Financial traders would definitely admit, that there are ways to predict particular actions of competing traders, starting with a simple guess based on extensive practical experience, or by observing actions of other traders for a period of time and then making particular conclusions. However, the field of financial market forecasting is attracting increasingly more attention from the specialists of mathematical analysis and data mining.
Researchers are making attempts to discover fragments of useful information by analyzing the patterns in market events initiated by individual traders. Researchers Lajos Gergely Gyurkó, Terry Lyons (Oxford-Man Institute of Quantitative Finance, University of Oxford) and Mark Kontkowski, Jonathan Field (Man Group Plc) presented their idea in a paper which was published today at arXiv.org.
Their main idea is to use different parameters of market events, which would seem relatively insignificant probably to most of us. Such events may be the time when particular order was made, also the time, when it was cancelled (if cancelled at all), best ask price, best bid price, number of orders and similar. Authors of the study named the sets of such data as financial signature.
From mathematical point of view, the trading-related data can be considered as a complex oscillatory system the dynamics of which can be described using the theory of rough paths. Algorithms developed on the basis of this theory can be successfully used to analyze streams of financial data and make substantiated predictions about future financial events.
In their work, researchers presented some practical examples of application of new technique: one associated with WTI crude oil future market (NYMEX) and two associated with FTSE 100 Index futures market listed on NYSE Liffe. According to authors, any financial trading data is noisy, i.e. contains information of little practical value. Therefore, developed algorithms at first extract useful data fragments using mathematical classification criteria, such as Kolmogorov-Smirnov and ROC curve.
Later this ‘filtered’ data must be fed into prediction-making algorithms which produce forecasts of financial events with increased degree of accuracy compared to other existing systems of statistical forecasting.
Sadly, scientists do not provide very detailed conclusions in their paper, but, apparently, new financial signature analysis technique gives significant results. The only sad side is the fact, that perhaps entire stock market nowadays is controlled by algorithms, so this study probably will benefit the high-frequency trading only.
Source: Technology.org. story by Alius Noreika