RESEARCH OUTPUT AND FINDINGS
4.6 Stability and Events Analysis
4.6.1 Behaviour Trading Determinant Model
Due to some events (like events one & two) having some small sample sizes, and in order to keep consistency in the models, only variables deemed important are regressed
127 See Equation 4.1
128 See Equations 4.10.1, 4.13, 4.14.2
Table 4.19
183 in the models. This means the removal of some unimportant variables like the three information variables and sentiment data129. The behaviour model of Equation 4.1 is thus changed to:
Δ NP = t+1 ϕ0 + ϕ 1 R +t ξt (4.20)
Only significant recursive coefficients for the futures returns (with significant t ratios130 after adjusting for structural breaks) in Equation 4.20 are graphically displayed in Appendix 6.16, together with two standard error bands around the estimated coefficients. The highest coefficient estimates of R can be found in Canadian dollars, Eurodollars, t
British pounds, Treasury bonds, Japanese yen and gold131. The occurrence of relatively higher coefficient estimates suggests that large players tend to rely more on actual returns
t
R to change their net positions next month than large players in agricultural futures
markets. Moreover, the coefficient estimates of R between hedgers and speculators tend t
to bear a negative relationship132. This is supportive that the futures market is a zero-sum game, and that for every long position there should a short position to net it off. The S&P500 R coefficient for hedgers appears to be negative on average. The fact that t
hedgers were net short during the 2000 burst compared to large speculators who were net long, suggests that following hedgers during that period would have led to less losses and possibly profits than trend-chasing with speculators.
In checking the stability of the behaviour model, most of the markets appear to be stable with rare occasions of structural breaks. It is important to neglect the instability of the coefficient estimates in early stages of the graph, since Δ NP would be highly t+1
129 Information variables are removed due to their insignificance as shown in earlier parts of the study.
Sentiment data is removed since they were exhibiting a bullish behaviour, which was quite predictable.
130 Significant t ratios are those independent variables being siginificant at 10% significance level. 131 Markets like British pounds, Canadian dollars, Eurodollars, Swiss francs, Treasury bonds, S&P500,
copper, live cattle and live hogs all had negative coefficients for hedgers’ return coefficient estimates.
132 Except for crude oil, Japanese yen and heating oil where both hedgers and speculators tend to add to
184 sensitive toRt
133. Those markets with significant breaks in their returns coefficient
estimates are crude oil, cotton, Eurodollars, soybean, wheat (Chicago) and cocoa (for speculators); and corn, Japanese yen, soybean and cocoa (for hedgers)134. This is consistent with Cheung and Wong (2001) that macroeconomic announcements have a smaller impact on the gold market than on the Eurodollars and Japanese yen. As further asserted by Fung and Patterson (2001), the Eurodollar, although influenced substantially by domestic US news, is an international asset that is traded globally and thus more readily reflects changes in risk premiums among different Eurocurrency rates in the international financial market. Table 4.21 in Appendix 6.16 shows that while all breaks for hedgers’ returns coefficients estimates are trending upwards, speculators’ returns breaks are heading in both directions. Results show that hedgers’ actual returns coefficient estimates go up for corn, Japanese yen, soybean and cocoa after the major economic event brought more stability to previous economic conditions. For instance, soybeans and corn returns have more effect on net positions of hedgers after the end of the US long period of tightening interest rates. The same analogy can be concluded with cocoa and Japanese yen returns bearing more effect on net positions of hedgers at the start of the temporary revival from Japanese recession.
On the other hand, the effect of speculators’ returns on next-month net positions, as expected, is backed by positive feedback behaviour, where speculators take more long positions when major economic events are an indication of easing economic conditions, and take more short positions where events tend to show tightening economic conditions. For instance, speculators took less long positions in Eurodollars after the LCTM near financial collapse, but then took more long positions after the buyouts occurred to save LTCM from affecting financial markets. The same analogy can be applied to the upwards jumps in returns coefficients occurring due to more favourable economic conditions like the end of US tightening interest rates, introduction of the Euro currency,
133 To ensure consistency throughout this event analysis, any instability before sample 49 is rejected for
analysis. This allows us to analyse any structural break starting with US Fed tightening of interest rates, which occurred in sample 50.
134 While there were more structural breaks in the 29 markets, only those structural breaks that match any
185 and downward breaks due to less favourable economic conditions like LTCM near collapse, Russian crisis, EM slump, US tightening interest rates, and Japanese recession. The only exceptions would be Eurodollars returns coefficient estimates which jumped at the start of US tightening of interest rates. This can be explained by speculators going more net long in Eurodollars, as an alternative to less attractive US dollars. More importantly, t statistics show that only soybeans, cotton, wheat (Chicago) and cocoa have significant return coefficient estimates (all from speculators). This supports BIS (1995– 2001) reports on these major economic events that the eight named events do not affect significantly futures markets in the US, except in four markets above at a specific point in time.