Forecasting of FX-Trading with Consideration for the Impact of News

This paper is concerned with a forecast of propriety of FX trading by quantitative analysis. The quantitative analysis in this paper uses a Naive Bayes model and a pair wise EPDW matrix. The Naive Bayes model employs a binomial distribution for the likelihood. Both the prior and posterior probabilities are assumed to be a beta distribution, so a conjugate distribution turns into the beta distribution. The denominator of the basic formula of Bayes becomes constant, so that their computation is very easy, unlike the MCMC-method, etc. As an initial value of the subjective probability of Bayes, we use the probability which is carried out the trade of the last term of a log it model. The forecasting value, which is the propriety of the trade by the expected value of the posterior probability of this naive Bayes model, has diverged largely with the actual value which becomes clear later. Therefore, in order to improve this value, this paper is introduced an influence of news such as economy, politics, disaster, and war. By a pair wise EPDW matrix, the degree of influence to trading of the news is acquired as Pnews (probability by news). As a result, the modified forecasting value E[p]mdfy is produced by compounding E[p] and Pnews. This E[p]mdfy is extremely close to the actual value. It is also shown in this paper that, in the trading of USD/JPY in 2011, the rate of coincidence of the forecasting value and the actual data becomes a good result of 91.67%.