This paper examines the feasibility of applying data mining techniques to testing market efficiency hypothesis using a high frequency, up to one thousand of a second, electronic brokerage data. Results suggest the existence of a pattern of negative autocorrelation in returns of DEM/USD over relative short lags (less than 40 seconds). However, this pattern is not feasible by two reasons: (1) the structure of autocorrelation pattern is inconsistent and changes too rapidly (2) the largest potential speculative profit is smaller than the regulated tick size. These results indicate that dealers have engaged in any potential profitable speculations based on past price information.
Lo, Melody and Hsieh, Chang T.
"Mining the FX electronic inter-dealer market,"
Journal of International Information Management: Vol. 12
, Article 5.
Available at: http://scholarworks.lib.csusb.edu/jiim/vol12/iss1/5