Australian Agribusiness Review - Vol. 8 - 2000
When using Bookmarks, use your browser BACK BUTTON to return to the main text.
Do Canadian Pork Imports Influence New South Wales Pigmeat Prices?
October 11th, 2000
G.R. Griffith and H-S (Christie) Chang
Scientist, NSW Agriculture Beef Centre, Armidale, NSW
This paper is a revised version of a section of a paper for the Productivity Commission Inquiry in Nov. 1998.
In September 1989 the Australian Government announced an in-principle decision to lift the existing ban on importation of unprocessed pork, specifically for Canadian product. The decision was confirmed in July 1990, the formal protocols were signed soon after and imports from Canada began arriving in August 1990. In the first year, import levels were generally minor. However, from July 1991 there was a sustained increase in volumes. Total imports for 1991/92 were over 4000 tonnes compared with about 1000 tonnes for the preceding year (APC 1998a). This increase in imports coincided with a dramatic fall in farm prices for pigs in early 1992 (APC 1998a).
Since the late 1980s, the pig industry had been a vocal critic of the decision to allow in Canadian imports on the grounds of possible disease risk and that Canadian producers were heavily subsidised. The large fall in farm prices in early 1992 heightened this concern and the Australian Pork Corporation was instructed to prepare and submit a case for the imposition of countervailing duties. On receipt of the submission, the Australian Customs Service initiated a dumping and subsidy inquiry which was reported in November 1992. The finding was that frozen pork imports from Canada had not caused and did not threaten to cause, material injury to the Australian pig industry (ACS 1992). On appeal, this negative finding was upheld by the Anti-Dumping Authority (ADA 1993), and also by the Federal Court following a subsequent appeal. Another period of high import levels and low prices in early 1995 and a rally by pig producers in Canberra led to an Industry Commission inquiry into pigmeat imports (IC 1995), while a similar set of circumstances in late 1997 and early 1998 led to another inquiry by the Productivity Commission to determine whether safeguard actions against imports were justified (PC 1998).
While there had been some coincidental opposing movements in import volumes and domestic pig prices (see Figure 1), and theory would predict that there may be a causal relationship between these two variables, there had been few formal attempts (ABARE 1995; Griffith 1995) to evaluate whether the observed changes in pig prices were caused by imports or by other factors in the pig market or the broader meat market. Neither of these studies found evidence of imports depressing domestic prices and the Industry Commission concluded that imports had little effect on pig prices (IC 1995, p.18). However, evidence presented to, and accepted by, the Productivity Commission (PC 1998, p.78) indicated that "…imports have played a far greater role in setting domestic pigmeat prices since 1996…".
The objective of this study was to test whether or not there exists a statistically significant causal relationship between the level of imports of Canadian pork into Australia and the level of prices in the New South Wales pigmeat market, using the most recently available data. The New South Wales market was chosen as most imports come in through the port of Sydney.
2 - Method of Analysis 1
2.1 The concept of causality
To test the causal relationship between Canadian pork imports and New South Wales pigmeat prices the notion of causality developed by Granger (1969) is used. Causality is formally defined as follows: "variable X is said to cause variable Y if current values of Y can better be predicted by using past values of X, than by not doing so, given that all other relevant information including past values of Y is used in both cases" (Granger 1969, p 426). Four options are possible (i) if X causes Y and Y causes X, the variables have a simultaneous causal relationship; (ii) if neither X causes Y nor Y causes X, they are independent; (iii) if X causes Y, but the converse is not true, then causality can be inferred where X is the leader and Y is the follower; and (iv) the opposite conclusion would hold for Y causing X. So if variable X was defined as pigmeat imports and variable Y was defined as the domestic price, statistical evidence supporting the third option would lead to the conclusion that imports influenced domestic prices.
There are a number of causality tests available based on time series methods (eg Pierce 1977) as well as various forms of regression models. However, considerable controversy surrounds the causality literature and the various methodological critiques associated with different aspects of causality tests (see Zellner 1979).
2.2 Causality in VAR models
The way that causality tests have often been implemented (the so-called "Granger" and "Sims" models) implies pair wise causality. That is, while other relevant market information is allowed to help explain variation in the dependent variable, this is not usually done in a formal manner and interest is primarily in the causal effect of one variable on another with little regard for the inter-relationships between other variables. Thus, these are partial tests.
Vector Autoregressive (VAR) models have been increasingly used in recent years to overcome this problem. A VAR model is "the unconstrained reduced form of a dynamic simultaneous equations model of a market or industry" (Sims 1980). This means that a VAR model comprises a set of equations explaining variations in the endogenous variables (the variables jointly determined in this market, like farm price, retail price, production, etc) as linear functions of their own lagged values and the lagged values of the other endogenous variables. Current and lagged exogenous variables (the variables determined completely outside this market, like consumer income, wage rates, etc) may also be included in the equations. The following is an example of one equation from a VAR model, explaining variation in the particular endogenous variable Y1;
(1) Y1t = Cy1 + SaiY1t-i + SbiY2t-i + SciY3t-i +…+ SdiX1t-i + SeiX2t-i +…+ u1t
Note - If you are viewing this using a Netscape browser, several of the symbols used in the equation above may appear incorrectly. Please refer to the version below for clarification.
It is recommended that you use IE 5.0 or better to view this site.
where Y2, Y3, etc = other endogenous or jointly determined variables in the underlying structural market model, t = current time period, i = the chosen lag structure, Y jt-i = variable Yj lagged i periods, Cy1 = constant term for this equation, X1, X2, etc = exogenous variables reflecting "other relevant information" and u1t = the random error for this equation. a, b, c, d, e and the constant term Cy1 are parameters to be estimated. Seasonal dummy variables also may be added if monthly or quarterly data are used. The "causal" variable of interest may be either one of the endogenous Y variables or one of the exogenous X variables. So if variable X1 was defined as pigmeat imports and variable Y1 was defined as the domestic farm price for pigs, this equation would say that the current value of the farm price was determined by past values of the farm price; past values of the wholesale price, the retail price and pigmeat production; past values of other influencing variables like retail prices of other meats or farm prices of inputs; and seasonal dummy variables. Similar equations would be specified to explain Y2, Y3, Y4 (wholesale prices, retail prices, pigmeat production). This set of equations is the VAR model.
The VAR model is estimated with and without the "causal" variable (in this case X1), and standard F tests or c2 tests are used to assess whether there is a statistically significant causal relationship in this market by comparing the error sums of squares from the with and without versions of (1) (Griffiths et al. 1993; Enders 1995). Because of the way the interactions between all relevant variables are included in this model, it is a more complete test of causality than the pair wise tests given by the simpler Granger and Sim’s models.
2.3 Specification choices
In using the procedures outlined above, and with an understanding of the data potentially available, a number of technical issues in implementation arise. In particular, it is necessary to choose (a) the appropriate lag length, (b) an appropriate pre-filter if required, and (c) the best way of accounting for all the other relevant information in the regression models.
A) Lag lengths
A prominent issue regarding tests of causality is the determination of the appropriate finite lag length that ensures well-behaved equation error terms required to satisfy the underlying assumptions of the concept of Granger causality. Causal inferences have been shown to be sensitive to the choice of lag length (Thornton and Batten 1985; Nakhaeizedah 1987), so with little prior information, various lag lengths are reported here, from a one-period lag through to a six-period lag. The six month lag structure would seem to be near the upper bound of possible causal influences.
It is apparent that long lag structures in models with many endogenous variables quickly expand the number of coefficients to be estimated, particularly if monthly data are used and 11 monthly dummy variables also have to be added to each equation. Therefore, there are often trade-offs between available sample periods and feasible lag lengths.
Granger’s definition assumes all variables are "stationary"; that is, the mean and covariance are not functions of time. However, most time series are non-stationary. Monthly commodity prices, in particular, have trend components because of inflation, storage costs, and other carrying charges (Grant et al. 1983). In order to make the Granger concept of causality operational, Sims and various other authors introduced pre-filters to transform the variables to stationarity before the series is analysed.
Differencing is often an appropriate filtering method for an analysis of agricultural prices, as differencing removes most of the autocorrelation and non-stationary influences. In the larger study (Griffith 1998) on which this paper is based, a formal investigation of the stationarity properties of the various series was conducted. For most series, the raw data implied a unit root and non-stationarity in the raw data, which was confirmed by the fact that all series when first differenced were stationary. Most of the causality tests reported in that study were based on first-differenced data.
As an alternative, a new procedure for undertaking causality testing in a VAR framework developed by Rambaldi and Doran (RD)(1996) is used here. This procedure does not require stationarity transformations of the data (that is, the use of first-differenced data) as long as an additional lag length is allowed for series which are first-difference stationary, and that the VAR system is estimated in a systems framework. Thus, for lag lengths of six, variables lagged up to seven need to be included in the equations. Some degrees of freedom are saved and the problem of interpreting the first-differenced results is avoided by using the RD procedure.
C - All Other Relevant Information
One of the assumptions underlying the Granger causality test is that "all other relevant information, including past values, is used". When the potential causal relationship includes a market-determined price variable, then it would be expected that many other factors in the market would have a potential influence on this price, so other variables need to be explicitly included in the regression models to account for such effects. In this study, other relevant variables were sought to help explain movements in domestic pig prices. The variables chosen were pigmeat production levels in NSW as an endogenous supply effect, and the retail prices of beef and lamb as exogenous demand effects.
The following data series were collected and used in the model.
IMPGCN: imports of pigmeat into Australia, fresh, chilled and frozen, from Canada, tonnes, July 1990 to April 1998, provided by APC (1998a,b).
PAPGNW: dressed carcase price of GH2, 60-70kg pigs, NSW markets, cents/kg, January 1988 to August 1998, collected by NSW Meat Industry Authority, provided by AMLC (1996) and MIA staff.
PWPGNW: dressed carcase price for baconer class pigs, delivered to Sydney retail shops, cents/kg, January 1988 to January 1998, collected by ACNeilson, provided by MATFA (1998) and AMLC staff.
PRLENW: retail price of pork legs, Sydney retail outlets, cents/kg, January 1988 to January 1998, collected by ACNeilson, provided by MATFA (1998) and AMLC staff.
PRRUNW: retail price of beef rump steak, Sydney retail outlets, cents/kg, January 1988 to January 1998, collected by ACNeilson, provided by MATFA (1998) and AMLC staff.
PRLLNW: retail price of lamb leg, Sydney retail outlets, cents/kg, January 1988 to January 1998, collected by ACNeilson, provided by MATFA (1998) and AMLC staff.
PDPGNW: production of pigmeat in NSW, tonnes carcase weight, January 1988 to June 1998, collected by ABS, provided by AMLC (1996) and APC (1998b).
Unfortunately, the ACNeilson wholesale and retail price data have not been published by MATFA since January 1998, restricting the data set to 89 observations.
Given the specification choices and the available data, the following is the specific estimating equation for PAPGNW in the model when lags are set to 6 periods:
(2) PAPGNWt = CPA
+ a1PAPGNWt-1 + a2PAPGNWt-2 +…+ a7PAPGNWt-7
+ b1PWPGNWt-1 + b2PWPGNWt-2 +…+ b7PWPGNWt-7
+ c1PRLENWt-1 + c2PRLENWt-2 +…+ c7PRLENWt-7
+ d1PDPGNWt-1 + d2PDPGNWt-2 +…+ d7PDPGNWt-7
+ e0IMPGCNt + e1IMPGCNt-1 + e2IMPGCNt-2 +…+ e7IMPGCNt-7
+ f0PRRUNWt + f1PRRUNWt-1 + f2PRRUNWt-2 +…+ f7PRRUNWt-7
+ g0PRLLNWt + g1PRLLNWt-1 + g2PRLLNWt-2 +…g7PRLLNWt-7
+ h1D1 +h2D2 + h3D3 +…+h11D11
where D1, D2, etc. are monthly dummy variables and the other variables are as previously defined. In the full VAR model, there are similar equations for PWPGNW, PRLENW and PDPGNW. All variables are in levels rather than differences and note that current period values of the three exogenous variables are included in the equation. In this 6-period lag model, a total of 64 coefficients have to be estimated. Even in the 1-period lag model, 29 coefficients have to be estimated. The test procedure involves estimating the full model with and without the IMPGCN variables and testing whether their omission significantly impacts on the models explanatory power. If it does, imports "cause" prices.
One other aspect of the estimation procedure that requires some discussion is the choice of sample period. It is evident from Figure 1 and other APC data (APC 1998a,b), and from the commentary in the various Inquiries (IC 1995; PC 1998), that while the volume of Canadian imports has fluctuated substantially from month to month, there has been a sustained increase in the last few years. Further, there is anecdotal evidence that during this period, impacts on prices have developed. Running the regression models over the full sample period may well mask some of the impacts on prices if these happen to be occurring mainly in the latter part of that sample. To examine such a possibility, the sample is broken down into two shorter sub samples and the models are run on these as well as on the full sample. These are the last five years of the data (60 observations) and the last three years of the data (36 observations). Given the discussion above about lag lengths however, degrees of freedom considerations restrict the length of some of the lags able to be tested in the shorter samples.
The four-equation VAR model of the form set out in (2) was estimated by a Seemingly Unrelated Regression estimator using data for the period September 1990 to January 1998 as described above. The TSP econometrics package was used (Hall et al. 1997). Different versions of the model were run to test the different lag lengths and the different sample periods.
Generally, each equation estimated had good summary statistics with an R2 around 0.95 for the price variables and around 0.85 for production, no evidence of autocorrelation and a good ratio of significant coefficients, especially on the shorter lag variables.
The results of implementing the causality tests described above are reported in Table 1, with a c2 statistic being computed. The 5% critical value is 7.81 for the four restrictions imposed. Over the full sample period, none of the test statistics were significant. Thus, over the whole period since Canadian imports have been allowed into Australia, they have had no significant influence on NSW farm, wholesale or retail pigmeat prices or NSW pigmeat production. There was a tendency for some influence to be occurring at the longer lag lengths, but not at statistically significant levels, and the tests are less reliable when degrees of freedom are low.
Table 1: c2 Tests of the Impact of Imports Using the Rambaldi and Doran VAR Model, Different Sample Periods and Different Lags
(a) Sample Period 1990:9 - 1998:1 (max n=89)
(b) Sample Period 1993:1 - 1998:1 (max n=60)
(c) Sample Period 1995:1 - 1998:1 (max n=36)
* Significant at the 5% level.
However, over the more recent shorter sample periods, imports were found to significantly influence variations in NSW prices and production. Over the most recent five year sample (1993:1-1998:1), this occurred at lag lengths of 2, 3 and 4, and over the most recent three year sample (1995:1-1998:1) this occurred at lag length 1 (the only lag length where an estimate could be obtained).
Therefore the story seems to be that when imports were first allowed, volumes were small and rather irregular (see Figure 1), and there was little consistent impact on the domestic market. However, as the trade became established, volumes grew and became more regular. As imports became regarded as an established source of supply, the domestic market had to adjust to the presence of imports and prices fluctuated as import quantities varied. These adjustments often took several months.
As imports continue to increase, it is likely that the domestic market will continue to be effected. In addition, imports are now allowed from other suppliers and further approvals are being sought. However, the safety net of greater pigmeat exports has recently come into play and it is hoped that this will likely take some of the adjustment pressure off the domestic market. Unfortunately, these data are not yet available in sufficient quantities to allow that to be shown in statistical analyses like that described above.
So, in answer to the question posed in the title to this paper, yes, Canadian pork imports do influence NSW pigmeat prices. This evidence was the basis of the Productivity Commission’s finding that safeguard measures can be justified under the WTO criteria (PC 1998, p. xxxi).
Anti-Dumping Authority (1993), Review of the Australian Customs Service Negative Preliminary Finding on Frozen Pork from Canada, Report No. 90, AGPS, Canberra.
Australian Bureau of Agricultural and Resource Economics (1995), The Impact of Imports on the Australian Pig Industry, Submission to the Industry Commission Inquiry, ABARE, Canberra.
Australian Customs Service (1992), Report and Preliminary Finding No. 92/10, Frozen Pork from Canada, AGPS, Canberra, 27 November.
Australian Meat and Livestock Corporation (1996), Australian Meat and Livestock Industry Statistics, Market Intelligence, AMLC, Sydney.
Australian Pork Corporation (1998a), PigStats97 - Australian Pig Industry Handbook, APC, Sydney, and previous issues.
Australian Pork Corporation (1998b), Imports and Exports Faxlist, APC, Sydney, monthly.
Enders, W. (1995), Applied Econometric Time Series, Wiley, New York, Chapter 5.
Feige, E.L. and Pearce, D.K. (1979), "The casual causal relationship between money and income: some caveats for time series analysis", Review of Economics and Statistics 61, 521-33.
Granger, C.W.J. (1969), "Investigating causal relations by econometric methods and cross spectral methods", Econometrica 37, 424-438.
Grant, W.R., Watua Ngenge, A., Brorsen, B.W. and Chavas, J.P. (1983), "Grain price interrelationships", Agricultural Economics Research 35 (1), 1-9.
Griffith, G.R. (1995), A Causality Analysis of the Impact of Canadian Pork Imports on NSW Pig Prices, NSW Agriculture Submission to the Industry Commission Inquiry, NSW Agriculture, Armidale.
Griffith, G.R. (1998), "The Impact of Canadian Pork Imports on New South Wales Pigmeat Prices", in Pigmeat and Pigmeat Industries: Safeguard Action Against Imports, Modelling Reports, NSW Agriculture, Armidale.
Griffiths, W.E., Hill, R.C. and Judge, G.G. (1993), Learning and Practicing Econometrics, Wiley, New York, Chapter 21.
Hall, Bronwyn H. et al. (1997), Time Series Processor Version 4.4 User's Guide, TSP International, Palo Alto, California.
Industry Commission (1995), Pigs and Pigmeat, Research Report, AGPS, Canberra.
Meat and Allied Trades Federation of Australia (1998), "People, places, prices", The Australian Meat Industry Bulletin (various issues).
Nakhaeizadeh, G. (1987), "The causality direction in consumption- income process and sensitivity to lag structures", Applied Economics 19, 829-838.
Pierce, D.A. (1977), "Relationships - and lack thereof - between economic time series, with special reference to money and interest rates", Journal of the American Statistical Association 72, 11-22.
Productivity Commission (1998), Pigmeat and Pigmeat Industries: Safeguard Action Against Imports, Report No. 3, AusInfo, Canberra.
Rambaldi, A. and Doran, H. (1996), Testing for Granger Non-causality in Co integrated Systems Made Easy, Working Papers in Econometrics and Applied Statistics No. 88, University of New England, Armidale.
Sims, C. (1980), "Macroeconomics and reality", Econometrica 48, 1-48.
Thornton, D.L. and Batten, D.S. (1985), "Lag-length selection and Granger causality between money and income", Journal of Money Credit and Banking 17, 164-178.
1 - This section is rather technical and most readers can skip directly to the results.