Recently QUIVINCO has been moving towards higher frequency trading system research. The requirement for low latency algorithm calculation had become pressing. The company has responded with development of its first parallel algorithm using the libgomp library for OpenMP with the GCC compiler. OpenMP technology allows existing C code to easily take full advantage of modern multi-core CPUs to accelerate computing intensive tasks.
The CoDesk application is beginning to take shape. The initial focus has been on the persistence and application state layers. The persistence layer now consists of DAOs implemented with Hibernate/EJB3. The application State is shared between client JVMs using distributed EHCache via Terracotta server. With these layers now operational, development of a “Session Facade” style layer is the next milestone.
A DAX strategy based on the company’s unique continuous market profile type indicator is now undergoing real-time testing. The backtest results show excellent linear returns over nearly 6 years of data with modest drawdowns. The indicator features only 2 parameters, making an unlikely candidate for over-fitting. The figures show a consistent 43% of trades win with a 2:1 profit to loss on winners versus losers. Watch this space.
QUIVINCO have successfully developed a data cleanse solution for the NeuroShell platform. The 2 pass statistical filter removes outlying prices from historical data and replaces them with linear projections. This has been an important milestone in developing strategies based on DTN intra-day market data.
QUIVINCO has developed NeuroShell Trader indicators implementing Market Profile technology. Testing of the prototypes has proven very promising on some markets. The vast majority of market timing indicators incorporate a period constant, while markets periodicity actually varies. This discrepancy usually results in a tendency for conventional indicators to produce long runs of losing or winning trades, and therefore poor equity curve linearity. The new indicator uses statistical measures instead of periodicity, thereby helping to reduce the effect of discordant constant selection. Currently the project is migrating to OpenCL for enhanced performance.
QUIVINCO has developed a low-latency publish-subscribe Java API for internal application event management called PubSub. Delivering over 1 million events per second, and delivering both content based and topic based message routing for events, this new light-weight package will provide the event framework for the distributed trade risk desk product CoDesk.
The new research facilities are now operational and consist of an overclocked i7 975 extreme PC with RAID1 and a DTN market data feed. The system is currently being used to develop a NASDAQ 100 futures strategy for the retail trader market and a stock porfolio managing strategy for cash rich institutions.