Firm characteristics and the ability to exercise market power: empirical evidence from the iron ore market

Following Schumpeter (1942), one may assess the finding of prices above marginal costs in light of dynamic competition as being the result of, e.g., goods of higher quality that, in the long run, may not lead to welfare losses but even to increasing consumer welfare.

Hereafter, when speaking of non-competitive behavior or the exercise of market power, we refer to a situation in which prices exceed marginal costs, i.e., we argue within a static competition framework.

Other measures include the concept of conjectural variations, the Hall–Roeger approach or the H-statistic [for recent applications of these approaches see e.g., Gunning and Sickles (2013), Christopoulou and Vermeulen (2012) and Huang and Liu (2014), respectively].

Throughout this paper, when speaking of firms exercising market power no explicit assumption is made about the way the market power is exerted. Neither does the methodology applied here rely on any game theoretical model.

The monotonicity and concavity restrictions are tested ex post after the estimation.

For the case of companies operating in multiple countries, the country with the most production activity is chosen.

‘SEC Form 20-F’ is a necessary form to file with the Securities and Exchange Commission (SEC) of the USA if the company is listed on the stock market in the USA.

For example, for Sesa Sterlite, only data on capital expenditure and total segment assets was available.

For the Ukraine, figures from UKRstat (2013) had to be used instead as data was not available from the OECD.

For 4 of the 10 companies, the fiscal year ends in June instead of December. Hence, without adjustment, different time periods would be compared. To adjust for these cases, two consecutive years are averaged, e.g., the average of the results for July 2004 to June 2005 and for July 2005 and June 2006 would form the value for the year 2005. A consequence of this adjustment is that the second half of 2004 and the first half of 2006 are included in the value for 2005. Furthermore, it reduces the number of observations from 100 to 96.

We thank an anonymous reviewer for raising this issue.

Note that only changes in the shadow cost may bias our markup estimates since we include firm fixed effects in our preferred specification, extracting the fixed part of markups. Another option to control for potential incentives from cooperation is to include time fixed effects as suggested by Puller (2009). Unfortunately, this is not feasible with the data at hand given the (highly) unbalanced nature of our panel data set and the resulting low number of observations in some years.

As the first-year observation of Atlas Iron may be an outlier, we re-estimated our model without this observation for a robustness check. The results confirm our previous findings.

For a systematic review on the treatment of endogeneity in SFA models see, e.g., Amsler et al. (2016, 2017).

In addition to our stochastic frontier model, we also estimate conventional OLS models. Using likelihood ratio (LR) tests, we evaluate whether a markup component exists at all. The LR tests have the null hypothesis: \(\lambda = 0\) with \(\lambda = \sigma _{u}/\sigma _{v}\) (Coelli et al. 2005). For all BC95 specifications, the null hypotheses that the OLS model is sufficient can be rejected at any conventional level of significance. Hence, the stochastic frontier model is preferred.

In case of BHP Billiton the estimated negative cost elasticities are in all likelihood due to the fact that we had to approximate the capital variable. Data on capital was only available on the total company level but not on the iron ore business segment level. Therefore, we used two alternative approximation approaches based on asset and revenue shares to proxy the capital variable. The estimation results for the two approaches do not differ significantly. Furthermore, all models were also estimated without the respective observations. The estimated coefficients are very similar to the coefficients presented in Table 4 and all cost elasticity estimates show positive values as required by economic theory. Therefore, in order to have more degrees of freedom, we opted to leave all observations in the frontier estimation and exclude the ones with negative cost elasticity estimates from the second-stage Lerner indices analysis. All estimation results are available from the authors upon request.

Given the differing magnitude between the two model specifications, the overall correlation of Lerner indices across specifications should be examined. The calculated Pearson correlation coefficient of 0.38 illustrates only a moderate correlation of Lerner indices across both specifications. This further stresses the importance of considering unobserved heterogeneity in the analysis.

LKAB is the only company (with large-scale operations) that is engaged in underground mining, which is associated with higher costs than production in open pit operations (Hellmer 1996).

Note that these figures are calculated in FE units in order to allow for comparison.

The main producer in the USA, Cliffs Natural Resources, however, does not seem to follow this hypothesis. Its average annual growth rate of production over the period 2000–2012 amounts to 10.3% and is therefore even larger than the rate for VALE. The time-varying Lerner indices, however, remain rather flat.

Hence, LKAB’s cost disadvantage may be outweighed by lower pelletizing costs due to a high magnetite fraction in the deposit (Hellmer 1996).

We thank an anonymous reviewer for suggesting the comparison of our estimated Lerner index values with individual profit-to-sales ratios. Full results on these figures are available from the authors upon request.

Note that this specification is not equivalent to individual fixed effects in the markup term, although each country is represented by one firm only. In contrast to individual fixed effects the reference group consists of a set of firms sharing the same characteristic (i.e., production in Australia) instead of one individual firm as in the fixed effects specification.

Time and firm indices are dropped for notational convenience.

As the marginal effects in the model without fixed effects (BC95) are negligible, we do not discuss them in the following. The results are available from the authors upon request.

This definition is used by Heij and Knapp (2014) and stems from the ship broker Braemar Seascope.