Using similarity measures in prediction of changes in financial market stream data—Experimental approach

Abstract

In this study, we experimentally investigated the possibilities of using selected similarity measures for predicting future price directions in the market. The basic premise for this approach was the common assumption relating to the technical analysis, namely that “history repeats itself,” and the “instrument price reflects all factors that have an impact on its value.” This approach has been studied extensively in many publications. We purport that the subjective interpretation of the chart by the decision-maker should be taken into account. As every decision in the market in the case of manual trading or decision support systems is eventually made by a human, it is necessary to emphasize that the same situation in the market may be interpreted in a different manner by two different decision-makers. Our goal is to use the proposed similarity measure to identify past situations that occurred in the market, and invest accordingly.Under these assumptions, we tested the usefulness of selected measures proposed in the literature, as well as the measure proposed by us, on 21 financial instrument datasets divided into three groups (stock companies, currency pairs, and stock indexes). Moreover, we statistically verified the prediction efficiency for different financial instruments, including stocks, currency pairs, and stock indexes. The statistical verification demonstrated that the proposed approach exhibited higher predictive strength than the classical measures proposed in the literature.