Simply the appearance of historical price data contains lots of information. Our trading system is able to see historical prices in a similar way to humans, it can detect visual features of price windows from which it is able to make profitable trades.
The main feature of this system is the computer vision algorithm, a deep convolutional neural network. A history of outputs from the network is combined into trading signals to initiate long and short positions.
A unique feature to this system is a dynamic take profit linked to the neural network. All open positions are closed at a take profit level based on network measures. This adds additional profit to certain trades compared to using a static take profit.
The trading strategy has performed exceptionally well during back-testing. Over a trading period between 17th October and 13th December on the EUR/USD currency pair, an amount of 322 trades were opened, of which 86.3% were closed profitable, generating an average daily return of 3.7%. Additionally, a common measure for risk-adjusted returns, the Sharpe ratio, was calculated to be 12.1, considered to be unusually high.
Analysts: Jim Öhman & Jan Mueller
