Water demand estimation in industry and agriculture

In crop production and other agricultural activities, especially in arid and semi arid areas, water is an important input which usually cannot be obtained through market activities.

Therefore, it is categorized as a non-market good and its value has to be estimated through direct and indirect non-market valuation methods. Young(2005) divided all em-pirical economic valuation techniques of water into the two broad categories of inductive and deductive techniques. Inductive techniques employ inductive logic by statistical or econometric procedures in order to generalize from individual observations. In contrast, deductive methods use logical processes to draw a particular conclusion from a general idea. Deductive techniques are built on constructed models and empirical assumptions suitable for the proposed projects or policy. Inductive techniques are constructed from statistical analysis and are based on observed transactions, responses to questionnaires, or secondary data from government reports. Their accuracy depends on the represen-tativeness of the samples, appropriate statistical distribution and the functional forms

which are used to fit the data. Inductive methods have the advantage of reflecting actual economic behavior, but they are weak in predicting future behavior. Future behavior and valuation may need to be forecasted by developing conclusions based on sample pa-rameters (Young, 2005, p.45). Moreover, inductive methods demand high statistical and computational skills. Some examples of inductive valuation methods are econometric es-timations of production and cost functions, hedonic valuation methods, observations of water market transactions, and contingent valuation methods (CVM). In contrast, deduc-tive techniques use empirical studies of production or consumption processes, published government reports and experts’ opinions. The accuracy of the results of deductive rea-soning depends on the validity of the model specification and database employed for the model. They are flexible and they can reflect any desired future economic and techni-cal conditions. Although these methods are fast and inexpensive, caution must be used with their application as any simple process may result in conceptually incorrect results.

These broad categorizations do not reject the fact that for any deductive method, there must be some initial empirical and inductive steps, just as an inductive approach needs some deductive reasoning to proceed (Young,2005, p.44-46). Some examples of deductive valuation methods include residual imputation methods, changes in net rents, mathemat-ical programming approaches, and computable general equilibrium (CGE) models. Since inductive reasoning has been followed in this study, the literature review focuses mainly on econometrical empirical studies.

As it was mentioned above, econometric models are one of the main tools for valuation of water and estimation of the demand for water when farm-level microeconomic data in agriculture or the same data for industrials enterprises are available. Three main methods for estimating the demand function for water are the reduced form approach, deriving demand function from the cost function (Shephard’s lemma) or the profit maximizing input demand function (Hotelling’s lemma).

Aggregated regional or district-level data have been regularly used for demand estima-tion. Nieswiadomy (1985,1988) represent one of the first attempts to estimate irrigation water demand. In the latter of these two studies, the water demand elasticity was es-timated at only -0,25 based on panel data from the period of 1970-1980 for cotton and

grain sorghum production on the high plains of Texas. Ogg and Gollehon (1989) applied the reduced form approach to estimate the water demand elasticity for field crops with the help of county level data. The results showed that irrigation water was highly in-elastic. Renzetti (1992) has done one of the most comprehensive studies on industrials water demand among Canadian manufactures using cost function approach. The results showed inelastic demand (-0,38). One of the prominent irrigation water demands studies has been done by Moore et al. (1994) in the central plains of the United States. They checked three different models of short-run input use. These models were the allocatable fixed input model, the variable input model and the satisficing model. Profit maximiza-tion, and as a result Hotteling’s lemma, was considered for direct estimation of irrigation water demand. They concluded that the fixed allocatable input model provides a better explanation for multi crop water use. Moore et al. (2000) estimated the supply functions for multioutput irrigators in Pacific Northwest of US by using a tobit model in order to analyse the economic welfare of the producers. Farm level survey data (1986, 1990) has been used for the study. This experiment predicted increases in the water pumping costs and decreases in the producers’ surplus. One of the recent econometric studies on water demand is Schoengold et al. (2006). They estimated water demand by using a linear function. Water consumption was regressed on water price, wage, farm characteristics and fuel price without any discussion about underlying cost or profit functions. Water price was significant with a negative sign. They conclude that better management alone can result in significant water conservation. Mullen et al. (2009) used a Heckman model with profit function to analyse the demand structure for water and water decision issues for corn, cotton, peanut and soybean in Georgia, USA from panel data sets in 1988, 1994, 1998 and 2003 (USDA-Farm and Ranch Survey). Pumping costs were used as a proxy for water price. Results indicated that the water demand was modestly affected by water price (with elasticities between -0.01 and -0.17), but more so by crop price (with elasticities between 0.5 and 0.82). Results also suggest adoption of lower pressure irriga-tion systems does not necessarily lead to lower water applicairriga-tion rates on corn, cotton, peanuts, and soybeans. The coefficient for the Mills Lambda Ratio was not statistically significant for any of the Heckman models, raising doubts over the applicability of this

model to irrigated crop production in southwest Georgia. In other words, the first stage of the Heckman model, namely crop selection, does not appear to be significantly affected by water-related decisions. Finally, it must be mentioned that the possible lack of farm level data is the major issue with this approach.