Trading with NARX neural networks – running

Real-time prediction of financial time series is important in the so-called "high frequency trading", where one has to trade successfully on intraday data. As a predictor we use the NARX type neural network. Since the success of trading (achieving maximal profit) is not reached by minimizing the mean square error, we introduce new objective functions, what are more suitable for optimizing profit, but those are still simple learning algorithms. The goal is implementing these learning algorithms on GPU and on BUDE's cluster super computer. The prediction algorithms must be tested on FOREX and S&P 500 time series.

Project owner:
Ceffer Attila (Híradástechnikai Tanszék)
Híradástechnikai Tanszék (VIK-HIT)

Konzulens: Dr. Levendovszky János