Identification of strategies to improve goat marketing in the lowlands of Ethiopia: a hedonic price analysis

ABSTRACT This article aims at identifying factors that determine market prices of goats and analyse potential mechanisms by which smallholder goat producers could maximize their benefits. Data on 357 farm households and 2103 goat transactions were collected in three major goat markets in the lowlands of Ethiopia. Hedonic price models adjusted for heteroscedasticity were employed to analyse the observed price data. Model results showed the relative importance of different factors in determining goat prices. Animal attributes including age, sex, live weight, body condition and presence of horn as well as types of buyer and market outlet targeted and time of selling were found to be important. Particularly, goats marketed during festive periods where demand for meat increases (e.g. Ethiopian New Year) command higher prices. These results imply that interventions such as systematic selection schemes targeting traits demanded by the market, improved linkages to markets, easy access to market information systems and creating conducive environment including incentive mechanisms can enhance smallholder farmers’ and pastoralists’ ability to take advantage of seasonal and spatial price changes and become market responsive with effective marketing strategies. Such changes can be potent in improving the livelihoods of smallholder farmers and pastoralists.


I. Introduction
Willingness to pay (WTP) for a commodity forms the basis for the process of price discovery in any given market. Economic theory on stated and observed preferences establishes that WTP depends on the perceived utility from the different attributes or consumable features of the commodity in question (Adamowicz et al. 1997;Ben-Akiva et al. 1994;Lancaster 1966;Richter 1966;Rosen 1974;Sen 1971). This has been empirically proven to be the case for quality-differentiated commodities studied in different parts of the world (Kassie, Abdulai, and Wollny 2009;Scarpa et al. 2003). In rural markets of Ethiopia also, analyses of stated and revealed preferences for agricultural commodities in general and livestock in particular have shown that attributes play a key role in determining observed prices (Kassie, Abdulai, and Wollny 2011;Ouma, Abdulai, and Drucker 2007;Terfa et al. 2013). Such studies on livestock revealed that the effects of animal attributes on price formulation varies across species, breeds and markets (Ayele et al. 2006;Kassa, Haile, and Essa 2011;Teklewold et al. 2009). This shows that targeted and rigorous studies for each livestock breed and species across market and/or regions are needed.
Hedonic modelling using revealed prices is an important analytical framework in identifying factors that determine price formulation and variability (Ayele et al. 2006;Kassie, Abdulai, and Wollny 2011;Terfa et al. 2013). For goats, analysis of revealed prices should be based on the fundamental assumption that the observed price of a goat is a composite of the implicit values of its attributes. Therefore, producers who target their production and marketing towards consumers that attach high utility to specific attributes possessed by their flock are expected to fetch higher premium.
In theory, the price of a goat in a competitive livestock market is mainly determined by consumers' utility for the animal's attributes. Therefore, individual consumer's or supplier's attributes are not important (Oczkowski 1994;Rosen 1974). In such markets, each attribute of the goat can be evaluated by buyers when they decide to buy a goat and attach an implicit price for each attribute (Orrego, Defrancesco, and Gennari 2012;Rosen 1974). Prices determined in such market settings are expected to be uniform across all buyers and sellers. However, past studies by, inter alia, Ayele et al. (2006), Kassie, Abdulai, and Wollny (2011) and Andargachew and Brokken (1993) conclude that animal markets in Ethiopia are not competitive as the behaviour of an individual consumer or producer may have a significant effect on prices. Therefore, unlike the case in competitive markets, the implicit prices of goats in Ethiopian markets are a function not only of the goat attributes but also of the individual consumer and/or supplier attributes. With such market imperfections, each individual consumer evaluates each of the goat attributes, attaches an implicit price for every attribute and decides how much she/he is willing to pay for the goat after a long and intricate valuation process.
There is large body of published literature on determinants of observed livestock price in Ethiopian markets (Adugna 2006;Andargachew and Brokken 1993;Ayele et al. 2006;Kassa, Haile, and Essa 2011;Kassie, Abdulai, and Wollny 2011;Teklewold et al. 2009;Terfa et al. 2013). Among these, Teklewold et al. (2009) and Ayele et al. (2006) particularly analysed the determinants of goat prices. However, these studies assumed that the random component of the hedonic price model had constant variability over the set of covariates considered in the study. The experience documented by studies which used similar data generation processes, i.e. one-off cross-sectional data (Kassie, Abdulai, and Wollny 2011;Terfa et al. 2013), shows that this assumption is questionable. The studies also had limited spatial scope and more importantly excluded some animal attributes such as sex and live weight which were found by past studies (Jabbar 1998;Kassie, Abdulai, and Wollny 2011) to be very important in influencing consumer utility and producer decisions. Terfa et al. (2013) and Kassie, Abdulai, and Wollny (2011) have also shown that smallholder farmers/pastoralists take heed of the temporal pattern of market prices of their animals. In contrast, Adugna (2006), Andargachew and Brokken (1993), Ayele et al. (2006), Kassa, Haile, and Essa (2011) and Teklewold et al. (2009) reported that smallholder farmers/pastoralists can hardly gauge the price volatility over time in the markets they conduct their transactions. Such contradictions characterize the literature on small ruminant marketing in the developing world in general and in Ethiopia in particular. Based on goat price data collected from three main livestock markets in the lowlands of Ethiopia, this study intends to contribute towards developing a meaningful analytical framework that could help in identifying the challenges and potential interventions for market development.
Biological clocks, climatic patterns and lack of resources limit farmers' ability to fully exploit seasonal price variations and tap into special niche markets to their own advantage. Therefore, the main objective of this study was to investigate the determinants of revealed prices of goats and in particular to look into the relative importance of the attributes of the animal in setting the price within local markets. Such a study is worthwhile because the benefits of its outcomes are manifold, including the following: (1) smallholder goat producers can utilize the information generated to make optimal production and marketing decisions and take advantage of market opportunities; (2) Federal, regional and local governments can make use of the results to influence microeconomic behaviour through changing market incentives; and (3) Researchers and extension agents can utilize the findings to guide farmers' and pastoralists' decisions and emphasis of future research and extension.
The remainder of the article is structured as follows. The next section describes the data, and Section III outlines the specification and estimation of the different econometric models used in the study. Section IV summarizes and discusses model results. Section V presents concluding remarks and few empirical recommendations.

II. Data and variable selection
Data for this study come from a household survey conducted in the Ethiopian lowlands in 2014. The sample was drawn from among households associated with three major goat markets namely, Bati, Dire Dawa and Yabello which are located in the North, East and Southern lowlands of Ethiopia, respectively ( Figure 1). Given that these markets are important as the main sources of goats for abattoirs and exporters, their inclusion into this study was purposive. Particularly, Yabello is the single most important market where abattoirs and exporters find goats that fulfil the minimum requirement of export in sufficient quantities. Therefore, considering the extent of abattoirs and exporters' market participation, the livestock market in Yabello can be considered as an export oriented, while those in Bati and Dire Dawa are more geared towards catering for domestic demand. In terms of human population sizes, Dire Dawa takes the lead followed by Bati and Yabello.
We generated and employed both qualitative and quantitative data in the study. The qualitative data were generated through focus group discussion (FGD) and key informant interviews (KIIs). The sample respondents were selected randomly from a sampling frame prepared based on the residents' list acquired from the Kebele 1 administration offices. The respondents were visited at their residence, and the discussion was guided with a checklist of semi-structured questions. The FGDs and KISs were employed to collect data on goat marketing strategies of households in the study districts. The sample farm households associated with the Bati market (found in the Amhara region) came from all villages found in Bati district and other kebeles from an adjacent district called Kallu. Sample households associated with the Dire Dawa market (located in the Somali region) came from Shinille district and other kebeles from the adjacent district called Errer. Sample farm households that are associated with the Yabello market (found in Oromiya region) came from Yabello district. While residents of Bati are predominantly sedentary, the residents of Yabello and Shinille districts are pastoralists. A total of 20 kebeles were randomly selected from the districts where the market places are located as well as other adjacent districts from which goat producing farmers come to the three market places under consideration. Subsequently, a total sample of 357 households -118 from Bati, 124 from Yabello and 115 from Shinillewas drawn for this study. A separate data collection format was used to collect primary data from the three sample livestock market places during fasting, festive and normal (two weeks after a major festivity) periods. Fasting periods were categorized into Christian fasting where no animal productsmeat and dairyare consumed (i.e. Adventfasting of the prophets and Lent, and the fasting of the assumption of the Holy Virgin Mary) and Muslim fasting (Ramadan). The festive periods considered for this study were Easter, Ethiopian New Year, Ramadan, Eid al Fetir, Meskel (Finding of the True Cross) and Christmas. The transaction-level data were captured in the markets on 2103 transactions (498 from Dire Dawa, 686 from Bati and 919 from Yabello) from 1551 buyers. The transaction data were collected before and after important social occasions (e.g. religious and/or cultural holidays) in the markets. The transaction per individual respondent ranged from 1 animal (by 63.5% of the respondents) to 29 animals (by 1.4% of the respondents). Physical measurements were carried out for some animal attributes, while visual evaluations were used for others. The live weight for each animal was measured using 100 kg Salter balance. Age was approximated by the method of dentition, which is based on type and number of teeth following Yami and Mekel (2009) estimation guideline. Body condition was determined based on visual evaluation of certain physical characteristics using scale 1 to 3 where 1 denoted emaciated (poor condition) and 3 denoted well fed (good condition) (Nicholson and Butterworth 1986).

III. Methods
Hedonic pricing models have been extensively used in the estimation of implicit prices of attributes characterizing differentiated products (Al-Hassan, Larvoe, and Adaku 2014; Anglin and Gencay 1996;Bartik 1987;Edmeades 2006;Ekeland, Heckman, and Nesheim 2004;Goodman 1978;Heckman, Matzkin, and Nesheim 2005;Kassie, Abdulai, and Wollny 2011;Lancaster 1966;Lawal, Mohammed, and Musa 2016;Lucas 1975;Oczkowski 1994;Parmenter and Pope 2013;Rosen 1974;Terfa et al. 2013). Hedonic pricing modelsindirectly measure the utility consumers derive from different attributes of products and estimating, in monetary terms their valuation of the different attributes. The underlying assumption in hedonic pricing models is that the attributes of a given product determine its price where the different attributes of the product are evaluated and combined by the buyer or seller to form a price for the product.
The theoretical foundations of the hedonic pricing analysis are Samuelson's revealed preference theory (Samuelson 1938) and Lancaster's characteristics theory of value (Lancaster 1966). The revealed preference theory, in general terms, shows that the preferences of consumers can be revealed through their purchasing habits. Lancaster's theory of value postulates that rather than having preferences over market goods directly, agents have preferences over the characteristics or attributes that these goods embody which provide utility for the consumer (Blow, Browning, and Crawford 2008). In this framework, buyers are assumed to derive value or satisfaction from the attributes of a product and not just the product regardless of its constituent attributes (Melton et al. 1978).
Model structure in general and functional form in particular are critical in building an accurate and consistent econometric model (Brown and Ethridge 1995;Cropper, Deck, and McConnell 1988). This is even more so with hedonic price equation estimations because functional form of the hedonic price equation is unknown and economic theory places few restrictions on the form of the hedonic function (Cropper, Deck, and McConnell 1988).
Based on simulated data, Cropper, Deck, and McConnell (1988) showed that regardless of the goodness-of-fit of the model, the linear and a quadratic functional forms give the smallest mean square error of the true marginal value of attributes, whether the attributes are unobserved variable or proxied by other variables. On the other hand, when choosing a functional form and the set of variables to include, it is important to bear in mind the almost-inevitable problem of collinearity. Moreover, high collinearity makes the choice of a very flexible functional form less attractive because the interactive terms of a flexible functional form often lead to greater collinearity (Haab and McConnel 2002).
In studying estimation of implicit housing prices using hedonic price models, Bin (2000) argues that the log-linear functional form is a benchmark parametric specification for hedonic price models. Therefore, following Haab and McConnel (2002), Bin (2000) and Cropper, Deck, and McConnell (1988), this study adopted the log-linear functional form for analysis of hedonic goat prices. The model is specified as follows: where X is the vector of independent variables including goat attributes, market place and season attributes and socioeconomic characteristics of market actors; β is a vector of parameters to be estimated; and ε is the error term which is assumed to be iid normal. The conditional distribution of the errors given the matrix of explanatory variables has zero mean [Ε(ε) = 0], constant variance [V(ε) = σ 2 ] and zero covariance [Ε (εX) = 0]. These assumptions may not hold in estimating parametric models using OLS (Verbeek 2004). Therefore, we conducted conventional regression diagnostics to check for any violations of the basic regression assumptions. A test using normal probability plot revealed that the shape of the probability density function of the price variable is not normal. To overcome this problem, log-transformation based on the quantilenormal distribution was made on the price variable. Then, model specification and normality tests were conducted using Pregibon test for linearity (linktest) and Shapiro-Francia test methods, respectively. The tests confirm that there are no problems related to model specification and normality of the distribution of the dependent variable, both of which are consistent with theoretically sound models.
Understanding the fundamental features of each independent variable is important to estimate robust OLS. A test on the variance inflation factor (VIF) rejected linear correlation among independent variables since the mean VIF is less than the rule of thumb threshold value of 10 (Leahly 2001).
We conducted heteroscedasticity test for each market and for the pooled data. Both the Breusch-Pagan and White tests rejected the hypothesis of constant variance of residuals of the models estimated based on each market and pooled market data sets. In the presence of heteroscedasticity, OLS estimates are still unbiased, but possibly inconsistent, and hence, the usual tests of significance are generally inappropriate and their use can lead to incorrect inference (Long and Ervin 2000). This requires the use of the robust estimation procedure through the derivation of an alternative estimator that is efficient. Alternatively, OLS can be used with adjusted SEs that are consistent but not efficient (Verbeek 2004).
The first option employed in this study is the heteroscedastic-consistent covariance matrix (HCCM) estimator that provides consistent estimator of the covariance matrix. Alternative covariance matrix estimators taken into account are HC0, HC1, HC2 and HC3 (MacKinnon and White 1985;White 1980). The difference among these alternative covariance matrix estimators lies in the SEs where the one with the largest SE is more robust. Cai and Hayes (2008) and Davidson and MacKinnon (1999) provide evidence that least square residuals from HC0 and HC1 tend to be very small. This implies that the estimates from HC0 and HC1 become less robust. While HC2 and HC3 are the best possible covariance matrix estimators, the superiority of one over the other lies in its properties when testing coefficients that are most strongly affected by heteroscedasticity (Long and Ervin 2000). We adopted both the HC2 and HC3 estimators, and following White (1980) and Davidson and MacKinnon (1999), they are specified as: whereh ii is x i 0 ðX 0 XÞ À1 x i : As expected, the coefficients of the variables for HC2 and HC3 estimators are the same. The difference between the two estimators is apparent in their SEs. Therefore, the efficiency differences of HC2 and HC3 estimators are clearly based on the SEs of coefficients, particularly that are mostly affected by heteroscedasticity. The HC2 SEs for all coefficients were found to be lower than those for HC3, which is consistent with Kassie, Abdulai, and Wollny (2011) and Terfa et al. (2013). The t-values of the former are inflated possibly leading to erroneous rejections of the null hypothesismaking inference from HC2 unreliable. In a situation like this, MacKinnon and White (1985) show that HC3 outperforms HC2. Therefore, in this study, the HC3 estimation results were used for inferences.
The second option to deal with unknown form of heteroscedasticity employed in this study is the feasible general least square (FGLS) (Long and Ervin 2000). FGLS SEs do not simply relax constant variance assumption unless correct model specification for heteroscedasticity is employed (Cameron and Trivedi 2010). In our estimations, FGLS could not significantly change the distribution of the residuals. Therefore, following Cameron and Trivedi (2010), we employed weighted least square (WLS) estimation procedure which is similar to FGLS but uses robust SEs that do not rely on a model for heteroscedasticity. Considering a general case of heteroscedasticity given as 6 4 3 7 7 5 Because Ω À1 ¼ P 0 P, P is n × n matrix whose 1diagonal element is 1= ffiffiffiffi ffi ψ i p .
By premultiplying P on y and X, we get :: x nk = ffiffiffiffiffi ψ n p 2 6 6 4 3 7 7 5 The WLS estimation will therefore be an OLS on y* and X*.
The estimated parameters are however interpreted in terms of the untransformed model.
The third option used in this study is structural heteroscedastic-in-mean (SHM) estimator due to Barrett, Bellemare, and Osterloh (2003), which they used in their study of the determinants of price and price variability in Northern Kenya. Kassie, Abdulai, and Wollny (2011) also used it in their estimation of implicit prices of indigenous cattle traits in central Ethiopia. Because SHM is a derivative of the generalized autoregressive conditional heteroscedasticity-inmean (GARCH-M) model, it is useful to analyse the relationship between asset price and volatility (Barrett, Bellemare, and Osterloh 2003;Kassie, Abdulai, and Wollny 2011). The SHM regression model is a multivariate regression of price and its SD simultaneously on overlapping set of explanatory variables to check the presence of spatial and seasonal price variability. The model can be specified as: where σ is the conditional SD of the natural logarithm of priceallowing for the existence of a direct correlation between price levels and variability (Barrett, Bellemare, and Osterloh 2003) and α is its coefficient. Ζ is a vector of selected exogenous variables in X. λ is a vector of parameters, and υ is iid error term where the two equations are estimated simultaneously. WLS and full information maximum likelihood (FIML) estimators yield consistent and efficient estimates of parameters of interest, whereas OLS estimator of HCCM provides only asymptotically consistent parameters (Barrett, Bellemare, and Osterloh 2003;Cai and Hayes 2008).
Variables that were included as explanatory variables in all of the hedonic price models can be broadly categorized in to three, namely: goat attributes, market locations and market agent socioeconomic characteristics. Other variables used to explain the variation in purchasing and/or selling decisions include the type of festivities that are observed during the different seasons of the year.

IV. Results and discussion
Goat production and marketing in Ethiopia Using an elevation of 1500 m above sea level as a crude cut-off point, Ethiopia is broadly divided into highlands (39%) and lowlands (61%). The lowlands are dominated by arid to semi-arid climates (i.e. up to 180 growing days and 700 mm of precipitation per year). The lowlands are home to a diverse array of pastoralists who depend, to a high degree, on livestock for their sustenance (Jahnke 1982;Pratt and Gwynne 1977). Out of the country's total livestock population, the lowlands constitute 100% of camels, 73% of goats, 25% of sheep and 20% of cattle (Alemayehu 2007).
Goats contribute significantly to the livelihoods of a large portion of the population in low-input and smallholder production systems where they serve as sources of subsistence, cash income and social capital. They are also known to be among the most liquid assets that can be used as store of value and a quick source of cash by millions of keepers in rural Ethiopia (Ayalew 2000;Tibbo 2006). Consequently, goats play vital roles in ensuring food security among rural households, often being the only asset possessed by poor families (Peacock 2005). Moreover, goats are critical as a safety net for major shocks such as crop failure or family illness (Ayalew et al. 2003).
Markets have long been a feature of East African pastoralist systems (Akililu 2008). Of course, the role of markets can differ according to environmental conditions and the nature of farmers' and/or pastoralists' assets (Alary et al. 2012). For instance, in dryland areas, livestock are the major marketable assets and a major social capital serving as source of sense of security for their keepers (Binswanger and McIntire 1987;Turner and Williams 2002). The multifaceted functions and agroecologically diverse challenges of livestock rearing make livestock marketing strategies very complex (Alary et al. 2012). The most widespread and serious concerns of pastoralists in the African and West Asian regions are the problems of low and variable livestock producer prices (Desta 1999;Smith, Barrett, and Box 2001). In addition to the non-market orientation of the livestock production systems, the poor accessibility of the markets, lack of market information, lack of market infrastructure, lack of collective actions and poorly designed government interventions are among the key causes of unrewarding markets for the rural communities. In the East African lowlands including Ethiopia, these market challenges and complex marketing strategies lead to consistently low marketed offtake rates (Desta 1999).
FGDs and KIIs revealed that goat is the major marketable asset for both smallholder farmers and pastoralists in all the study areas. Farmers and pastoralists are strategic in marketing their goats under challenging circumstances. Extents of cash needs, price of goat and access to inputs for goat production, they said, are crucial factors for decision on goat transactions. Normally, farmers sell when they are in need of cash, goat prices are high and feed is scarce and buy goats when conditions are the opposite. Most households are responsive to price changes in general and purchasing prices in particular. On the contrary, there are situations where farmers/pastoralists decide to sell or buy when prices are not favourable. For instance, they could sell when they are faced with dire cash needs and buy when they have surplus cash without much regard to the level of goat prices, which is consistent with the findings of Winrock-International (1983). The justification is that goat is the main source of cash in times of need and a means of wealth accumulation at times of plenty. Likewise, flock restocking that is the process of reestablishing flocks, if and when it is needed, takes place often regardless of prices in order not to miss the critical period. To some extent, feed shortage is also a cause to sell goat, whereas feed abundance is a cause for buying more goats.
Cash needs are among the most important factors that determine smallholder goat producers' marketing decisions. When there are high cash needs and surplus cash, smallholder farmers and pastoralists are less responsive to price changes. Having surplus cash to spend does not carry as much urgency as dire cash needs in certain conditions. As a result, 36% of households made transactions when they have high cash needs and 12% when surplus cash is available regardless of price. Cash needs vary across seasons, the classification of which follows rainfall distribution and length. In Bati district which is an agro-pastoralist area, cash needs become high during the main rainy season (locally known as Kiremt) and immediately after the crop harvest season (locally called Meher). The Kiremt season is from July to September and is characterized by shortage of food and feed. The Meher season is from October to January and it is the main crop production period. This is a period of relatively better food and feed availability, while there are still high cash requirements to pay for agricultural labour, government taxes, children schooling and many cultural and traditional holidays. During the Kiremt season, demand for goat decreases while supply of goat increases resulting in low goat prices. As a result, smallholder farmers who have surplus cash are at an advantage, while those farmers who have cash shortage are at a disadvantage. In contrast, price of goat is high during the Meher season. The main reason is high demand for goats for weddings and sacrifice-locally known as 'sedeka'. Thus, smallholder farmers who opt for selling their animals during the Meher season maximize the monetary benefit that can be obtained from goat production. From point view of feed availability, smallholder farmers who buy goat during Meher can also maximize their future benefits.
Unlike Bati, the farming system in both Shinelle and Yabello districts is purely pastoralist where almost all the agricultural lands are not cultivated but covered with natural vegetation, which serves as forage. As the availability of forage is influenced by the amount of rainfall, pastoralists in Shinille district design their marketing strategies in such a way that they sell goats to fetch better price and to satisfy high cash needs during rainy seasons. During the rainy season, goats command higher price due to the high demand and also possibly due to good body conditions. During drought season, goats command low price essentially because of low demand and high supply of goats with poor body conditions as many farmers try to sell as a way of dodging the feed shortage for their flocks. Re-establishment of flocks through purchases in the aftermath of droughts is the other marketing strategy, which is implemented only if the drought causes high damage on the flock. In such circumstances, prices of goats go up due to low supply of goats and/or high demand for goats.
Marketing strategies of pastoralists' in Yabello district is dictated not only by cash needs and availability but also by availability of feed. Cash needs are high for purchasing of food commodities and hiring of labour during drought seasons and immediately after the main rainy season, respectively. In contrast, cash becomes relatively surplus during the cool dry season (i.e. from May 15 to July 15). This shows that even if pastoralists develop good marketing strategies factoring various criteria, recurring and severe droughts usually put them at a disadvantage. In the face of such precarious production and marketing conditions involving high risks, it is important to identify important factors at the disposal of farmers/pastoralists, which they can incorporate in their marketing strategies in order to make use of the available market opportunities to their own advantage.
Summaries of data collected on important variables show that more than 60% of households sell goats when goat prices are high and buy goats when goat prices are low. Liquidity constraints are also important in farmer/pastoralist sales decisions where 35% of farmers sell when they are in need of cash ( Figure 2).

Empirical analysis results
The implicit prices of live weight, age group, coat colour and presence of horn were estimated using the models discussed above. The models included the seasonal variables to control for the changes that happen in price due to time of purchase. The average price of goats ranged from 660 Birr in Yabello to 712 Birr per head in Bati market. The live weight of the goats ranged from 19 kg in Dire Dawa to 25.6 kg in Yabello. This shows that although the goats in Yabello are bigger on average, the prices they fetch are not as high the domestic consumer-dominated markets of Dire Dawa and Bati (Table 1). Most of the goats brought to the markets are quite young such that more than 75% are less than three years old. Similarly, white and mixed goat color goats dominate the market. An important attribute of goats is having a horn or not and 81% of the goats transacted had horns. Three-fifth of the goats purchased by the sample households were male goats and of medium body condition (Table 1).
The results of the alternative heteroscedastic hedonic price models are comparable where the signs of the coefficient estimates (particularly of the statistically significant variables) are identical and their magnitudes show only slight differences across the models. Most of the animal attributes (i.e. weight, sex, age and condition) are significant at 1% level of statistical error in all models (Table  2). Moreover, the coefficient estimates of the linear and quadratic terms of weight and the dummy variables for sex, age, body condition, color and presence of horn have similar signs in  all models. Market place has similar signs and significance levels in all models. The dummy variables representing the main occasions of animal transactions (Easter, Ramadan and Ethiopian New Year) have similar signs and significance levels across all models. The dummy variables representing types of buyers have also similar signs and significance levels in all models.
Test for heteroscedasticity showed that the constant variance assumption was violated in both the specific market and pooled data models. For the sake of simplicity, the pooled data are used to estimate alterative heteroscedastic hedonic price models. In addition to evaluating the direction and magnitude of parameter estimates of HCCM, WLS and SHM models, we employed the most appropriate and more informative Akaike and Bayesian information criteria. The FIML estimator was appropriate for estimating SHM. OLS estimation was employed to analyse hedonic price model using HCCM. Heteroscedastic hedonic model was also analysed using WLS estimator. Based on the result of Akaike and Bayesian information criteria, the modified SHM model showed an improvement over the single equation models including HCCM and WLS, while the WLS showed an improvement over HCCM. Therefore, only the results from the modified SHM model estimated using the FIML are discussed in this article. Tables containing results from the other models are attached as an appendix.  SEs in parentheses. *p < 0.05, **p < 0.01, ***p < 0.001.

Implicit price of attributes
Buyers attached significant implicit price for some goat attributes. Live weight has a positive and significant influence on observed goat prices which is consistent with previous studies on goats in Ethiopia (Ayele et al. 2006;Dossa et al. 2008). The quadratic term for weight, however, has a small but negative significant effect on price of goat showing that weight has a positive impact up to a certain maximum level beyond which it will have a negative effect (Table 2). Specifically, goats up to 58 kg command higher price and then price starts to decline as weight increases. This is consistent with smallholder farmers' and pastoralists' traditional practice of castrating male animals targeting weight gain by keeping it for long period with little change in feeding. Castrated male goats, on average weighting 37 kg, have a 15% price premium over intact males. Under traditional production system, body condition is used as a good indicator of meat quality and goats that are marketed with good condition command prices that were about 6.7% higher than those in poor condition. Such price premium is consistent with other previous studies (Ayele et al. 2006;Teklewold et al. 2009). The only attribute found to be important in all models in determining the price of goats was age. Goats that are marketed at the age of less than one year have no significant price discount over those marketed at the age of between 1 and 3 years. A possible explanation for this reason is that the purchasing criteria for abattoirs' that target the export market are male goats that are less than one year and have weight of 13-30 kg leading yearlings to fetch price as high as those aged between 1 and 3. Smallholders and pastoralists are aware of this unique opportunity and hence often target the export market with male goats that fall in the desired weight range. In contrast, goats that are marketed at age 1 have significant price discount over those marketed at ages of between 3 and 4 years and ages of above 4 years. Because of lighter weight of young animals, some local buyers compromise quality of meat, and they resort to aged goats. Closer look at the coefficients of some attributes reveals the importance of the attributes beyond the amount and quality of meat. For example, intact males have a 9% price premium above females, although the average weight of both intact males and females is the same (22 kg) ( Table 2). Moreover, while presence of horn does not have any bearing on the quality and quantity of meat, model results show that goats with horn have price premium over goats that are not horned.
Effects of the types of market agents, location and marketing seasons on price. The econometric models estimated showed that the observed price of goats is a function of both the goat attributes and individual buyer attributes. Buyers' objective plays an important role in price formationfor instance, those who buy for consumption and traders target to optimize their satisfaction and profit, respectively. Traders bought goats at discount price over consumers. Unlike traders, butchers/restaurants often bought big goats aiming a profit making through value addition at a price premium over consumers. Farmers/pastoralists bought goats for stock replacement and flock re-establishment at a price discount over consumers. They often purchase female goats at or below first kid delivery age.
Occasiondenoting time of transactionwas found to be an important determinant of goat price. Occasion also shows the interplay of market demand and supply forces. Some marketing strategies are widely applicable and some others are area specific. In the process of designing marketing strategies, the role of the type of the production system is limited; instead, modes of rainfall and seasons of the year play vital role. In all study areas, producers identified festive seasons as periods of high prices for goats, and hence, they try to exploit the opportunities these periods avail as much as they can. These results are consistent with the local practice as smallholder farmers and pastoralists have clear understanding of how timing of transaction (festive verses non-festive and rainy verses dry seasons) plays a vital role in price formation and hence often target opportune periods for goat purchases and sales. Even though data on whether the transactions took place during the dry or rainy periods were not collected, the high goat prices during specific festive and fasting events such as the Ethiopian New Year, Meskel and Ramadan all of which take place during the main rainy season also reflect the importance of rains in price formulation. For example, goat marketed around the Ethiopian New Year (celebrated nationally) command the highest price. Ramadan is the second most expensive marketing season next to Ethiopian New Year. This is attributed to the high demand for meat during Ramadan both in the country and in the Middle East countries that are the main destinations for Ethiopian goat exports. Goats marketed around Christmas and Meskel had a price discount compared to prices during fasting periods. Ethiopians have a culture of celebrating these festivals together where they commonly slaughter cattle collectivelya practice locally known as 'kircha' than slaughtering goats at individual household level.
The parameter estimates for market locations in the models imply spatial price variability. Bati market is where goats fetch the lowest price, whereas Dire Dawa market is where goat prices are highest (Table 2). These spatial price variations are a reflection of the number of participant buyers and sellers in the market. Bati market is essentially a buyer's marketwith many sellers and few buyers, whereas Dire Dawa market is relatively a seller's marketfew sellers and many buyers.

V. Conclusions and recommendations
By employing effective marketing strategies, smallholder and pastoralist goat producers have great potential to exploit market opportunities to their own advantage. Using goat marketing data from three main livestock markets in the Ethiopian lowlands, this study provides empirical evidence on smallholder livestock producers' marketing behaviour. One of the main findings of this study is that smallholder farmers and pastoralists do speculate the temporal pattern of goat market prices and develop their marketing strategies to maximize their benefits. However, extreme weather conditions often reduce their ability to implement their optimal marketing strategies.
Goat price formulations are influenced by a number of animal attributes, particularly those that are related to the amount and quality of meat. The heterogeneity in the types of buyers has also a significant effect on price formation reflecting the lack of competitiveness in the Ethiopian goat markets. Goats marketed during typical festive periodsparticularly Ethiopian New Year and Ramadanfetch premium prices showing the importance of both national and international demand. Goat markets in the lowland areas of Ethiopia are exposed to price variability. While sedentary and pastoral goat producers were observed following price signals for their marketing 60% of the time, they were also selling at times of low goat prices or purchase at times of high prices about 40% of the time. This shows that risk management has a huge bearing on their marketing strategies, thereby reducing producers' ability to take advantage of market opportunities for specific traits and during specific periods.
Empirical results from both the econometric models based on individual surveys and qualitative analysis of data from FGDs and KIIs gave consistent results regarding marketing behaviour of households. Goat keepers know the seasons during which enough feed is available and hence they are under no pressure to sell. These seasons also coincide with times of relative cash abundance and hence goats command high prices. On the other hand, goat keepers identified seasons where cash requirement and feed shortage are very high overlapping with the periods in which they are compelled to sell. Obviously, these are periods when goat prices go down to their lowest levels. This is because supply of goats with poor body conditions increases due to producers' intentions to minimizedrought induced livestock production risk. These marketing strategies result in variations of goat demand and supply and hence prices across seasons. Goats with good body conditions generally command price premiums. However, supplying goats with good body conditions is often challenging under arid production conditions and more so in cases of drought.
Analysis of hedonic prices and qualitative market information provide analogous evidences that producers cannot fully exploit market opportunities that are available during specific seasons for specific animal attributes. This is because of one or a combination of the following reasons: (1) they are relying on traditional production systems; (2) agricultural products in general and meat demand are mostly price inelastic; and (3) farmers keep goats for multiple purposes including wealth accumulation and hedging and to ameliorate pressing cash needs in order to avoid the need to sell cattle. All these empirical evidences suggest that households' marketing strategies do consider not only higher prices but also the need for securing some non-cash but highly liquid assets such as goats which can serve as store of value. This does not, however, preclude the possibility of sales if and when conducive environment is created for taking advantage of high prices from goat selling. As a result, goat supply is seasonal due to price speculation by smallholder farmers and pastoralists.
A number of smallholder producer-appropriate technologies are available that can help goat producers overcome some of their major constraints. For instance, kidding periods can be planned using simple synchronization methods. Moreover, using community-based breeding strategies, farmers can produce goats that possess traits that are demanded by specific niche markets in specific periods ). Development of different feed resources can also be used to overcome feed shortage problems and increase feeding efficiency. Access to market information is essential for farmers to maximize their benefit by enhancing competitiveness in the market while also making the farmers responsive to seasonal and spatial changes in the goat market. Therefore, the introduction of different goat production and forage technologies and networking and institutionalization of markets so as to provide market information and building market infrastructure can improve the existing production conditions and marketing strategies of smallholder and pastoralist goat producers, thereby help improve their livelihoods.

Disclosure statement
No potential conflict of interest was reported by the authors.