The goal of this master thesis was to develop a method for real-time classification of market trading data at the Foreign Exchange (FX) department at the Skandinaviska Enskilda Bank (SEB). The characteristics in the market data sets were analyzed using Principal Component Analysis (PCA). The analysis showed that the principal component subspaces for two different types of market data, normal and abnormal, for the EUR/USD instrument where significantly different. The result from the PCA naturally led into the construction of a Single-class detector, for detecting if quote updates were normal or abnormal based on training data. The market data sets were shown to possess multicollinear characteristics, resulting in low-rank properties of the covariance matrices. To overcome this problem the solution was to transform the data using PCA, resulting in full-rank properties of the covariance matrices of the transformed data. This vital step made it possible to classify quote updates for the EUR/USD instrument. The project resulted in a classification algorithm which is able to successfully classify if a quote update is normal or abnormal with respect to training data in real-time. The algorithm is versatile in the sense that it can be implemented on any market for any currency pair, and can easily be extended to classify the relative behaviour between several currency pairs in real-time.
Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:uu-256469 |
Date | January 2015 |
Creators | Lundberg, David |
Publisher | Uppsala universitet, Avdelningen för systemteknik |
Source Sets | DiVA Archive at Upsalla University |
Language | English |
Detected Language | English |
Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
Format | application/pdf |
Rights | info:eu-repo/semantics/openAccess |
Relation | UPTEC F, 1401-5757 ; 15039 |
Page generated in 0.0025 seconds