| Abstract | This paper introduces a new hybrid deep quantum neural network for financial predictions, the QuantumLeap system. This system consists of an encoder that transforms a partitioned financial time series into a sequence of density matrices; a deep quantum network that predicts the density matrix at a later time; and a classical network that measures, from the output density matrix, the maximum price reached by a security at a later time. The deep quantum network is isomorphic to a deep classical network and is computationally tractable. A hybrid deep network is associated with each time stride, allowing for parallelisation of the learning process. The classical network is a learnable measurement apparatus which infers, from the output density matrix, the maximum price reached by a security for a given time. Experimental results associated with 24 securities clearly demonstrate the accuracy and efficiency of the system in both the regression and extrapolation regimes. |
|---|
