Managing price risks using and local polynomial kernel forecasts

Abstract

This study contributes to understanding price risk management through hedging strategies in a forecasting context. A relatively new forecasting method, nonparametric local polynomial kernel (LPK), is used to forecast prices and to generate ex ante hedge ratios. The selective multiproduct hedge based on the LPK price and hedge ratio forecasts is in general found to be better than continuous hedging, no hedging and alternative forecasting procedures. Selective multivariate hedging using the LPK is found to improve hog producer's expected returns. The findings indicate that combining hedging with forecasts, especially when using the LPK procedure, can improve price risk management.

Notes

¹Forward pricing is not the only alternative for managing pricing risk. Floor pricing through the options market provides a minimum price while allowing the producer to take advantage of any higher prices. Forward pricing usually provides more price protection if prices fall, but precludes gains from higher prices.

²Another potentially important input is soybean meal. However, the amount of soybean meal consumed per hog is approximately only 10% of total value of feed grains while corn comprises around 85%. Hence, soybean meal is not included in the analysis.

³Another framework to trigger the selective hedging decision was examined where the hog producer hedges when the cash price is forecast to fall. The results are not significantly different from those explained above. The results are available on request.

⁴Certainly, the use of contracts in hog production is increasing in the U.S. An estimated 83% of hogs in 2001 were transacted under various types of contracts (Grimes and Meyer, 2001). For many types of contracts, producers are not subject to fluctuations in input and output prices. Nevertheless, about 17% of hogs transacted rely on the spot market, and some contracts have windows allowing for price risks.

⁵Several approaches, such as additive modelling, partially linear modelling and modelling with interactions, have been proposed to deal with this. See Fan and Gijbels (1996) for more details about these approaches.

⁶The final live hog futures contract is December 1996 and the first lean hog futures contract is February 1997. Lean hog prices are converted to live hog prices by multiplying by 0.74 to get overall hog hedge ratios, reflecting the average carcass yield of 74% from live hogs.

⁷The optimal bandwidth was re-calculated whenever the model was re-estimated for each forecast. VAR procedures are not described for brevity. See Clements and Hendry (1998).

⁸The VAR(2), which reflects two lags for corn and hog price, produces the smallest forecast errors among various VAR models considered. For brevity, only results of VAR(2) are presented.

⁹The BEKK GARCH (1,1) model is based on Baba et al. (1989).

¹⁰The term one-to-one hedge is used instead of NAÏVE hedge to differentiate from NAÏVE price forecast.

¹¹Sources are the USDA references listed in the bibliography.

¹²These procedures are not needed to forecast price in the case of the continuous hedge. Thus, similar to Table 3, the results for the continuous hedge presented in Tables 4 and 6 are generated by using the hedge ratio forecast only.