Asymmetric transmission between terminal and shipping point prices for selected fruits Byeong-Il Ahn* and Hyunok Lee** June 2012 *Department of Food and Resource Economics Korea University Anam-dong Seongbuk-gu, Seoul, 136-701 South Korea and **Department of Agricultural and Resource Economics University of California, Davis Davis, CA 95616 This research was funded in part by California Department of Food & Agriculture (CDFA) Specialty Crop Block Grant CFDA #10.170 1 Asymmetric transmission between terminal and shipping point prices for selected fruits Price transmission has drawn widespread interest from economists. Previous studies analyzed price relationships in both input and output markets, among different links along the supply chain, and across nodes in spatially disperse markets. A variety of products—from agricultural commodities to petroleum products—has been the subject of empirical work. One common assumption used in studies of price transmission is symmetry of responses to shocks. That is, the magnitude of price transmission across markets or nodes does not depend on the direction (up or down) of the initial price shock (Wohlgenant, 2001). A few studies applied to agricultural commodities have attempted to investigate empirically this assumption using a more general framework that separates the regimes by the direction of initial price shock and allows the possibility of non-symmetric transmission (Karrenbrock, 1991; Azzam, 1999; Meyer and von Cramon-Taubadel, 2004; Kaufmann, 2005; Ahn and Kim, 2008). Following this line of literature, commonly referred to as asymmetric price transmission, the present study adds to the price transmission literature on specialty crops by investigating the structure of price transmission in the context of the vertical market chain for fruit markets in the United States. Focusing on the initial shipping point and terminal (wholesale) links in the marketing chain, we examine short-term as well as cumulative price responses of terminal prices to changes in shipping point prices. In addition to providing empirical evidence related to asymmetry, previous studies on asymmetric price transmission explored its interpretations. If the markets were efficient (and some additional conditions are met), a price shock in one market affects the price of the related market in a symmetric fashion. This suggests that the test of asymmetry could be used to investigate market efficiency, and that the evidence of asymmetry would be consistent with a market with asymmetric transaction costs, market power or some other deviation from perfect competition (Meyer and von Cramon-Taubadel, 2004; Carmen and Sexton, 2005; Koutroumanidis et al., 2009). 2 A number of studies of agricultural commodities suggested that the main driver of asymmetric price transmission in a vertical marketing chain is market power (McCorriston et al., 1998; Azzam and Schroeter, 1995; Chen and Lent, 1992; Bunte and Peerlings, 2003; Carmen and Sexton, 2005). A party with market power can influence the price to increase profits. Under such a situation, market participants with market power exploit the situation differentially depending on the direction of the initial price shock. We explore the implications of our test results for the underlying market structure of marketing chains considered in this study. Of course, evidence of asymmetry does not necessarily imply market power. Carlton (1989) suggests that unless the change in marginal cost (the procuring price from the upstream marketing chain) is sufficiently large, retailers do not implement a price change due to a menu or re-pricing cost. Bettendorf and Verboven (2000) supported this claim empirically in the context of vertical markets for coffee. Other explanations include Ball and Mankiw (1994) who focused on asymmetric adjustment of nominal prices during the time of inflation, and Reagan and Weitzman (1982) who showed how competitive industries result in asymmetric price transmission through their inventory holding behavior. Finally, Bailey and Brorsen (1989) pointed out that asymmetric (or imperfect) information about costs may cause asymmetric price transmission.1 We formally test the asymmetry of price transmission between the shipping point and terminal prices using weekly price data spanning from 1998 to 2011. The model separates the two regimes of the initial price shock and incorporates time lags of both explanatory and dependent variables, which enables us to investigate the price adjustment process between prices in different stages of the marketing chain. The empirical models are discussed in the next section. Data description and preliminary tests on data and model specification, including the tests on lag order, unit-root, causality and cointegration follow. We then report the model estimation results, and conclude the paper with a summary and implications. 1 For more discussion, see Peltzman (2000) who investigated various plausible causes for price asymmetries, such as market concentration, inventories, inflation-related asymmetric "menu costs" of price changes, or the fragmentation of the marketing chain. His study supports none of these causes, except for the level of fragmentation of the marketing chain. 3 Empirical Model The salient features of the models used in testing asymmetric price transmission involve the segmentation of the regimes, differentiated by the sign of the initial price shock. Early studies adopting regime segmentation include Wolffram (1971) and Houck (1977).2 Allowing the possibility of non-instantaneous price adjustments, Ward (1982) and Boyd and Brorsen (1988) extended Houck’s model by incorporating lagged explanatory variables in the vertical marketing chains of fresh vegetables and pork in the United States.3 The length of lags in this framework corresponds to the duration of the price adjustment, and the coefficients on the lagged explanatory variables indicate the magnitude of their impacts on the dependent variable. The main drawback of this approach relates to the time series properties of the data. Recognizing that this simple approach is not consistent with common time series properties of the data, Borenstein et al. (1997) and von Cramon-Taubadel (1998) applied a cointegration method to the tests of asymmetric transmission between the crude oil and retail gasoline prices and between producer and wholesale prices in the German pork market, respectively. As a more comprehensive time-series approach, Krivonos (2004) adopted an error correction model and characterized the long-run equilibrium price transmission in the world versus producing-county coffee markets. Developing the empirical model starts with defining the relationship between the current and lagged prices in a marketing chain simply composed of the downstream and upstream d markets. Let Pt , denoting the downstream price at time t, depend on its own lagged values u and the contemporaneous and lagged upstream prices, P . Then, a typical autoregressive distributed lag (ADL) model with the lag length of n can be specified as: n n Pt a0 a P bi Pt ui t . This equation assumes a symmetric relationship d i 1 d i t i i 0 2 Even though the typical asymmetric specification originates from Wolffram and Houck, the basic conceptualization of asymmetric price transmission goes back to Farrell (1952), who first investigated empirically the irreversibility behavior of the demand function of habitual consumption goods. 3 Houck’s (1977) model was developed to test the asymmetry in supply response in the U.S. milk and pinto beans markets by extending the basic model concept by Wolffram (1971), who segmented the initial price shock into increasing and decreasing phases. 4 between the changes in explanatory variables and the dependent variable. This symmetric relationship is more immediate when the equation is expressed in differences, n1 n1 Pt ai P bi Pt ui , where ∆ signifies a change from the value of the previous d i 1 d t i i 0 period. To incorporate the possibility of asymmetric transmission, we need to separate the explanatory variables depending on the direction of the change, which can be specified using binary variables, Ai Ai Bi and Bi : (Model 1) 1 if Ai 0 n1 n1 n1 n1 i 1 i 1 i 0 i 0 Pt d ai Ai Pt di ai Ai Pt di bi Bi Pt ui bi Bi Pt ui et 1 if Pt d i 0 1 if Pt ui 0 1 if Pt ui 0 , Ai , Bi Bi otherwise otherwise otherwise otherwise 0 0 0 Pt d i 0 Equation (1) allows two types of asymmetric price transmission tests. First, we can test for u u short-term asymmetric price transmission between Pt i and Pt d . If Pt i were symmetrically transmitted to Pt d , estimated coefficients bi and bi would be the same. Thus, asymmetric price transmission exists with respect to the ith lagged upstream price if two coefficients are significantly different from one another. The second test is for cumulative n 1 asymmetric price transmission. If the influences of B P i 0 i u t i n 1 dependent variable are symmetric, the cumulated coefficients b i 0 i B P i 0 b i 0 n 1 n 1 and i i u t i on the would be the same as . Thus, the hypothesis of symmetric cumulative price transmission would be rejected if these sums were significantly different from one another. The short-term symmetry implies cumulative symmetry, but not vice versa. Further, the short-term asymmetry does not imply cumulative asymmetry. Similar tests between the contemporaneous and lagged dependent variables apply. That is, the short-term price adjustment is said to be asymmetric if the data reject H 0 : ai ai and the cumulative price adjustment is asymmetric if the data 5 reject H 0 : n 1 n 1 i 1 i 1 ai ai . Note that tests on asymmetric price transmission provide information about market efficiency (or inefficiency) and market inefficiency is indicated by the asymmetry on the b coefficients. n 1 Further, asymmetry confirmed with further statistical evidence of b i 1 n 1 greater than b i 1 i i being significantly is termed positive asymmetry, which is consistent with the fact that higher profits are earned by downstream participants than what they could have earned under the efficient market (Carman and Sexton, 2005). Another alternative model specification relates to the time series nature of the price variables. Although equation (1) can capture the cumulative effects in price transmission, these models essentially do not consider the effects of price transmission when price variables deviate from d u their long run path. In general, differenced variables (such as Pt d , Pt i and Pt i ) tend to be stationary, however, the original price variables may meander without showing the constant mean or variance over time. Although the prices are not stationary, if a linear relationship between these price variables is stable and the residual from this linear relationship is white-noisy, we say that cointegration exists between these variables (Anders, 1995). If the existence of cointegration is identified, the asymmetric price transmission model can be extended to specify the long-run adjustments by introducing the errorcorrection terms, as in von Cramon-Taubadel and Loy (1996). Given the error correction model can be extended based on the results of the cointegration test, the presentation of the error correction model will be deferred until we perform the cointegration test. Data One distinct characteristic of fresh fruits is perishability, which surely contributes to shortterm fluctuation of market prices. This implies that price transmission can be relatively in short term and the data used to examine price transmission necessarily have to reflect such 6 short terms. In light of this, we searched for time series price data with short intervals. The Marketing Service at the U.S. Department of Agriculture provides weekly prices of major agricultural commodities at various marketing channels. We have chosen fresh strawberries, apples, table grapes and fresh peaches as representative fruits, and for each of these fruits, we collected weekly prices at the shipping point and terminal market. In terms of selecting the location of the shipping point and terminal market, we picked the shipping point in the region that is associated with the largest production and the terminal market that likely handles the largest volume. Obviously, the terminal prices and shipping point prices correspond to the upstream price and downstream price in our model, respectively. While our data period spans from 1998 to 2011, depending on its season, the data series for each fruit begins and ends in different months. Data details are provided in the appendix. These two price series tend to move together, and fluctuate considerably over the time period investigated. Price fluctuations are larger in the second half of the period for both series. The time pattern of the price fluctuation suggests the possibility of non-stationarity of price series, which violates the time series property of constant mean and variance. The comovement of these series further suggests the possibility of cointegration, as is often manifested by the parallel pattern of non-stationary variables. The time series properties of the data will be formally investigated next. Preliminary tests In this section we first consider the choice of lag orders and causality and then investigate time series properties of the data. We select appropriate lag orders based on statistical criteria, and perform the causality tests to identify the relationship between shipping point and terminal prices. We conduct unit root tests on the variables used in empirical equations to avoid spurious regression and erroneous interpretation of estimated results. We also test for cointegration between the wholesale and factory prices for the possibility of including long-run relations of these time series vectors in the empirical estimation. 7 Lag order choice and causality tests The following vector autoregressive (VAR) model is used to determine the lag orders and conduct the causality test: PtT T a1T b1T PtT1 a2T b2T PtT 2 akT bkT PtT k etT S S S S S S S S ... S S S S Pt a1 b1 Pt 1 a2 b2 Pt 2 ak bk Pt k et Within the VAR formulation expressed as above, the optimum lag order is selected using the Akaike’s information criterion (AIC) and Schwartz Bayesian Information Criterion (SBIC). We considered up to the fifth lag, and the lag order that produces the minimum information criteria is selected (Anders, 1995). Table 1 presents the values obtained under the two information criteria for each order of lags considered. Using these criteria, we selected the lag order one for apples, lag order four for table grapes, lag order three for strawberries, and lag order four for peaches.4 In specifying the empirical equations, our underlying assumption is that the current terminal price is influenced by the shipping point prices, consistent with the usual assumption that downstream prices are affected by upstream prices.5 However, this is not always the case, and an easy example is the case of market power. Under the influence of market power, the party with market power influences the price to his/her advantage and thus acts as an initiating party in price causality. Thus, with no prior knowledge about the industry, we need to empirically evaluate causality. To check the causality between prices, we used the estimation results from the VAR model. k k i 1 i 1 From the first equation of the VAR model ( Pt T d aid Pt Ti bid Pt Si ), we can say that 4 Due to the space limitation, we do not present the full estimation results of the VAR model. 5 The opposite direction of causality is, of course, plausible as in Koutroumanidis et al., which deals with a market where imports represent a large share of domestic consumption, and the import price leads the consumer price. They find that under such market conditions, causality is from downstream to upstream markets, i.e., the consumer price affects the producer price. 8 T P S causes P if we reject the null hypothesis H0: b1d b2d ... bkd 0 . Likewise, using k k i 1 i 1 the results of the second equation of the VAR model ( Pt S u aiu Pt Ti biu Pt Ti ), PT is said to cause P S , if we reject the null hypothesis H0: a1u a2u ... aku 0 . Table 2 shows the causality test results. The test results show that except for strawberries, shipping point prices cause terminal prices, which is consistent with the underlying assumption upon which our model specification is based.6 However, for strawberries, the causality of both directions is strongly rejected, implying that the direction of price causality is inconclusive. This invalidates the specification of either VAR equation and any statistical results obtained from our asymmetric regression cannot be valid. Unit-root tests Unit-root tests are essential in checking the spurious regression and the existence of cointegration. For the regression to be statistically meaningful, all the variables in the regression have to be stationary. Specification tests concerning equation (1) include spurious regression. For each variable used in equation (1), we conduct a unit root test by adopting the Augmented Dickey-Fuller (ADF) test. The ADF test with an intercept and trend is based n on the following form of regression: Yt 0 t t Yt 1 j Yt j ut , where Y is the i 1 variable that is subjected to the unit-root test. The null hypothesis for testing the unit root is 0 . If the absolute value of the ADF test statistic is greater than the absolute value for the critical point, the hypothesis of unit root is rejected. Test results are provided in the appendix. The hypothesis of unit root is rejected at 1% significance for every variable tested, strongly indicating little possibility of spurious relationships for regression equation (1). The validity of an error correction model requires a cointegration test on time series Pu and Pd. This test is conducted in two steps. The stationarity (or non-stationarity) properties of these 6 Note that the findings of causality may result from the omission of external variables which affect both factory and wholesale prices. To check this possibility, we estimated the structural VAR (SVAR) by including price indices of recycled timber and furniture as input and output prices that may affect the fiberboard market (note that retail prices of fiberboard were not available). The estimation results obtained from the models including these exogenous variables also found that that factory prices still Granger-cause the wholesale prices. 9 time series are inspected first, and upon the evidence of non-stationarity, a cointegration test is performed. We first apply the augmented Dickey-Fuller (ADF) test to each of time series Pu and Pd using the same regression form employed in the previous unit root tests.7 Based on the estimated ADF test statistics for shipping point and terminal prices (Table 3), the null hypotheses of unit root process is rejected for both variables, implying that both time series variables are stationary. Having obtained the evidence of stationarity, we do not proceed with the Engle-Granger cointegration test, which thus precludes in the context of modeling the error correction model approach. Estimation results of asymmetric price transmission Table 4 reports the results obtained from estimating model (1). All coefficients except one (which is statistically insignificant) related to shipping point prices are positive, while all lagged own-price effects except those for apples are negative. Positive shipping point price effects indicate that the terminal price moves together with shipping point prices both current and lagged, meaning a rise (fall) in shipping point price induces an increase (reduction) in terminal price. While this finding is consistent with our intuition, negative lagged own price effects are interesting. Except in the case of apples, previous terminal prices affect the current terminal price in the opposite direction. Combined with the findings on positive shipping point price effects, the significance of negative lagged own price effects is that the lagged own prices work as a dampening factor, even though positive shipping point price effects may dominate. We have found distinct patterns of lagged price effects for each fruit. The lagged price effects we obtained do not conform to the usual expectation that prices effects taper down as the order of lag increases. For apples, one period lagged shipping price effect associated with a negative price change is much larger than the current price effect (0.08 vs. 0.34). For peaches, the second lagged shipping point price has the largest price effects among all lags for both positive and negative changes, and for own price effects, lagged price effects tend to 7 Recall that we test the non-stationarity of the differenced variables such as Pt d and D Pt u to check the spurious regression for models 1 and 2. For the test for cointegration, we have to check the non-stationarity (unit root) of the original variables Pu and Pd. 10 get larger as the lag increases. F-test statistics for asymmetry tests are reported in Table 4 and the summary interpretation of the test results is provided in Table 5. The null hypothesis that the effect of the current shipping point price is symmetrically transmitted to the current terminal price is rejected for apples and peaches but not for table grapes. These asymmetry effects are, however, positive for apples and negative for peaches, meaning that for apples the terminal price responds with a larger margin to a shipping point price increase than to a decrease and the reverse is true for peaches. Nevertheless, as indicated by the magnitude of the estimated coefficients for b0 and b0 , positive asymmetry for apples is substantial while the negative asymmetry for peaches is relatively marginal. Asymmetry of cumulative FOB price effects is also positive for apples, but inconclusive for peaches. For table grapes, the symmetry of the cumulative effect of shipping point prices cannot be rejected. The hypothesis of symmetric cumulative lagged own price effects is strongly rejected for apples but cannot be rejected for table grapes and peaches. We have also examined asymmetric price transmission in the context of quantile regression. Quantile regression, which was introduced by Koenker and Bassett (1978), extends the regression model to conditional quantiles of the response variable. Quantile regression is useful when the rate of change in the conditional quantile, expressed by the regression coefficients, depends on the quantile. In addition to the fact that the quantile regression estimates are more robust against outliers in the response measurements, the quantile regression approach produces a comprehensive analysis of the relationship between variables. We considered the quantiles at the 20% increment, and summarized the asymmetry test in Table 6. A distinct pattern of asymmetry emerges from table 6. Negative asymmetry is, in general, found mostly at low levels of quantiles and positive asymmetry mostly at high levels of quantiles. That is, at a low level of price change the terminal price responds more to a decline in the FOB price than to an increase. On the other hand, at a relatively larger price change, the terminal price responds more to an increase in the FOB price than to a decrease. These 11 patterns are pronounced especially for apples and table grapes. In comparison with the mean value estimation (under the usual least square estimation method), quantile estimations tend to produce negative asymmetry associated with changes in FOB price results at a relatively low quantile range. There may be a number of explanations for positive asymmetry at a high quantile level. For instance, the profit-extracting behavior of terminal marketers would be a possibility, which is allowed only under certain market conditions. Another possibility has to do with the operating structure of terminal marketers. Suppose that a large increase in the FOB price occurred after a serious supply shock. A contraction in supply causes a reduction in total product quantities at the terminal market. It is easy to imagine that the terminal market may operate in less than full capacity in this case. This likely increases per-unit operating cost, which results in an increase in terminal price exceeding the increase in FOB price. Regarding the negative asymmetry at a relatively low level of price change, explanations are not immediate. Conclusions This article has developed and tested hypotheses of asymmetric price transmission in fruit prices using the appropriate statistical tools and after tests to account for time series properties of the data. The econometric results have found evidence of asymmetric price movements between shipping point and terminal markets for fruits. We are not able to say definitively what drives the asymmetry but provide some potential rationales and ideas for further investigation. 12 References Anders, W. (1995), Applied Economic Time Series, John Wiley & Sons. Andrews. D. W. K, (1993), “Tests for parameter instability and structural change with unknown change point,” Econometrica 61(4): 821-856. Andrews, D.W.K., W. Ploberger 1994. “Optimal tests when a nuisance parameter is present only under the alternative,” Econometrica, 62 (6): 1383–1414 Azzam, A. M. (1999) “Asymmetry and rigidity in farm-retail price transmission” American Journal of Agricultural Economics 81: 525-533. Azzam, A. M., & Schroeter, J.R. (1995) “The tradeoff between oligopsony power and cost efficiency in horizontal consolidation: An example from beef packing” American Journal of Agricultural Economics, 77(4): 825–836. Bailey, D. Y. and B. W. Brorsen (1989): “Price asymmetric in spatial fed cattle markets.” Western Journal of Agricultural Economics, 14: 246-252 Ball, L. and N. G. Mankiw, (1994). “Asymmetric price adjustment and economic fluctuations,” Economic Journal 104:247-261. Bettendorf, L. and F. Verboven (2000): “Incomplete transmission of coffee bean prices: evidence from the Dutch coffee market”, European Review of Agricultural Economics, 27, 1-16. Borenstein, S., C Cameron and R. Gilbert (1997), "Do gasoline prices respond asymmetrically to crude oil prices?" Quarterly Journal of Economics 112: 305-339. Boyd, M.S. and Brorsen, B.W. (1988), “Price asymmetry in the U.S. pork marketing channel” North Central Journal of Agricultural Economics, 10: 103-109. Carlton, D. W. (1989). “The theory and practice of how markets clear: Is industrial organization valuable for understanding macroeconomics” In R. Schmalensee & R. Willing (Eds), Handbook of industrial organization. Amsterdam: North Holland, 909-946. Carman, H. F. and R. J. Sexton (2005), "Supermarket fluid milk pricing practices in the Western United States” Agribusiness, 21(4): 509-530. Farrell, M. J. (1952). “ Irreversible demand functions” Econometrica, 20(2): 171–186 Hansen, B. (1997). “Approximate asymptotic p-values for structural change tests,” Journal of Business and Economic Statistics, 15: 60–80 Houck, J.P. (1977). “An approach to specifying and estimating nonreversible functions”, American Journal of Agricultural Economics, 59, 570-572. 13 Karrenbrock, J. D. (1991). “The behaviour of retail gasolin prices: symmetric or not?”, Federal Reserve Bank of St. Louis Review 73: 19-29. Kaufmann R. K. and C. Laskowski (2005) “Causes for an asymmetric relation between the price of crude oil and refined petroleum products” Energy Policy (33): 1587–1596. McCorriston, S., Morgan, C. W., & Rayner, A. J. (1998) “Processing technology, market power and price transmission” Journal of Agricultural Economics, 49(2): 185–201. Meyer, J. and von Cramon-Taubadel, S. (2004), "Asymmetric price transmission: A survey" Journal of Agricultural Economics 55(3): 581-611. Reagan, P. B. and M. L. Weitzman (1982): “Asymmetries in price and quantity adjustments by the competitive firm”, Journal of Economic Theory, 27: 410-420 von Cramon-Taubadel, S. (1998) "Estimating asymmetric price transmission with the error correction representation: An application to the German pork market" European Review of Agricultural Economics 25: 1-18. Ward, R. W.(1982) “Asymmetry in retail, wholesale and shipping point pricing for fresh vegetables” American Journal of Agricultural Economics 62:205-212. Wohlgenant, M.K. (2001) “Market margins: Empirical analysis” Chapter 16 in Handbook of Agricultural Economics, Volume 1. B. Gardner and G. Rausser, eds., Elsevier Sceince B.V. Wolffram, R. (1971) Positivistic measures of aggregate supply elasticities: Some new approaches: some critical notes” American Journal of Agricultural Economics 53:356359. 14 01/10/1998 05/16/1998 09/19/1998 01/23/1999 05/29/1999 10/02/1999 02/05/2000 06/10/2000 10/14/2000 02/17/2001 06/23/2001 10/27/2001 03/02/2002 07/06/2002 11/09/2002 03/15/2003 07/19/2003 11/22/2003 03/27/2004 07/31/2004 12/04/2004 04/09/2005 08/13/2005 12/17/2005 04/22/2006 08/26/2006 12/30/2006 5/5/2007 9/8/2007 1/12/2008 5/17/2008 9/20/2008 1/24/2009 5/30/2009 10/3/2009 2/6/2010 6/12/2010 10/16/2010 2/19/2011 6/25/2011 0 01/10/1998 05/02/1998 08/22/1998 12/12/1998 04/03/1999 07/24/1999 11/13/1999 03/04/2000 06/24/2000 10/14/2000 02/03/2001 05/26/2001 09/15/2001 01/05/2002 04/27/2002 08/17/2002 12/07/2002 03/29/2003 07/19/2003 11/08/2003 02/28/2004 06/19/2004 10/09/2004 01/29/2005 05/21/2005 09/10/2005 12/31/2005 04/22/2006 08/12/2006 12/02/2006 3/24/2007 7/14/2007 11/3/2007 2/23/2008 6/14/2008 10/4/2008 1/24/2009 5/16/2009 9/5/2009 12/26/2009 4/17/2010 8/7/2010 11/27/2010 3/19/2011 7/9/2011 10/29/2011 Fig. 1.a-d. weekly prices ($/lb) for table grapes, fresh apples, fresh peaches, and fresh strawberries at terminal market and shipping point (FOB) Table grape terminal and FOB prices 3.5 3 2.5 2 1.5 1 0.5 Grape terminal prices ($/lb) terminal apple prices Grape FOB prices ($/lb) Fresh apple terminal and FOB prices 1.2 1 0.8 0.6 0.4 0.2 0 FOB apple prices 15 01/10/1998 05/30/1998 10/17/1998 03/06/1999 07/24/1999 12/11/1999 04/29/2000 09/16/2000 02/03/2001 06/23/2001 11/10/2001 03/30/2002 08/17/2002 01/04/2003 05/24/2003 10/11/2003 02/28/2004 07/17/2004 12/04/2004 04/23/2005 09/10/2005 01/28/2006 06/17/2006 11/04/2006 3/24/2007 8/11/2007 12/29/2007 5/17/2008 10/4/2008 2/21/2009 7/11/2009 11/28/2009 4/17/2010 9/4/2010 1/22/2011 6/11/2011 10/29/2011 05/23/1998 10/03/1998 02/13/1999 06/26/1999 11/06/1999 03/18/2000 07/29/2000 12/09/2000 04/21/2001 09/01/2001 01/12/2002 05/25/2002 10/05/2002 02/15/2003 06/28/2003 11/08/2003 03/20/2004 07/31/2004 12/11/2004 04/23/2005 09/03/2005 01/14/2006 05/27/2006 10/07/2006 2/17/2007 6/30/2007 11/10/2007 3/22/2008 8/2/2008 12/13/2008 4/25/2009 9/5/2009 1/16/2010 5/29/2010 10/9/2010 2/19/2011 7/2/2011 11/12/2011 Fresh peach terminal and FOB prices 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 peach terminal prices strawberry terminal prices peach FOB prices Fresh strawberry terminal and FOB prices 4 3.5 3 2.5 2 1.5 1 0.5 0 strawberry FOB prices 16 Table 1. Lag order choice Apples Table grapes Strawberries Peaches Criteria Lag1 Lag 2 Lag 3 Lag4 Lag 5 AIC -10.2381 -10.2600 -10.2901 -10.3130 -10.4519 SBIC -10.1991 -10.1947 -10.1983 -10.1945 -10.3064 AIC -3.3526 -3.6092 -3.6886 -3.8060 -3.7008 SBIC -3.2993 -3.5159 -3.5507 -3.6190 -3.4588 AIC -1.9766 -2.1087 -2.1231 -2.1201 -2.1510 SBIC -1.9339 -2.0360 -2.0189 -1.9831 -1.9801 AIC -6.1912 -6.4914 -6.8734 -6.9510 -7.009 SBIC -6.1230 -6.3708 -6.6944 -6.7072 -6.6920 Choices of lag order: Apples = 1, table grapes = 4, strawberries = 3, peaches = 4 17 Table 2. Granger Causality test results Apples chi2 test statistic Pr.( | chi2 | > critical value)1 d.f. 0.0026 0.9591 1 30.8340 0.0000 1 0.7550 0.9444 4 P ) → Terminal bT b2T ... bkT 0 PT Price( ) (H0: 1 ) T P H1: Terminal price( ) → Shipping point 41.3703 0.0000 4 a1S a2S ... akS 0 ) PS H1: Shipping point Price( ) → Terminal 11.4412 0.0096 3 166.11 0.0000 3 1.1427 0.8874 1 102.2311 0.0000 1 Causality T H1: Terminal price( P ) → Shipping point S Price( P ) S S S (H0: a1 a2 ... ak 0 ) S H1: Shipping point Price( P ) → Terminal T Price( P ) Table grapes T T T (H0: b1 b2 ... bk 0 ) T P ) → Shipping point a S a2S ... akS 0 ) (H0: 1 H1: Terminal price( Price( PS ) S H1: Shipping point Price( Strawberries Apple Price( PS ) (H0: Price( PT ) (H0: b1T b2T ... bkT 0 ) T P ) → Shipping point a S a2S ... akS 0 (H0: 1 ) H1: Terminal price( Price( PS ) S P ) → Terminal b1T b2T ... bkT 0 H1: Shipping point Price( Price( PT ) (H0: ) 1. If the probability that chi2 test statistic is greater than the critical value is less than 0.05, we reject the null hypothesis (H0), and the alternative hypothesis (H1) is accepted. 18 Table 3. Unit root test results Variable ADF test statistic 99% critical value PT -4.0922 -3.9710 PS -4.5792 -3.9711 PT -8.1362 -3.9740 PS -14.4843 -3.9751 PT -6.2987 -3.9712 PS -5.2867 -3.9727 PT -16.0236 -3.9780 PS -13.4687 -3.9799 Apples Table grapes Strawberries Peaches 19 Table 4. Estimation results Table grapes Apples Coefficient Regressor b0 b1 Std. Error Coefficient estimate1) Std. Error Coefficient estimate1) Std. Error 0.0002 0.0009 0.0084 0.0080 0.0053 0.0054 0.0600 0.8324*** 0.0702 0.2030 0.1291 0.0879 0.7168*** 0.1841 0.2647** 0.1301 D Pt D Pt S1 0.2015*** 0.0689 0.2646*** 0.0855 0.5035*** 0.1395 DPt S1 0.3446*** 0.0907 0.6088*** 0.1488 0.7430*** 0.1223 D Pt S2 0.2896*** 0.0908 0.0651 0.1808 D Pt S2 -0.0032 0.1442 0.1068 0.1148 b0 b1 Coefficient estimate1) D Pt S 0.6677*** Peaches b2 b2 S 0.0796 b3 D P 0.3751*** 0.1001 0.0792 0.1856 b3 D Pt S3 0.1344 0.1231 0.3420*** 0.0973 b4 D Pt S4 0.0174 0.0904 0.1947 0.1880 b4 D Pt S4 0.1490 0.1074 0.1925** 0.0788 S t 3 D PtT1 0.0229 0.0433 -0.1128* 0.0650 -0.0433 0.1058 DPtT1 0.0758 -0.2913*** 0.0769 -0.0371 0.0988 D PtT2 -0.2388*** 0.0724 -0.1345 0.1218 DPtT2 -0.1027 0.0720 -0.1696** 0.0840 a3 D PtT3 -0.3485*** 0.0826 -0.1446 0.1255 a3 DPtT3 -0.0003 0.0611 -0.1666** 0.0756 a4 D PtT4 -0.0199 0.1207 -0.1948 0.1690 a4 DPtT4 -0.0279 0.0517 -0.1946*** 0.0556 D2) Dumm -0.0224*** y 2 0.2551 R -0.1160*** 0.0233 -0.0259* 0.3939 a1 a1 a2 a2 0.4064*** 0.0066 0.4291 0.0149 20 F test results (table 4 continued) Test stat. (d.f.) Null hypothesis b0 = b0 n 3) n bi = b a a i 0 n i 1 i i 0 n = i 1 n b b i n0 i 0 n i n1 i 1 i i 9.700 (1,693) 0.001 18.237 (1,693) 0.000 Test stat. (d.f.) 0.322 (1,348) 0.293 (1,348) 1.344 (1,348) Test stat. (d.f.) Pr(|F|>c) 0.5887 2.919 (1,240) 0.088 0.2471 0.0263 (1,240) 0.871 Pr(|F|>c) 0.570 0.8692 1.7791 1.0455 0.4242 1.6058 1.6488 0.0229 -0.7201 -0.5171 0.4064 -0.4222 -0.5679 i i a a Pr(|F|>c) i i 1) The levels of statistical significance are denoted with *for 1%, ** for the 5% and *** for the 1%. 2) Dummy variable is set to 1 when terminal price is lower than shipping point price 3)H0: b0 = b0 is not tested when one of the estimated coefficients is not statistically significant 21 Table 5. Summary of asymmetry test results Cumulative short term shipping point price terminal price shipping point price terminal price previous terminal price current terminal price Apples Positive APT positive APT Negative APT table grapes No APT No APT No APT Peaches Negative APT Inconclusive No APT Table 6. Tests of asymmetry using quantile regressions Short term Apples Table grapes Peaches Cumulative shipping point previous terminal price terminal price current price terminal price Negative APT Negative APT Level of Quantile shipping point price terminal price 20% Negative APT 40% No APT Negative APT Negative APT 60% - - - 80% Positive APT Positive APT No APT 20% Negative APT Negative APT No APT 40% No APT No APT No APT 60% Positive APT Positive APT No APT 80% Positive APT Positive APT No APT 20% No APT No APT No APT 40% Negative APT No APT No APT 60% Positive APT No APT No APT 80% No APT No APT No APT 22 Appendix DATA DETAILS Fresh apples: Data period: from Jan 10, 1998 to Oct 1, 2011 Variety: red delicious Unit: $/pound Other details: o Retail prices: region=Northwest of U.S.; o terminal prices: region=Seattle, grade=WaExFcy, product origin=Washington, size=88s or mid-range package=carton tray pack; o Shipping point prices: region=Washington state, other specifications are the same as what are reported in the terminal market. Fresh peaches Data period: from 5/23/1998 to 2/1/2012 Variety: “various yellow flesh available.” Unit: $/pound Other details: o Retail prices: region=Northwest U.S. o Terminal prices: region=Los Angeles, size=42s, package=carton 2 layer tray pack o Shipping point prices: region=Central and Southern San Joaquin Valley California, size=40-42s, “preconditioned”. Table grapes Data period: from 1/10/1998 to 12/242011 Variety: red/white seedless Other details: o Retail prices: region=Northwest U.S. o Terminal prices: region=Los Angeles, variety=Thompson seedless, size=large, product origin=California and imports, package=all containers o Shipping point prices: regions=Coachella Valley and Chile imports Fresh strawberries Data period: from 1/10/1998 to 2/18/2012 Unit: $/lb Other details: o Retail prices: region=Northwest of U.S. 23 o Terminal prices: region=Los Angeles, size=medium to large, origin=Oxnard and Salinas–Watsonville, package=flats 12-pt baskets o Shipping point prices: region=Oxnard and Salinas–Watsonville, and other specifics are the same as reported in terminal market. Quantile Estimation results: Apples 40% quantile 20% quantile Coefficient1) Std. Error Std. Error Coefficient1) 0.0000 0.0000 0.0000 0.0000 D Pt S 0.0000 0.0037 0.0000 D Pt S 0.2500*** Na 0.0000 0.0001 0.0010 Na 1.3636*** 0.0266 0.0837 0.0437 0.0941 Na 0.0000 D Pt S1 0.0000 0.0053 0.0000 0.0013 Na 0.3030*** 0.0933 DPt S1 0.2500 0.2595 0.2044*** 0.0091 Na 0.0000 0.0207 D PtT1 0.0000 0.0027 0.0000 0.0007 Na 0.0909 0.2034 DPtT1 0.1461 0.4782*** 0.0133 Na 0.0000 0.0175 0.0001 Na -0.0076 0.0051 Pr(|F|>c) Test stat. (d.f.) Test stat. (d.f.) Pr(|F|>c) 0.010 Na 0.000 Na b0 b0 b1 Std. Error Std. Error Regressor b1 80% quantile Coefficient1) Coefficient a1 a1 Dumm 0.0000 0.0003 0.0000 y F test results D2) Test stat. (d.f.) Null hypothesis b0 = b0 n n 0.0000 0.0000 Na 0.5000 0.2480 Na 0.0000 0.0000 0.0000 Na 0.0909 0.8750 04782 Na 0.0000 a i 0 n = i 1 n b b a a i n0 i n0 i n1 i i i i i 7.775 (1,693) 35.934 (1,693) 0.005 0.000 6.600 (1,693) 1295 (1,693) Pr(| F|>c ) a i Test stat. (d.f.) 233.5 (1,693) 0.195 (1,693) 1.666 b i 0 n Pr(|F|>c) 3) bi = i 1 0.8750*** Coefficient1) 60% quantilee 0.0216 0.000 0.658 i i 1 levels of statistical significance are denoted with *for 1%, ** for the 5% and *** for 1) The the 1%. 2) Dummy variable is set to 1 when terminal price is lower than shipping point price 3)H0: b0 = b0 is not tested when one of the estimated coefficients is not statistically significant 24 Note: Estimation at 60% quantile was not possible because of data limitation. 25 Quantile Estimation results: Table grapes 40% quantile 20% quantile Coefficient Coefficient1) Regressor -0.0141*** Coefficient1) Std. Error Coefficient1) Std. Error Coefficient1) Std. Error 0.0041 -0.0008 0.0036 0.0038 0.0039 0.0183*** 0.0048 0.2322 0.8187*** 0.1213 1.0969*** 0.0954 0.1698 0.3291* 0.1723 0.4048** 0.2027 0.1103 0.2354 0.2080 0.9487*** 0.1488 0.6977*** b1 b1 b2 b2 b3 b3 b4 b4 a1 a1 a2 a2 a3 a3 a4 a4 D Pt D Pt S D Pt S1 DPt S1 D Pt S2 D Pt S2 D Pt S3 D Pt S3 D Pt S4 D Pt S4 D PtT1 DPtT1 D PtT2 DPtT2 D PtT3 DPtT3 D PtT4 DPtT4 D2) Dummy b0 b0 80% quantile Std. Error 0.4881** S 60% quantile 0.1284 0.1088** 0.0474 0.1028 0.0995 0.2269** 0.5527*** 0.1567 0.3516*** 0.0805 0.2715 0.1761 0.2081 0.1329 0.1428 0.0895 0.0654 0.0532 0.1722* 0.0995 0.3806*** 0.0939 0.1672 0.1360 0.1551 -0.0893 0.1577 -0.1655 0.1783 0.1493 0.1715 0.1240* 0.0674 0.1338* 0.0789 0.2456** 0.0949 0.1171 0.1566 -0.0031 0.0713 0.0429 0.0740 0.1554 0.1108 0.0240 0.0767 0.0056 0.0547 0.0601 0.0587 0.0718 0.0494 -0.0581 0.0716 0.0221 0.0577 0.0414 0.0646 0.0484 0.0910 -0.0553 0.0388 -0.0942** 0.0373 0.0177 0.1840 0.1040 0.0787 -0.1322* 0.0793 -0.0556 0.0435 -0.0351 0.0435 -0.1275 0.1021 0.0565 -0.2068 0.0418 -0.0477 0.3621 ** -0.1382 0.0928 -0.0420 0.0380 -0.0966* -0.0932** 0.0412 -0.0360 0.0335 -0.0016 0.0402 -0.0275 0.0841 -0.1595 0.1632 -0.1684** 0.0655 -0.1844*** 0.0609 -0.1155** 0.0509 0.0346 0.0415* 0.0251 0.0176 0.0500 0.0071 0.0437 -0.0125 *** -0.0152 0.0625 0.0695 0.0679 0.1116** 0.0466 0.0261 0.0956 -0.0012 0.0478 -0.0060 0.0382 -0.0063 0.0229 -0.0113 0.0547 -0.0367 0.0381 -0.0255** 0.0106 -0.0407*** 0.0146 -0.0563*** 0.0134 Test stat. (d.f.) Pr(|F|>c) Test stat. (d.f.) Pr(|F|>c) Test stat. (d.f.) Pr(|F|>c) F test results Test stat. (d.f.) Null hypothesis b0 = b0 n b i 0 i 3) n b = i 0 n a i 1 i 1 n b b a a i n0 i n 0 i n1 i 1 8.909 (1,348) 0.003 0.2031 (1,348) 0.652 i n ai = Pr(|F|>c) i 0.460 (1,348) 1.236 (1,348) 1.244 (1,348) 0.497 0.267 0.265 4.750 (1,348) 4.437 (1,348) 0.344 (1,348) 0.030 0.036 0.557 8.986 (1,348) 21.515 (1,348) 0.019 (1,348) 0.5039 0.7918 1.2876 2.0218 1.9225 1.2042 0.5957 0.6512 -0.3681 -0.2350 -0.1517 -0.1922 -0.2390 -0.0561 -0.0254 -0.1592 i i i i 0.003 0.000 0.890 1) The levels of statistical significance are denoted with *for 1%, ** for the 5% and *** for the 1%. 2) Dummy variable is set to 1 when terminal price is lower than shipping point price 26 3)H0: b0 = b0 significant is not tested when one of the estimated coefficients is not statistically 27 Quantile Estimation results: Peaches Coefficient Regressor -0.0093** D Pt S 0.1360 D Pt S 0.2333 ** D Pt S1 0.4703 *** DPt S1 0.8956 D Pt S2 0.2039 D Pt S2 0.0381 D Pt S3 0.2582 * D Pt S3 0.2883 ** D Pt S4 0.2788 ** D Pt S4 0.2499 D PtT1 -0.2877 DPtT1 0.0571 ** D PtT2 -0.4044 DPtT2 -0.1046 D PtT3 -0.0511 DPtT3 -0.1888 D PtT4 -0.0584 DPtT4 -0.0979 b0 b0 b1 b1 b2 b2 b3 b3 b4 b4 a1 a1 a2 a2 a3 a3 a4 a4 Dum my D2) Null hypothesis b0 = b0 n b a in 0 i n0 ai = i 1 n bi i n0 b a a i n 0 i -0.0221 Std. Error 0.0046 40% quantil Coefficien t1) e -0.0008 Std. Error 60% quantil Coefficien t1) e Std. Error 80% quantil Coefficien t1) e Std. Error 0.0034 0.0025 0.0035 0.0201** 0.0082 0.0846 0.1214 0.1611 0.1408 0.0914 0.0897 0.1570* 0.2564 0.3402** 0.1501 0.1791 0.1406 0.2617 0.1652 0.2046 0.5144*** 0.0880 0.4995*** 0.1102 0.7591*** 0.1483 0.2203 0.6243*** 0.1636 0.4765*** 0.1789 0.4048* 0.2437 0.2111 0.0595 0.1160 0.2744 0.2157 0.1231 0.3128 0.2267 0.1323 0.1403 -0.0394 0.1091 0.1279 0.1316 0.1695 0.1389 0.1328 0.0420 0.1170 0.0002 0.3101 0.1695 0.1188 0.0813 0.0934 0.1026 0.1385 0.1510 0.1127 0.1751 0.1271 0.1114 0.1126 0.1418 0.1886 0.1211 -0.0056 0.0741 -0.0028 0.0642 0.1610 0.1748 0.3924 0.0045 0.0597 -0.0270 0.0577 0.2944 0.6304 0.2033 0.0657 0.1086 0.0985 0.1000 -0.0974 0.1627 0.1851 -0.3004 0.1820 -0.1920*** 0.0705 -0.0203 0.3771 0.1406 -0.0249 0.0726 -0.0231 0.0862 -0.0918 0.0984 0.0751 0.0259 0.1398 0.1137 0.1134 -0.1048 0.1922 0.1204 -0.0272 0.0674 -0.0431 0.0675 -0.0995 0.1193 0.1073 -0.0481 0.0994 -0.0038 0.0782 0.0877 0.2026 0.0407 -0.1809 0.1131 0.0838 -0.0860 0.0524 -0.0692* 0.0143 -0.0053 0.0108 -0.0130 0.0124 -0.0230 0.0174 Test stat. (d.f.) Pr(|F|>c) Test stat. (d.f.) Pr(|F|>c) Test stat. (d.f.) Pr(|F|>c) F test results Test stat. (d.f.) Pr(|F|>c) 3) n bi = i 1 20% quantil Coeffici eent1) i i 0.582 0.446 (1,240) 0.9618 0.327 (1,240) 1.3471 0.901 0.343 (1,240) 0.671 0.413 (1,240) 0.9794 0.980 0.323 (1,240) 0.0922 0.7616 (1,240) 1.0842 0.006 0.936 (1,240) 1.482 0.225 (1,240) 1.1456 1.7051 1.2100 0.7067 1.0939 0.2570 0.8016 0.3180 0.1091 i n1 i 0.3341 0.0725 0.0370 0.4696 1 1) Thei levels of statistical significance are denoted with *for 1%, ** for the 5% and *** for the 1%. 2) Dummy variable is set to 1 when terminal price is lower than shipping point price 3)H0: b0 = b0 is not tested when one of the estimated coefficients is not statistically significant i 28