To charge, or not to charge?

The two-clock instrument keeps those meanings apart. Move the registration clock, and the left-hand reading changes with the selected month. The right-hand reading shows the latest fleet observation whose reference date is not later than that month. It does not interpolate missing stock observations, move a January count to December, or treat a missing early snapshot as zero. This is a retrospective inspection of the frozen research edition, not a reconstruction of everything that had already been published at each historical date.

Monthly readings are volatile. August 2026 recorded 68,930 BEVs among 212,563 newly registered passenger cars, a 32.43% share. The trailing-year result is lower because it includes earlier months. It is calculated by dividing the sum of BEV registrations by the sum of all passenger-car registrations, rather than averaging twelve percentages with different market sizes. A spectacular month can be real without being a reliable description of the whole year.

Definitions also decide which cars enter the story. BEVs are battery-electric vehicles. Plug-in hybrids remain a separate category; conventional hybrids are not quietly added to either. A new registration is not necessarily a household's first electric car. It can be a replacement, an additional vehicle, a rental car, a dealer registration or a manufacturer's car. A domestic change of keeper does not itself add a car to the national fleet. We can observe these administrative categories without claiming to observe the personal decision behind each one.

The dispersion is not only a matter of levels. Denmark gained 16.66 percentage points between 2024 and 2025, Vietnam 21.01 points and Thailand 8.52. Canada fell 5.62 points, Jordan 8.89 and Nepal 6.00. Some of the steepest recent trajectories do not have enough consecutive history to enter the model contest. Vietnam remains visible as an observed market even though this design does not fit it. Excluding difficult histories from a forecast sample must not turn into excluding them from the descriptive story.

A popular repair is to start every country's clock when its BEV share first reaches 5%. That produces a useful comparison, but it does not establish a tipping point. Twelve of the sixty countries never cross 5% in their available histories; twenty-eight never cross 10%. Among the countries that do cross, subsequent paths still spread widely. The research also distinguishes a first annual crossing, confirmation in consecutive years and a later reversal. An annual observation cannot identify the exact day a market crossed a threshold, or establish that it stayed above it.

The phase-space figure gives the common-curve claim a harder test. The horizontal coordinate is the share in one year; the vertical coordinate is its change in the following year. A stable logistic process with common parameters would imply an orderly relationship between penetration and subsequent growth. The actual observations include very different gains, pauses and reversals at similar shares. Spotlight a country to follow its ordered path through that field. The resulting differences could reflect national parameters, policy, prices, supply or measurement. The picture challenges a universal common curve without pretending to identify which explanation wins.

Even a rising share deserves a denominator check. In the exact German record, BEV registrations increased by 164,533 between 2024 and 2025 while non-BEV registrations fell by 124,273. The total market grew slightly. The supplied Shapley decomposition assigns 4.86 percentage points of the share gain to additional BEVs and 0.71 points to fewer other new cars, averaging the two possible orders of changing those counts. That is an accounting decomposition, not a causal allocation. It nevertheless improves on treating every percentage-point increase as the same economic event.

Across the 48 eligible countries, Gompertz lowers one-year country-equal mean absolute error from 2.94 to 2.65 percentage points, roughly a 10% improvement. Logistic reaches 2.96 points and does not improve the overall average. At two years, Gompertz records 4.26 points against 4.73 for naive; logistic reaches 5.63. That supports keeping Gompertz in a short-horizon forecasting toolkit. It does not establish that every country follows Gompertz, or that the model with the best average error is the best model for each market.

The uncertainty is more interesting than a league table suggests. Country-cluster resampling gives Gompertz a one-year improvement of about 0.29 points, with a 95% interval of 0.02–0.61 points. When the sensitivity exercise preserves common forecast-origin shocks and gives origins equal weight, the estimated improvement is about 0.34 points but the interval includes a small deterioration. The figure plots model-minus-naive errors, so improvements appear on the negative side. The two exercises change both dependence assumptions and weighting; they are not two interchangeable confidence intervals around an identical estimand.

The apparent advantage also depends on the evaluation window. Among 24 eligible European peers, one-year errors are 2.59 points for Gompertz, 2.66 for logistic and 3.18 for naive. Restricting the all-country evaluation to 2023–2025 narrows Gompertz's advantage to 3.68 against 3.73 points. This is why an average gain should travel with its country sample, period and benchmark. Otherwise a useful conditional result can acquire the reputation of a general law.

Germany makes the horizon problem particularly clear. Gompertz's six-month error is 4.29 points, compared with 5.80 for seasonal naive. At twenty-four months it is 9.60 points, compared with 6.18. Freezing the model in December 2022 or December 2023 lets us inspect the paths that those fits produce before later observations arrive. The same family of curve that helps at a short horizon can miss a policy-sensitive sequence badly enough that a simple seasonal benchmark does better farther out.

The supplied interval diagnostics are not reassuring enough to decorate these paths with narrow probability bands. Nominal 80% international intervals cover only about 55–66% of realized outcomes across the reported comparisons. Only three German twenty-four-month origins have enough prior forecast errors to receive an interval. The charts therefore retain model disagreement and explicit assumption ranges rather than passing them off as calibrated certainty. All of these backtests use revised history; a complete archive of contemporaneously available releases and publication lags was not reconstructed.

Yet the aggregate rise is not merely a compositional illusion. The research constructs a balanced panel of 53 countries for 2020–2025 and adds an explicit residual for the uncovered market. Between 2024 and 2025, increasing penetration within markets contributes 2.93 percentage points to the global rise; changing market weights contributes 0.38. Together they reconcile to the 3.31-point increase. Approximately 88% of that year's rise comes from within-market changes. Weight changes can also work in the opposite direction: their contribution was negative in 2024.

Holding the 2020 weights fixed produces a 2025 BEV share of 16.82%, slightly above the actual 16.51%. This is a useful accounting countercheck, not a policy experiment in which somebody prevented national car markets from changing size. The exercise includes the residual geography and inherits the source's rounded shares and inferred total-market counts. Regional aggregates such as Europe and the EU are not added to their constituent countries; doing so would count the same market twice.

The distinction between BEVs and plug-in hybrids matters here too. The world's combined BEV-plus-PHEV share reached 25.0% in 2025. The additional 8.49 percentage points are PHEVs, not battery-electric sales. Combining the categories can be appropriate for a question about plug-in infrastructure or reported electric-vehicle sales, provided the definition stays visible. It cannot support an unannounced switch from a BEV claim to an all-plug-in headline.

The policy chronology separates events that are often compressed into one date. The decision to end the Umweltbonus was taken on 13 December 2023, announced on 16 December, and followed by the application deadline at the end of 17 December. Approved support was retained. In the new program, qualifying private registrations became eligible retroactively from 1 January 2026, while the application portal opened on 19 May. An announcement, a registration's eligibility date, an application and a disbursement are different events. The first actual payment date is not verified in the supplied research.

Private registrations tell a particularly strong descriptive story. Over January–August 2026, BEVs represented 34.77% of private registrations and 21.56% of commercial registrations. In the same months of 2025, the corresponding shares were 18.55% and 17.63%. Private BEV registrations rose from 116,490 to 241,506; commercial BEVs rose from 218,472 to 273,699. The widening gap is consistent with a program aimed at private buyers, but consistency with a mechanism is not identification of its effect.

Commercial registrations cannot simply become the untreated control group. They include dealers, manufacturers, rental companies and other businesses, with different product mixes and incentives. They also interact with private demand through demonstrators and used vehicles. The accessible joint keeper-by-powertrain table starts in 2024, so it cannot supply a fictional pre-termination comparison for 2023. The source's short-window contrasts and placebo dates expose sensitivity to timing and seasonality rather than establishing a causal event study.

Market structure adds context without filling the missing price data. VW, BMW, Mercedes, Škoda and Audi together account for about 40.0% of August 2026 BEV registrations; several of those brands share corporate ownership. SUVs account for 46.2% of that month's BEVs, while minis and small cars together account for 24.8%. These are composition statistics, not evidence that comparable electric and combustion cars have equal transaction prices or uses. A subsidy can affect buyers, pricing and product decisions at the same time, which is precisely why the absent transaction and order data matter.

The 400 district records account for 49,436,020 passenger cars and 2,031,870 BEVs. KBA separately reports an unallocated category with 50,467 cars and 2,390 BEVs. Adding it reconciles the district data to the national stock. The atlas does not put that residual into an invented place. Nor does it silently treat changing districts as a constant geography: Hanau becomes separate in 2026, and its relationship to Main-Kinzig prevents a simple reuse of the older census geography for both units.

The regression comparison excludes those two areas, additional conservative boundary-risk cases in Thüringen, and a Trier coding mismatch in which identical-looking codes do not represent identical jurisdictions. The final matched sample contains 395 areas. In the cross section, the charging coefficient is about 0.848 percentage points per unit of log(1 + public charging points), with a state-cluster standard error of 0.140. In the January 2023–January 2026 first-difference specification, the coefficient on charging growth is about 0.003, with a standard error of 0.170.