Comparison of the estimation results

As discussed in Chapter 6.2, the competitor coefficients constitute a link between the oligopolistic PTM model and the residual demand model. Since both contain an influ-ence that we call the markup channel, we expect the coefficients to be related. Firstly, the competitor coefficients in the residual demand model should lie above those in the PTM model. Figure 6.1 shows that this condition is mostly fulfilled for the SMC coeffi-cients in the WMP²⁹ market. In thirteen out of sixteen cases, the SMC coefficients are larger (or equal) in the residual demand model than in the PTM model. Secondly, we hypothesized that the difference between the SMC coefficients of the models is almost constant across the destinations. As Figure 6.1 shows, this is not the case. Nonetheless, the coefficients exhibit a correlation coefficient of 0.43 across the models.

Figure 6.1: Comparison of competitor coefficients for WMP

Source: own calculations

There are a lot of possible reasons why the expectation of constant differences is not fulfilled and the SMC coefficients are only loosely linked across the models. One rea-son is that a constant difference can only be expected when the price (as the dependent variable in the residual demand model) would be measured in the export country’s cur-rency. Furthermore, Australia – as an additional foreign competitor – was considered in the residual demand model but not in the oligopolistic PTM model. This means that the EU’s coefficients in the oligopolistic PTM model could be influenced by the omission of Australia when the variables are correlated. Finally, for most of the destination

29 Since the destination-specific version of the oligopolistic PTM model is only estimated for WMP, the results for the other products cannot be compared.

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

0.9 EU´s SMC coefficient in the PTM model EU´s SMC coefficient in residual demand model

tries the number of lags introduced in the oligopolistic PTM model is larger than in the residual demand model. This further complicates any comparison.

However, this raises the question of why the optimal lag numbers are so different across the models. The most likely explanation for this is that New Zealand’s SMC and ex-change rate – that are not part of the residual demand model – cause the need for more lags. Contrary to this explanation, however, we understand that a larger lag number in the oligopolistic PTM model (Chapter 4.5.2) primarily has an influence on the EU’s coefficients. Alternatively, in the residual demand model a significant further lag of a certain variable could be overshadowed by a lack of significance of the lags of the large number of other coefficients. Since most of the variables can be expected to be correlat-ed, we always used the same lag number for all variables included in the model. In addi-tion to the lag number, a further difference in the results is that for the residual demand model quarterly dummies are significant, whereas for the PTM model they are not. This means that when New Zealand`s marginal cost is accounted for, the price does not show any seasonal anomalies. Conversely, when only competitors’ marginal costs are consid-ered, seasonality is an issue. This supports our explanation in Chapter 5.5 of seasonal differences in competition, probably due to seasonality in the milk production.

Since a PTM model usually only provides proof of whether market power is existent or not, these results can hardly be compared with the estimate of the RDE. In general, only information on whether market power is supported by the models can be compared, not the extent of the market power. However, the case is different for the oligopolistic PTM model. In Chapter 4.5.2 we hypothesized that a large difference between | | and can be seen as an indication of a large markup. We tested for this hypothesis by regressing the UV surcharges on the coefficients’ values. Since this procedure delivered significant results, the fitted values of the regression could be used for a comparison with the opti-mal markup indicated by the RDE estimates. This is done for WMP in Figure 6.2. In addition to the fitted values and the optimal markup, the UV surcharges are displayed in the figure. As shown in Figure 6.2, the fitted values are very close to the actual UV sur-charges, with the exception of Singapore where a large difference is depicted. Interest-ingly, the RDE estimates support the large fitted value for Singapore. Besides this, the optimal markups indicated by the RDE estimates and the fitted values of the UV sur-charges do not show a similar pattern nor are they correlated at all. Moreover, the PTM results indicate the presence of market power for each destination country while the RDE estimates only support the existence of markups for single destinations. This result

could be expected for two reasons. Firstly, the RDE estimates already showed no link-age to the UV surcharges, and secondly, the value of the EU’s SMC coefficient in the residual demand model is not connected to the RDE estimates. Remember that the com-petitors’ coefficients constitute a link between the approaches.

Figure 6.2: Comparison of market power indicators³⁰ for WMP

Source: own calculations

A number of reasons can be considered for the lack of agreement between the two des-tination-specific measures of Fonterra’s market power. Both measures may not show the actual markup. For the RDE, it is discussed in Chapter 5.3.2 that the usage of OLS (or SUR) can be expected to bias the RDE estimates in the direction of rejection of market power – we therefore speak of conservative estimates. Such a bias can be differ-ently pronounced across the destination countries. However, even when the RDE esti-mates would be unbiased, according to the discussion in Chapter 5.2, they show the optimal markup of the firm and not the actual markup which can differ from the former.

In the case of the fitted values (PTM approach), it is not clear at all whether these values can be considered a good approximation of the markup. If quality premiums play an important role on the market, the assumptions of the OLS error term are surely violated since quality premiums are only positive surcharges and not zero on average. If the quality premiums are uncorrelated with the variables of the oligopolistic PTM model, they only bias the constant of the OLS model – i.e. the level of the fitted values. If, however, the quality premiums are correlated with the SMC or the exchange rate, the

30 Since the theoretical coefficients for China and the Philippines in the PTM model are outside the range which is consistent with theory, they are not used to calculate the regression results in Table 4.4 and no fitted values are displayed in Figure 6.2.

0%

5%

10%

15%

20%

25% UV surcharge

fitted values PTM model absolute value of the RDE

coefficients are biased too. Indeed, such a correlation is discussed in Chapter 4.5.2 and is considered to be possible. Therefore, the results in Table 4.4 could – in an extreme case – just imply that PTM is a good indicator of the existence of different qualities.

Even if this case is not likely for the dairy markets, it illuminates the possible problems with regard to fitted values in Figure 6.2. A further general problem is that PTM may prove the existence of a markup but a markup can also exist without observing PTM.

This is the case when the (residual) demand elasticity in the destination country is con-stant. This situation may be most likely when no competitors exist in the destination country, since the results in Chapter 4.5.2 show that PTM is mainly caused by the EU’s existence as a competitor. However, because the destination-specific results for WMP also show that PTM takes place in almost all single destinations analyzed, this problem seems to be negligible in our dataset. Moreover, another observation suggests that the fitted values may be not as bad in indicating market power as might be supposed. When the fitted values are compared with New Zealand’s import market shares, a correlation coefficient of 0.33 is exhibited. Although this value is not significant (p-value = 0.24), it is more promising than the lack of correlation in the case of the RDEs.

Nonetheless, since both indicators of market power in Figure 6.2 face limitations, we prefer using them to define a possible range for the markup. The actual markup should at least be as high as the minimum of both the UV surcharge and the absolute value of the significantly negative RDE. Conversely, the actual markup should not exceed the minimum of both the UV surcharge and the fitted value of the PTM model.

However, there is one clear similarity between the approaches. When the results are compared across the products, they do not change a lot – this holds true in both ap-proaches. For the residual demand approach, this applies for the average RDE and the share of significant RDEs. In the case of the non-destination-specific version of the oli-gopolistic PTM model, this is true for the PTM and competitor coefficient. There is only one exception: SMP. For SMP, a non-destination-specific PTM does not take place, but the RDE estimates imply the same degree of market power as for the other products. When the market for SMP is compared to the other markets, there are only two noticeable differences. Firstly, SMP can be expected to be more homogeneous and secondly, the US – as a third foreign competitor – is active in the market.