Essays on the Efficiency of Non-Genetically Modified ( Non-GM ) and Conventional Soybean Features Markets

This dissertation explores issues related to efficiency, how efficiently markets transmit information, of non-genetically modified (GM) soybean and conventional soybean futures markets at the Tokyo Grain Exchange (TGE). The first manuscript examines how efficiently non-GM and conventional soybean futures markets react to an announcement to change the contract unit, suppliers, and expiration date on the conventional soybean contract. Box and Tiao's intervention analysis is used for this purpose. The result reveals that the price premium for non-GM soybeans (the price difference between the two soybean contracts) and the volumes of non-GM soybeans increase after the announcement and this effect remained after the announcement. Hence the two soybean futures markets did not respond quickly to the announcement and there was an informational inefficiency after the change occurred. The second manuscript explores the market linkages between the non-GM and conventional soybean, and the com futures markets at the TGE in the presence of unknown breaks. Bai-Perron multiple structural change test and Johansen cointegration tests are used for this purpose. The results reveal that cointegration relationships exist between the two soybean futures prices and between the non-GM soybean and corn futures prices. Yet the breaks found in the soybean futures markets affected these price linkages, and there were periods where the two soybean and corn futures markets were not efficient. The third manuscript tests if the two soybean futures markets fully reflect available information by testing the market efficiency of the two soybean futures markets. This manuscript also investigates the causality of this long-run relationship to find out if it is the spot price or the futures price that first incorporates new information into the market. Johansen cointegration tests are used for these purposes. The results suggest that both non-GM and conventional soybean futures markets are efficient but the non-GM soybean market is inefficient compared to the conventional soybean market. The test on the causality of the long-run relationship showed that both of the soybean futures markets are led by the spot price for the spot and futures prices to move together in the long-run. ACKNOWLEDGEMENTS I would like to thank my advisor James L. Anderson and the faculty at the Department of Environmental and Natural Resource Economics at the University of Rhode Island for supporting me in writing this dissertation. I would also like to thank Christian Vye of the university's Information Technology Services for giving me the opportunity to work as his graduate assistant and providing financial assistance. I am also grateful to the Tokyo Grain Exchange (TGE) for the price data and valuable information about the TGE soybean futures markets. Especially, I appreciate Takahiro Ueyanagi of the TGE for providing the data and explaining the TGE soybean futures markets. I would also like to give special thanks to Frank Asche, and Nobumori Yagi for providing the opportunity to comment on the dissertation. Finally, I would like to express my gratitude to my family (my parents, Natsuki and Tsutomu Aruga, my grandmother, Toyo Hirasawa, and sister, Mika Toyomura) and fellow graduate students at my department for supporting me throughout the PhD program. I also thank Katy Meigs, a family friend, for editorial assistance.

thank Christian Vye of the university's Information Technology Services for giving me the opportunity to work as his graduate assistant and providing financial assistance. I am also grateful to the Tokyo Grain Exchange (TGE) for the price data and valuable information about the TGE soybean futures markets. Especially, I appreciate Takahiro Ueyanagi of the TGE for providing the data and explaining the TGE soybean futures markets. I would also like to give special thanks to Frank Asche, and Nobumori Yagi for providing the opportunity to comment on the dissertation.
Finally, I would like to express my gratitude to my family (my parents, Natsuki and Tsutomu Aruga, my grandmother, Toyo Hirasawa, and sister, Mika Toyomura) and fellow graduate students at my department for supporting me throughout the PhD program. I also thank Katy Meigs, a family friend, for editorial assistance.

PREFACE
This dissertation is composed of three manuscripts and a set of supporting appendices. The objective is to address issues related to the efficiency of the non-genetically modified (GM) and conventional soybean futures markets at the Tokyo Grain Exchange (TGE). Efficiency here means information efficiency such that the prices always fully reflect available information. More and more food products are using genetically modified organisms (GMOs) throughout the world, and concern about such products is spreading. However, not much is known about how a segregated market for non-GM food functions as a source of providing effective information to the market participants. This dissertation examines how such a market for non-GM food transmits price information efficiently through the case of the TGE non-GM soybean futures market, the world's first individual futures market for a non-GM commodity.
The first manuscript examines how efficiently the non-GM and conventional soybean futures markets react to new information by testing the effect of an announcement to change the contract unit, suppliers, and expiration date on the conventional soybean futures contract. The result reveals that the price premium for non-GM soybeans (the price difference between the non-GM and conventional v soybean futures prices) and the volumes of non-GM soybeans increase after the announcement and this effect remained for at least a month. Hence it is concluded that the two soybean futures markets did not respond quickly to the announcement and there was an informational inefficiency after the change occurred.
The second manuscript focuses on the linkage between the non-GM and conventional soybean futures markets to find out if these markets are co integrated so that they provide valuable information to each other. The linkages between these two soybean futures markets and the com futures market are also investigated and effects of unknown breaks on the co integration, if any, are tested as well. The Johansen cointegration test suggests that a market linkage exists between the non-GM and conventional soybean futures markets and between the non-GM soybean and com futures markets but that they were not cointegrated during periods with breaks.
Hence these markets are efficient when the effect from the breaks is not apparent but they become inefficient when the breaks are affecting the three markets.
The third manuscript tests for market efficiency of the non-GM and conventional soybean futures markets at the TGE to see if the two soybean futures markets fully reflect available information. Both soybean futures markets turned out to be efficient (do provide efficient information) but the non-GM soybean futures VI market was inefficient compared to the conventional soybean futures market. In this manuscript the causality of the long-run relationships between the spot and futures prices of non-GM and conventional soybeans were also investigated in order to find out whether it is the spot price or the futures price that first incorporates new information to the market. In both soybean futures markets it was the spot price that led the spot and the futures prices to move together in the long-run.
Through these manuscripts the dissertation finds out that the non-GM soybean and conventional soybean futures markets do s~tisfy the market efficiency condition. However, there were some periods where the prices of the two markets did not respond quickly to known and unknown breaks, and hence, these markets are not perfectly efficient.

t.1 Introduction
Many regions and countries, including the European Union, Australia, New Zealand, and Brazil, now require labeling for genetically modified (GM) food products (Huffman 2003). Japan has followed this trend.  revealed that Japanese consumers have a higher preference for non-GM food over GM food. As more consumers became concerned about GM food products in Japan and demanded regulation, the Japanese government issued a law to require labeling for GM food products as of April 2001(TGE 2003. This law imposed mandatory labeling for most of the GM food products (TGE 2003) so that consumers can identify products containing genetically modified organisms (GMOs). 1 For example, one of the world's largest soy sauce companies, Kikkoman, decided to use only non-GM soybeans for its product (Kikkoman 2006).
To meet the increasing demand on non-GM food products, on May 18, for non-genetically modified (GM) soybeans. Since the opening of the non-GM soybean futures market, it has been known that the price of non-GM soybeans is higher than the price of "conventional soybeans,'' which contain both non-GM and GM soybeans .  defines the price difference between the prices of non-GM and conventional soybean futures contracts as the price premium for non-GM soybeans. He argues that this premium should represent the marketing and production costs of segregating non-GM soybeans.2 It is also known that the price premium for non-GM products exists in the demand side as well. Wachenheim and Wechel (2004) find that consumers are willing to pay a premium for non-GM products using experimental auction and it is arguable that the price premium for the non-GM soybeans at the TGE is also driven from the demand side.
However, in July and August 2002, there were trading days when the conventional soybean price reached a higher price than the non-GM soybeans 2 The segregation costs include various costs of preserving the identity of the non-GM soybeans from the seed level to the distribution level (Bullock, and Dequilbet 2002). 4 on the last day of trading. The TGE suggests that there was market inefficiency involved in the soybean futures markets during these periods and that this may have driven the price premium to become negative. 3 To cope with the problem of the price premium becoming negative, which was beyond market expectations, the TGE made a major change in the specification for conventional soybeans on October 29, 2002(TGE 2002

5
• Change in the last day of trading for conventional soybeans. Before this change, the last day of trading for all conventional and non-GM soybean contracts was two business days before the end of the month. After the change, the last day of trading for conventional soybeans was changed to fifteen business days before the end of the month .
What can be expected from the first change is that the volume of trading for non-GM soybeans would rise. After the change in the contract unit for the conventional soybean futures contract, traders have to trade 50 mt of soybeans to obtain conventional soybeans, so small traders who were trading less than 50 mt of conventional soybeans would have to shift their trade to non-GM soybeans if they wanted to continue their trading at their previous volume. 7 Thus the change may attract traders who want to trade soybeans in smaller amounts to the non-GM soybean futures market. This shift of traders from the conventional soybean futures market to the non-GM soybean futures market may drive the price of non-GM soybean futures contracts to rise after 7 The contract unit for non-GM soybeans stayed the same (10 mt).
6 the change.
The second change, the one that widens the suppliers for the conventional soybeans may also increase the comparative price of non-GM soybeans, since the market participants may expect the total amount of available conventional soybeans at the TGE to become larger than the non-GM soybeans after the change is conducted. The suppliers for the non-GM soybeans remain only from the United States while the conventional soybeans will be supplied from countries in the Southern Hemisphere in addition to the United States after the change. Thus the difference in the stock availability between conventional and non-GM soybeans may become more apparent to market participants, and this may affect the soybean prices.
Finally setting the last day of trading for conventional and non-GM soybeans on different dates will help segregate the two soybean futures markets and make it easier for investors to distinguish their portfolios for the two types of soybeans. The change may separate the market participants so that they trade the two soybeans on different days, and this may strengthen the distinction between the two soybean futures markets: one market for soybeans that require 7 labeling under the JAS law and the other for soybean products such as soybean oil and soy sauce that do not require labeling. 8 The objective of this paper is to examine how efficiently the TGE non-GM and conventional soybean futures markets react to an announcement by testing the influence of the above mentioned specification change on the price premium for non-GM soybeans, and on the trading volumes of non-GM and conventional soybeans. There are still few studies using the TGE non-GM soybean futures price, and there are not any when it comes to how an announcement from the TGE, such as this specification change might affect the market prices.  explains about this new market for non-GM soybean futures at the TGE and computes the price premium for non-GM soybean contracts. Bullock and Desquilbet (2002) shows the price premium of Their study showed that the specification change of the S & P 500 futures contracts did not change the contract volumes. These previous studies on the effects of announcements on futures markets use the Box and Tiao's (1975) intervention analysis, but these studies are focused on financial futures products.
The reaction to the announcement may be different in the commodity futures market. Previous studies using the intervention analysis only tests the reaction for the period before and after the event but this study uses this method to also find out how long the effect from the announcement lasted after the event. This will be done by creating individual dummy variables for each specific period where the impact may have lasted.
It is important to find out how the TGE soybean futures market reacts to an announcement such as this specification change. If the market did not respond quickly to the specification change and the effect of the change remains for a certain period, it would suggest that it took some time for market prices to reflect the new information. If the market is fully efficient, all available information, including public information should immediately be reflected in the price (Fama 1991). Thus ifthe effect from the announcement stays in the market it means that there is an informational inefficiency in the market. 9 Although the specification change may increase the price of non-GM soybean futures contracts as explained above, this increase should occur only for a short period of time if the market is fully efficient. If the market is efficient the price should adjust quickly to the level before the announcement due to the buying and selling activities of the arbitrageurs.
In the following section I will describe the data used in the study and According to Fama (1991) typical results in event studies using daily data suggest that if the market is efficient prices often adjust within a day after an announcement occurs.
provide more explanation on the changes that was conducted for the conventional soybean futures contracts. In the third section the details of the method used for this research will be explained. The fourth section will show the results of the investigation. In the last section, I will present the conclusions of the study.

Data
The data used for the analysis are obtained from the TGE via online and personal negotiations with the TGE (TGE 2008 be used for the analysis. The price unit is provided in yen per mt.   second-nearest contracts are either two-month-ahead or three-month-ahead futures contracts, the third-nearest contracts are four-month-ahead or five-month-ahead futures contracts, and so on. The difference between the daily prices of conventional and non-GM soybeans for the second-nearest futures will be the second-nearest price premium, that for the third-, fourth-, fifth-, and sixth-will be the third-, fourth-, fifth-, and sixth-nearest price premiums respectively.   "" "" "" "" "" "" "" "" "" 0 0 0 0 0 9 9 0 0 0 0 0 9 0 0 0 0 0 9 0 0 C: _a .!. <5. >. c: ::; bo 6. ti > 6 c: _a .!. .!. >. c ::; bo 6. The daily data on the volumes of conventional and non-GM soybean futures contracts are converted to actual volumes traded in mt. The volume data provided by the TGE are the total number of contracts. To get the actual volume of soybeans traded on a certain day, this volume data is multiplied by the contract unit. Since the contract unit for conventional soybeans increased from 30 mt to 50 mt after October 29, 2002, the volume data before the change are multiplied by 30 whereas the data after that date are multiplied by 50. The volume data on non-GM soybeans are multiplied by 10 through the study period. As seen in figure 1.2, it seems that the volume on the non-GM soybean contract increased more than that of the conventional soybean contract after the specification change took place in November 2002.

Methodology
The Box and Tiao's (1975) intervention analysis is used to test the effects of the specification change on the price premium and the volume traded for the non-GM and the conventional soybean futures contracts. This analysis takes into account of the effect of an announcement on a given response variable using the autoregressive moving average model (Doukas and Rhaman 19 86). It also allows the observed autocorrelation in the model residuals to be removed, which improves the statistical testing (Guzhva 2008;Larker, Gorden, and Pinches 1980). As suggested by Larker, Gorden, and Pinches ( 1980), this method is a more appropriate method for testing effects on financial markets from an announcement compared to the cumulative abnormal returns (CAR) measure, which is often used in event studies when the exact date of the event is unknown (Tsay, Alt, and Gordon 1993).
When using an intervention analysis the impact to be tested must be an event in the strict sense and the time when that event occurred has to be specified a priori  in which CD is the impact of the interruption on the series. 12 The impact is analyzed using the step function: where S is the step input, and t 0 is the time period during which the intervention occurs.
(3) The details are explained in Appendix A.

12
w can be also interpreted as the coefficient of I in equation (3) To avoid biased estimates of autocorrelation functions (ACFs) and partial autocorrelation functions (PACFs), only observations before the intervention is used to estimate the ARIMA model. In Box and Tiao's intervention analysis, it is assumed that the same model identified for the pre-intervention series applies to the post-intervention autocorrelation behavior (Tsay and Hung 1994). Assuming there was no intervention effect before If the pattern of ACFs shows that the response series are nonstationary, the series will be differenced to remove its trend and make the series stationary. An augmented Dickey-Fuller (ADF) test is conducted to test this . Then the estimated ACFs and PACFs are compared with various theoretical ACFs and PACFs and the final order of the autoregressive and the moving average elements are determined by the extended sample autocorrelation function (ESACF) (Tsay and Tiao 1984), and the minimum information criteria (MINIC) (Hannan and Rissanen 1982). 13 At the estimation stage the coefficients of the parameters of the model are estimated. The coefficients are estimated using the maximum likelihood estimation. The log-likelihood function uses the covariance matrix of the vector calculated from equation (1). 14 The stationarity and the significance of the model are tested as well.
At the diagnostic stage the residuals of the model are tested as to whether or not they are white noise. The statistic used for this test is the 13 These are done by using SAS software (SAS 2008).
14 The details of the process and the functions can be seen in Box and Tiao (1975) 20 Box-Pierce Q statistic: Q == T L~=l r~ where T is the number of observation and rk is the autocorrelation of the kth variable where t 0 is November 1, the day when the event occurred, and toec_F and

t.4 Results
The results of the ADF test conducted on the data before the specification change for the conventional soybean futures contract (from ' SB and non-GM represent the conventional and non-GM soy beans. The parenthisis is the order of the aut oregress ive, int egrated, and moving average component s of the ARlMA model. 11°he 2nd through 6th represent the second-nearest to sixth-nearest futures contracts.
The orders of the ARIMA model used for the analysis are given in 'T he esiimat es are th e coefficients of the input variables and the values in parentheses are the I-values. 'P rem ium 2 through 6 are the price premiums of second-nearest to sixth-nearest futures contracts. 'The coefficient on the mov in g average was not significant at the 5% level.
•Statistically significanl at the 5% level Table 1.5 shows the estimated coefficients for the input variables (Nov. -Mar.) of different contract months, which represent the effect of the event.
For example, the model of the price premium for the second-nearest futures contract with an input variable Nov is where vrre is the price premium at time t, and Nov is the input variable  . guable that this increase in the volumes traded for non-GM soybeans raised 1 s ar the price of non-GM soybean futures contracts and that this contributed to the increase in the price premium for non-GM soybean futures contracts.

t.5 Conclusions
The change in the contract specification for conventional soybeans that

29
The results from the length of the impact on the price premium for non-GM soybeans suggest that the effect on soybean futures prices from the event lasted for three to four months. However, the event only had an impact for a month on the volume of non-GM soybeans. These results revealed that the specification change remained in the market after the announcement. This implies that there was an informational inefficiency in the market since it took at least a month for the price and the volume to adjust to the levels before the change occurred.
In conclusion the announcement from the TGE on the specification change for the conventional soybean futures contract did affect the price premium between the conventional and non-GM soybean futures contracts. It is also found from the study that this effect remained three to four months in the price premium and for a month in the volume of non-GM soybeans. Hence the two soybean futures markets did not respond quickly to the announcement and there was an informational inefficiency after the change occurred.

Abstract
The market linkages among the non-genetically modified ( Pa nies obtaining soybeans for these products can use the conventional com soybeans. Thus from the demand side perspective, these different soybeans may belong to different markets and may not be related to each other. However, some traders may be purchasing non-GM soybeans for the same purpose as conventional soybeans since there are no legal barriers on using non-GM soybeans for oil or processing. If many traders were substituting non-GM soybeans for conventional soybeans, the non-GM soybean price would show a substitutive movement with the conventional soybean price, and the two price series would have a cointegration relationship, that is the prices move together and do not take apart within the series tested.
The objective of this paper is to determine whether or not these two soybean futures markets are cointegrated so that they share valuable price information in the presence of breaks in the markets. This will be investigated by testing the cointegration between the non-GM and conventional soybean futures prices. Studying this price linkage is important since markets that are not cointegrated often convey useless price information and can distort the decisions of market participants . If a cointegration does exist between the two soybean futures markets it would imply that the price discovery process of either one of the soybean futures markets provides valuable information for the other . It would mean that the non-GM and conventional soybean futures markets are economically linked and price information of these markets could be used for cross-hedging, which would justify the introduction of this new non-GM soybean futures contract.
There are various studies analyzing the price relations of commodity futures markets, but most of these studies focus on testing for market efficiency  or finding spatial linkages of futures markets of different regions and locations (Xu and Fung 2005). However, some studies investigate the price linkages among different commodity futures contracts to find out whether the commodity futures institution is transmitting information efficiently among different contracts. This study also examines the price linkages of different futures contracts within the TGE to pursue this objective.  analyze the cointegration among the prices of com, azuki beans, soybeans, and sugar futures traded at the TGE to find out whether these commodity futures are linked because of common economic fundamentals or because of herd behavior 40 by market participants. They used the cointegration method and found that these four commodity futures that are traded at the TGE are interdependent and that this interdependency is due to common economic fundamentals.  examined price discovery on the Chicago Board of Trade (CBOT) for the U.S. grown com, wheat, oats, soybean, soybean meal, and soybean oil futures prices by using pair-wise cointegration tests and found out that long-run linkages exist among these markets.
Besides the price linkage between the two soybean futures markets, this paper will also test for the linkage between the two soybean and com futures prices traded at the TGE. Testing these market linkages is meaningful since the two soybeans and corn are mostly imported from the United States so that these commodities can be affected from the U.S . farm policy. It is also important to study these linkages since they can be substitutes. A previous study on testing linkages between the TGE soybean and com futures markets found that they are cointegrated  but this study was conducted before the TGE soybean futures market was split into the non-GM and conventional soybean futures markets. It could be that the cointegration result between the 41 soybean and corn futures prices will be different after the non-GM soybean futures market opened at the TGE.
Most of the previous studies on price linkages between certain commodity futures markets do not consider the effects of unknown breaks on the price linkages but this study will consider this and test how such breaks will affect them.

Cointegration Test
The Johansen cointegration test  is used for testing the price linkages of non-GM soybean, conventional, and com futures prices at the TGE. Some studies have used the  test for examining the price linkages (Goodwin and Schroeder 1991) but Johansen method is more efficient since it can analyze the variables of the interests as endogenous in the model and is more useful in a multivariate framework. Enders 46 005) suggests that the Engle and Granger procedure can give different test results based on which variable will be taken as the dependent variable. Johansen method has been used for examining linkages among different markets (Asche, Bremnes, and Wessells 1991; but there are few studies applying this method on the TGE soybean and corn futures markets.  is one of those few using this method to test for the price relations between the TGE soybean and corn futures markets. The time series data of the non-GM soybean, conventional soybean, and com prices have to be integrated at the same order for the series to be cointegrated. So before performing the cointegration tests, the three price series are tested for their stationarity by the augmented Dickey-Fuller (ADF) test . Then bivariate Johansen cointegration tests  are used for testing the linkages between the prices of non-GM soybean, conventional soybean, and corn futures contracts.
Let Yt be the n x 1 vector of the non-stationary variables, and k be the order of the vector autoregressive process. Then the vector autoregressive model used for the Johansen cointegration test is denoted as the following:

47
(1) where Yt is the endogenous variables of interest (prices of soybeans and corn), ni is a n x n matrix of parameters, and Ut denotes a normally distributed n-dimensional white noise process.2 Converting this model into the error correction model leads to ~ Yt = rrvt-1 + If:l rjl'.l Yt-i + ut (2) variables is integrated of the same order by assumption, whether the variables of interest become co integrated depends on the rank of the IT matrix. The rank of a matrix is equal to the number of its significantly positive characteristic roots, which is called the eigenvalue.
Using this eigenvalue, the trace and maximum eigenvalue tests are performed to determine the number of cointegrating vectors (Asche, Bremnes, and Wessells 1991 ). The trace test tests the null hypothesis of at most r positive eigenvalues exist in the IT matrix against the alternative hypothesis that there are 2 The model assumes that it does not contain deterministic terms. 48 more than r positive eigenvalues, where r is the rank of the I1 matrix. The test statistic for this test is (3) where Tis the number of observations, and Xi is the estimated i th eigenvalue from the TI matrix. The maximum eigenvalue test determines whether there are r or r + 1 co integrated vectors in the I1 matrix. The null hypothesis of having exactly r positive eigenvalues is tested against the alternative hypothesis of having exactly r + 1 positive eigenvalues. The test statistic for the maximum eigenvalue test is (4)

Bai-Perron Multiple Structural Change Test
The Bai-Perron (1998) method is used for determining whether the price series contain unknown breaks. For a long time  test has been the major method for determining structural change in a time series data but this test is not adequate when the breakdate is unknown . Quand (1960), Andrews (1993), and Andrews and Ploberger (1994) develop a method based on the Chow test for testing structural breaks when the break is unknown 49 but these methods were limited to testing for only one structural break.
Furthermore these methods had deficiency in identifying the breakpoints when the series were nonstationary . Bai-Perron test overcomes these problems and is very useful for finding breaks when the potential break date is unknown and the series tend to have more than one break ).
The first stage of Bai-Perron test considers ifthe price series contain unknown breaks using the "double maximum test." This test uses the maximum f-statistic that is calculated from the global minimum of the sum of squared residuals of them-partitioned multiple regression models: Yt = z~8i + ut where j = 1, · · ·, m + 1 (5) where Yt is the dependent variable at time t, Zt is a vector of covariates, 8i is the corresponding vector of coefficients, m is the number of breaks, and Ut 1s the disturbance at time t (Bai, and Perron 2006). The unweighted double maximum (UDmax) test statistic is obtained by calculating various F-statistic when the series are divided into one through m breaks. This statistic is compared to the critical values provided by Bai and Perron (2003b ). The f-statistic can decrease as m increases, and if this is the case, the marginal p-values will decrease as m increases. Hence Bai-Perron provides the weighted double maximum (WDmax) test to take in account of this change in the F va. lue as the size of m increases by multiplying a weight component to the UDmax test statistic (Bai, and Perron 1998). When these tests do not reject the null hypothesis of having no structural breaks in the series, there will be no significant evidence of a break in the series.
In the second stage, if there happens to be an unknown break in the first stage, the number of appropriate potential breaks is identified by testing the null of l breaks versus the alternative of l + 1 breaks. The null hypothesis of l breaks is rejected in favor of the l + 1 breaks if the overall minimal value of the sum of squared residuals of a model with . l + 1 breaks is sufficiently smaller than that of the l breaks model (Bai and Perron 2003a). Since minimizing the sum of squared residuals is equivalent to maximizing the F-statistic of the model, the test statistic used for this test is called the supF(l + lll) test statistic and the critical values are provided by Bai and Perron (1998).

z.4 Results
The results from the ADF unit root tests indicate that in every contract month, conventional and non-GM soybean, and corn futures prices all had a unit root. However all series became stationary after taking the first differences (table   2.3). Thus the three price series are all integrated of order one, 1(1 ).    Food Others reason for this may be because corn and soybeans are used for different purpose in Japan. As seen in table 2.5, of the total demand for corn and soybeans in Japan, com is used for livestock meal and processing but soybeans are mostly used for processing and food. 6 The other possible reason is that more participants of the com market at the TGE may have been arbitraging between the non-GM soybean contracts rather than between the conventional soybean contracts since between 2003 and 2007, the annual average of the trading volumes for non-GM soybeans were larger than the conventional soybeans, which implies that the non-GM soybean futures market was more active than the conventional soybean futures market during these periods. There is a whole separate market for soybean meal in Japan but soybean meal futures contracts no longer exist at the TGE (TGE 2008a). For the price premium of the second-nearest futures contract, the UDmax and WDmax tests do not reject the null hypothesis of having no breaks in the series, which imply that there are no breaks in this series. On the other hand, the double maximum tests for the price premiums of the third-through sixth-nearest futures contract rejected the null hypothesis and suggested that the series do contain unknown breaks. Since the result of double maximum tests identified the existence of the breaks in the price series of third-through sixth-nearest futures contract we need to look into the results of the supF(l + 11 l) test statistic to identify the optimal number of breaks for these series.
The supF(l + lll) test for the price premiums of third-and fourth-nearest futures contracts show that three breaks is the optimal number of breaks for these series. The null hypothesis of having two breaks is rejected in 58 favor of three breaks for these series. On the other hand the null hypothesis is not rejected for premiums 5 and 6, which suggests two breaks is appropriate for the fifth-and sixth-nearest futures contracts. From the results of these tests, the optimal number of breaks for each contract months is determined and each of them is split into periods identified by the breaks, which is shown in table 2.7. The contract unit was changed from 30 metric tons (mt) to 50 mt, suppliers were changed from six U.S . states to all U.S. states and Brazil, and the last day of trading changed from two business days to fifteen business days before the end of month for the conventional soybeans (TGE 2002) 59 seen in figure 2.2, the change in the price premium is small compared to the changes in 2007 and 2008. 8  As shown in manuscript one, the impact from the 2002 specification change only lasted for three to four months at most and did not change the price premium permanently.

9
There was also a shift from soybean acreage to corn acreage in 2007 and this may also affected the soybean stock to decrease for this year (OMNICO Corp. 2007). non-stationary before differencing but are stationary after differencing. Again the AlC is used to identify the most appropriate lag length for the VAR model. Here too the cointegration equations tested assume no deterministic trends but include intercepts.  Note: • denotes sigi ificance at 5%. SB, NG, and CO are the futures prices of conventional soybeans, non-GM soybeans, and com. The numbers after the SB , NG, and CO rep resent t he second-to sixth-nearest futures contracts.
Tables 2.8 and 2.9 give the results for the third-nearest to sixth-nearest futures contracts. As seen from these tables, in all different contract months, conventional and non-GM soybeans were not cointegrated after the breaks in Noce: Y denotes that the two pric.es are cointegraled and N indicates that they are not cointegrated. SB, NG , and CO denote conventK>nal soybean, non-GM soybean, and com futures contracts. 2nd to 6t h represent the sec.end-nearest to si.'1h-neares t futures contracts.
:!~ates that the trace test did not reject acoint egration relationship between SB and CO , but the ma\lmum eigenvalue te~t rejected this relations hip .
ues th:i: the trace test rejected acointegration relat ionship between NG and CO, but the ma'limum eigenvalue test did not rejected this relationship . Table 2.10 gives the summary of the co integration tests conducted on 63 each period for different contract months. Here, too, it can be seen that the breaks that occurred in late 2007 and July 31, 2008, both had a large impact on the price relations between the non-GM and conventional soybeans, and com. Thus it can be concluded that these breaks did affect the co integration relationships of these price series.

Conclusions
Testing for the co integration relationships between the prices of non-GM and conventional soybeans, and com using the data for the whole period revealed that a cointegration relationship exists between the non-GM and conventional soybean futures prices and for the non-GM soybean and com futures prices. This result implies that the non-GM and conventional soybean futures market, and the non-GM soybean and com futures markets are linked and have an influence on one another. Hence these markets can share valuable price information and price information in these markets can affect the decisions of participants in these futures markets. This implies that the price discovery process of the non-GM soybean futures market offers valuable information to the participants in the conventional soybean and corn futures markets and that cross-hedging is possible among these futures markets.
One of the possible reasons that the non-GM soybean market is cointegrated with the conventional soybean and corn futures markets is that the non-GM soybeans can be substitutes for these commodities. Most of the conventional soybeans and some of the corn traded at the TGE are used for producing oil but it is also possible to use the non-GM soybeans for oil. The other reason for these markets to be cointegrated is that the traders may be participating in these futures markets for arbitrage purposes. The cointegration found between the non-GM and corn markets may be related to the activities of arbitragers since These breaks found on the price relationship between the conventional and non-GM soybeans also had an impact on their cointegration price relationship, and that between the two soybean and com prices. The break found in late 2007 changed the cointegration relationship between the conventional and non-GM soybean futures prices. The two soybean futures prices were cointegrated for the period before this break but they were not cointegrated for the period after this break occurred. The cointegration test conducted for the period after the break that was found in late 2008 also showed an effect on the cointegration relationship between the conventional soybean and com futures prices. These prices were not cointegrated even for the whole period used in this study but the result of the trace test for the period after this break suggested that these prices are cointegrated. As mentioned in the introduction, 2008 was a dramatic year in terms of world economic crisis and it is reasonable to believe that this break had affected the price relationship of these commodities.
In conclusion a cointegration relationship exists between the non-GM and conventional soybean futures markets, and between the non-GM soybean and com futures markets. However, the breaks found in these markets affected these relationships. Hence, the price information of these markets can be valuable when the breaks are not affecting the price relationship between the markets but it can become useless when the breaks are affecting the three markets. In this sense, the TGE soybean and com futures markets are not efficient.
The Tokyo Grain Exchange (TGE) is the world's first futures market to create a separate futures market for non-genetically modified (non-GM) soybeans (Parcell 200 l ). Since May 18, 2000, the soybean futures market at the TGE has been split into two different markets: conventional and non-GM. The main reason for opening this new market for non-GM soybeans was to meet the increasing demand for non-GM soybeans in Japan (TGE 2003). According to , Japanese consumers show a high preference for non-GM food over GM food and concerns toward GM products have been spreading.
It is known that the futures market provides an important role in facilitating price discovery of commodities and to hedge price risk . Fontenbery and Zapata (1997) state that the futures market has to be efficient for the price of the market to be able to accurately reflect market participants' supply and demand expectations for a future delivery period. Thus, to find out if the two soybean futures markets are functioning to play the above mentioned roles, this paper will investigate the efficiency of these markets.
Market efficiency here means "speculative efficiency" as defined by Bilson ( 19 g 1) where prices fully reflect available information so that there is no strategy for participants in the market to make consistent profits from the market. It is important for a market to be efficient since traders engaged in an efficient market can trade at lower transaction costs due to fewer searches for extensive information . Furthermore, if a market is efficient the futures price becomes a reliable source for forecasting and hence the market provides reliable information for price discovery .
There are many studies investigating the market efficiency of the commodity futures market, such as those of wheat, soybeans, rice, nonferrous metals, and so on .
The results of these previous studies on market efficiency of commodity futures markets vary, and whether a certain commodity futures market is efficient depends highly on the market itself.   concluded that this market is efficient. So far no testing has been conducted on market efficiency for the TGE soybean futures market and the result of this study will be valuable for understanding whether the newly developed non-GM soybean futures market provides effective information for its price discovery process.
This paper also examines whether it is the spot price or the futures price that causes the two prices to move together in the long-run at the TGE soybean futures market. When market efficiency holds in a market, it means that the spot and futures prices are "close together," never drifting far apart .
Most studies on testing market efficiency only examine the existence of the long-run relationship between the spot and futures prices and do not look further to find out the causes of this long-run relationship but this paper will also study how this long-run relationship was achieved. If the test results show that it is the futures price that leads the long-run relationship it will be the futures price that first incorporates new information to the market, and vice versa if the spot price leads the futures price. The result of this test will reveal whether it is the spot price or the futures price that plays an important role in the price discovery process.

77
Many studies on the price relationship between the spot and futures prices have shown that it is the futures price that leads the spot price, but there are some studies that reveal the opposite case, and so far, there is not any agreement as to which price binds the spot and futures prices to move together in the long-run . The argument for the futures price to lead the spot price is that the futures market has lower transaction cost and is easier for shorting compared to the spot market so the futures price should respond quicker to new infonnation than the spot price . On the other hand the supporters of the spot price leading the futures price believe that if the difference between the spot and futures market diminishes quickly (converges quickly to equilibrium), and if traders cannot perceive this difference and are more aware of the cointegration relationship between the spot and futures prices, the long-run equilibrium tends to be led by the spot price ).
In the following section the details of the data used in the study are explained. In the third section the model and the methods used in this research will be discussed. The fourth section presents the results of the analysis. Finally, 78 in the last section conclusions and implications from the study wi ll be explained.

Data
The data used in the analysis is obtained from the TGE (TGE 2008). The monthly futures price wi ll be extracted by taking the average price of each month by using the daily price data of the TGE non-GM and conventional soybeans.' The terms of the data taken are from June 2000 to October 2008.

Dec.
Feb. Ap r.

Dec. Feb
Ap r.
Dec. Feb Apr. Jun . Ju l.
A ug . Aug. Oct.
As seen in table 3 There is no organized cash market for the non-GM soybeans, and it is common in practice to use the closest contract price as a proxy for the cash price when it is not available (Asche 2002). Thus the second-nearest futures price is used as the spot price in this study.

Methodology
The market efficiency is tested under the following model: where St is the spot price at time t, Ft-l,t is the futures price at time t -1 maturing at time t , Et is the error term, and a and b are constant coefficients.
If a == 0 and b = 1, the spot price at time t becomes equal to the futures price at some period prior to its contract maturity. When this condition holds, the futures price fully reflects available information and there is no chance for traders to consistently profit through their trades in the market. This is the hypothesis that will be tested to see if the conventional and non-GM soybean futures markets are efficient. This is examined by using the model used in  and through the use of Johansen cointegration method .
Many studies have used the   82 months (third-to sixth-nearest contracts).
As suggested by , the existence of a long-run relationship between the spot and futures prices needs to hold before doing further tests such as the market efficiency test or the causality test. If the spot and futures prices do not show cointegration relationships, it will mean that both prices are generated independently and it is impossible for one to provide any information for predicting the other . Thus the market efficiency condition, a = 0 and b = 1 is tested after the cointegration relationship is found between the spot and futures prices. This condition is tested by putting these restrictions on the cointegrating vector. The causalities of a long-run relationship between the spot and futures prices in the non-GM and conventional soybean futures markets are also tested using the Johansen procedure by imposing restrictions on the so-called speed of adjustment parameters in the Johansen framework (Johansen and Juselius 1990).

The Johansen Cointegration Test
The Johansen cointegration procedure used in this study is based on the following vector error correction model (VECM):

83
( 2) where Xt is the n x 1 vector (x 1 v Xzv · · ·, Xnt)', p is the order of the vector autoregressive process, Et is a normally distributed n-dimensional white noise process, n = -I+ Ii=l nj, and ri = -If=i+l n/ In this research the vector (xlt, Xzt• · · ·, XntY consists of the spot and futures prices of the soybean futures prices at the TGE: ( 3) where St is the spot price and Ft-i to Ft_ 4 are the prices of third-nearest to sixth-nearest futures contracts. Whether equation (2)   This is because other parts of equation (2) will be stationary since difference of the X variables will be integrated of the same order by assumption.

84
where xi is the estimated values of the unit roots obtained from the estimated n matrix, T is the number of usable observations, and r is the number of possible cointegrating vectors. The appropriate lag length for the VAR model is determined based on the Akaike information criteria (AIC).

Restriction Testing
The market efficiency condition in equation (I) is tested by imposing restrictions on the cointegrating vector in the Johansen procedure. For restriction testing, Johansen defines the n matrix as n = af3' where f3 is the matrix of cointegrating vector and a is the speed of adjustment parameters that is outside the cointegrating relationship. The following VECM is used in the study to test these restrictions: where S is the spot price, and F is the futures price at some period before the contract maturity. A cointegration between the spot and futures prices is a necessary condition for market efficiency. Thus a cointegration between the test variables needs to be verified before performing the restriction test.
If equation (6)  x;' = (St, Ft-i-1)'. The test statistic used to test this restriction is: where Xi and Xi denote the ordered characteristic roots of unrestricted and restricted models. This test statistic follows an asymptotic x 2 distribution with degrees of freedom equal to the number of co integrating vectors.
The causality of the long-run equilibrium between the spot and futures prices is tested by implementing the restriction on the a matrix and testing whether af3' (St-l• Ft_ 1 )' is stationary. Defining a' = (a 1 , a 2 ) , if a 1 * 0, the deviation from the long-run equilibrium will be mainly adjusted by the change in the spot price, while if a 2 * 0 the deviation will be adjusted by the change in the futures price. This would mean that if a 1 = 0, there are no changes in the spot price due to change in the long-run relationship between the spot and futures prices and all corrections to reach the long-run equilibrium are done through the changes in the futures price. If a 2 = 0, the changes in the equilibrium will be 86 adjusted by the spot price. Thus, if the results of the restriction test suggest a 1 == O, the spot price will lead the futures price and vice versa when a 2 == 0.
However, if a 1 * 0 and a 2 * 0, there will be no price leadership. It cannot be a 1 == a 2 == 0 since this would mean that there is no long-run relationship between the spot and futures prices and would contradict the assumption that there is a cointegration relationship.
This test is often known as the weak exogeneity test since, if a 1 = 0 or a 2 == 0, it will mean that the corresponding variable does not respond to discrepancy from the long-run equilibrium relationship. So in this study when a 1 = 0, as seen from equation (6), the spot price becomes weakly exogenous for the futures price and will imply that the spot price leads the futures price while the futures price is weakly exogenous for the spot price if a 2 == 0, and in this case the futures price will lead the spot price.

Results
The result from the ADF unit root tests indicate that in every contract month, conventional and non-GM soybean futures prices all had a unit root (  • denotes s i~i ficance at 5% . The values inside the parenthisis are the p-\'alues . SB is the conventional soy bean price mid LR is the likelihood ratio explain ed in equaion (7). The s pot price (Spot SB) fo r the comentional soy beans is the price of second-nearest futures contract as exp lained in the dat a section. 14.87 (000)* Note: • dalotcs si1'J,ificancc at 5% . The \'alues insldc the parcnlhisis are the p-vaJues. NG is the non-GM soybean price and LR is the likelihood ratio c:-:p laincd in equa.ion (7). The spot price (Spot NG) fo r the non-GM soy beans is the price of so::ond-ncarest futures contract as c:q>lained in the data section.
The results show that for both conventional and non-GM soybeans, the 90 spot and futures prices at different contract months are all cointegrated of order one: the null hypothesis of no cointegration relationship, that is, r = 0, is all rejected at the 5% significance level as seen in the colums of the trace and maximum eigenvalue tests. Hence we can proceed for testing the market efficiency condition, a = 0 and b = 1, and the causality between the spot and futures prices by imposing restrictions in the Johansen cointegration procedure.
The results of the restriction tests conducted in the Johansen cointegration framework show that the hypothesis a = 0 and b = 1 is not rejected for the conventional soybean series for all different contract months, which suggests that the market efficiency condition holds between the spot and futures prices for this market. On the other hand, this condition is met only for the fifth-and sixth-nearest contract months for the non-GM soybean futures market and was rejected for the third-and fourth-nearest contracts. It is known that at the TGE more distant contracts are more active than the nearby contracts, and this could be the reason why the market efficiency condition did not hold for the 91 nearby third-and fourth-nearest futures contracts . 5 Thus the test result for the market efficiency condition can be summarized as the conventional soybean futures market is efficient while the non-GM soybean futures market is somewhat inefficient.
Finally the result of the causality test, which can be seen in the last column of the tables, for both conventional and non-GM soybean futures markets, the null hypothesis of the spot price being weakly exogenous for the futures price is not rejected, except for the case of the third-nearest non-GM soybean futures price. However the opposite case is denied for all tests conducted between the spot and futures prices for different contract months. This suggests that spot price leads the futures price, which implies that the spot price is the one that binds the spot and futures prices to move together in the long-run and that new information is first incorporated into the spot price at the TGE soybean futures markets.
5  explain that the reason why the more distant contracts are more active at the TGE is because of their trading system, which is called ' itayose-hoh' or single fixed-price auction. In this system the contracts are auctioned in the order of the expiration of the contract.
Thus the nearest contracts are auctioned first and then the second-nearest futures contracts are auctioned, and this continues until the furthest contracts are auctioned so that more information is always available for the further contracts (Booth and Ciner I 997).

Conclusions
The Johansen multivariate and bivariate cointegration tests revealed that the spot and futures prices of the TGE conventional and non-GM soybeans are cointegrated. This result revealed that these prices move together in the long-run.
Cointegration is a necessary condition for the market to be efficient so that the prices in the market fully reflect available information and no traders can profit consistently from the market. However this is not a sufficient condition to conclude that a market is efficient.
I therefore tested as to whether the futures price at some period before its maturity will be equal to the spot price in the long-run. This test showed that the sufficient condition for market efficiency does hold for the conventional soybean futures market. On the other hand, it failed for some contract months for the non-GM soybean futures market. This implies that the non-GM soybean futures market is relatively inefficient compared to the conventional soybean futures market. Some investors may be profiting consistently through their trades at the market. The possible reason for this inefficiency may be that there is no organized cash market for the non-GM soybeans, and that the only closest cash market available is the nearest futures contract, which has low liquidity.
Furthermore, the non-GM soybean futures market is new compared to the conventional soybean market so that its historical price information may not be as valuable as the conventional soybean market.
non-GM soybean futures market had been one fifth of the conventional soybean futures market, the non-GM soybean futures market may have attracted more speculators to the market and this may have increased the number of short-term traders in this market.

7
The result of the causality of the long-run relationship suggested that for both conventional and non-GM soybean futures markets, it is the spot price that leads the spot and futures prices to move together. As mentioned in the introduction when the long-run equilibrium is led by the spot price it is argued in previous studies that this occurs because the market participants believe that the spot and futures price are strongly linked and that they converge quickly to equilibrium . This could be the case with the two soybean futures markets at the TGE, since both these markets showed a strong cointegration relationship in both the multivariate and bivariate cointegration tests.
7 Until Oct~ber, 2008, the minimum contract unit for the non-GM soybeans at the TGE had been 10 mt while that for the conventional soybeans had been 50 mt (TGE 2008).

Appendix A. Explanation on the ARIMA Expression
Say y is the response series, p is the order of the autoregressive part, q is the order of moving average part, and other notation is the same as the ones used in the main text. Then ARIMA(p, 0, q) can be expressed as Yt = ( <P1Yt-1 + <P2Yt-2 + ··· + <t>p-1Yt-p+1 + <t>pYt-p) + where B is the backshift operator, <PCB) is the autoregressive operator