Can we ignore spatial dependence when evaluating mergers?

There are two refineries in the Czech Republic, both belonging to Unipetrol. These refineries produce about 40–50% of the gasoline consumed in the Czech Republic; the rest is imported from neighboring countries, mainly from Slovakia, Germany, Austria, and Poland. The biggest chains are usually integrated with refineries, either in the Czech Republic or in one of these countries: Benzina belongs to Unipetrol, Agip belonged to Eni, while Mol (including Papoil), Shell, and OMV belong to the eponymous companies. Similarly, Eurooil belongs to Čepro, a company wholly owned by the Czech Ministry of Finance, which manages gasoline pipelines and state gasoline reserves. These chains have their own distribution companies too. The smaller chains and the independent stations buy the gasoline from one of these refineries and are supplied by independent distributors. Most of these chains can easily switch their suppliers under the same conditions, as the price competition among their suppliers is strong (ÚOHS 2016).

For detail on Pumpdroid, see www.​pumpdroid.​com. Based on Google’s data, between 100,000 and 500,000 users have downloaded the app so far in 2020. Others use its iPhone version.

The same approach was taken for these reasons, e.g., by Kvasnička et al. (2018).

An alternative approach could be including all existing stations and estimating the SAR model with time-varying weight matrices. I do not use this approach because the prices are unreliable for most stations excluded from the balanced panel and the existence of these stations in some months is anyway not certain. Moreover, most of the excluded stations are relatively small and less important than the stations covered in the panel.

I use R packages fixest (Bergé 2018) to estimate the fixed-effects panel models FE and FE+ and splm (Millo and Piras 2012) to estimate the fixed-effects SAR panel model by the maximum likelihood method.

Driving distances are the fastest routes. They are found with GraphHopper (www.graphhopper.com), open source software for route planning, using Dijkstra and A* algorithms. The routing uses the road network provided by OpenStreetMaps (www.openstreetmap.org), a collaborative project which provides high-quality free maps. I use maps packed by geofabrik.de.

I do not include the number of competitors at greater distances. In the model with the number of competitors between 9 and 12 km, its parameter estimate is statistically insignificant and virtually zero and other parameter estimates are very close to the estimates presented in Table 3.

In this paper, I assume a typical exchange rate of 25 CZK per Euro.

For the formal definition and computation of direct, indirect, and total effect for SAR model, see LeSage and Pace (2009), pp. 34–39.

Czech association of petroleum industry and trade estimates gasoline consumption in the Czech Republic at around 2.1 billion liters each year, see https://​www.​cappo.​cz/​info/​vyvoj-spotreby-pohonnych-hmot-v-cr-v-roce-2018.