Accounting Information and the Prediction of Farm Viability

Abstract:

Farms make little use of accounting and until now have been largely excluded from the scope of accounting standards. However, we hypothesize that the use of accounting based information can significantly improve the explanation and prediction of farm viability/failure. Two dichotomous logit models were applied to sub samples of viable and unviable farms in Catalonia, Spain. One model included non-accounting-based variables, while the other also considered accounting-based variables. It was found that, when accounting variables were added to the model there was a significant reduction in deviance.

Background

A voluminous literature on failure prediction models has been developed since Beaver (1966) and Altman's (1968) seminal studies.

Research into failure prediction in agriculture is comparatively scarce. To the author's knowledge the earliest empirical studies in this area began with Reinsel and Brake (1966). Krause and Williams (1971), Bauer and Jordan (1971), Johnson and Hagan (1973) and Dunn and Frey (1976) used discriminant models to assess farm loan repayment. These studies were undertaken at a time in which the indebtedness of US farms was increasing and this was more and more difficult to manage (Murdock and Leistritz, 1988, p. xiii). The subsequent agricultural crisis and high incidence of farm and agricultural bank failures, coupled with unexpected losses by agencies providing loans to farmers during the mid 1980s, stimulated new research (Ibid.). Shepard and Collins (1982) explained farm failure at the macroeconomic level, while Grisley (1985), Griffis (1988) and Lins et al. (1987) measured the financial health of farms.

Subsequent research into attempts to cope with the financial crisis of US farms involving explicative or prediction models can be classified in three groups. Some analysts focused on the economic viability of farms: Kauffman and Tauer (1986) and Smale et al. (1986) used binomial logit models to do this, while Adelaja and Rose (1988) used a simultaneous-equation model. Another group used multinomial logit models (Lines and Zulauf, 1985), ordered logit models (Lines and Morehart, 1987; and Wadsworth and Bravo-Ureta, 1992) and a multiresponse ordered model (Carley and Fletcher, 1988) to explain and predict various degrees of financial health. The third group, from which we highlight the work of Mortensen et al. (1988), Turvey and Brown (1990) and Knopf and Schoney (1993), used binomial logit models to predict farm loan repayment. In the same group, Turvey (1991) compared the predictive accuracy of the linear probability model, discriminant analysis, logit and probit.

Download accounting information journal: ziddu