Developing the Criterion of Exchange Quotation Prediction Quality for Metal Trader Decision Support

• Published in: Vestnik Volgogradskogo gosudarstvennogo universiteta. Serii͡a︡ 3, Ėkonomika, pravo Ėkonomika, pravo, 2019-03 (1), p.116-124
• Main Author: Paytyan, Karen
• Format: Article
• Language: Russian
• ISSN: 1998-992X
• DOI: 10.15688/jvolsu3.2019.1.12

The main purpose of this article is to present a criterion for assessing the quality of forecasting nickel exchange prices to support the decision-making of a metal trader, taking into account the specifics of their commercial activities. We consider the commercial operation of a metal trading company, which consists in purchasing a metal with a view to its resale in 14 days. Most of these operations occur in nickel-rich alloys, which suggests that their prices are largely determined by nickel quotes on the LME (London Metal Exchange). Therefore, metal traders need forecasting models with a long lead time series with a high degree of volatility. One of the first assumptions suggests that in order to ensure profitable trading, it is necessary, first of all, to correctly predict the direction of price change. On the other hand, it is intuitively clear that the better the model, the more likely it will be to guess the further trend. However, it remains an open question as to what accuracy of the forecast will provide the so-called break-even point to the metal trader. This work is devoted to the search for this facet. Another assumption that the quality of any model depends on the degree of volatility of the predicted time series helps to find the exact point from which we can talk about the possibility of successful trading. Taking into account these provisions, a criterion has been developed for assessing the quality of forecasting models, which makes it possible to state with high probability that the use of a forecast that meets it will ensure commercial success for the metal trader. It is also important to note that the quality of the model can be judged only after comparing its accuracy with the characteristics of the considered time series. The prediction error itself does not give an exhaustive picture of the quality of the applied model.