Do Higher Gasoline Prices Reduce Road Traffic in Copenhagen?

1 Results

This analysis examines whether fuel price fluctuations are associated with lower road traffic in Copenhagen, using traffic counting station data from 2011 to 2013. In a model with road fixed effects, weekday controls, local rainfall, and a smooth time trend, the estimated elasticity of traffic with respect to the real gasoline price is about −0.559. This implies that a 1% increase in the real gasoline price decreases traffic by about 0.56%. When I test lagged and smoothed fuel prices, the negative relationship is still present in most checks, but the magnitude moves across specifications. So the exact size should be read with caution, even if the overall direction is fairly stable. Part of the larger effect may reflect Copenhagen's strong alternatives to driving, including cycling and public transport.

Table 1 — Regression estimates across model specifications

Specification	Elasticity	Std. error	p-value	R²
Preferred: real gasoline price, quadratic trend	−0.559	0.119	<0.001	0.976
Alternative price measure: nominal gasoline, quadratic trend	−0.403	0.110	<0.001	0.976
Alternative fuel type: nominal diesel, quadratic trend	−0.173	0.114	0.128	0.976
All specifications include road fixed effects, weekday fixed effects, and daily rainfall. Bold row indicates the preferred specification. Standard errors are clustered at the road level. The estimation sample contains 53,467 road-day observations from 2011 to 2013.

2 Method

The historical traffic counts from Copenhagen's fixed counting points in 2011 to 2013 are matched to the historical weather data from the nearest DMI weather station, before removing holidays and holiday-adjacent days. Real-prices for fuel are constructed using Statistics Denmark's CPI (PRIS1) and OK's fuel pricedata. The main model is:

\[ \log(\text{Traffic}_{it}) = \alpha_i + \delta_{d(t)} + \beta \log(RGP_t) + \gamma \,\text{Rain}_{it} + \theta_1 t + \theta_2 t^2 + \varepsilon_{it} \]

where $\text{Traffic}_{it}$ is the daily traffic count at road counting station $i$ on day $t$, $\alpha_i$ are road fixed effects capturing time-invariant differences across road counting stations, and $\delta_{d(t)}$ are weekday fixed effects capturing systematic differences between Mondays, Tuesdays, …, Sundays. $RGP_t$ is the real gasoline price on day $t$, $\text{Rain}_{it}$ is daily rainfall at the weather station matched to road station $i$, and $t$ and $t^2$ are linear and quadratic time trends allowing for non-linear development over time. The coefficient $\beta$ is the elasticity of traffic with respect to the real gasoline price, and $\varepsilon_{it}$ is the error term. Additional robustness checks use lagged and moving-average fuel prices, as well as alternative time controls and a city-day specification.

3 Robustness checks

The main result also looks fairly stable when the model is changed in a few simple ways. The negative relationship is still present when the gasoline price is shifted by one day, when prices are averaged over the previous week, and when traffic is aggregated to the city level instead of individual counting sites. At the same time, the size of the estimate becomes smaller in some alternative versions, which suggests that the exact magnitude should be interpreted with some caution even though the overall pattern remains negative in the most relevant checks.

Table 2 — Robustness checks

Specification	Elasticity	Std. error	p-value	R²
Real gasoline, main trend model	−0.559	0.119	<0.001	0.976
Real gasoline, lag 1	−0.290	0.122	0.018	0.977
Real gasoline, lag 7	−0.119	0.129	0.359	0.977
Real gasoline, MA7	−0.256	0.129	0.047	0.977
Real gasoline, MA14	−0.153	0.137	0.263	0.978
Real gasoline, MA30	0.120	0.152	0.433	0.978
Real gasoline, month fixed effects	−0.112	0.177	0.526	0.978
City-day real gasoline, trend	−0.703	0.234	0.003	0.623
Lag 1 and lag 7 shift the real gasoline price by one and seven days, respectively. MA7, MA14, and MA30 denote 7-, 14-, and 30-day moving averages. The city-day specification aggregates traffic to the daily city level. The first row repeats the main specification for comparison. Lag 1 and lag 7 shift the gasoline price by one and seven days, respectively. MA7, MA14, and MA30 denote 7-, 14-, and 30-day moving averages. The city-day specification aggregates traffic to the daily city level.

4 Figures

Figure 1

Weekly total traffic and real gasoline price, 2011–2014

The figure is shown for the full 2011–2014 period. The regression models are instead restricted to 2011–2013, because traffic measurement rules and coverage changed from 2014 onward, lowering the recorded traffic counts and making them less comparable to the earlier years.

Figure 2

Average traffic by weekday

Figure 3

Average traffic by rain intensity

Figure 4

Spatial distribution of traffic counting sites