TY - JOUR
T1 - An algorithmic framework for frequent intraday pattern recognition and exploitation in forex market
AU - Goumatianos, Nikitas
AU - Christou, Ioannis T.
AU - Lindgren, Peter
AU - Prasad, Ramjee
PY - 2017/12
Y1 - 2017/12
N2 - We present a knowledge discovery-based framework that is capable of discovering, analyzing and exploiting new intraday price patterns in forex markets, beyond the well-known chart formations of technical analysis. We present a novel pattern recognition algorithm for Pattern Matching, that we successfully used to construct more than 16,000 new intraday price patterns. After processing and analysis, we extracted 3518 chart formations that are capable of predicting the short-term direction of prices. In our experiments, we used forex time series from 8 paired-currencies in various time frames. The system computes the probabilities of events such as “within next 5 periods, price will increase more than 20 pips”. Results show that the system is capable of finding patterns whose output signals (tested on unseen data) have predictive accuracy which varies between 60 and 85% depending on the type of pattern. We test the usefulness of the discovered patterns, via implementation of an expert system using a straightforward strategy based on the direction and the accuracy of the pattern predictions. We compare our method against three standard trading techniques plus a “random trader,” and we also test against the results presented in two recently published studies. Our framework performs very well against all systems we directly compare , and also, against all other published results.
AB - We present a knowledge discovery-based framework that is capable of discovering, analyzing and exploiting new intraday price patterns in forex markets, beyond the well-known chart formations of technical analysis. We present a novel pattern recognition algorithm for Pattern Matching, that we successfully used to construct more than 16,000 new intraday price patterns. After processing and analysis, we extracted 3518 chart formations that are capable of predicting the short-term direction of prices. In our experiments, we used forex time series from 8 paired-currencies in various time frames. The system computes the probabilities of events such as “within next 5 periods, price will increase more than 20 pips”. Results show that the system is capable of finding patterns whose output signals (tested on unseen data) have predictive accuracy which varies between 60 and 85% depending on the type of pattern. We test the usefulness of the discovered patterns, via implementation of an expert system using a straightforward strategy based on the direction and the accuracy of the pattern predictions. We compare our method against three standard trading techniques plus a “random trader,” and we also test against the results presented in two recently published studies. Our framework performs very well against all systems we directly compare , and also, against all other published results.
KW - Data mining
KW - Forex
KW - Hidden intraday patterns
KW - Pattern recognition
KW - Template grid method
UR - http://www.scopus.com/inward/record.url?scp=85018823471&partnerID=8YFLogxK
U2 - 10.1007/s10115-017-1052-2
DO - 10.1007/s10115-017-1052-2
M3 - Journal article
AN - SCOPUS:85018823471
SN - 0219-1377
VL - 53
SP - 767
EP - 804
JO - Knowledge and Information Systems
JF - Knowledge and Information Systems
IS - 3
ER -