## Electric Vehicles, Tax incentives and Emissions: Evidence from Norway

## Source details

**Canonical URL:** [Electric Vehicles, Tax incentives and Emissions: Evidence from Norway](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021162-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021162-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021162-print-pdf.pdf.json)

---

### Abstract and Purpose
- Objective: Econometrically estimate emission savings from EV usage at the household level and compute back-of-the-envelope cost effectiveness of Norway’s tax incentives.
- Data scope: Universe of cars and households in Norway over the 2010-2019 period (with some gaps for the last year).
- Key takeaway preview:
  - Household-level emission savings from additional EVs are limited on average, yielding high implicit abatement costs of Norway’s tax incentives relative to emission savings.
  - Emission savings can be much larger if EVs replace the dirtiest cars.
  - Norway’s experience can inform other countries’ climate mitigation policies.

### Context, Scope, and Data
- Policy context and scope of emissions studied:
  - EVs are exempt from VAT and exempt from weight-based component and green component of one-off motor vehicle registration taxes; combined incentives can amount to more than 40 percent of the pre-tax price depending on assumptions.
  - EV owners benefit from not paying fuel excises/taxes, lower annual vehicle license fees, reductions/exemptions from certain tolls and parking fees, and past access to special road lanes.
  - Analysis focuses on reduction of exhaust CO2 emissions from conventional cars due to acquisition of EVs; excludes other exhaust emissions (NOx, PM), emissions during EV production, non-exhaust particle emissions during operation, indirect effects on electricity exports or generation abroad, and effects on share of renewable electricity in Norway.
- Data sources and coverage:
  - Norway’s passenger car registry and tax records, linked via individual identifiers and aggregated to households.
  - Time span: 2010-2019 (with some gaps for the last year); individual tax records available 2010-2018.
  - Universe: all cars and individuals residing in Norway; cars owned by households represent 88 percent of cars.
  - Sample used in econometric analysis after cleaning: 16.4 million household-year-level observations.
  - Vehicle information: engine, fuel type, brand, model, year of registration, and kilometers driven per year (meter readings from inspections every four years plus Statistics Norway estimates for intermittent years).
  - Pre-tax price of new cars (available from 2012) merged based on brand/model and engine characteristics; averages computed when multiple prices exist.

### Stylized Facts
- Pace of electrification:
  - The share of battery electric vehicles in total sales of new cars reached over 50% in 2020.
  - At current trends, the transition to full electrification of Norway’s car fleet would take several decades, though could be substantially faster given rapid growth in new-car EV share and the 2025 goal for tailpipe emission-free new vehicles.
- Ownership and usage patterns:
  - Of households that owned EVs in 2018: 37 percent owned only EVs; 46 percent also had one conventional car in addition to EV(s); 17 percent had more than one conventional car in addition to EV(s).
  - Average annual driving distance of EVs is 18 percent lower than that of the median car, controlling for a range of unobserved household characteristics; gap appears to be narrowing over time.
- Distributional observations:
  - Combined CO2 emissions from passenger cars used by households in the highest income quintile are more than seven times higher than those of households in the lowest income decile.
  - 5 percent of passenger cars account for 15 percent of all emissions from passenger cars.
  - Poorer households are less likely to own EVs.
  - Median income of households that own EVs was above 900,000 NOK, around 50 percent higher than that of households that own conventional cars only.

### Econometric Strategy and Baseline Results
- Outcome and identification:
  - Outcome: household-level CO2 emissions from passenger cars (household-owned cars).
  - Identification: change in household emissions associated with EV ownership reflects degree of substitution of ICEV usage by EVs, driven by household preferences and behavior.
- Baseline regression (Equation (1)): Emissions_hmt regressed on NumberEV_hmt, household controls (including household income and net wealth), municipality-year fixed effects, and household fixed effects.
- Principal empirical findings:
  - Purchasing an additional EV lowers household-level emissions by 1.17 tCO2 annually (specification 1, Table 1).
  - Interpretation: one additional EV saves around half of the average car-owning household’s passenger car emissions (average = 2.3 tCO2 annually).
- Back-of-the-envelope cost-effectiveness (baseline assumptions):
  - emissions_saved (annual) = 1.17 tCO2
  - lifetime assumed = 15 years
  - lifetime emission savings = 15 × 1.17 tCO2 = 17.55 tCO2
  - subsidies (VAT exemption cost, averaged EV price) ≈ USD 12,500
  - implied cost per tCO2 ≈ USD 710
  - Caveats: ignores registration taxes and other incentives, assumes attribution of savings to VAT exemption only, ignores substitution to cheaper conventional cars, and does not estimate causal effect of tax incentives on EV uptake.
- Comparison with benchmarks:
  - Simulated marginal abatement cost in Norway up to around USD 420 (Fæhn et al. (2020)) — imperfect benchmark.
  - Norway’s Ministry of Finance (2021 budget) implicit cost of all tax benefits for EVs ≈ USD 1,400 per tCO2 saved.
  - Literature range (Gillingham and Stock (2018)) places these estimates in the upper range compared with many other interventions.

### Robustness, Heterogeneity, and Replacement Scenarios (Table 1 and Table 2 summaries)
- Robustness:
  - Spec. 2: including households that do not own any cars — coefficient on number of EVs remains statistically significant and similar.
  - Spec. 3: excluding observations where annual mileage was estimated — coefficient on number of EVs slightly increases.
  - Spec. 4: first EV has largest effect; additional EVs yield smaller marginal savings.
  - Spec. 5: emission savings from EVs increased after 2015.
  - Spec. 6: emission savings larger in households that only own EVs (full replacement) than in mixed households.
- Replacement conditional results (Table 2):
  - If number of ICEVs is held constant (EVs added without replacing ICEVs), estimated emission savings drop substantially (Table 2, spec. 1: no_ev = -86.83***).
  - If total number of cars is held constant (EV replaces ICEV), estimated emission savings increase to 1.9 tCO2 annually (Table 2, spec. 2: no_icev / no_car coefficient 1,906.29*** interpreted as 1.9 tCO2).
  - Replacing ICEVs in the 5th quintile → household emissions drop by 3,831.44*** (i.e., 3.8 tCO2 annually).
  - Replacing ICEVs in the 10th decile → household emissions drop by 4,649.06*** (i.e., 4.6 tCO2 annually).
  - Median-efficiency replacement scenario implies annual emission savings ≈ 0.6 tCO2 (spec. 6 interpretation).

### Heterogeneity by Household Income (Table 3 and Section C)
- Baseline differences:
  - Households in the richest quintile emit 0.4 tCO2 higher than those in the lowest quintile (omitted category).
  - Interaction of NumberEV with income quintiles: emission savings of EVs are almost identical across income groups in baseline.
- Conditional results:
  - Holding ICEVs constant: annual emission savings fall across all quintiles but slightly increase with household income.
  - Holding total number of cars constant (ICEV replacement): emission savings increase across all income groups and slightly increase with household income.
- Selected coefficient excerpts (Table 3 and regressions reported):
  - no_icev col (1): 1,788.37*** (1.72)
  - no_car col (2): 1,888.52*** (1.78)
  - Income quintile dummies (netinc2q to netinc5q) and interaction terms netincXq#no_ev reported in full in the source.

### Regression Results and Additional Details (Section 3 highlights)
- Dependent variable: emissions (Total CO2 emissions per household in kg).
- Sample and fit:
  - Observations 9,358,297
  - R‐squared 0.64 (col (1)), 0.65 (col (2)), 0.74 (col (3)), 0.74 (col (4))
  - Mun.‐Year FE YES; Household FE YES
- Household characteristic coefficients (selected):
  - head_age: ‐13.62*** (0.22); ‐13.55*** (0.22); ‐4.61*** (0.18); ‐3.89*** (0.17)
  - head_female: ‐60.95*** (2.38); ‐57.41*** (2.36); ‐39.60*** (1.96); ‐39.27*** (1.95)
  - no_adult: 431.19*** (1.94); 443.92*** (1.93); 80.47*** (1.56); 60.27*** (1.55)
  - no_child: 73.52*** (1.68); 82.26*** (1.66); 67.26*** (1.36); 66.51*** (1.35)
  - netwealth: ‐0.02 (0.01); ‐0.00 (0.01); ‐0.02** (0.01); ‐0.02** (0.01)
- Notes: Robust standard errors in parentheses. Significance: *** p<0.01, ** p<0.05, * p<0.1.

### Mileage Evidence (Appendix III)
- Empirical specification: OLS at car level with dependent variable annual driving distance (distance_km in thousand kilometers), controls include household-year FE and municipality-year FE.
- Results (Appendix Table 5):
  - ev coefficient col (1) kmyear: ‐2.20*** (0.01)
  - ev coefficient col (2) kmyear: ‐2.86*** (0.02)
  - ev_2018 coefficient (col (2)): 2.17*** (0.03)
  - Constant: 12.21*** (0.00); 12.20*** (0.00)
  - Observations 12,634,975; R‐squared 0.43 (both specs)
- Interpretation:
  - EVs are used significantly less than conventional cars by 2,200 km per year relative to a median annual driving distance of around 12,000 km.
  - In 2018, the difference in annual driving distance between EVs and conventional cars became smaller but remained negative.

### Descriptive Statistics (selected)
- no_ev mean 0.0425; sd 0.212; p5 0; p10 0; p50 0; p90 0; p95 0
- no_icev mean 1.232; sd 0.549; p5 1; p10 1; p50 1; p90 2; p95 2
- no_car mean 1.297; sd 0.550; p5 1; p10 1; p50 1; p90 2; p95 2
- co2_emi_hh mean 2,702; sd 1,910; p5 527.0; p10 854.7; p50 2,306; p90 5,003; p95 6,170
- head_age mean 52.13; sd 15.84; p5 28; p10 31; p50 51; p90 74; p95 79
- netinc mean 6.691; sd 10.23; p5 2.300; p10 2.800; p50 5.960; p90 10.88; p95 12.99
- netwealth mean 29.78; sd 169.9; p5 ‐6; p10 ‐1.600; p50 18.30; p90 62; p95 87.20
- Total CO2 emissions per household (All mean 2,736; p5 633.0; p10 923.0; p25 1,520; p50 2,326; p75 3,454; p90 5,019; p95 6,188)
- Total CO2 emissions per car (All mean 2,129; p5 512.9; p10 771.2; p25 1,290; p50 1,969; p75 2,725; p90 3,573; p95 4,233)

### Main Conclusions and Policy Implications
- Main conclusions:
  - Emission savings at the household level from purchasing EVs vary dramatically by replacement scenario: largest savings when EVs replace the most polluting ICEVs; much smaller savings if EVs replace fuel-efficient cars or merely increase household car stock.
  - Average household-level emission reduction per additional EV is limited (1.17 tCO2 annually under baseline), implying relatively large fiscal cost per unit of emissions abated under Norway’s tax incentives.
  - VAT exemption and related incentives are regressive in Norway’s context because higher-income households adopt EVs more frequently while producing larger emissions, concentrating benefits in upper income quintiles.
- Policy recommendations:
  - Recalibrate tax incentives to better target emission reductions, for example by encouraging replacement of low-value-high-pollution cars by EVs in a revenue-neutral way.
  - Possible instruments:
    - Targeted subsidies to scrap dirty cars when replaced by EVs.
    - Tax or regulatory measures such as capping the amount of the VAT exemption on high-end EVs and/or levying annual road tax on some of the most luxurious EVs.
  - Revenue neutrality could be achieved by increasing the tax burden on high-end EVs and returning additional tax revenue to households in ways that account for income levels and gasoline consumption patterns.
- Further research:
  - Combine current estimates with explicit econometric modeling of the decision to buy cars to better estimate how incentives change EV purchase behavior and cost effectiveness.

*IMF Working Paper WP/21/162 — Prepared by Youssouf Camara, Bjart Holtsmark, and Florian Misch (June 2021).*

### Section 1

### Electric Vehicles, Tax incentives and Emissions: Evidence from Norway

### Abstract and Purpose
- Objective: Econometrically estimate emission savings from EV usage at the household level and compute back-of-the-envelope cost effectiveness of Norway’s tax incentives.
- Data scope: Universe of cars and households in Norway over the 2010-2019 period (with some gaps for the last year).
- Key takeaway preview:
  - Household-level emission savings from additional EVs are limited on average, yielding high implicit abatement costs of Norway’s tax incentives relative to emission savings.
  - Emission savings can be much larger if EVs replace the dirtiest cars.
  - Norway’s experience can inform other countries’ climate mitigation policies.

### Introduction — Context and Motivation
- Policy relevance: EV promotion features in national Covid recovery programs and climate mitigation strategies.
- Conceptual point: Emission savings from purchasing an EV depend critically on which ICEVs are substituted and household behavior/preferences.
- Scope of emissions studied: Reduction of exhaust CO2 emissions from conventional cars due to acquisition of EVs; excludes:
  - other exhaust emissions (NOx, PM),
  - emissions during EV production,
  - non-exhaust particle emissions during operation,
  - indirect effects on electricity exports or generation abroad,
  - effects on share of renewable electricity in Norway.
- Norway-specific policy context:
  - EVs are exempt from VAT and exempt from weight-based component and green component of one-off motor vehicle registration taxes; combined incentives can amount to more than 40 percent of the pre-tax price depending on assumptions.
  - EV owners also benefit from not paying fuel excises/taxes, lower annual vehicle license fees, reductions/exemptions from certain tolls and parking fees, and past access to special road lanes.
  - Norway has one of the largest numbers of public charging stations per capita.
  - Policy goal: Norway aims for all new vehicles sold to be tailpipe emission-free by 2025.

### Data
- Sources: Norway’s passenger car registry and tax records, linked via individual identifiers and aggregated to households.
- Coverage:
  - Time span: 2010-2019 (with some gaps for the last year); individual tax records available 2010-2018.
  - Universe: all cars and individuals residing in Norway; cars owned by households represent 88 percent of cars.
- Sample used in econometric analysis after cleaning: 16.4 million household-year-level observations.
- Vehicle information includes engine, fuel type, brand, model, year of registration, and kilometers driven per year (meter readings from inspections every four years plus Statistics Norway estimates for intermittent years).
- Pre-tax price of new cars (available from 2012) merged based on brand/model and engine characteristics; averages computed when multiple prices exist.

### Stylized Facts (motivation for empirical analysis)
- Pace of electrification:
  - The share of battery electric vehicles in total sales of new cars reached over 50% in 2020.
  - At current trends, the transition to full electrification of Norway’s car fleet would take several decades, though the transition could be substantially faster given rapid growth in new-car EV share and the 2025 goal for tailpipe emission-free new vehicles.
- Ownership patterns:
  - Of households that owned EVs in 2018:
    - 37 percent owned only EVs,
    - 46 percent also had one conventional car in addition to EV(s),
    - 17 percent had more than one conventional car in addition to EV(s).
  - Implication: degree of substitution between EVs and ICEVs varies widely and affects emission outcomes.
- Usage intensity:
  - Average annual driving distance of EVs is 18 percent lower than that of the median car, controlling for a range of unobserved household characteristics.
  - There is some indication this gap is narrowing over time (possibly due to improvements in EV range).
- Distribution of emissions:
  - Combined CO2 emissions from passenger cars used by households in the highest income quintile are more than seven times higher than those of households in the lowest income decile.
  - 5 percent of passenger cars account for 15 percent of all emissions from passenger cars (reflecting poor fuel efficiency and high annual mileage).
- Income and equity:
  - Poorer households are less likely to own EVs.
  - Median income of households that own EVs was above 900,000 NOK, around 50 percent higher than that of households that own conventional cars only.
  - Trade-off noted: richer households generate more emissions (so EV adoption among them yields larger emissions reductions), while tax incentives (e.g., VAT exemption) concentrate benefits in upper income quintiles, raising equity concerns.

### Econometric Approach and Key Results (baseline)
- Outcome of interest: household-level CO2 emissions from passenger cars (household-owned cars).
- Identification logic: change in household emissions associated with EV ownership reflects degree of substitution of ICEV usage by EVs, driven by household preferences and behavior.
- Principal empirical finding (summary):
  - Purchasing an EV is associated with a significant decrease in household passenger car emissions, controlling for observed variables and unobserved effects.
  - Magnitude of average emission reduction is limited, implying relatively large fiscal cost per unit of emissions abated under Norway’s tax incentives.
  - Emission savings are substantially larger when EV purchases replace the dirtiest conventional cars.

### Policy Implications and Interpretation
- Cost-effectiveness:
  - Given limited average household-level emission savings, the implicit abatement cost of Norway’s tax incentives is relatively large.
- Targeting potential:
  - Emission savings increase considerably if policies encourage replacement of the dirtiest and/or highest-mileage ICEVs.
  - Policy options to increase realized emission savings include incentivizing scrapping of replaced ICEVs to prevent resale domestically or abroad.
- Equity considerations:
  - Tax incentives for EV purchases are regressive in Norway’s context because richer households benefit disproportionately; this mirrors broader literature on subsidies for durable goods.
  - Regressivity could be mitigated by returning additional tax revenue to households in ways that account for income levels and gasoline consumption patterns.

*IMF Working Paper WP/21/162 — Prepared by Youssouf Camara, Bjart Holtsmark, and Florian Misch (June 2021).*

### Section 2

### wpiea2021162-print-pdf - Section 2

### Estimation approach and baseline specification
- Household total CO2 emission (Emissions_hmt) is defined as the product of annual mileage and CO2 emissions per km, summed over all cars owned by household h in municipality m in year t.
- Baseline regression (Equation (1)) includes NumberEV_hmt (number of EVs owned by household h in municipality m in year t), household controls (including household income and net wealth), municipality-year fixed effects, and household fixed effects.
- Sample: all households that own electric vehicles or conventional cars (plug-in hybrids and some other car types omitted).

### Baseline empirical findings (Table 1)
- Purchasing an additional EV lowers household-level emissions by 1.17 tCO2 annually (specification 1, Table 1).
- Interpretation: one additional EV saves around half of the average car-owning household’s passenger car emissions (average = 2.3 tCO2 annually).
- Back-of-the-envelope cost-effectiveness calculation:
  - emissions_saved (annual) = 1.17 tCO2
  - lifetime assumed = 15 years
  - lifetime emission savings = 15 × 1.17 tCO2 = 17.55 tCO2
  - subsidies (VAT exemption cost, averaged EV price) ≈ USD 12,500
  - implied cost per tCO2 ≈ USD 710
  - Caveats noted: ignores registration taxes and other incentives, assumes attribution of savings to VAT exemption only, ignores substitution to cheaper conventional cars, and does not estimate causal effect of tax incentives on EV uptake.
- Comparison with benchmarks:
  - Simulated marginal abatement cost in Norway up to around USD 420 (Fæhn et al. (2020)) — noted as an imperfect benchmark.
  - Norway’s Ministry of Finance (2021 budget) implicit cost of all tax benefits for EVs ≈ USD 1,400 per tCO2 saved (broadly consistent with the paper’s VAT-only estimate).
  - Literature range (Gillingham and Stock (2018)) places these estimates in the upper range compared with many other interventions.

### Key coefficients and household controls (selected exact values from Table 1)
- no_ev (spec. 1): -1,171.96*** (standard error reported as (3.78) in table)
- head_age: -15.82*** (0.22)
- head_female: -84.00*** (2.38)
- no_adult: 535.78*** (2.01)
- no_child: 122.50*** (1.65)
- netinc: 0.01* (0.00)
- netwealth: -0.09 (0.12)
- Constant (spec. 1): 2,549.95*** (12.94)
- Observations (spec. 1): 9,358,297
- R-squared (spec. 1): 0.65
- Mun.-Year FE: YES; Household FE: YES

### Robustness and heterogeneity (Table 1 specifications 2–6)
- Spec. 2: includes households that do not own any cars — coefficient on number of EVs remains statistically significant and similar in magnitude.
- Spec. 3: excludes observations where annual mileage was estimated — coefficient on number of EVs slightly increases, suggesting mileage estimation would downward-bias emission-savings magnitude.
- Spec. 4: the first EV owned by a household has the largest effect; emission savings from additional EVs are smaller.
- Spec. 5: emission savings from EVs increased after 2015 (possible increased usage / technological improvements).
- Spec. 6: emission savings larger in households that only own EVs (full replacement) than in households that own a mix of ICEVs and EVs.

### Emission savings conditional on replacement of ICEVs (Section B; Table 2)
- Identity: NoCars_hmt = NoICEV_hmt + NoEV_hmt (Equation (3)) — regressions include only two of these variables at a time to avoid collinearity.
- If number of ICEVs is held constant (implying EVs are added without replacing ICEVs), estimated emission savings of one additional EV drop significantly (specification 1, Table 2: no_ev = -86.83*** in that specification).
- If total number of cars is held constant (implying EV replaces ICEV), estimated emission savings increase to 1.9 tCO2 annually (Table 2, specification 2: no_icev / no_car coefficient 1,906.29*** corresponding to interpretation in text).
- Replacement by ICEVs in different emission quintiles (Table 2, spec. 3):
  - Replacing ICEVs in the 5th quintile (highest-emitting quintile) → household emissions drop by 3,831.44*** (i.e., 3.8 tCO2 annually as described in text).
- Replacement by ICEVs in deciles (Table 2, spec. 4):
  - Replacing ICEVs in the 10th decile → household emissions drop by 4,649.06*** (i.e., 4.6 tCO2 annually as described in text).
- Categorizing ICEVs by CO2 intensity (spec. 5 and 6):
  - Replacing least fuel efficient ICEVs (top efficiency quintile) yields larger savings than baseline but smaller than quintile/decile total-emission replacements.
  - Specification 6 focusing on median-efficiency replacement scenario implies annual emission savings ≈ 0.6 tCO2 (textual summary).

### Heterogeneity by household income (Section C; Table 3 summary)
- Baseline: households in the richest quintile emit 0.4 tCO2 higher than those in the lowest quintile (omitted category).
- Interaction of NumberEV with income quintiles: emission savings of EVs are almost identical across income groups.
- When controlling for the number of ICEVs (holding ICEVs constant), annual emission savings fall across all quintiles but slightly increase with household income.
- When controlling for total number of cars (assuming ICEV replacement), emission savings increase across all income groups and again slightly increase with household income.

*Italic: Source — wpiea2021162-print-pdf - Section 2*

### Section 3

### wpiea2021162-print-pdf - Section 3

### Regression results: Emission savings and household income
- Dependent variable: emissions (Total CO2 emissions per household in kg).
- Sample size and fit:
  - Observations 9,358,297
  - R‐squared 0.64 (col (1)), 0.65 (col (2)), 0.74 (col (3)), 0.74 (col (4))
  - Mun.‐Year FE YES; Household FE YES
- Selected coefficient estimates (standard errors in parentheses):
  - no_icev:
    - col (1): 1,788.37*** (1.72)
  - no_car:
    - col (2): 1,888.52*** (1.78)
  - Income quintile dummies (netincXq):
    - netinc2q: 47.70*** (2.28); 45.45*** (2.28); 34.87*** (2.01); 34.51*** (2.00)
    - netinc3q: 171.08*** (2.94); 168.19*** (2.93); 74.07*** (2.53); 68.42*** (2.51)
    - netinc4q: 311.66*** (3.44); 317.70*** (3.43); 107.33*** (2.92); 96.78*** (2.90)
    - netinc5q: 430.75*** (4.05); 464.99*** (4.03); 132.53*** (3.42); 116.45*** (3.39)
  - Interaction terms netincXq#no_ev (effect of income quintile when no_ev):
    - netinc1q#no_ev: ‐1,200.34*** (14.92); ‐74.95*** (11.83); ‐1,845.49*** (11.70)
    - netinc2q#no_ev: ‐1,234.38*** (10.25); ‐129.67*** (7.73); ‐1,896.30*** (7.64)
    - netinc3q#no_ev: ‐1,140.46*** (8.19); ‐185.94*** (6.34); ‐1,969.65*** (6.29)
    - netinc4q#no_ev: ‐1,109.10*** (5.92); ‐181.21*** (4.58); ‐1,971.62*** (4.56)
    - netinc5q#no_ev: ‐1,227.43*** (5.11); ‐216.69*** (3.90); ‐2,001.69*** (3.85)
  - Household characteristics:
    - head_age: ‐13.62*** (0.22); ‐13.55*** (0.22); ‐4.61*** (0.18); ‐3.89*** (0.17)
    - head_female: ‐60.95*** (2.38); ‐57.41*** (2.36); ‐39.60*** (1.96); ‐39.27*** (1.95)
    - no_adult: 431.19*** (1.94); 443.92*** (1.93); 80.47*** (1.56); 60.27*** (1.55)
    - no_child: 73.52*** (1.68); 82.26*** (1.66); 67.26*** (1.36); 66.51*** (1.35)
    - netwealth: ‐0.02 (0.01); ‐0.00 (0.01); ‐0.02** (0.01); ‐0.02** (0.01)
  - Constant terms:
    - 2,409.31*** (12.88); 2,418.95*** (12.82); 503.46*** (10.42); 353.15*** (10.36)
- Notes:
  - Robust standard errors in parentheses.
  - Significance: *** p<0.01, ** p<0.05, * p<0.1.

### Main conclusions and interpretation
- Heterogeneity of emission savings from EV adoption:
  - Emission savings at the household level from purchasing EVs vary dramatically by replacement scenario:
    - Largest savings occur if EVs replace the most polluting ICEVs.
    - Much smaller savings if EVs replace fuel-efficient cars or merely increase the household car stock.
- Distributional aspects of VAT exemption:
  - The VAT exemption is regressive because higher-income households pick up EVs more frequently.
  - Emission savings of EVs are broadly similar across households of different incomes.
- Caveats and limitations highlighted:
  - Emission savings could increase in the future if EV maximum range increases and usage rises.
  - Aggregate emission savings depend on whether replaced cars are scrapped versus sold domestically or abroad for continued use.
  - Analysis ignores externalities from EV use and non-exhaust emissions.
  - Emissions during production are ignored.
- Cost effectiveness of VAT exemptions:
  - A simple back-of-the-envelope calculation using the average price of newly purchased EVs implies the implicit cost of VAT exemptions is high relative to average household-level emission savings.
- Policy-relevant observations on vehicle stock:
  - Among cars with annual emissions in the top quintile in 2018, 10 percent are older than 15 years and/or have odometer readings of almost 200,000 km, or more.

### Policy implications and recommendations
- Recalibration of tax incentives:
  - Consider targeting incentives to encourage replacement of low-value-high-pollution cars by EVs in a revenue-neutral way.
  - Possible instruments:
    - Targeted subsidies to scrap dirty cars when replaced by EVs.
    - Tax or regulatory measures such as capping the amount of the VAT exemption on high-end EVs and/or levying annual road tax on some of the most luxurious EVs.
  - Revenue neutrality could be satisfied by increasing the tax burden on high-end EVs.
- Further research recommended:
  - Determine more exact parameters for recalibration using data in this paper.
  - Combine current estimates with explicit econometric modelling of the decision to buy cars (as in Johansen, 2021) to better estimate how incentives change EV purchase behavior and cost effectiveness.

### Appendix highlights: variables and descriptive statistics
- Key variable definitions (selected):
  - head_age: Age of the head household
  - head_female: Dummy for the female head household
  - no_adult: Number of adults per household
  - no_child: Number of children per household
  - netinc: Net income of household in 100,000 NOK
  - netwealth: Net wealth of household in 100,000 NOK
  - netinc1q, ..., netinc5q: Dummy variable if household income is in the 1st, ..., 5th quintile for each year
  - no_ev: Number of electric vehicles (EVs) owned by the household
  - no_icev: Number of conv. cars (ICEVs) owned by the household
  - no_car: Number of cars owned by the household
  - emissions: Total CO2 emissions per household in kg
- Descriptive statistics (sample of baseline specification):
  - no_ev mean 0.0425; sd 0.212; p5 0; p10 0; p50 0; p90 0; p95 0
  - no_icev mean 1.232; sd 0.549; p5 1; p10 1; p50 1; p90 2; p95 2
  - no_car mean 1.297; sd 0.550; p5 1; p10 1; p50 1; p90 2; p95 2
  - co2_emi_hh mean 2,702; sd 1,910; p5 527.0; p10 854.7; p50 2,306; p90 5,003; p95 6,170
  - head_age mean 52.13; sd 15.84; p5 28; p10 31; p50 51; p90 74; p95 79
  - head_female mean 0.340; sd 0.474; p5 0; p10 0; p50 0; p90 1; p95 1
  - netinc mean 6.691; sd 10.23; p5 2.300; p10 2.800; p50 5.960; p90 10.88; p95 12.99
  - netwealth mean 29.78; sd 169.9; p5 ‐6; p10 ‐1.600; p50 18.30; p90 62; p95 87.20
- Total CO2 emissions per household over time (selected):
  - All mean 2,736; p5 633.0; p10 923.0; p25 1,520; p50 2,326; p75 3,454; p90 5,019; p95 6,188
  - 2018 mean 2,518; p5 549.3; p10 798.3; p25 1,325; p50 2,081; p75 3,198; p90 4,756; p95 5,942
- Total CO2 emissions per car over time (selected):
  - All mean 2,129; p5 512.9; p10 771.2; p25 1,290; p50 1,969; p75 2,725; p90 3,573; p95 4,233
  - 2018 mean 1,939; p5 459.1; p10 687.6; p25 1,150; p50 1,776; p75 2,492; p90 3,317; p95 3,959

### Appendix III: Annual mileage of EVs versus conventional cars
- Empirical specification: OLS at car level with dependent variable annual driving distance (distance_km in thousand kilometers). Controls: household-year FE and municipality-year FE.
- Results (Appendix Table 5):
  - ev coefficient:
    - col (1) kmyear: ‐2.20*** (0.01)
    - col (2) kmyear: ‐2.86*** (0.02)
  - ev_2018 coefficient (col (2)): 2.17*** (0.03)
  - Constant: 12.21*** (0.00); 12.20*** (0.00)
  - Observations 12,634,975; R‐squared 0.43 (both specs)
- Interpretation:
  - Specification 1: EVs are used significantly less than conventional cars by 2,200 km per year relative to a median annual driving distance of around 12,000 km.
  - Specification 2: In 2018, the difference in annual driving distance between EVs and conventional cars became smaller but remained negative.

*Source: wpiea2021162-print-pdf - Section 3*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021162-print-pdf.pdf_