decreases. Therefore, the coefficient for HHI and the coefficients for number of bidders
have the correct positive signs. The coefficient for number of bidders is correct because
the reference is the indicator variable for four bidders. Thus, as expected, results show
that the price spread between price of price of beef and cattle increases as the number of
bidders decrease.
However, while the coefficient for HHI and the coefficients for number of bidders
have the correct sign, the coefficient for HHI is least twenty times bigger than the
coefficients of indicator variables for number of bidders. This suggests that the number
of firms is more important in explaining price markups than the number of bidders for a
particular lot of cattle. Thus, an aggregate model (such as NEIO) seem relatively more
consistent with the experimental data than a disaggregate model (such as structural
auction model).
The null hypothesis that an aggregate model (M2) is the correct model (H01: bid1
= bid1 = bid1 = 0) is rejected at the 5 % level based on a likelihood ratio (LR) test, since
LR = -2[log-likelihood M1- log-likelihood M3] = 10.2 > χ32,0.05=5.99. The null hypothesis
that a disaggregate model (M1) is the correct model (H02 : HHI ≤ 0)is also rejected at the
at the 5% level based on a one tailed t-test (t = 1.98>1.75 = t16, 0.05). Thus, although size
of the coefficients showed that the number of firms in the experimental game is more
important than the number of bidders for a particular lot of cattle, the number of bidders
does contain some (unique) information about pricing behavior in the game. Results
suggest that both the number of firms and bidders should be considered in the estimation.
Thus, there is some gain from considering both traditional NEIO and auction measures of
market power within the same model.
26