Asymmetric pricing dynamics with market power: investigating island data of the retail gasoline market

Abstract

This paper investigates how and why a link between market power and asymmetric pricing occurs. Analyzing unique island panel data of the Korean gasoline market, we exploit geographic separation as a reliable measure of market power. Our findings confirm a positive correlation between market power and price-response asymmetry. The empirical results on sticky pricing behaviors suggest that tacit collusion is the main channel through which market power influences asymmetric pricing. In addition, we examine the effect of station heterogeneity on asymmetric pricing to provide further evidence of tacit collusion in a localized market.

Appendices

Appendix: A Discussions on market relevance to the focal point tacit collusion model

Although Borenstein et al. (1997) suggest that past prices can serve as a focal point on which gas stations may coordinate, the literature has not offered an answer to the stability of past prices as a focal point. By contrast, we could consider monopoly prices as a natural focal point at first glance, provided that stations make use of a focal point to resolve coordination problems.³² We therefore discuss specific market characteristics that possibly help explain why past price outperforms monopoly prices and can serve as more appropriate focal points in this gasoline market.

First, differentiated features of the gasoline market cause stations difficulty in arriving at a consensus on a particular monopoly price as a focal point without communication (or even with communication). Gasoline stations are spatially differentiated, and their attributes such as convenience store and car wash also differentiate them from others. Furthermore, different brand stations face changing wholesale prices over time. These market characteristics predict multiple monopoly prices across stations that fluctuate over time and make it impossible for stations to coordinate on a particular monopoly price as a focal point.³³

Second, a price war could easily be triggered because gasoline itself is nearly homogeneous, even though stations are differentiated and their advertised prices are easily observable by both competitors and consumers. In this environment, a small price decrease can attract many consumers in a localized market, which resembles the Bertrand competition. Consequently, changing prices would be a signal of a price war to nearby competitors and thus deteriorate the stability of the coordination.³⁴ For these two reasons, we consider past prices a more sustainable focal point in the retail gasoline market than monopoly prices, which fluctuate with cost changes.

B Calculating the CRFs

To formulate the cumulative price response to cost changes, consider a simplified version of the regression equation, as follows:

$$ \Delta p_{it} = \sum\limits_{j = 1}^{J} {\left( {\alpha_{j}^{ + } \Delta p_{i,t - j}^{ + } + \alpha_{j}^{ - } \Delta p_{i,t - j}^{ - } } \right)} + \sum\limits_{j = 0}^{J} {\left( {\beta_{j}^{ + } \Delta c_{i,t - j}^{ + } + \beta_{j}^{ - } \Delta c_{i,t - j}^{ - } } \right)} + \theta \eta_{i,t - 1} + X_{it} \gamma + \xi_{i} + \varepsilon_{it} , $$

(B1)

where $\eta_{i,t - 1} = p_{i,t - 1} - \delta_{i} - \phi c_{i,t - 1}$.

For a unit increase in wholesale price at time $t$ ($\Delta c_{it} = 1$), the cumulative adjustment of prices after $k$ period is now given by

$$ \begin{aligned} {\text{CRF}}_{0}^{ + } & = \beta_{0}^{ + } \\ {\text{CRF}}_{1}^{ + } & = {\text{CRF}}_{0}^{ + } + \beta_{1}^{ + } + \theta \left( {{\text{CRF}}_{0}^{ + } - \phi } \right) + \alpha_{1}^{ + } \hbox{max} \left( {0,{\text{CRF}}_{0}^{ + } } \right) + \alpha_{1}^{ - } \hbox{min} \left( {0,{\text{CRF}}_{0}^{ + } } \right) \\ \vdots \\ {\text{CRF}}_{k}^{ + } & = {\text{CRF}}_{k - 1}^{ + } + \beta_{k}^{ + } + \theta \left( {{\text{CRF}}_{k - 1}^{ + } - \phi } \right) \\ & \quad + \sum\limits_{i = 1}^{\hbox{min} (J,k)} {\left[ {\alpha_{i}^{ + } \hbox{max} \left( {0,{\text{CRF}}_{k - i}^{ + } - {\text{CRF}}_{k - i - 1}^{ + } } \right) + \alpha_{i}^{ - } \hbox{min} \left( {0,{\text{CRF}}_{k - i}^{ + } - {\text{CRF}}_{k - i - 1}^{ + } } \right)} \right]}. \\ \end{aligned}$$

The CRFs comprise four parts: (1) The cumulative changes in prices until previous period, (2) the current price changes directly affected by the past cost changes, (3) the reversion effects by the error correction term, and (4) the effects by the lagged price changes. Similarly, we can derive the CRFs from a negative cost shock.

To apply this formula to our regression model, we only need to modify the coefficients of $\alpha$’s and $\beta$’s. For example, we replace $\alpha_{k}^{ + }$ with $\alpha_{k}^{ + ,m} + \alpha_{k}^{ + ,is}$ and $\beta_{k}^{ + }$ with $\beta_{k}^{ + ,m} + \beta_{k}^{ + ,is}$ to calculate the CRFs for stations on isolated islands.

C Supplementary results

In Table 8, we consider a similar regression model to Eq. (1); however, the price and cost variables are now interacted with self-service or branded stations indicators. We use the estimation results to calculate CRFs of low- and high-cost stations presented in Fig. 5.

Table 8 Asymmetric price adjustment of low- versus high-cost stations

Figure 6 depicts the same results as in panel b of Fig. 3 separately by margin sizes. Moreover, we add the equal-tail probability intervals to each CRF. The results confirm that the intervals for each CRF do not overlap over the time horizon.

Fig. 6

CRFs by island types and margins. a CRFs over high-margin period. b CRFs over low-margin period