Threshold effects in price transmission: the case of Brazilian wheat, maize, and soya prices.

Readers are reminded that within the current framework, a test for symmetry is not equivalent to a test for a threshold, unless [theta] = 0 and, as noted above for the series used in this article there was considerable support for the Band-TAR model with overshooting ([theta] > 0). However, the support for the Band-TAR model for Brazil-U.S. wheat and Brazil-Argentine maize is curious in view of the fact that asymmetry was not supported in these two cases. Beyond the tricky question of interpretation of these results, these two cases are interesting in themselves. In the first case, Brazil was a large net importer of wheat throughout, however, Brazil imports the vast majority of these volumes from Argentina rather than from either the United States or countries trading regularly with the United States. In the second case, although Brazil trades maize on the open market, these volumes account for only a very small share of its total consumption of maize. Perhaps then, failure to find asymmetry, and therefore to provide reasonable support for threshold effects, between Brazilian and U.S. wheat prices reflects effective market segregation while failure to find asymmetry in the Brazil-Argentine maize price pair might simply reflect the minor importance played by regional international trade on the domestic maize market, and domestic price determination, within Brazil.

The posterior means of the threshold parameters, [lambda], presented in table 5 ranged from around 0.15 for the wheat data, 0.11 for the soya data and between 0.20 and 0.19 for the maize data. The interpretation of these results is that, for wheat, for example, both positive and negative price variation of up to 15% from the equilibrium price are acceptable to traders. Only if prices diverge more than 15% from equilibrium will arbitrage activity be triggered. In this case, the threshold band has an estimated "width" of 30% (15% above and 15% below equilibrium). In the case of the Brazil-U.S. price pair, this threshold is effectively much smaller from without since [theta] (1 - [theta]) effectively reduces that threshold bandwidth to 22% of the equilibrium price seen from outside the threshold. Arguably though, all of these thresholds are quite wide. However, in some cases the differential between the within-and out-of-threshold adjustment was not very large. In other cases, such as Brazil-U.S. soya, there is a very big difference between within-and out-of-threshold adjustment ([[pi].sub.i,u] = -0.7, [[pi].sub.i,l] = -0.24 for the Brazil price and [[pi].sub.i,u] = 0.19 and [[pi].sub.i,l], = 0.08 for the U.S. price). Moreover, while the nonthreshold (ML) results for soya suggests a rather "sluggish" adjustment, denoted by [[pi].sub.i,u]--[[pi].sub.i,l], in the table (-0.29 for the Brazil price and 0.07 for the U.S. price), the threshold models suggest that the out-of-threshold, [[pi].sub.i,u], adjustment of Brazilian soya prices is very high (around -0.7 and 0.19).

Within-Threshold and Out-of-Threshold Timing

An important final issue pertains to the timing of within- and out-of-threshold episodes. As discussed in the introduction, conformity with transfer cost modified arbitrage conditions can be seen as evidence of market integration, thus periodic violation of these arbitrage conditions is likely to be indicated by the duration of out-of-threshold episodes. If rents to arbitrage, R, defined as [R.sub.AB] = [p.sup.A.sub.t] - [k.sup.AB.sub.t] - [p.sup.B.sub.t] (or [R.sub.BA] = [p.sup.B.sub.t] - [k.sup.BA.sub.t] - [p.sup.A.sub.t]), remain positive for a prolonged period then either trade does not occur or significant temporary barriers to trade exist, which prevent the restoration of the LOP during that period. Examples of temporary barriers might include a divergence of official and real exchange rates, the imposition of emergency phytosanitary controls, regional military conflict and the like.

These within- and out-of-threshold episodes were analyzed using the Bayesian threshold estimates and are summarized in figure 2. The upper two windows concern wheat, the middle two windows concern maize, and the bottom window is with respect to soya. Figure 2 can be interpreted as follows: an upward spike signifies that the Brazilian price lies outside and above its threshold band and a downward spike signifies that the Brazilian price lies outside and below its threshold band. In each case, spikes indicate that, during the relevant period, prices are not in equilibrium and rents to arbitrage, net of transactions costs, are positive.

[FIGURE 2 OMITTED]

If one compares figure 2 with the time plots of exchange rates presented in figure 1, it is difficult to see much in the way of correspondence between periods of rapid change in the behavior of the exchange rates used and any out-of-threshold episodes. Furthermore, there appears to be little in the way of correspondence in the out-of-threshold episodes across all commodities in figure 2. These observations suggest that the out-of-threshold episodes do not appear to be caused by macroeconomic shocks.

As would be expected, there is a reasonable degree of correspondence of threshold episodes within the wheat and maize price pairs, respectively, (e.g., Brazilian-Argentine wheat and Brazilian-U.S. wheat but more markedly between Brazilian-Argentine maize and Brazilian-U.S. maize). Most of the price pairs alternate between being inside and outside of their thresholds in a transient pattern throughout the sample period. The exception here is the case of the Brazilian-U.S. wheat price pair, which lies outside and below its threshold for much of the period up to about 1997, then predominantly lies outside and above its threshold after this date. It is possible that, in this particular case, there has been a structural break within these series around 1997 that may be responsible for these results. If indeed there is a structural break here, then this might help to explain the model results for this price pair that failed to distinguish between the within- and out-of-threshold response even though both the estimated threshold parameters, [lambda] and [theta] supported the Band-TAR model. One possible explanation for the poor result for the Brazilian-U.S. wheat price pair may be the implementation of the Mercosur free trade area agreement. It may be that this agreement introduced an effective barrier to imports of wheat from the United States to the South American member states. We note also that Brazil was a net exporter of both maize and soya beans during the mid- to late 1990s and, as such, the price pairs for these crops, with respect to the U.S. prices, may have been immune from the potentially distorting effect of the Mercosur agreement.

Summary and Conclusions

This article reports estimated threshold error correction models for pairs of wheat, maize, and soya prices for Brazil, Argentina and the United States. It introduced a generalization on existing threshold models in which the prices could be attracted to either the edge of the threshold interval or to some point within this interval, therefore, encompassing both the Eq-TAR and Band-TAR models introduced by Balke and Fomby (1998). A Bayesian approach to the estimation of these models provided a tractable means of circumventing problems associated with jagged likelihood functions and parameter identification problems inherent in these models.

The evidence for the existence of threshold effects in the transmission of commodity prices was mixed, with evidence for thresholds found in three out of the five cases considered here. Wheat and soya price pairs appeared to have smaller thresholds than the maize price pairs. Both the standard and threshold error correction results suggested that causality flowed from Argentine and U.S. prices toward Brazilian prices. This was evident in the smaller error correction adjustment parameters and by the significance with which noncausality could be rejected (for all commodities and countries). In essence, this suggests that the long-run forces driving these prices were coming from the U.S. and Argentine markets, rather than from Brazil. Consequently, to the extent that differential within- and out-of-threshold adjustment occurs, this largely relates to the responsiveness of the Brazilian prices, rather than to how U.S. or Argentine prices adjust.