ABSTRACT
Our study assesses the performance of a threshold vector error correction model (TVECM) for the analysis of spatial price transmission. We specifically determine how inference about price regimes and the process of price adjustment to the established long-run equilibrium is affected when the stochastic transaction cost is exogenously or endogenously determined in fixed and flexible threshold models. The study first generates non-stationary but cointegrated price series and transaction costs based on true price transmission parameters: percentage of violations of spatial equilibrium, speed of price adjustment, threshold and inaction band using Monte Carlo simulations. Then, the generated data are used to estimate TVECMs under various assumptions on transaction cost variation and threshold specification. Finally, estimates of price transmission parameters are compared with the true parameters to evaluate the performance of model specifications. Our results indicate that the flexible TVECM accounting for transaction cost variation outperforms the standard fixed TVECM, which is consistent with findings from many previous studies in the literature. Our study also finds that although the flexible TVECM performs better than the fixed TVECM, it may not fully address the limitation of linear models when a real-world data represents a complex nonlinearity in the spatial price transmission.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 One can include several lagged transaction costs and test for proper lags, but we use only one-period lagged transaction cost in (10) for brevity.
2 EquationEquation 10(10) (10) focuses on the relationship between commodity price and transaction cost because our key objective is to examine the performance of threshold vector error correction model under various assumptions on transaction cost variation. However, as one of anonymous reviewers pointed out, it is possible to consider other factors such as local policy effects in (10). Once the time-relevant variation of other factors as well as transaction cost are filtered out, TVECM results could be compared to results from individual and time fixed effect models, which is beyond the scope of this paper and could be conducted as a future study.
3 When estimating TVECM, the supWald/LM test is usually conducted to test joint significance of the adjustment parameters across regimes after the estimated TVECM (Hansen and Seo Citation2002; Seo, 2006). However, the joint cointegration test is not needed in our study because our price data were generated to be cointegrated with one and two thresholds.