Abare Conference Paper 98.3

ABARE

ABARE CONFERENCE PAPER 98.3

Australia’s coking coal exports to Japan Price–quality relationships under benchmark and fair treatment pricing Anthony Swan, Sally Thorpe and Lindsay Hogan* Australian Bureau of Agricultural and Resource Economics 42nd Annual Conference of the Australian Agricultural and Resource Economics Society University of New England, Armidale, 19–21 January 1998

Given Japan’s dominant position in the regional coal market and the continuing relatively low profitability of Australia’s coal industry, the influence of the Japanese steel mills on coal pricing arrangements between Australia and Japan remains a policy issue in Australia. The Japanese steel mills introduced the ‘fair treatment’ pricing system in Japanese fiscal year (JFY) 1996, whereby coal contract information would be confidential but it was argued that coal would be priced according to its value in use. In this paper, Quandt’s switching regime model is used to test for structural change in price–quality relationships in the important Australia–Japan coking coal trade. There is statistical evidence that price–quality relationships changed fundamentally after JFY 1994 for semisoft coking coals when the soft coking coal category was merged with the semisoft coking coal category, and in JFY 1996 for hard coking coals when the fair treatment pricing system was introduced. It is concluded that coking coal price–quality relationships have become substantially less transparent in recent years.

* The authors wish to thank Andrew Dickson of ABARE for useful comments.

ABARE project 1387

1

ABARE

ABARE CONFERENCE PAPER 98.3

Introduction Australia is the world’s largest coal exporter, accounting for around 29 per cent of world coal exports in 1996, and Japan is the world’s largest coal importer, accounting for around 26 per cent of world coal exports (IEA 1997). In 1996-97, Australia’s coal exports were valued at A$7.9 billion or 9 per cent of Australia’s total merchandise exports. Japan remains Australia’s largest coal export market, accounting for 41 per cent and 53 per cent of Australia’s metallurgical and thermal coal exports, respectively, in 1996-97. Given the importance of coal in Australia’s merchandise exports, Japan’s dominant market position and the continuing relatively low profitability of the Australian coal industry, the efficiency of the Asia Pacific regional coal market remains a policy issue in Australia (Hogan, Thorpe, Graham and Middleton 1997). In recent years, there have been two major inquiries into the economic performance of Australia’s black coal industry. In the final report of the 1994 inquiry, Taylor (1994) noted that it is difficult to assess whether the collective buying practices of the Japanese steel mills and power utilities have resulted in lower coal prices to this market. The recommendations in the Taylor report were directed at increasing market transparency, increasing productivity and international competitiveness, and implementing an export diversification strategy (Taylor 1994). The recent inquiry by the Industry Commission, which is to report in early 1998, has emphasised an international benchmarking approach to assess productivity and international competitiveness issues more fully. The focus in this paper is on the recommendations of the Taylor report relating to the need to increase transparency in coal price determination, particularly with respect to the influence of the Japanese steel mills on coal pricing arrangements between Australia and Japan. From the mid-1980s, coking coal prices paid by Japanese steel mills, and broadly followed by Asian steel mills more generally, were based on the Japanese benchmark pricing system. Under this system, coking coals were grouped and priced by coal category, where coal categories included hard, soft and semisoft coking coal. The absolute price of a coal brand within a given coal category was negotiated relative to the benchmark coal with known quality characteristics. After Japanese fiscal year (JFY) 1994, the soft coking coal category was merged into the traditionally lower priced semisoft coking coal category. In JFY1996, Japan replaced the benchmark pricing system for coking coals with the ‘fair treatment’ pricing system whereby, it was argued, each coal brand would be valued according to the quality requirements of specific Japanese steel mills. Under this 2

ABARE

ABARE CONFERENCE PAPER 98.3 arrangement, however, coal prices and other contract details would remain confidential. The fair treatment system was introduced after substantial coking coal price increases in the JFY1995 negotiations, following several years in which coking coal prices were consistently reduced in real terms. Issues of market inefficiency arise under both the benchmark and ‘fair treatment’ systems because, first, coking coal categories are not uniquely defined in terms of coal quality attributes but prices are nonoverlapping between categories and, second, the international price negotiations are conducted sequentially by coal category and by coal importing nation which effectively reduces the number of buyers in the market at any point in time (Hogan, Thorpe, Graham and Middleton 1997). Confidentiality of price–quality–quantity information under the fair treatment system is an issue since coal price discovery during the annual negotiations, particularly for coal exporters, is made difficult. More generally, the information content of price signals is critical for efficient resource allocation in international coal trade. The objective in this paper is to test for structural change in the price–quality relationships implicit in the Australia–Japan coking coal trade between JFY1992 and JFY1997 using Quandt’s exogenous switching regime regression technique, described in Goldfeld and Quandt (1976) and Johnston (1991). Price–quality data are obtained from the Australian Coal Report (various issues), including the data for JFY 1996 and 1997 which represent estimates of actual settlements. Recent hedonic regression analyses of hard coking coal price–quality relationships include Chang (1995), Koerner (1996) and Hogan, Thorpe and Middleton (1997), although the last paper also includes other coal categories. The consistency of coal price–quality relationships by two or more major suppliers into Japan are examined in the first two papers, and coal price–quality relationships in Australia’s exports to Japan and other markets are examined in the third paper. Chang (1995) pools data for JFY 1993 and JFY 1994, Koerner (1996) undertakes separate regression analyses for JFY 1992 and JFY 1994, while Hogan, Thorpe and Middleton (1997) pool data for the period JFY 1989 to JFY 1996. None of these studies explicitly test for structural changes in coal price–quality relationships over time. The next section contains a brief description of coal pricing arrangements and the quality attributes of coking coals. Recent previous studies are described in the third section,

3

ABARE

ABARE CONFERENCE PAPER 98.3 followed by a description of the research method and data. Results are presented in the following section and concluding comments are given in the final section.

Coal quality and pricing arrangements Coal quality Coal is not a homogeneous product, and is classified into broad groups — hard, soft and semisoft coking and thermal — according to the end use and the manner in which the coal performs in that end use. The main quality characteristics of coking coal are total moisture (TM), inherent moisture (IM), ash (ASH), volatile matter (VM), sulphur (SULP), crucible swelling number (CSN), vitrinite reflectance (REFL), fluidity (usually specified as the natural logarithm of fluidity, LFLUID) and coke strength after reaction (CSR). Coal quality characteristics are described in appendix 1. Coking coal is primarily used to produce the coke required in traditional iron and steel making. While hard coking coal can replace other coal types in most end uses, the reverse is not true. Hard coking coal has few contaminants and is a necessary input in the production of strong coke. Typically, an Australian hard coking coal has a high crucible swelling number, moderate volatile matter, and low inherent moisture, ash and sulphur levels. Hard coking coal typically receives a price premium over other coals. The use of coal in steelmaking is discussed further in Scott (1994). Subject to technical limits, soft (when classified separately) and semisoft coking coals are used in the blending mix for coke making to minimise costs, although there is some weakening in the strength of the coke as a consequence. Soft coking coals have relatively low ash levels and historically received a price premium over semisoft coking coals. PCI (pulverised coal injection) coal is coal that is pulverised and injected into the base of the blast furnace. PCI reduces the volume of coke required in the traditional steelmaking process, although high quality coke must be produced (Thomson, Zulli, McCarthy and Horrocks 1996). Japan is a leader in the development and adoption of iron and steelmaking technologies that use lower quality coal types. Semisoft coking coal is distinguished from thermal coal in having lower contaminants such as ash. Some contaminants can be reduced by washing. Some semisoft coking coals are also used as PCI coals, although the quality requirements of PCI coals are not well known. Thermal coal is used mainly for combustion purposes in electricity generation, 4

ABARE

ABARE CONFERENCE PAPER 98.3 cement manufacture and general industry. Historically, consistent with relatively strong substitution possibilities, the prices for thermal coal used in electricity generation and for semisoft coking coal have tended to move together. Thermal coal used by general industries and for cement manufacture has the lowest price. The impact of coal quality on end use performance in power station applications is discussed in Skorupska (1993) and Carpenter (1995).

Coal pricing arrangements In the Asia Pacific regional market, coal price settlements between Australia and Japan are most commonly the first to be negotiated each Japanese fiscal year and have tended to signal overall market conditions in regional coal trade. Annual coking coal trade negotiations over price and tonnage are usually settled sequentially with suppliers from each exporting country and representatives of the joint purchasing cartel of Japanese steel mills. Thermal coal settlements between the Japanese power utilities and exporting countries are usually announced following the coking coal agreements. Pricing arrangements for coking coal trade are more complex than for thermal coal used in electricity generation. Historically, the benchmark prices for the hard, semisoft and soft coking coal categories were used in the price negotiations for other coals in the corresponding coal category. In principle, coal prices are determined broadly according to differences in their relative contribution to the iron and steelmaking process. In practice, under the benchmarking system, a ladder of relative coal prices was established that was relatively rigid within each coal type and between coal types. The Japanese steel mills abandoned the benchmark pricing system following the JFY 1995 negotiations when hard and semisoft coking coal prices increased by around US$5.65 and US$6.27 a tonne respectively (Australian Coal Report 1995). They argued that the changing quality specification requirements of the individual steel mills were not reflected in the relative price structure of coking coal. The emerging adoption of pulverised coal injection technology is one example where a new price formation process would need to reflect the change in value in use of particular quality specifications. The rationale that the Japanese steel mills provided for the introduction of the fair treatment system in JFY 1996 was that prices could be negotiated to better reflect the value in use for specific quality specifications and hence improve the efficiency of price formation. However, given that coal prices and other contract details remain confidential both during and after the price

5

ABARE

ABARE CONFERENCE PAPER 98.3 negotiations, it is unlikely that the efficiency of the market has been enhanced with the introduction of the fair treatment system. The electricity sector is the major end use market for thermal coal. All prices paid by the Japanese power utilities, expressed fob in US dollars, are the same in terms of energy content with a coal benchmark of 6700 kcal/kg on a gross air dried basis. In the future, efforts to deregulate the electricity sector in Japan may result in the emergence of price premiums and discounts for a range of end use attributes. In Hogan, Thorpe and Middleton (1997), various quality related price premiums and discounts are estimated for thermal coal used in general industry and cement manufacture. However, since published price–quality data are not available, these categories are excluded from the current study.

Recent empirical studies Hedonic regression analysis is used widely to examine price–quality relationships for a range of commodities. Berndt (1991) explains in detail several of the pioneering applications including those of Waugh, Griliches and Chow, and further elaborates on several key model misspecification issues. Rosen (1974) was one of the first to derive an underlying hedonic pricing theory in conducting his econometric analysis. In this section, only the most recent empirical analyses of coking coal price determination for Japan are discussed since these are most relevant to the current study. These studies include Chang (1995), Koerner (1996) and Hogan, Thorpe and Middleton (1997). Earlier studies, discussed in some detail in Chang (1995), include Porter and Gooday (1990), Koerner (1993) and Low, Dwyer, Jolly and Olejniczak (1993). Koerner (1996) updates the results from the earlier econometric research contained in Koerner (1993). Each study uses hedonic regression analysis to estimate price–quality relationships relevant to the Asia Pacific coal trade and, either directly or indirectly, is concerned with potential inefficiencies in regional coal market pricing. Both Chang and Koerner examine the hypothesis that Japan pays different prices for essentially the same hard coking coal from different regions of supply — Australia and Canada in Chang, and Australia, the United States and Canada in Koerner. In constructing quality adjusted prices of all major coal categories, Hogan, Thorpe and Middleton (1997) examine the hypothesis that Australia receives different prices for essentially the same product into different regions of demand.

6

ABARE

ABARE CONFERENCE PAPER 98.3 In all three cases there is some econometric evidence of inefficient pricing once coal prices are adjusted for differences in coal qualities. Both Chang and Koerner found the nonAustralian country intercept dummy variables were positive and significantly different from zero, indicating prices for both Canadian and US coals have been significantly higher than prices for similar Australian coals. In Hogan, Thorpe and Middleton, non-Japan regional intercept dummy variables were found to be significantly different from zero although these varied across coal types and years, indicating that there have been significant differences in the prices of Australia’s coal exports to Japan and other markets. The estimation period varies in each of the studies — JFY 1993 to JFY 1994 in Chang, JFY 1992 and JFY 1994 in Koerner, and JFY 1989 to JFY 1996 in Hogan, Thorpe and Middleton. None of the studies formally test for structural change in the regression coefficients for quality attributes between years. Notably though, Koerner estimates separate regression equations for each year, while Chang and Hogan, Thorpe and Middleton allow for changes in the intercept terms by year. The datasets used by Chang and Koerner are mainly sourced from the Australian Coal Report for Australia, while the sparser data for the United States and Canada were collected from Japanese sources. The Tex Report’s annual Coal Manual, published in Tokyo, is the main source of information on Japanese coking coal contracts. By contrast, Hogan, Thorpe and Middleton used the Australian government’s coal export controls data which represents the most comprehensive dataset available for hedonic regression analysis for Australia’s coal exports. Key results from the regression analyses in each of the three studies are presented in table 1. The results from Hogan, Thorpe and Middleton (1997) refer to a single pooled analysis over hard, soft and semisoft coal types. Pure dummy variable parameters included in the regression equations are not shown in the table for brevity. Only the quality attributes included in the preferred model or models from each study are included in the table. Each quality variable is standardised to common units so that each slope coefficient is the estimated implicit price of a quality variable. These adjustments, which can be readily derived, are required since Chang’s preferred equation is a linear–log specification. A linear hedonic regression equation, such as that used in Hogan, Thorpe and Middleton (1997), may be written in its simplest form as: (1)

Pj = a0 + Σi ai QCij 7

ABARE

ABARE CONFERENCE PAPER 98.3 where Pj is the coal price in coal shipment j; QCij is the value of quality characteristic i in coal shipment j; a0 is an intercept term; and ai is the coefficient for quality characteristic i. An implicit price of a coal attribute is the marginal change in the coal price given a marginal change in the coal quality. From equation 1, the implicit price for quality characteristic i is the coefficient for that variable, that is: ∂P/∂QCi = ai

(2)

The corresponding linear–log regression equation, such as that given in Chang, may be written as: (3)

Pj = a0 + Si ai ln QCij

where ln is the natural logarithm. The implicit price for quality characteristic i is: ∂P/∂QCi = ai/QCi

(4)

which typically is evaluated at the mean of quality characteristic i to give a point estimate. This is the approach used in table 1 to adjust the coefficients, with the exception of fluidity, in Chang’s preferred model to obtain estimated implicit prices that can be compared across

Table 1: Main results from previous studies

Metallurgical coal Notation Coal type Estimation period a No. observations Adjusted R squared b Quality variables c Total moisture Inherent moisture Ash Volatile matter Less than 30% 30% or higher Sulphur Crucible swelling number Vitrinite reflectance Log fluidity

Chang (1995) Hard 1993–94 34 0.89

TM IM ASH VM VM–KINK SULP

CSN REFL LFLUID

Koerner (1996) Hard 1992 38 0.95

Hard 1994 38 0.93

Insignificant

Insignificant

–US$0.12

Insignificant

US$0.36

–US$6.89

Insignificant

US$0.53 US$6.58 US$0.57

Hogan, Thorpe and Middleton (1997) Hard 1989–96 1255 0.96

Soft 1989–94 1255 0.96

Semisoft 1989–96 1255 0.96

Insignificant –US$2.25 –US$0.46

Insignificant –US$2.25 –US$1.81

Insignificant –US$0.97 –US$0.17

Insignificant

Insignificant

Insignificant

–US$3.64

US$0.18 –US$0.23 Insignificant

US$0.23

US$0.23

US$0.23

US$6.02 US$0.41

a Based on Japanese fiscal years. b Unadjusted R squared for Koerner (1996). Only one adjusted R squared is generated for the pooled regression in Hogan, Thorpe and Middleton (1997). c Insignificant denotes a variable that is not significantly different from zero at the 5% significance level.

8

ABARE

ABARE CONFERENCE PAPER 98.3 studies. The coal price is measured in nominal US dollars a tonne in Chang (fob) and Koerner (cif), and in JFY 1995 prices in Hogan, Thorpe and Middleton (fob). The four quality variables included in Chang’s regression analysis are VM, SULP, CSN and LFLUID. With the exception of SULP, all quality parameters are statistically significant from zero at the 5 per cent level and have the correct sign. The four quality variables included in Koerner’s JFY 1992 regression equation are ASH, SULP, REFL and LFLUID. With the exception of ASH, all quality parameters are statistically significant at the 5 per cent level and have the correct sign. A similar regression analysis is also conducted for JFY 1994 and, except for an insignificant coefficient for sulphur, the results are similar. Notably, Koerner does not include CSN in the regression model arguing, among other things, that it is an alternative physical measure to LFLUID of the coking characteristic of a coking coal. Koerner also argues that LFLUID is related to VM, which may explain why he does not also include that variable in the regression. Unlike Chang, Koerner also includes REFL in the regression model arguing that this variable is a measure of the available carbon in a coking coal. Hogan, Thorpe and Middleton (1997) have information on six quality characteristics for Australia’s coking coal shipments, including TM, IM, VM, ASH, SULP and CSN. They impose a linear functional form on the coking coal regression equation arguing that the high degree of correlation between each of the quality variables and squared and crossproduct terms rules out the use of more general functional forms in ordinary least squares regression. Notably, Hogan, Thorpe and Middleton (1997) conduct RESET tests for general model misspecification, particularly in functional form, and find no evidence of this. RESET tests pick up curvature in the underlying model. Hogan, Thorpe and Middleton (1997) also undertake several tests for departures from homoskedasticity and find no evidence of this. Hogan, Thorpe and Middleton (1997) find that IM, ASH and CSN are significant explanators of hard, soft and semisoft coking coal prices. SULP is also an important explanator of soft coking coal prices, and a kinked relationship for VM, whereby the implicit price of semisoft coking coal is positive for coals with a VM less than 30 but is negative thereafter, is an important explanator of semisoft coking coal prices. The negative kink is introduced through variable VM-KINK, a dummy variable relationship for VM that is defined in section 4. Intercept dummies for soft and semisoft coking coals are also found 9

ABARE

ABARE CONFERENCE PAPER 98.3 to be significant. This suggests that the categorisation of coal by type is a key explanator of coking coal price differentials. Hogan, Thorpe and Middleton (1997) provide some caveats to their research. Omitted variables are not a problem if they contain no new information. However, Hogan, Thorpe and Middleton note that their model may suffer from omitted variable bias because variables considered important to the end use performance of coking coals are excluded from the regression. The omitted variables are REFL, LFLUID and CSR. These variables are not recorded in the export controls database. Notably, Chang omits REFL and both Koerner and Chang omit CSR from the regression models. The two key quality characteristics that the Japanese steel mills nominated under the fair treatment system are CSR and fluidity (Kahraman, Coin and Reifenstein 1997). Based on unpublished research, Kahraman, Coin and Reifenstein (1997) state that both factors have been significant determinants of price for several years, even though there is no substantive evidence for high fluidity levels to necessarily imply a high level of value in use. As Berndt (1991) and Kahraman, Coin and Reifenstein (1997) note, however, hedonic regressions are concerned with consumer perceptions of end use quality attributes rather than actual performance attributes.

Research method Hedonic regression equations and data Separate hedonic regression equations are estimated for hard and semisoft coking coals. In each case, the dependent variable is a measure of the real coal price. The set of explanatory variables comprise coal quality characteristics and a set of annual dummy variables. Following Hogan, Thorpe and Middleton (1997), a linear functional form based on equation 1 is adopted to avoid multicollinearity problems associated with the inclusion of quadratic or higher order polynomial terms. The dataset used in this study comprises a cross-section of prices and quality characteristics for hard and semisoft coking coal brands exported from Australia to Japan in the years JFY 1992 to JFY 1997. The data were obtained from the Australian Coal Report’s Coal year (1992 and 1994–97). Coal prices are fob and quoted in US$ a tonne. Real prices are derived by deflating the published nominal prices by the US consumer price index with a base year of JFY 1995. Annual dummy variables (D1992, D1993, …, D1997) are included to cater 10

ABARE

ABARE CONFERENCE PAPER 98.3 for general changes in market conditions over time. Since coal prices are in JFY 1995 US dollars, DJFY95 was excluded to avoid perfect multicollinearity. The quality variables included in the regression equations, with notation and expected sign shown in brackets, are excess moisture (EM, –), inherent moisture (IM, –), ash (ASH, –), volatile matter (VM, nonlinear), sulphur (SULP, –), crucible swelling number (CSN, +), vitrinite reflectance (REFL, +), log fluidity (LFLUID, +) and coke strength after reaction (CSR, +). Excess moisture (EM) is equal to total moisture less inherent moisture. All other quality characteristics are as defined in appendix 1. With the exception of log fluidity, the regression coefficient on a quality variable is the implicit price for an extra unit of the quality attribute within the coal category. To avoid perfect multicollinearity, fixed carbon content is not included directly in the regression equations since, to a first order approximation, it is equal to 100 per cent less VM, ASH and IM. Following Hogan, Thorpe and Middleton (1997), two linear segments for volatile matter with a maximum at 30 is hypothesised for the semisoft coking coal equation. Both VM and VM-KINK are included in the regression model where VM-KINK is a dummy variable defined by: (5)

VM-KINK = (VM – 30) if VM ≥ 30 =0 elsewhere

The implicit price for VM is expected to be positive up to 30 and negative for higher levels of VM. Published information on coke strength after reaction is only available for hard coking coal. This could indicate that for semisoft coking coal, marginal changes in CSR do not have any influence on price. It is assumed in this paper that CSR does not have an impact on semisoft coking coal prices and thus does not lead to biased regression results. Although the prices of all coals were reported for 1994, associated quality specifications were not. However, as quality specifications for these brands are largely constant between years, 1994 quality specifications were equated to the average of the respective specifications in adjacent years. The number of observations for the hard and semisoft coking coal categories totalled 103 and 82, respectively, between JFY 1992 and JFY 1997. Soft coking coal was not included 11

ABARE

ABARE CONFERENCE PAPER 98.3 Table 2: Summary statistics for hard coking coal data Mean value

Standard deviation

Minimum value

Maximum value

Real coal price (in JFY 1995 US$) JFY 1992 US$/t JFY 1993 US$/t JFY 1994 US$/t JFY 1995 US$/t JFY 1996 US$/t JFY 1997 US$/t

54.6 51.0 45.8 50.2 51.4 50.0

1.2 1.1 1.1 1.0 1.2 1.2

51.8 48.3 43.1 47.6 49.1 47.7

55.6 52.0 46.7 51.1 53.0 51.7

Quality variables Excess moisture % Inherent moisture % Ash % Volatile matter % Sulphur % Crucible swelling number no. Vitrinite reflectance % Log fluidity log ddpm Coke strength after reaction %

7.4 1.4 8.4 25.3 0.6 7.6 1.2 6.9 61.8

1.1 0.4 1.1 4.3 0.1 1.3 0.2 1.2 7.7

6.0 1.0 6.2 17.2 0.4 5.0 0.9 3.9 50.0

9.0 2.0 9.9 34.0 0.9 9.0 1.7 9.2 74.0

Mean value

Standard deviation

Minimum value

Maximum value

Real coal price (in JFY 1995 US$) JFY 1992 US$/t JFY 1992 US$/t JFY 1993 US$/t JFY 1994 US$/t JFY 1995 US$/t JFY 1996 US$/t JFY 1997 US$/t

54.6 44.6 40.9 36.5 42.2 42.2 39.0

1.2 1.3 1.1 1.1 1.3 1.4 1.5

51.8 43.0 39.2 34.8 40.1 40.1 37.1

55.6 47.0 43.1 38.6 45.9 45.9 42.7

Quality variables Excess moisture % Inherent moisture % Ash % Volatile matter % Sulphur % Crucible swelling number no. Vitrinite reflectance % Log fluidity log ddpm

6.5 2.3 9.5 30.0 0.6 4.5 0.9 4.3

1.0 0.7 0.7 7.1 0.2 1.8 0.3 2.2

5.0 1.0 7.9 15.5 0.3 2.0 0.7 0.0

9.0 3.5 11.3 36.6 1.3 8.5 1.7 7.6

Variable

Unit

Table 3: Summary statistics for semisoft coking coal data

Variable

Unit

12

ABARE

ABARE CONFERENCE PAPER 98.3 in the analysis since only 15 observations were available for this category over the period of analysis. Summary statistics of the quality and price variables used in this analysis are listed in tables 2 and 3 for hard and semisoft coking coals, respectively.

Structural change tests The key objective in this paper is to test for structural change or price–quality regime switches over the period JFY 1992 to JFY 1997. A maximum likelihood estimation technique, based on Quandt’s exogenous switching regime method, is used to test for structural change (Goldfeld and Quandt 1976; Johnston 1991). With this method, the coefficients for the quality characteristics may vary between hypothesised regimes and the distributions of the error terms are assumed to be independent normal random variables with zero mean and constant but different variances. Given the hedonic regression equations described above, the maximum likelihood method involves choosing the regime switch year, including the possibility of no switch, that maximises the likelihood of observing the data sample as measured by the log likelihood function. For completeness, tests are conducted for structural change at the beginning of JFY 1994, 1995 and 1996. Log likelihood function estimates are derived using OLS estimates of the relevant parameters for each of these hypothesised regime switches. A statistically definitive likelihood ratio test of the null hypothesis of no structural change or regime switch in the dataset is not available. However, Goldfeld and Quandt (1976) report that the use of a Chi-square (3) distribution has been found to be an acceptable approximation in some applications. The results for this approximate test are reported in section 5. A Chow test is commonly used to test the null hypothesis that the regression parameters are jointly identical between regimes. However, the validity of this test rests on the assumption that the error variances are common and constant for the regimes (Greene 1993). Greene (1993) describes a test which can be applied to Quandt’s switching regime regression model of heteroscedasticity across regimes, with homoscedasticity within each regime. Greene’s likelihood ratio statistic for a test of the null hypothesis of homoscedasticity is identical to the likelihood ratio test for Quandt’s switching regime regression model. In this case, the likelihood ratio test statistic is asymptotically distributed as a Chisquare (1) distribution. The results from this test are also reported later.

13

ABARE

ABARE CONFERENCE PAPER 98.3

Other diagnostic tests Following Hogan, Thorpe and Middleton (1997), a set of Ramsey RESET tests are conducted to test for general specification error, and a set of LM tests are also conducted to test for the presence of nonconstant error variances. RESET test statistics (2), (3) and (4) are derived from artificial regressions of the coal price on the exogenous variables and an increasing number of polynomial terms in the fitted value of the dependent variable (SHAZAM 1993). A test of the null hypothesis of no model misspecification error is an F test that all of the coefficients on the fitted values of the dependent variable in the artifical regression are jointly zero. Due to the power terms in the regression, RESET tests tend to be very useful in detecting a misspecified functional form. The LM tests for departures from homoscedasticity are also based on artificial regressions. Those considered here are based on regressing the estimated squared errors on the fitted value of the dependent variable. In the LM (1), (2) and (3) tests, the fitted variable is expressed in linear, quadratic or log quadratic form, respectively. In each case, the LM test statistic, which has an asymptotic Chi-square (1) distribution, is computed as the number of observations multiplied by the unadjusted R squared from the artificial regression.

Results Structural change test results The log likelihood function estimates for the assumptions of no structural change and structural change at the beginning of JFY Table 4: Log likelihood function estimates 1994, 1995 and 1996 are given in table 4. using Quandt’s exogenous switching regime From these results, it is likely that there method for hard and semisoft coking coal equations a has been a structural shift in price–quality relationships for both hard and semisoft Hard Semisoft coking coal, although the timing of these Switching point b coking coal coking coal shifts differs. JFY 1996 is the most likely No switch –98.21 –83.71 switch point for the change in valuation JFY 1994 –95.78 –77.80 regime for hard coking coal, while JFY JFY 1995 –92.53 –69.96 1995 is estimated to be the most likely JFY 1996 –83.51 –71.94 a Bold indicates the maximum log likelihood function estimate. switch point for the change in valuation b The hypothesised regime switching point or timing of the structural change is at the beginning of the year indicated. regime for semisoft coking coal. 14

ABARE

ABARE CONFERENCE PAPER 98.3 The likelihood ratio test statistic is 29.39 for the hard coking coal equation and 27.50 for the semisoft coking coal equation. Based on the Chi-square (3) distribution, the null hypothesis of no structural change is rejected at the 5 per cent significance level for both hard and semisoft coking coal equations. Based on Greene’s test using the Chi-square (1) distribution, the null hypothesis of homoscedasticity across regression groups is rejected at the 5 per cent significance level for both hard and semisoft coking coal equations. Thus, the application of Quandt’s switching regime model is appropriate in this study, rather than Chow’s test for structural change. Overall, these results indicate that the structure of price–quality relationships for the Australia–Japan coking coal trade changed fundamentally for semisoft coking coal with the merging of the soft coking coal category into the semisoft coking coal category after JFY 1994 and for hard coking coal with the subsequent adoption of the fair treatment system in JFY 1996.

Estimated regression equations Following Chang (1995) and Koerner (1996), the full regression results are reported in this section including variables not significantly different from zero at the 5 per cent level. A major reason for this approach is that multicollinearity is a potential problem in this type of study which, if present, lowers the t-values and increases the risk of excluding significant variables. This may particularly be the case for the hard coking coal equation in the second period. Omitted variables bias is considered to be a greater problem than the inclusion of irrelevant variables. The full regression results are also strictly consistent with the structural change tests and associated diagnostic tests reported above. Hard coking coal equations The regression results for the hard coking coal equations are given in table 5. Consistent with the structural change test results, there are two sets of regression results corresponding to the estimation periods JFY 1992 to 1995 and JFY 1996 to 1997. Notably, the adjusted R squared is reduced from 0.97 in the first period to 0.74 in the second period. All RESET and LM tests for both equations are accepted at the 5 per cent significant level. For the first period, six quality characteristics are found to have a significant impact on hard coking coal prices: excess moisture, inherent moisture, log fluidity, crucible swelling number, coke strength after reaction and sulphur. Each coefficient for these quality characteristics has the expected sign and all are significantly different from zero at the 5 15

ABARE

ABARE CONFERENCE PAPER 98.3 per cent level (the level used in the remainder of this section for two tail t tests unless otherwise specified). In the second period, the coefficients of only three quality characteristics — inherent moisture, volatile matter and sulphur — are found to be significantly different from zero. However, the t-values for crucible swelling number and coke strength after reaction are significantly different from zero at the 10 per cent significance level (or, equivalently, at the 5 per cent level using one tail t tests). In contrast to the results in the first period, excess moisture and log fluidity are not significant determinants of hard coking coal prices in the second period. Conversely, volatile matter is found to be a significant determinant of hard coking coal prices in the second period, but not the first.

Table 5: Regression results for hard coking coal equations JFY 1992–95 Variable EM IM ASH VM SULP CSN REFL LFLUID CSR D1992 D1993 D1994 D1997 Constant

Estimated coefficient

t-value

–0.64 –1.90 –0.11 –0.01 –2.26 0.56 0.57 0.73 0.06 4.37 0.73 –4.43

–4.73 –3.79 –0.87 –0.11 –3.60 4.63 0.66 3.91 2.63 21.20 3.60 –22.31

46.22

LM tests LM test 1 LM test 2 LM test 3

Estimated coefficient * *

* * * * * * *

15.14 *

Number of observations Adjusted R squared RESET tests RESET test 2 RESET test 3 RESET test 4

JFY 1996–97 t-value

–0.39 –4.30 0.21 0.48 –2.43 0.41 2.18 –0.22 0.07

–1.28 –3.89 * 0.71 4.27 * –2.18 * 1.73 ** 1.67 –0.89 1.82 **

–1.40 38.94

–5.70 * 8.26 *

69 0.97

34 0.74

F(1,55) F(1,54) F(1,53)

0.43 0.21 0.38

F(1,22) F(1,21) F(1,20)

4.06 2.67 1.99

Chi square(1) Chi square(1) Chi square(1)

0.05 0.05 0.05

Chi square(1) Chi square(1) Chi square(1)

0.28 0.29 0.27

* Statistically significant at the 5 per cent significance level (2 tail test). ** Statistically significant at the 10 per cent significance level (2 tail test).

16

ABARE

ABARE CONFERENCE PAPER 98.3 The interpretation of the estimated coefficients for the quality variables is summarised as follows. • For each one percentage point increase in excess moisture, the hard coking coal price (in JFY 1995 US dollars) falls by an estimated US$0.64 a tonne in the first period. For each one percentage point increase in inherent moisture, the hard coking coal price falls by an estimated US$1.90 a tonne in the first period and US$4.30 a tonne in the second period. • For each one percentage point increase in coke strength after reaction, the hard coking coal price is estimated to increase by US$0.06 a tonne in the first period and, at the 10 per cent significance level, US$0.07 a tonne in the second period. • For each one unit increase in the crucible swelling number, the hard coking coal price is estimated to increase by US$0.56 a tonne in the first period and, at the 10 per cent significance level, US$0.41 a tonne in the second period. • For each one unit increase in log fluidity, the hard coking coal price is estimated to increase by US$0.73 a tonne in the first period. • For each one percentage point increase in sulphur, the hard coking coal price is estimated to fall by US$2.26 a tonne in the first period and US$2.43 a tonne in the second period. • For each one percentage point increase in volatile matter, the hard coking coal price is estimated to increase by US$0.48 a tonne in the second period. Overall, price–quality relationships for the Australia–Japan hard coking coal trade have changed substantially since JFY 1992. Compared with the first estimation period JFY 1992 to 1995, in JFY 1996 and 1997, the extent to which differences in coal quality explain differences in coal prices has been reduced markedly. The number of quality characteristics that are found to be significant determinants of hard coking coal prices has also been reduced, and the mix of quality characteristics has changed. Compared with Chang (1995), Koerner (1996) and Hogan, Thorpe and Middleton (1997), a larger number of quality characteristics are found to be significant determinants of hard coking coal prices, particularly in the first estimation period which is the most comparable time period. The quality characteristics included in the three previous econometric exercises vary across the studies, but each represents a subset of the quality characteristics used in the current study. 17

ABARE

ABARE CONFERENCE PAPER 98.3 The one exception to this is the use of total moisture rather than excess moisture in Hogan, Thorpe and Middleton (1997), and total moisture was not included in their final equation. Based on goodness of fit, excess moisture was preferred to total moisture in the current study. The estimated equations in the three previous studies may therefore have problems associated with omitted variables bias. However, given the relatively high adjusted R squared in these studies, particularly Koerner (0.95 and 0.93) and Hogan, Thorpe and Middleton (0.96), there is a possibility that some properties of the coal are measured by more than a single variable. For example, crucible swelling number, coke strength after reaction and vitrinite reflectance are all indicators of coking properties of the coal. As a consequence, exclusion of some quality characteristics may not be as serious as would otherwise be the case. Semisoft coking coal equations Two sets of regression results for the semisoft coking coal equations are given in table 6 corresponding to the estimation periods JFY 1992 to 1994 and JFY 1995 to 1997. The adjusted R squared is reduced from 0.97 in the first period to 0.84 in the second period. All RESET and LM tests for both equations are accepted at the 5 per cent significant level. For the first period, four quality characteristics are found to have a significant impact on semisoft coking coal prices with the correct sign, including excess moisture, log fluidity, crucible swelling number and sulphur. Coke strength after reaction was not included in the semisoft coking coal equations since this information was not published for this coal category. In the second period, the coefficients of three quality characteristics — excess moisture, volatile matter (for coal with a volatile level higher than 30) and crucible swelling number — are found to be significantly different from zero. Log fluidity and sulphur are not significant determinants of semisoft coking coal prices in the second period. Similar to the hard coking coal equations, volatile matter is found to be a significant determinant of prices of semisoft coking coal (with high volatile matter levels) in the second period, but not the first. The interpretation of the estimated coefficients for the quality variables is summarised as follows.

18

ABARE

ABARE CONFERENCE PAPER 98.3 • For each one percentage point increase in excess moisture, the semisoft coking coal price (in JFY 1995 US dollars) falls by an estimated US$0.61 a tonne in the first period and US$0.92 a tonne in the second period. • For each one unit increase in the crucible swelling number, the semisoft coking coal price is estimated to increase by US$0.27 a tonne in the first period and US$0.57 a tonne in the second period. • For each one unit increase in log fluidity, the semisoft coking coal price is estimated to increase by US$0.35 a tonne in the first period. • For each one percentage point increase in sulphur, the semisoft coking coal price is estimated to fall by US$1.31 a tonne in the first period. • For each one percentage point increase in volatile matter above 30, the semisoft coking coal price is estimated to fall by US$0.40 a tonne in the second period. Table 6: Regression results for semisoft coking coal equations JFY 1992–95 Variable EM IM ASH VM VM_KINK SULP CSN REFL LFLUID D1992 D1993 D1996 D1997 Constant Number of observations Adjusted R squared RESET tests RESET test 2 RESET test 3 RESET test 4 LM tests LM test 1 LM test 2 LM test 3

JFY 1996–97

Estimated coefficient

t–value

–0.61 0.48 –0.33 –0.10 0.01 –1.31 0.27 –0.53 0.35 8.16 4.47

–3.34 1.64 –1.39 –1.43 0.08 –2.34 3.93 –0.91 3.25 34.46 19.10

44.02

Estimated coefficient *

* * * * *

15.32 *

t–value

–0.92 0.14 0.26 0.05 –0.40 –0.68 0.57 –0.73 0.09

–2.63 * 0.32 0.74 0.51 –2.11 * –0.48 4.68 * –1.13 0.48

–0.27 –3.43 43.39

–0.86 –10.55 * 11.68 *

43 0.97

39 0.84

F(1,30) F(1,29) F(1,28)

0.03 0.35 0.77

F(1,26) F(1,25) F(1,24)

1.44 1.01 0.68

Chi square(1) Chi square(1) Chi square(1)

1.31 1.31 1.30

Chi square(1) Chi square(1) Chi square(1)

0.66 0.69 0.62

* Statistically significant at the 5 per cent significance level (2 tail test).

19

ABARE

ABARE CONFERENCE PAPER 98.3 The structural changes in the price–quality relationships for the Australia–Japan semisoft coking coal trade since JFY 1992 appear to be substantial, but less so than for hard coking coal trade. Compared with the first estimation period JFY 1992 to 1994, in the second period JFY 1995 to 1997, the explanatory power of coal quality in explaining differences in coal prices has been reduced markedly, but by less than occurred for the hard coking coal equations. Similar to the hard coking coal equations, the number of quality characteristics that are found to be significant determinants of semisoft coking coal prices has been reduced, and the mix of quality characteristics has changed.

Conclusion In this paper, Quandt’s exogenous switching regime method was used to test for structural change in price–quality relationships for Australia’s hard and semisoft coking coal exports to Japan in the period JFY 1992 to 1997. For semisoft coking coal, structural change is estimated to have occurred at the beginning of JFY 1995, which follows the reclassification of soft coking coal into the semisoft coking coal category. For hard coking coal, structural change is estimated to have occurred a year later at the beginning of JFY 1996, which coincides with the introduction of the fair treatment pricing system. Throughout the full period, there are relatively more quality characteristics that are significant determinants of hard coking coal prices than of semisoft coking coal prices. In each case, however, the number of significant quality variables is lower in the recent period and the mix of significant quality variables changes. Consistent with the findings of Kahraman, Coin and Reifenstein (1997) that fluidity is a poor indicator of the coking properties of coal, log fluidity is not a significant determinant of either hard or semisoft coking coal prices in the recent period. Notably, the explanatory power of the equations, as measured by the adjusted R squared, is reduced substantially between the two periods from 0.97 to 0.74 for hard coking coal and from 0.97 to 0.84 for semisoft coking coal. Under the fair treatment pricing system, in operation in both JFY 1996 and 1997, final hard and semisoft coking coal prices and other contract information remain confidential. The substantial reduction in the explanatory power of quality characteristics may be consistent with increased difficulties, particularly for coal exporters, in obtaining information about coal price–quality outcomes during the annual negotiations. However, this conclusion is dependent on the assumption that coal price-quality data published since the beginning of JFY 1996 are reliable estimates of actual coal settlements. 20

ABARE

ABARE CONFERENCE PAPER 98.3

Appendix 1: Coal quality Coal quality characteristics refer to either the composition of the coal or properties of the coal that directly influence its ability to perform the function required in the end use of the coal. Using proximate analysis, coal is comprised of fixed carbon, volatile matter, ash and inherent moisture. Fixed carbon and coal volatile matter can be further divided into the elements carbon, hydrogen, nitrogen, sulphur and oxygen using ultimate analysis. Vitrinite reflectance (R0max or Rvmax) is an accurate measure of coal rank and is also used to evaluate coking coal blends. Coke quality increases with coal rank. Older, higher rank coals tend to have a lower inherent moisture content due to their longer exposure to heat and geological pressure. Similarly, volatile matter tends to decrease as the rank of a coal increases. By contrast, the fixed carbon content and calorific value of coal tends to rise with the rank of the coal. The crucible swelling number indicates the capacity of the coal to expand when subjected to a standardised heat and is used to evaluate the coking properties of the coal. Coking coal is determined primarily by its coking properties. The crucible swelling number ranges from 0 to 9 and tends to be positively related to coke quality. That is, a crucible swelling number of 9 implies the coal has very good coking properties. The fixed carbon content of the coal influences the specific energy content or calorific value of the coal. The calorific value is the amount of heat released by the complete combustion of the coal under specified conditions and is usually measured in kilocalories per kilogram (kcal/kg). The fixed carbon content is measured as a percentage of the air dried coal sample. Volatile matter is the proportion of the air dried sample that is released in the form of gas or vapour during a standardised heating test. Volatile matter is a positive influence on the ability of the coal to sustain combustion, but is inversely related to coke yield (Roberts and Callcott 1984). A high volatile matter content — generally exceeding 30 per cent of the air dried coal — increases the potential risk of spontaneous combustion. The total moisture content refers to the water in coal, and is the sum of inherent (or air dried) moisture and excess moisture. Excess moisture can be removed by coal preparation. Transport costs increase directly with moisture content. Extremely low total moisture increases the risk of spontaneous combustion. 21

ABARE

ABARE CONFERENCE PAPER 98.3 Ash is the residue remaining after the complete combustion of all coal organic matter and oxidation of the mineral matter present in the coal. Ash is therefore the incombustible material present in the coal and is measured as a percentage of the air dried coal sample. A higher ash content results in both higher transport and handling costs per unit of energy contained in the coal and waste that, in general, requires disposal. For coking coals, ash remains in the coke and becomes incorporated in blast furnace slag. Coal contracts typically contain penalties for ash outside the specified range. Sulphur increases operating and maintenance costs by causing fouling and corrosion of metal surfaces. Coal contracts typically contain penalties for sulphur outside the specified range. Fluidity of coking coals increases up to a point of maximum fluidity when subjected to increasing temperature. Some coking coal users pay a premium for increased fluidity despite the fact that there is little evidence of a positive correlation with the quality of the coke produced (Kahraman, Coin and Reifenstein 1997). Coke strength after reaction is strongly positively related to coke quality. For the purpose of blast furnace applications, the coal used must be able to form large angular coke that retains its form despite constant abrasion and collision in the blast furnace.

22

ABARE

ABARE CONFERENCE PAPER 98.3

References Australian Coal Report 1992, Coal 1992, Sydney. —— 1994, Coal 1994, Sydney. —— 1995, Coal 1995, Sydney. —— 1996, Coal 1996, Sydney. —— 1997, Coal 1997, Sydney. Berndt, E.R. 1991, The Practice of Econometrics: Classic and Contemporary, AddisonWesley, Massachusetts. Carpenter, A.M. 1995, Coal Blending for Power Stations, IEA Coal Research, International Energy Agency, London. Chang, H.S. 1995, ‘Examining hard coking coal price differentials: a hedonic pricing approach’, Resources Policy, vol. 21, no. 4, pp. 275–82. Goldfeld, S.M. and Quandt, R. 1976, Studies in Nonlinear Estimation, Ballinger, Massachusetts. Greene, W.H. 1993, Econometric Analysis, Second Edition, Macmillan, New York. Hogan, L., Thorpe, S. and Middleton S. 1997, Quality Adjusted Prices for Australia’s Black Coal Exports, ABARE report to the Department of Primary Industries and Energy, Canberra. Hogan, L., Thorpe, S., Graham, P. and Middleton, S. 1997, ‘Coal price–quality relationships and the outlook for coal’, in Outlook 97, Proceedings of the National Agricultural and Resources Conference, Canberra, 4–6 February, vol. 3, Minerals and Energy, ABARE, Canberra, pp. 249–68. International Energy Agency (IEA) 1997, Coal Information 1996, IEA/OECD, Paris. Johnston, J. 1991, Econometric methods, Third Edition, McGraw-Hill, Singapore. Kahraman, H., Coin, C. and Reifenstein, A. 1997, ‘Technical factors determining comparative coking coal prices in the Japanese market’, in Outlook 97, Proceedings of the National Agricultural and Resources Conference, Canberra, 4–6 February, vol. 3, Minerals and Energy, ABARE, Canberra, pp. 277–80.

23

ABARE

ABARE CONFERENCE PAPER 98.3 Koerner, R. 1993, ‘The behaviour of Pacific metallurgical coal markets: the impact of Japan’s acquisition strategy on market price’, Resources Policy, vol. 19, no. 1, pp. w 66–79. —— 1996, Behaviour of Pacific energy markets: the case of the coking coal trade with Japan, Pacific Economic Papers, no. 252, Australia–Japan Research Centre, Australian National University, February. Low, J., Dwyer, G., Jolly, L. and Olejniczak, J. 1993, Price formation for coking coal exports to Japan: a hedonic approach, ABARE paper presented at the 22nd Conference of Economists, Economic Society of Australia, Murdoch University and Curtin University, Perth, 27 September to 1 October. Porter, D. and Gooday, P. 1990, The effect of coal quality on Japanese coking coal contract prices, ABARE paper presented at the 19th Conference of Economists, Economic Society of Australia, University of New South Wales, Sydney, 24–27 September. Roberts, O.C. and Callcott, T.G. 1984, Net Carbon in Coking Coal, Published Report 84–17, ACIRL, Sydney. Rosen, S. 1974, ‘Hedonic prices and implicit markets: product differentiation in pure competition’, Journal of Political Economy, vol. 82, no. 1, pp. 34–55. Scott, D.H. 1994, Developments Affecting Metallurgical Uses of Coal, IEA Coal Research, International Energy Agency, London. SHAZAM 1993, User’s Reference Manual Version 7.0, McGraw-Hill, Canada. Skorupska, N.M. 1993, Coal Specifications – Impact on Power Station Performance, IEA Coal Research, International Energy Agency, London. Taylor, R. 1994, Study of the Queensland and New South Wales Black Coal Industry: A Report to the Australian Coal Industry Council, Canberra, November. Thomson, A., Zulli, P., McCarthy, M. and Horrocks, K. 1996, ‘Pulverised coal injection in ironmaking blast furnaces’, Australian Coal Review, October, pp. 42–6.

24