A Framework for Predicting U.S. Catfish Prices: The Leading Indicators Approach  

Catfish prices in the United States (U.S.) have been quite volatile over the past few years. A simple but robust price prediction model could help farmers to better understand determinants of fish prices and long-run price trend. In this paper, we propose a leading indicators model to reveal dynamic adjustments in long-run catfish prices in the US market. This paper highlights how US catfish prices are influenced by multiple key factors.

We followed a three pronged approach. Firstly, we set up an algorithm (leaps-and-bounds model) to select useful "leading indicators" for extracting information about long-run catfish price formation. A leading indicator is a variable whose value can be obtained in advance of a related variable for prediction. Secondly, after identifying the appropriate indicators, we specified and estimated a model using Autoregressive Distributed Lags (ADL) algorithm of Kripfganz and Schneider (2015). Finally, we used in-sample parameters to predict out-sample values. Also, we checked the robustness of our models with multiple approaches.

Our leading indicators approach has a subtle link with the approach of using time series based historical simulations, but can be viewed as an alternative of time series based historical simulations. A unique contribution of this paper is the selection of the best subset of leading indicators. In addition to the identification of the target variable and its leading indicators, we also considered how to select a proper subset from a full model of indicators for parsimony of the prediction model. In this paper, the best prediction model was obtained by comparing prediction performance across combinations of leading factors.

We find that pond-bank price of catfish in a particular month is a function of  four "leading" indicators (four- month lag catfish price, four-month lag seafood price index, three-month lag gasoline price, and current  fish feed price) and some of their derivatives. Our results indicate predicted values well approximate actual in-sample price movements.