Three essays on the applications of neural networks and genetic algorithms in agricultural economics

Applied economic researchers are always looking for new techniques to improve the understanding and interpretation of economic theory and behavior. This study provides three essays which examine two such techniques, neural networks and genetic algorithms. Neural networks are a class of mathematical functional forms and/or associated estimation algorithms. A benefit of neural networks is their flexibility and universal function approximation. Genetic algorithms are an optimization technique that encourages improvement over an iterative process with minimal mathematical structure.; The first essay examines the use of neural networks in hedonic pricing. The study examines derivative impacts and out-of-sample prediction for farmland prices between a feedforward backpropagation neural network and an ordinary least squares regression. Results indicate the two methods are similar in derivative impacts and price prediction. It is concluded that ordinary least squares is a robust technique in this case for evaluating farmland prices.; The second essay examines the use of genetic algorithms to select parameter estimates that satisfy theory considerations for cost functions and output-compensated factor demands. The genetic algorithm uses a feedforward neural network as its underlying structure. Results of the genetic algorithm are compared to an econometrically estimated system of equations. The results indicate that the genetic algorithm can be used to impose regularity conditions, while providing better in-sample fit. Genetic algorithms may provide useful in cases with concerns about estimation efficiency when imposing restrictions.; The third essay examines various techniques for the financial classification of farms. The study uses ordinary least squares, ordered multinomial logit, neural networks, and genetic algorithms. The study analyzes financial ratios for a sample of farms. OLS results indicate that solvency, financial efficiency and profitability have significant impact on a farm's financial performance. Liquidity was insignificant. The ordered multinomial logit model was used to predict financial performance. It did moderately well for predicting good and fair financial performers, but had difficulty predicting poor financial performers. Results of the neural network were poor as the model overgeneralized, predicting 95% of all farms as fair financial performers. The genetic algorithm lacked consistency and was inconclusive. Future work is needed on the latter two approaches.