Trading Activity and Volatility

In the same spirit as Roth et al. (2003), who found a positive relation between volatility and open interest for both hedgers and speculators, Equations 4.24 and 4.25 are

190 regressed to see whether there is such a relation between volatility and net positions137. Volatility is proxied as standard deviation in Equation 4.24, and variance in Equation 4.25.

t

NP

σ

ε

t

NP

σ

ε

Findings upon regressing Equation 4.24 show that volatility, when measured as standard deviation, has a mixed effect on the trading activity of hedgers and speculators, where trading activity is measured as current net positions. Recursive estimates of volatility had a significant negative effect on net positions of hedgers for soybean oil, British pounds, copper, live hogs and Swiss francs; and only a significant positive effect for Japanese yen. On the other hand, recursive estimates of volatility had a significant positive effect on net positions of speculators for copper, Japanese yen, wheat138 (Chicago); and a significant negative effect for corn, soybean, cocoa139, cotton, pork bellies, sugar, S&P500, and Treasury bonds. The boundaries within which recursive estimates of volatility lie are much broader in currency markets like Japanese yen, British pounds and Swiss francs; and financial markets like S&P500. This is consistent with Christie-David and Chaudhry (1999) who reported that more liquid financial instruments show longer volatility persistence following macroeconomic announcements.

Recursive estimates of variance from Equation 4.25 have a significant and negative effect on hedgers’ trading activity in soybean oil, cotton, wheat (Chicago, Minnesota), soybeans, silver and S&P500; and a significant positive effect in crude oil, heating oil, Swiss francs and platinum at 10% significance level. Recursive estimates of volatility have a significant positive effect on net positions of speculators for Japanese

137 _{Due to the high correlation between net positions and volatility, there were too many breaks occurring}

within the model, making it impractical to test for any significant structural break.

138 _{Stationary after first level differencing.}

191 yen140, coffee and cotton; and a significant negative effect in crude oil, wheat (Minnesota), pork bellies, and sugar. The boundaries between which the coefficient estimates of variance lie are smaller than those of standard deviation. This can be attributed to variance which is expected to be the squared of standard deviation. The significant negative relationship observed between standard deviation and net positions, and between variance and net positions, is consistent with Peck (1981), Bessembiner and Seguin (1992), but inconsistent with Roth et al. (2003) who found a positive relation between net position and open interest for both hedgers and speculators. However, it is also important to note that Roth et al. (2003) pointed out that the positive significance of open interest and volatility is highly sensitive to the volatility measure used, particularly for hedgers’ trading activity.

Conclusion

The trading determinant model suggests hedgers are positive feedback traders in most markets. Speculators’ behaviour is inconclusive due to higher trading frequency level suspected. Hedgers also have superior market timing abilities than speculators, on a monthly basis. Hedging pressure effects are mostly insignificant, suggesting no transfer of risk from hedgers to speculators. The negative market timing of hedgers in heating oil and Japanese yen, and positive feedback trading behaviour, suggest large hedgers destabilized futures prices in these markets, suggesting a need to look again at CFTC’s stringent position limits imposed on speculators. Overall, information variables are insignificant in determining monthly trading decisions.

The decomposed mean equation in the performance section, with a higher number of expected net positions, suggests hedgers are more prone in setting an expected net positions at the start of the month in determining actual returns rather than readjusting their net positions althroughout the rest of the month. The lower number of significant

192 expected net position variables in determining volatility further added support that hedgers are better informed, where their trading at the start of the month has less effect on the volatility than compared with speculators. Higher expected volatility of hedgers in crude oil and heating oil added support to their destabilizing features. The GARCH model suggested the importance of both lagged volatility and news of volatility from previous month in determining actual volatility. Both players’ volatility had a tendency to decay over time in response to shocks, supporting the informative traits of these large players. The PARCH model, by capturing more significant negative impact of variables, is a better model than the GARCH model for both hedgers and speculators. Expected idiosyncratic volatility and unexpected volatility had a mixed effect on returns, supporting the poor measure of idiosyncratic volatility as a measure of risk. Both variance- and standard deviation- based models, under normal and t distribution, reported the returns tend to be leptokurtic. The (PARCH, t) model represented actual returns less accurately, while the (GARCH, normal) and (PARCH, normal) models ranked first in explaining hedgers’ and speculators’ actual returns. The (PARCH, normal) model ranked first in forecasting one-month return. Idiosyncratic volatility poorly forecasted volatility in forecasting one-month returns.

The trading determinant model, mean equation model, and risk/return relationship model were all stable over the 10 years. Major economic events had little or no significant effect on hedgers’ and speculators’ trading decisions, and risk attitude. Lastly, the trading activity and volatility relationship model showed stronger effect of volatility on speculators’ trading activity, particularly in financial and currency markets. All models tend to capture more structural breaks, where risk was proxied as standard deviation, due to the higher sensitiveness of standard deviation to futures prices than variance.

193

CHAPTER 5