79 Stochastic Frontier Analysis for Austrian Agriculture Productivity Analysis Martin Kniepert and Johannes Sauer 1 Abstract - Efficiency analysis is one of the major tools helping to improve economic productivity. In Austria the Data Envelope Analysis has been applied often in recent times, and also methods comparing similar farms with respect to some distinct criterion are implemented. The Stochastic Frontier Analysis could be another tool. It became more prominent in recent years in international literature, but has not been implemented for the Austrian agriculture. This paper first explains the principle differences of these three approaches to show the distinct character of the Stochastic Frontier Analysis. It then outlines some possibilities to use this method. Data from the Farm Accountancy Network are used to discuss examples. 1 EFFICIENCY ANALYSIS Efficient use of resources is very much at the heart of any economic reasoning. It is essential for the competitiveness of firms, for the justification of public spending e.g. on agricultural policy programs, and for a sustainable use of resources in general. Microeconomic theory offers a strong basis for this kind of analysis. Methods could closely build up on it, like the Data Envelope Analysis (DEA) following an optimisation approach (Hambrusch et al., 2006) or the Stochastic Frontier Analysis (SFA) relying on econometric evaluations. Other methods – namely matching - following similar intentions stay closer to descriptive evaluations (Coelli et al., 2005). Each of these approaches has its advantages and drawbacks. The SFA has – to the knowledge of the authors – not yet been implemented in research on agriculture in Austrian. Considering the prominence of the SFA in the agricultural economics literature and its acceptance as quasi text-book standard it should be worthwhile to look at its conceptual distincition and its empirical usability. Following Kumbhakar and Lovell (2000:1ff) the usual production, profit or cost functions as known from standard microeconomics textbooks are characterised by the assumption that firms optimise successfully. The fact, that individual firms are empirically not found exactly on an estimated function is then attributed solely to measurement errors or influences outside the reach of the firm (weather etc.). In this sense, the deviation of firms 1 Martin Kniepert is from the University of Natural Resources and Applied Life Sciences Vienna, Institute of Sustainable Economic Development, Vienna, Austria (martin.kniepert@boku.ac.at). Johannes Sauer from the University of Manchester, UK, has been guest professor at the University of Natural Resources and Applied Life Sciences Vienna during the winter term 2010/11 (johannes.sauer@manchester.ac.uk) from estimated functions is assumed to show the typical normal distribution of statistical noise. SFA maintains the existence of normally distributed statistical noise but makes an additional assumption for differently distributed technical inefficiencies (e.g. half-normal distributed). Thereby, the analysis of production, cost, and profit is recast from traditional functions towards frontiers. The differences between SFA and DEA then boil down essentially to two characteristics (Passel et al., 1996 or Kebede, 2001): 1. The SFA as an econometric approach attempts to distinguish noise from inefficiency. The DEA as programming approach is non-stochastic and lumps both effects together. It then calls the result inefficiency. 2. The SFA is parametric, and confounds the effects of misspecification with inefficiency. The DEA is non-parametric and less prone to this type of specification error. The statistical problem of an appropriate identification of control and treatment groups is tackled by the matchingapproach, selecting data for farms which can be compared pairwise, showing differences only with respect to e.g. a particular policy instrument. This should help to limit the identification bias (Kirchweger et al., 2011). The advantage of the SFA could here be seen in its comprehensive – less cumbersome – approach, its disadvantage in its restriction to somefunctional specification and in the danger of the usual econometric fallacies like the mentioned identification problem. It is obvious that each of these approaches can be most promising, justdepending on the research question addressed, when encountering different data availability, or different research backgrounds. SFA might also find a place in this context in Austria. DATA FROM AUSTRIA In principle, the data situation in Austria allows extensive econometric work. Bookkeeping data (as part of the European Farm Accountancy Data Network – FADN) are available now for seven years following the same consistent and detailed coding system. Earlier data are available as well, though showing different coding and conceptual features. Each year more than 2.000 sets of farm data can be evaluated. The data can pooled or combined to panel data. Fluctuation of farms requires decisions on whether or not tobalance panel data sets.