Modelling Winegrape Prices in Disequilibrium

Modelling Winegrape Prices in Disequilibrium By Associate Professor Edward Oczkowski Working Paper No. 36/0 October 00 About the Author Associate Professor Edward Oczkowski is a lecturer in Economics in the School of Management at Charles Sturt University, Wagga Wagga. E-mail: eoczkowski@csu.edu.au Phone: +61 6933377, Fax: +61 6933930. 1

Charles Sturt University Faculty of Commerce Working Paper Series Editors: Dr Pamela Mathews Professor Reg Mathews Dr Arthur Sweeney The Faculty of Commerce Working Paper Series is intended to provide staff and students with a means of communicating new and evolving ideas in order to encourage academic debate. Working papers, as the title suggests, should not necessarily be taken as completed works or final expressions of opinion. All working papers are subject to review prior to publication by one or more editors or referees familiar with the discipline area. Normally, working papers may be freely quoted and/or reproduced provided proper reference to the author and source is given. When a working paper is published on a restricted basis, notice of such restriction will appear on this page.

Table of Contents Abstract 4 1. Introduction 5. Theoretical and Econometric Framework 5 3. Winegrape Pricing and the Australian Market 8 4. Winegrape Price Equation Estimates and Discussion 13 5. Conclusion 19 Data Appendix 0 References 1 3

Abstract Fluctuations in winegrape production and prices have been well documented. Survey data from winemakers point to some substantial demand and supply winegrape imbalances. This paper presents an econometric model of winegrape prices which recognises the existence of demand and supply imbalances in the Australian market. Modelling is based on regional and variety winegrape data produced by the Australian regional winegrape survey. A markets in disequilibrium framework is employed to motivate modelling price changes as responding to variations in excess demand/supply. The disequilibrium price equation provides estimates of regional and varietal price discounts/premiums and a measure of the speed of disequilibrium price adjustment. Disequilibrium estimates are compared to those from a conventional equilibrium price equation model. 4

1. Introduction Pricing in wine markets has been researched extensively in recent years with the estimation of hedonic price functions, see for example, Nerlove (1995) and Oczkowski (001). A hedonic price function relates the price of a product to its characteristics. In contrast, very little attention has been paid to winegrape pricing, Golan and Shalit (1993) is a notable exception. This paper adds to the paucity of winegrape pricing studies by examining winegrape prices in the Australian market. It has been long recognised that the Australian winegrape market suffers from demand and supply imbalances, Spencer (00). As a consequence this paper employs a disequilibrium approach to modelling prices by assuming prices only partially adjust to market imbalances, see Quandt (1988). Given the absence of technical winegrape data, the modelling focuses upon regional and variety price differences through a price reduced form specification. In the next section, the theoretical and econometric framework is outlined. Section three describes the data and some key characteristics of the market. Section four presents and discusses the results, and section five concludes.. Theoretical and Econometric Framework We employ and briefly present the hedonic framework outlined by Nerlove (1995) which is based on the repackaging model of quality differences of Fisher and Shell (1971) and Muellbauer (1974). Put simply, a commodity has various model types denoted by Z, each Z has an associated quality index which is a function of a set of characteristics. For the typical demander, the quality index appears directly in the utility function and therefore permits demanders to purchase numerous model types and/or more than one unit of a specific model type. In a competitive market, the representative demander is assumed to maximise utility subject to a budget constraint given market determined prices P(Z). This program results in a demand function: d( Z) = ϕ[ P( Z), Y, A( Z)] (1) where, d(z) is the quantity demanded for model type Z, Y is a vector of demander characteristics and A(Z) is a quality index of demander preferences for Z. 5

Analogously, the representative supplier is assumed to maximise profits subject to a stated technology given market determined prices P(Z). This program results in a supply function: s( Z) = ψ[ P( Z), W, B( Z)] () where, s(z) is the quantity supplied for model type Z, W is a vector of supplier characteristics and B(Z) is an index of the difficulties/costs of supplying Z. For given P(Z) aggregation across all demanders and suppliers results in market demand D (Z) and market supply S(Z) functions for each model type Z. These aggregate functions are dependent on the factors in eqns (1) and (). Assuming the existence of a competitive equilibrium process resulting in D ( Z) = S( Z), an equilibrium reduced form price function can be written as: P ( Z) = G[ A( Z), B( Z), Y, W ] (3) thus, P(Z) is a function of indexes of demand preferences for and difficulties in supplying Z, and demander and supplier characteristics. As an alternative the market process may not result in equilibrium trading at each session, D( Z) S( Z). A price adjustment scheme based on excess demand/supply could be specified: P ( Z) = F[ D( Z) S( Z)] = H[ A( Z), B( Z), Y, W ] (4) Consider now, the econometric linear representation of this disequilibrium hedonic market for model type Z: D S P = α P 1 = β P 1 = P + α X P + β X 1 D S + u + u = λ( D 1 S ) + u 3 (5) (6) (7) D S where, D, and P are market demand, supply and price respectively; X, X are S demand and supply regressors ideally reflecting demand preferences and supply production difficulties for Z, and demander and supplier characteristics; u are contemporaneously i independent error terms with zero means and constant variances; α, β, λ suitably define parameters. It is expected that α 1 < 0, β 1 > 0 and λ > 0, prices rise (fall) given excess demand (supply). 6

In most situations D and S are directly unobservable and some assumption is needed relating demand and supply to the observed quantity transacted, Q. The minimum condition is typically employed, Q = min( D, S ), and the system of equations can be consistently estimated via maximum likelihood methods if normal errors are assumed, see Quandt (1988, ch). For some markets however, quantity transacted information maybe unavailable 1 and/or the minimum condition may represent an unreasonable assumption. For the analysis of these markets, some information on the parameters in the system represented by eqns (5) - (7) can be gained by working through the reduced form for price. Substituting (5) and (6) into (7) gives: P where = µ P 1 µ = [1 λ( α β )] u + µλα X = µ [ u 3 1 D 1 + λ( u µλβ X 1 1 u S + u )] (8) Equation (8) can be re-written as: * P = µ P + ( 1 µ ) P + µ u3 1 (9) * where, P defines equilibrium price ( D = S ) with 0 µ 1, µ = 0 implies equilibrium and µ = 1 implies infinitely slow market clearing. This partial adjustment interpretation of the disequilibrium model is due to Bowden (1978) and for some reason (adjustment costs, incomplete information, contractual arrangements, etc) assumes that price rigidity results in prices only partially adjusting to equilibrium in each period. The fraction of the t disequilibrium gap closed after t periods is: (1 - µ ). For examples of disequilibrium modelling using this reduced form approach, see Ito and Ueda (1981) for business loans and Anas and Eum (1984) for housing. Given classical error term assumptions, eqn (8) can be consistently estimated via least squares and a unique parameter identified for the speed of adjustment coefficient µ. The remaining coefficients in (5) and (6) cannot be uniquely gained from eqn (8). Despite this, qualitative 1 The practice of most hedonic studies is to ignore quantity information, to the despair of Nerlove (001). To some extent the unavailability of quantity information may have led to this practice. Goldfeld and Quandt (1981) present some Monte Carlo evidence supporting the use of µ = 0 as a basis for testing the equilibrium hypothesis. 7

signs can be inferred for the parameters associated with the demand/supply regressors in (8). For example, with µ > 0 and λ > 0, a positive demand shifter ( α > 0 ) will result in a positive coefficient for the associated D X regressor in (8). For regressors common to both demand and supply, implied signs for coefficients in (8) can only be inferred if α have opposite signs. For example, if > 0 (a positive demand shifter) and < 0 (a negative supply shifter) then the sign on the coefficient of the associated common regressor in (8) would be positive, i.e., µλ( β ) > 0. In cases where and have similar signs, α the larger (absolute value) coefficient determines the sign of the coefficient in (8). α β α β and β As an alternative to the linear specification of eqns (5) (9), a double log specification may be more appropriate, where D and S are specified in multiplicative form and: ( P / P 1) = (D / λ S ) u 3 (10) this implies that excess demand in percentage terms leads to a constant percentage change in prices. This system results in a double log reduced form price equation: S ln( P ) = µ ln( P + µλα ln( ) µλβ ln( X ) + (11) 1) D X u with similar interpretations for the speed of adjustment coefficient µ, as in the linear system. 3. Winegrape Pricing and the Australian Market In Australia, over 1400 wineries produce wine from grapes supplied by their own vineyards and from approximately 7000 independent growers. 3 There are approximately 40 wine regions producing about 50 different winegrape varieties. About two-thirds of grapes are grown in warm inland irrigated regions, the remainder are grown in cool dryland regions. 4 Wineries purchase approximately 75% of their grapes from independent growers. It has been estimated that half of these purchases take place under contract and the remainder on the spot market. Contracts vary significantly in their length and composition. Contracts are mainly specified for quantity supplied with varying conditions specified for price determination, Scales, Croser and Freebairn (1995, p11). 3 Background information is soured from: Australian and New Zealand Wine Industry Directory (00), Scales, Croser and Freebairn (1995), NSW Government Review Group (1996) and the Australian Regional Winegrape Crush Survey form. 4 This distinction is made in Scales, Croser and Freebairn (1995), NSW Government Review Group (1996) and is employed by industry commentators, Stanford (00). The distinction is employed for data analysis. The warm inland irrigated regions consist of: Riverina, Murray Valley, and the Lower Murray Zone (Riverland). 8

Various motivations have been examined and confirmed as explantors for price rigidity, market inefficiency and disequilibrium trading in markets generally, e.g., contractual arrangements, industrial concentration and incomplete information, see for example, Carlton (1986) and Borenstein and Shepard (00). Some of these motivations may apply to the winegrape market. First, as previously suggested, individual contracts for supplying grapes do exist, in some regions (Riverland) 80-90% of trades are conducted under contracts, NSW Government Review Group (1996, p59). Second, the market dominance of large conglomerate winemakers is well known, the largest four makers produced 70% of case wine sales in 001. Concerns for domineering winemaker bargaining power in price negotiations with growers, has been expressed in various inquires, Scales, Croser and Freebairn (1995, p10), NSW Government Review Group (1996, p40) and has resulted in the establishment of grower associations. Thirdly, it has been argued that growers posses incomplete market information compared to winemakers. Wineries deal with numerous growers across various regions and therefore have more knowledge about general growing costs and conditions and demand requirements, Scales, Croser and Freebairn (1995, p17) and NSW Government Review Group (1996, p56). Taylor (000) provides a non-technical summary of the reasons for winegrape market inefficiencies, alluding to: imprecise quality measurement; high sunk costs; perishability of grapes; discontinuous annual trading sessions; and asymmetrical market intelligence. The recognition of market imbalances has motivated the industry through the Australian Bureau of Agricultural and Resource Economics (ABARE) to measure and forecast demand and supply imbalances, Spencer (00). The publication of these projections commenced in 1989, see Croser and Strachan (1997) for an evaluation of the accuracy of these projections. Finally, there has been frequent industry comment on the existence and consequences of market imbalances, see for example, Stanford (00) and Marlowe (00). Given this reasonably strong motivation for modelling prices in disequilibrium, we now consider the data. Data from the annual Australian Regional Winegrape Crush Survey (ARWCS) is employed for modelling prices. The hedonic model type Z of section two, is defined by a particular region/variety combination. Some aggregation among varieties and regions was necessary for minor regions and varieties, to avoid having too many small 9

hedonic model types which might lead to modelling problems 5. Weighted average price data provided by wineries, relating to the 000 and 001 vintages is employed for all states, excluding Tasmania and Queensland. 6 In total, 413 region/variety combinations for two years exist on winegrapes from 8 regions on 34 unique varieties. Consistent with previous analyses, a broad distinction is made to facilitate modelling: warm inland irrigated consists of 106 region/variety combinations, encompassing 3 regions and unique varieties; cool dryland consists of 307 region/variety combinations, encompassing 5 regions and 3 unique varieties. As a precursor to disequilibrium modelling, it will prove informative to describe some of the data collected by the ARWCS. In addition to weighted average price information, data on grapes purchased, crushed and preferred, is collected from wineries. The difference between preferred and crushed tonnage is used to measure excess demand in the various state reports from the AGWCS. Price changes (between 000 and 001) based on this excess demand measure, are suggested to have occurred, see Stanford (00, pp49-50). However, as pointed out by Croser and Strachan (1997) the demand measure of tonnes preferred is not consistent with the standard notion of demand employed in economics. 7 Conventionally, demand is the amount demanded by wineries given actual (not ideal) prices, weather conditions, contracts, preferences, etc. It is not clear that the concept of ideal vintage means the same to all wineries and certainly preferred tonnes cannot meaningfully be related to observed actual demand determinants in any modelling process. We investigated the correlation between the ARWCS excess demand measure (tonnes preferred less tonnes crushed) and price changes, as predicted by the conventional model, i.e., eqns (7) and (10) with λ > 0. In qualitative terms, in only about half the cases (50.8%) was the excess demand sign consistent with the price change direction as predicted by the conventional model. Contrary to expectations, in 33.9% of cases a price increase occurred with an excess supply and in 15.3% of cases a price fall occurred with an excess demand. The non-parametric nominal measure of correlation between price changes and excess 5 Regions are combined with closely related geographical areas if less than six varieties are recorded for a region. This resulted in the merger of only two regions with other regions. Varieties are combined into other red or white varieties if a variety is recorded for less than six regions. This resulted in the combination of 17 minor white varieties and 17 minor red varieties, see the data appendix. 6 Data is inaccessible for Queensland and Tasmania, these states however, are estimated to produce only 0.6% of total Australian production in 001, Australian and New Zealand Wine Industry Directory (00). 7 The ARWCS defines tonnes preferred as: grapes you would have preferred to crush this year in an ideal vintage not affected by weather, fixed contract commitments, etc. 10

demand is statistically insignificant, Phi = 0.054 (p = 0.76). 8 actually estimated (naively by OLS) the When eqns (7) and (10) are R s for warm/cool regions and linear/log models, range from 0.0004 to 0.0. In the case of the log model for cool regions, a negative (λ) price adjustment coefficient is estimated. Clearly, employing the excess demand measure, based on tonnes preferred information from the ARWCS as an indicator of price changes, is fraught with danger. The model defined by eqns (5) (7) with the Q minimum condition, cannot be directly employed for the ARWCS data even given price and quantity transacted information. First, the quantity minimum condition is inappropriate for the market. Given the existence of longterm contracts for quantity supplied, in many situations wineries are buying more than they prefer for some particular seasons, (Q > D ). Even given the limitations of the tonnes preferred data, some broad support for this contention can be gained from the ARWCS data. For the 001 vintage about one quarter of all trades are characterized by an excess supply of more than 5% of tonnes crushed. This implies that many traders are forced to trade more than they wish, this violates the involuntary trading notion underpinning the minimum condition. Secondly, insufficient information exists on the separate demand and supply determinants for equations (5) and (6). Apart from price and quantity information only a weighted average baume measure of sugar content is collected for traded grapes. Taylor (000) bemoans the lack of industry-wide quality information on grapes. Only recently have attempts been made to collect better winegrape quality information, see Swinburn (001). 9 Given these deficiencies, our modelling relates to reduced form price estimates for eqns (8) and (11), which make no assumption about quantity transacted and cannot isolate separate demand/supply determinants. As regressors we employ baume and separate dummy variables for each variety and region. 10 Despite the lack of technical winegrape information, the variety 8 These qualitative correlations are even weaker when prices are deflated by the CPI. 9 Swinburn (001) reports a quality indicator number (QIN) for only 1 parcels of grapes in the Sunraysia/Riverland. The QIN combines measures of baume, color, ph and TA levels. Studies of winegrape markets in other countries have employed information such as: sugar, total acid, tartaric acid, malic acid, date of harvest and grape weight, see Golan and Shalit (1993). 10 Oczkowski (1994) successfully employed region and variety dummy variables in the analysis of retail wine prices. 11

and regional premiums/discounts provide important industry-wide information. A priori arguments can be mounted that particular regions and varieties have specific influences on both demand and supply, resulting in an expected sign for the coefficients in eqns (8) and (11), i.e., µλ( α β ) can be signed a priori. For example, given prices, demand preferences and costs of production are such that grapes in the Margaret River region are expected to have above average influence on demand ( α > 0) and a below average influence on supply ( β < 0) compared to other regions, resulting in a positive reduced form price coefficient, µλ( α β ) > 0. Effectively, incorporating these dummy variables summarizes and captures the differences which exist between regions and varieties in demand preferences and the difficulties in producing winegrapes, i.e., A(Z) and B(Z) in section two. The specifications of eqns (8) and (11) assume that current price appears in the supply equation. The existence of an active spot market and the non-harvesting of grapes, suggests some supply flexibility to current prices. Moreover, recent multi-sectoral equilibrium models of the wine industry, assume a current price effect on grape supply, for example, Zhao, Anderson and Wittwer (00) assume a supply elasticity of 0.4 for premium grapes. Beyond an immediate price impact, supply variations based 4-5 year lagged prices may be important given the lags between vine planting and commercial harvest. These arguments suggest that one-year lagged prices may not significantly influence supply, even so, we need to recognize that the inclusion of P in S does pose some modelling difficulties. 1 If also depends upon in eqn (6), then the reduced form price equation maintains the S same regressors as eqn (8), that is, it is observationally equivalent to the reduced form without P 1 P 1 in S. This causes problems for the identification of µ. It can be shown that if S also depends upon P 1 (with a coefficient β 3 ) then the coefficient on P 1 in (8) is now: µ ( 1 λβ 3). Thus interpreting the coefficient on P 1 as µ leads to an estimation bias equal to: (true less estimated) = µ µ = µ µ 1 λβ ) = ( ˆ µ 1)[( β /( α β 1 β )] > 0. That is, ˆ ( 3 3 1 3 1

we underestimate µ if β 3 > 0. Interestingly, as ˆ µ 1 the bias approaches zero. We will make comment later on the size of any likely biases for our models. 11 4. Winegrape Price Equation Estimates and Discussion Table 1 presents the summary measures for eqns (8) and (11) for the cool and warm region distinctions, based on: baume, region and variety variables. The data appendix lists the data sources and key descriptive statistics. The RESET tests and equilibrium hypothesis µ = 0 is overwhelmingly rejected in all models. 1 R values indicate that the linear specification is preferred. The R values are suitably high. All models suffer from heteroscedasticity problems and all resulting estimates employ Whites heteroscedastic consistent covariance matrix (HCCM) estimator. Estimates for the speed of adjustment coefficients differ significantly between regions but not between functional forms. The Table 1: Disequilibrium Price Equation Summary Measures Cool Dryland (n = 307, d.f = 56) Warm Inland Irrigated (n= 106, d.f = 78) Linear Log Linear Log R 0.854 0.845 0.955 0.954 RESET() -0.44-3.39* 1.80 0.94 Hetero 18.4* 154.4* 100.6* 63.5* µˆ 0.63* 0.691* 0.95* 0.916* * denotes significant at a 5% level. R for log models based on predictions for raw price data. RESET() test is distributed as N(0,1) and based on Whites HCCM. Hetero is the Breusch- Pagan test, disturbed as χ with 50 (cool) and 7 (warm) d.f. µˆ is the disequilibrium speed of adjustment price coefficient. 11 The inclusion of P in also makes the equilibrium price reduced form observationally equivalent to eqn 1 S P 1 (8). For the equilibrium reduced form, the coefficient on is disequilibrium specifications, where positive coefficients are expected for a1 = β 3 /( α1 β1). P 1 Unlike the, for the equilibrium model a < 0 1 is expected (co-web model) and therefore can be meaningfully distinguished from the disequilibrium specifications. 1 Likely biases for the interpretation of µ resulting from the possible existence of lagged price in the supply equation can be approximated. Based on assumptions made in Zhao, Anderson and Wittwer (00) about elasticities [0.4 for supply (current and one-year lagged) and -0.8 for demand] estimates of the possible underestimation of µ are: 0.09 for cool and 0.0 for warm regions. The estimated positive speed of adjustment coefficient suggests that an equilibrium reduced form equation resulting from a supply equation based on lagged prices, is unlikely to be consistent with the data. 13

Given the RESET test results, linear specifications are presented for the disequilibrium and equilibrium models for the cool and warm regions in tables and 3. 13 In addition to parameter estimates, sample mean elasticities are presented for lagged price and baume variables and percentage changes from average prices are presented for the variety and region dummy variables. 14 Consider the main features of the disequilibrium estimates. The speed of adjustment coefficient is significantly larger for warm regions. For warm regions after 4 years (the time between vine planting and harvesting) only 7% of the disequilibrium gap is closed, while for cool regions 84% is closed. The additional price rigidity in warm regions may reflect the higher degree of industrial concentration in these regions. Even though the average number of wineries purchasing grapes in each of the warm and cool regions is approximately equal (7. in warm and 7.1 in cool, per individual region), in warm regions an average of 476 tonnes is purchased per winery in an individual region, which compares to 8 tonnes in cool regions. 13 The equilibrium model assumes µ = 0 and therefore only the lagged price is omitted from the disequilibrium specification. The RESET test also points to the superior performance of the linear specification for the equilibrium models. R values for the linear equilibrium specifications are: cool = 0.70 and warm = 0.57. 14 The method of Kennedy (1984) is employed to estimate coefficients for all regions/varieties without the need to omit a control. The method is implemented by running two suitably defined regressions, see Oczkowski (1994). The mechanics of the approach are as follows. Consider Y = β 0 + β1d 1 + β D + γx + u, where D 1 + D = I are dummy variables. At the means of the data: Y = β 0 + β1p1 + β P + γx, where P1 and P are the proportion of non-zeros in the dummies. Impose β 1 P1 + β P = 0, which implies βˆ0 = Y γˆ X. Thus ˆβ 0 is the average intercept for the entire sample. The predicted value for Y (at X ) for a particular dummy (say 1 = 1) is: ˆ β + ˆ γx + ˆ β = Y + ˆ β, thus dummy coefficients measure differences from Y if evaluated at D 0 1 1 X. In percentage terms from Y, the impact of a dummy variable (evaluated at X ) is: *{( Y + ˆ β Y }/ Y = 100*( ˆ β / ). 100 This expression is evaluated in tables and 3 as the % impact. 1) 1 Y 14

Table : Winegrape Linear Price Equation Estimates: Cool Dryland Disequilibrium Equilibrium Coeff t-ratio % impact Coeff t-ratio % impact Constant 06.1 1.18 6.3*.1 Lagged Price 0.63* 1.66 0.65 Baume 1.86 1.65 0.14 5.38*.1 0.513 Varieties Chardonnay 19.3 0.81 1.5 41.1 1.40 3. Chenin Blanc -46.1* -6.9-19. -506.3* -6.1-39.6 Colombard -183.7* -4.9-14.4-485.8* -10.9-38.0 Muscadelle -10.0-1.46-8.0-95.0* -.03-3.1 Muscat Blanc -90.* -7.63 -.7-683.* -15.4-53.4 Pedro Ximenes -39.7* -5.59-5.8-717.8* -1.9-56.1 Riesling.8 0.08 0. -45.5* -5.47-19. Sauvignon Blanc 6.3 0.67.1-17.8-0.4-1.4 Semillon -7.4* -.30-5.7-171.5* -3.87-13.4 Traminer -134.7* -.70-10.5-343.3* -3.81-6.8 Verdelho -6.0-0.83 -.0-75.3* -.1-5.9 Pinot Gris 74. 0.7 5.8 98.3 1.00 7.7 Other White -74.3-0.88-5.8-164.1-1.03-1.8 Cabernet Franc -9.8-0.75 -.3 57.7 0.91 4.5 Cabernet Sauvignon 80.5*.55 6.3 06.4* 4.6 16.1 Grenache -67.7-1.59-5.3-15.* -.1-11.9 Malbec -16.7-0.38-1.3 131.1*.1 10.3 Mataro -15.9-1.3-9.8-154.7* -3.60-1.1 Merlot 18.1 0.50 1.4 191.5* 5.35 15.0 Petit Verdot 31.6*.88 4.4 713.*.91 55.8 Pinot Noir 5.4 0.14 0.4 78.7 1.36 6. Ruby Cabernet -105.7* -.38-8.3-94.6-0.93-7.4 Shiraz 73.8*.44 5.8 0.6* 4.68 15.8 Sangiovese 67.1 0.80 5. 3.9* 3.5 18. Other Red 187.1*.70 14.6 86.4* 3.80.4 Regions ACT & Sth NSW 93.3*.3 7.3 11.3*.47 8.8 Cowra -38.8-1.04-3.0-198.* -3.30-15.5 Hunter Valley 0.7 0.0 0.1-13.1-0.3-1.0 Mudgee 193.8*.90 15.1-108.9* -.3-8.5 Orange 140.6*.87 11.0 146.6*.19 11.5 King Valley -4.0* -5.4-18.9-310.9* -5.73-4.3 Mornington Peninsula 7.9 0.. 369.1* 3.11 8.9 Port Phillip other -4. -0.66-3.3 151.5 1.71 11.8 Yarra Valley 90.4 1.79 7.1 370.5* 7.09 9.0 Rutherglen. 0.37 1.7-165.5* -.15-1.9 Western Victoria -96.9-1.54-7.6-59.9-0.68-4.7 Central Victoria -73. -1.81-5.7-51.5-1.00-4.0 Adelaide Hills 130.0*.84 10. 317.5* 6.54 4.8 Clare Valley -4.8-0.1-1. -86.3* -.14-6.7 Mount Lofty other -6.8* -.15-4.9-53.8* -7.93-19.8 Barossa Valley 3.6 0.79.5-49.8-0.75-3.9 Eden Valley 115.7*.44 9.0 166.7*.08 13.0 Langhorne Creek -85.4* -3.3-6.7-70.5* -5.40-1. McLaren Vale 43.4 1. 3.4 117.9 1.8 9. Fleurieu Zone other 19.3 0.60 1.5 66.9 1.13 5. Coonawarra -3.3-0.66 -.5 130.8 1.41 10. Padthaway 15.7 0.33 1. 38. 0.37 3.0 Great Southern WA -8.1-0.41 -. 113. 1.76 8.9 Margaret River 106.5* 3.79 8.3 179.1* 4.40 11.7 Swan District & other WA -11.8* -.90-8.8-33.3* -5.19-18. * denotes significant at a 5% level. T-ratios are based on Whites HCCM. % impact are: elasticities for lagged price and baume at sample means of data and the % premium or discount from the average price for varieties and regions. 15

Table 3: Winegrape Linear Price Equation Estimates: Warm Inland Irrigated Disequilibrium Equilibrium Coeff t-ratio % impact Coeff t-ratio % impact Constant 9.5 0.41-381. -0.87 Lagged Price 0.95* 19.9 0.933 Baume 0.18 0.03 0.005 68.97 1.9 1.798 Varieties Chardonnay 60.0* 4.94 1.6 167.8* 5.04 35.1 Chenin Blanc -0.5-0.05-0.1-44.0-1.13-9. Colombard 16.0* 1.98 3.3-45.0-1.79-9.4 Muscadelle -5.* -4.71-5.3-90.0* -4.10-18.8 Muscat Blanc 14.7*.38 3.1-70.3-1.50-14.7 Pedro Ximenes 0.0 0.00 0.0-171.6* -4.4-35.9 Riesling 37.*.5 7.8 59.5 1.51 1.5 Sauvignon Blanc -6.6-0.3-1.4 1.1 0.58 4.4 Semillon 1.7 1.59.7 6.4 1.03 5.5 Traminer 14.3 0.59 3.0 114.3* 4.16 3.9 Verdelho 35.5 1.08 7.4-41.3-1.1-8.6 Other White 9.7 1.17.0-75.0-1.91-15.7 Cabernet Franc 33.1* 3.0 6.9 88.6*.01 18.5 Cabernet Sauvignon -78.9* -3.67-16.5 136.3*.61 8.5 Grenache -19. -0.89-4.0-3.3-0.37-4.9 Malbec -17.3-0.60-3.6 46.3 1.73 9.7 Mataro -.5-1.69-4.7.5 0.43 4.7 Merlot -14.8-1.10-3.1 146.8*.76 30.7 Petit Verdot -55.1* -3.18-11.5 151.9*.80 31.8 Pinot Noir 5.0 1.64 5. 8.6* 4. 47.8 Ruby Cabernet -5.8* -.9-5.4 80.6 1.41 16.9 Shiraz 9.4 0.31.0 173.* 3.17 36. Sangiovese -1. -0.04-0.3 79. 1.9 16.6 Other Red -18. -1.3-3.8-46.4-1.3-9.7 Regions Riverina 0.7*.8 4.3-48.6* -4.07-10. Murray Valley (Vic/NSW) -1.1-0.17-0. 10.5 0.56 0.1 Lower Murray Zone (SA) -9.3* -.14-1.9 16.6 1.1 0.3 * denotes significant at a 5% level. T-ratios are based on Whites HCCM. % impact are: elasticities for lagged price and baume at sample means of data and the % premium or discount from the average price for varieties and regions. The baume effect is positive as expected, but is more economically important for cool regions, it is practically zero for warm regions. 15 For cool regions the main variety premiums/discounts are: petit verdot (+4.4%), minor reds (+14.6%), cabernet sauvignon (+6.3%), pedro ximenes (-5.8%), muscat blanc (-.7%) and chenin blanc (-19.%). For cool regions the main region premiums/discounts are: Mudgee (+15.1%), Orange (+11.0%), Adelaide Hills (+10.%), King Valley (-18.9%), Swan District (-8.8%) and Western Victoria (-7.6%). For warm regions the main variety premiums/discounts are: chardonnay (+19.3%), riesling (+7.8%), verdelho (+7.4%), 15 It might be hypothesised that non-linear specifications for the baume effect may be more appropriate, ie., very high sugar content in some grapes may be less preferred. Alternative non-linear specifications for baume proved unsuccessful. 16

cabernet sauvignon (-16.5%), petit verdot (-11.5%) and ruby cabernet (-5.4%). For warm regions, Riverina commands a 4.3% premium and Riverland a 1.9% discount. The estimated differences in variety impacts between cool and warm regions for the disequilibrium model are noteworthy. For four varieties (colombard, muscat blanc, cabernet sauvignon and petit verdot) statistically significant sign reversals occur. For example, in cool regions cabernet sauvignon commands a 6.3% premium, but in warm regions attracts a 16.5% discount. In total 14 (8 whites and 6 reds) sign reversals occur for varieties between cool and warm regions, this constitutes about 56% of all varieties. Recall these premiums and discounts are estimates of µλ( α β ) and therefore reflect both demand and supply influences. The dominant influence determines whether a premium or discount exists. The influences are a complex combination of physical and sensual characteristics of varieties and regions. Broadly, factors such as vine vigour, yield, grape flavour, regional climate and soil, might be important influences. For example, in warm regions grape yield tends to be high and flavour reduced. Petit verdot requires a relatively high number of heat degree days to fully ripen (Beeston, 00), this may partially explain its premium in cool regions and discount in warm regions. A similar argument, but to a lesser extent, applies to cabernet sauvignon. For regions, while Orange and Adelaide Hills are relatively cool regions commanding premiums, the King Valley and Swan District are relatively warm regions attracting discounts. Contrary to this argument Mudgee, the region with the highest estimated premium is a relatively warm region. Other factors, such as the relative high cost of grape producing land, given its proximity to Sydney, may partially explain its higher than average price. Consider the major differences between the disequilibrium and equilibrium estimates. Immediately noticeable differences are the substantially larger price premiums and discounts for the varieties and regions in the equilibrium model. This difference is illusory. As pointed out in Oczkowski (1998) disequilibrium models have more elaborate dynamic effects than equilibrium models. In tables and 3 for the 17

equilibrium models the estimated impacts are both short and long-run given the absence of lagged prices in the specification. For the disequilibrium models the presented impacts are only immediate, after k years the cumulative impact of a variable with coefficient γ is given by: γ (1 µ ) /(1 µ ). For example, after 4 years the coefficient should be multiped by:.44 for cool and 4.31 for warm regions. This recognition makes many of the percentage impacts comparable after 4 years. For example, cabernet sauvignon in cool regions has an immediate disequilibrium impact of 6.3% which translates to 15.4% after 4 years, while the equilibrium premium remains at 16.1%. k + 1 There are a significant number of sign reversals between the equilibrium and disequilibrium estimates. For cool regions 11 (% of all estimates) sign reversals occur (4 variety and 7 region). For warm regions 16 (59% of all estimates) reversals occur (13 varieties and 3 (all) regions). Given the rejection of the equilibrium hypothesis for both regions, these results clearly point to the inaccuracies of the equilibrium estimates. The differences between equilibrium and disequilibrium estimates shed interesting information on warm region price differences. During 000, some concern was expressed that for no logical reason Riverina prices were lower than prices in other warm regions (AGW, 000). The equilibrium estimates in table 3 support this contention, suggesting that the Riverina prices are discounted (-10.%) compared to other warm regions (+0.1% and +0.3%). In contrast however, the disequilibrium estimates point to a Riverina premium (4.3%) compared to slight discounts in the other regions (-0.3% and -1.9%). The summary statistics on prices helps explain the difference between disequilibrium and equilibrium estimates. Even though the average price for 001 was lower in the Riverina ($43 per tonne) than the other warm regions ($491), the average price change between 000 and 001 was higher for the Riverina (+$1 per tonne) compared to a fall in other regions (-$1 per tonne). Given its more complex dynamics by recognising market imbalances, the disequilibrium model captures these price changes and therefore provides a theoretically more consistent and hence meaningful representation of regional premiums/discounts. 18

5. Conclusion This paper provides the first estimates of region and variety price premiums and discounts for the Australian winegrape market employing a disequilibrium modelling approach. The estimates are particularly useful in identifying where current demand preferences and production difficulty differences lie in the market. The disequilibrium approach to modelling produces substantially different estimates to the conventional equilibrium approach. The distinction between cool and warm regions points to interesting findings on relative price rigidity in the markets, which raises some concerns about market dominance and efficiency. Future research should focus more directly on wine grape quality measurement. Technical winegrape information on characteristics such as grape acid and colour should be collected more widely and could be usefully employed using the outlined disequilibrium approach. Similarly, survey consumer preference data on wines might usefully be incorporated into future winegrape pricing models. 19

Data Appendix Data is from the Australian Regional Winegrape Crush Survey published on web-sites of the various state wine industry associations: New South Wales: www.nswwine.org.au/news.htm Victoria: www.nre.vic.gov.au/ South Australia: www.phylloxera.org.au/statistics/utilisationsurvey.html Western Australia:www.agric.wa.gov.au/programs/hort/viticulture/wine/Default.asp Data is from the 000 and 001 vintages covering 413 region/variety combinations; n =307 for cool dryland regions and n = 106 for warm inland irrigated regions. Price ($ per tonne): 001 Cool: mean = 178.8 std = 413.6 Warm: mean = 477.9 std = 177.0 000 Cool: mean = 163.8 std = 448.9 Warm: mean = 48.1 std = 195. Change Cool: mean = 15.03 std = 18.4 Warm: mean = -4.16 std = 51.5 Baume (weighted average): Cool: mean = 1.53 std = 1.13 Warm: mean = 1.46 std =1.13 Other Red Wines: Barbera, Muscat Brown, Saint Macaine, Rose Cross, Gamay, Tarrango, Touriga, Tempranillo, Carina, Rubired, Zante Currant, Carignan, Nebbiolo, Chambourcin, Durif, Zinfandel, Meunier. Other White Wines: Crouchen, Italia, Sultana, Sylvaner, Orange Muscat, Biancone, Canada Muscat, Merbein Seedless, Taminga, Waltham Cross, Emerald Riesling, Muscat Gordo Blanco, Doradillo, Marsanne, Trebbiano, Viognier, Palomino. 0

References Anas, A. and S.J. Eum (1984) "Hedonic Analysis of a Housing Market in Disequilibrium." Journal of Urban Economics, 15, 87-106 Australian Grapegrower and Winemaker (AGW) (000) "Riverina Growers not Happy with Wineries 000 Prices." No. 435, 6. Australian and New Zealand Wine Industry Directory (00) 0th edn, Adelaide: Winetitles. Beeston, J. (00) The Wine Regions of Australia. nd edn, Crows Nest: Allen and Unwin. Borenstein, S. and A. Shepard (00) "Sticky Prices, Inventories, and Market Power in Wholesale Gasoline Markets." RAND Journal of Economics, 33, 116-139. Bowden, R. (1978) "Specification, Estimation and Inference for Models of Markets in Disequilibrium." International Economic Review, 19, 711-76. Carlton, D.W. (1986) "The Rigidity of Prices." American Economic Review, 76, 637-658. Croser, B. and S. Strachan (1997) "Premium Winegrape Supply and Demand Projections - An Appraisal of the ABARE Study." Australian Grapegrower and Winemaker, No. 399,, 14-0. Fisher, F.M. and K. Shell (1971) "Taste and Quality Change in the Pure Theory of the Cost of Living Index." In Z. Griliches (ed) Price Indices and Quality Change: Studies in New Methods of Measurement. Harvard University Press: Cambridge MA. Goldfeld, S.M., and R.E. Quandt (1981) "Single Market Disequilibrium Models: Estimation and Testing." The Economic Studies Quarterly, 3, 1-8. Golan, A. and H. Shalit (1993) "Wine Quality Differentials in Hedonic Grape Pricing." Journal of Agricultural Economics, 44, 311-31. Ito, T. and K. Ueda (1981) "Tests of the Equilibrium Hypothesis in Disequilibrium Econometrics: An International Comparison of Credit Rationing." International Economic Review,, 691-708. Kennedy, P.E. (1986) "Interpreting Dummy Variables." The Review of Economics and Statistics, 68, 174-175. Marlowe, R. (00) "Talking Points." Australian Viticulture, 6 (3), 3. 1

Muellbauer, J. (1974) "Household Production Theory, Quality and the Hedonic Technique." American Economic Review, 64, 977-994. New South Wales (NSW) Government Review Group (1996) Review of the Legislation Establishing the Wine Grapes Marketing Board: Final Report. Sydney: NSW Government. Nerlove, M. (1995) "Hedonic Price Functions and the Measurement of Preferences:`` The Case of Swedish Wine Consumers." European Economic Review, 39, 1697-1716. Nerlove, M. (001) "Zvi Griliches, 1930-1999: A Critical Appreciation." Economic Journal, 111, F4-F448. Oczkowski, E. (1994) "A Hedonic Price Function for Australian Premium Table Wine." Australian Journal of Agricultural Economics, 38, 93-110. Oczkowski, E. (1998) "The Impact of Market Clearing Assumptions and Dynamics on Demand Elasticities." Economics Letters, 59, 39-43. Oczkowski, W. (001) "Hedonic Wine Price Functions and Measurement Error." Economic Record, 77, 374-38. Quandt, R.E. (1988) The Econometrics of Disequilibrium. Blackwell Publishers: Oxford. Scales, W., Croser, B., and J. Freebairn (1995) Winegrape and Wine Industry in Australia: A Report by the Committee of Inquiry into the Winegrape and Wine Industry. Canberra: AGPS. Spencer, D. (00) Australian Wine Grape Production and Winery Intake. ABARE Research Report 0.1, Canberra: ABARE. Stanford, L. (00) "Red Surge Stamps its Character on 001." Australian Viticulture, 6 (1), 45-51. Swinburn, G. (001) "Are Grape Prices a Good Indicator of Quality within a Region?" Australian Grapegrower and Winemaker, No. 446, 1-. Taylor, P. (000) "As the Winegrape Market Develops, Price Determination is Reviewed." Australian Viticulture, 4 (4), 47-57. Zhao, X, Anderson, K. and G. Wittwer (00) "Who Gains from Australian Generic Wine R&D and Promotion?" CIES Discussion Paper, No. 004, Adelaide University.