumption Assumption 3 (convex hull): after demeaning, Gotthard's pre-treatment residuals lie inside the donor range
Section 4.2, Assumption 3; verified in Table 1 and Section 5.
domain assumption Assumption 4 (no spillover): the GBT did not affect traffic on donor crossings
Section 4.2, Assumption 4; only tested by excluding San Bernardino in Section 6.2.
domain assumption Assumption 5 (no external shocks): no unit-specific shocks to road traffic during 2013-2019
Section 4.2, Assumption 5; not tested.
domain assumption Parallel trends in residuals: absent treatment, Gotthard's deviations from its own seasonal means would follow the weighted donor average
Core identifying assumption introduced by the residual transformation in Section 5; untestable with T0=4.
ad hoc to paper Bootstrap validity: drawing seven donor units with replacement 200 times yields valid confidence intervals
Section 6.1; no coverage theory or simulation evidence is provided.
domain assumption Leisure car occupancy of 1.89 persons per vehicle
Section 7, from FSO/ARE microcensus; used to convert vehicle reductions into person counts.

how reviews work

0 comments

Cite this review

Pith. "Pith review of The Effect of the Gotthard Base Tunnel on Road Traffic: A Synthetic Control Approach." pith.science (2026). https://pith.science/paper/72227M5U

@misc{pith2026250521129,
  author       = {Pith},
  title        = {Pith review of: The Effect of the Gotthard Base Tunnel on Road Traffic: A Synthetic Control Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/72227M5U}},
  note         = {Machine review of arXiv:2505.21129}
}

read the original abstract

The opening of the Gotthard Base Tunnel in 2017, the longest railway tunnel in the world, marked a milestone in Swiss transport policy. The tunnel, a part of the New Rail Link through the Alps, serves as a key instrument of the so-called "modal shift policy," which aims to transfer transalpine freight traffic from road to rail. The reduction in travel time by train between northern and southern Switzerland raised expectations that a substantial share of tourist-oriented passenger traffic would also shift from car to rail. In this paper, we conduct a causal analysis of the impact of the Gotthard Base Tunnel's opening at the end of 2016 on the number of cars using the parallel Gotthard motorway section in the subsequent years. To this end, we apply the synthetic control and the synthetic difference-in-differences methods to construct a synthetic Gotthard motorway section based on a weighted combination of other alpine road crossings (a so-called donor pool) that did not experience the construction of a competing rail infrastructure. Our results reveal only a modest but statistically significant decline in the number of cars between the actual and the synthetic Gotthard motorway in the short run. Given the consistently strong and increasing demand for the new rail connection through the Gotthard Base Tunnel, we infer a substantial induced short-run demand effect resulting from the rail travel time savings.

Figures

Figures reproduced from arXiv: 2505.21129 by the authors.

**Figure 2.** Development of Daily Rail Traffic Volumes: Gotthard vs. overall Long-Distance [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Average Daily Road Traffic at the Gotthard Road Tunnel (G [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Demand development of the Gotthard motorway section and the synthetic counterpart [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]

**Figure 5.** Figure 5: Demand development of the Gotthard motorway section and the synthetic counterpart [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]

**Figure 6.** Figure 6: Decomposition of the 2019 public transport share [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

[1]

Using synthetic controls: Feasibility, data requirements, and methodological as- pects,

Abadie, A. (2021): “Using synthetic controls: Feasibility, data requirements, and methodological as- pects,” Journal of Economic Literature , 59(2), 391–425. Abadie, A., A. Diamond, and J. Hainmueller (2010): “Synthetic control methods for comparative case studies: Estimating the effect of California’s tobacco control program,” Journal of the American sta...

work page 2021
[2000]

23 Figure B.5: Boxplot of Hourly Vehicle Counts at the G ¨oschenen Counting Station, by Day Type (2019, Values <

work page 2019
[2019]

Figure B.4: Boxplot of Hourly Vehicle Counts at the G ¨oschenen Counting Station, by Day Type (2016, Values <

work page 2016

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Using synthetic controls: Feasibility, data requirements, and methodological as- pects,

Abadie, A. (2021): “Using synthetic controls: Feasibility, data requirements, and methodological as- pects,” Journal of Economic Literature , 59(2), 391–425. Abadie, A., A. Diamond, and J. Hainmueller (2010): “Synthetic control methods for comparative case studies: Estimating the effect of California’s tobacco control program,” Journal of the American sta...

work page 2021

[2] [2000]

23 Figure B.5: Boxplot of Hourly Vehicle Counts at the G ¨oschenen Counting Station, by Day Type (2019, Values <

work page 2019

[3] [2019]

Figure B.4: Boxplot of Hourly Vehicle Counts at the G ¨oschenen Counting Station, by Day Type (2016, Values <

work page 2016