Artificial Neural Network for the implementation of a Trading System

Abstract

This thesis presents the powerfulness of neural network models for technical analysis as regards to Intesa stock. Their application consists in prediction for future price of the stock and rely on a trading system to a buying and selling trading signal. Recently, numerous deep learning algorithms have been proposed to intervene in traditional forecast methods. Great advantage derives from their application directly on raw data and their quite well prediction in capturing stock's volatility.

This work shows that these algorithm can capture with a quite good and acceptably approximation mean and volatility of stock price in order to do predictions. On these, predicted trading signals provide evidence not only in an anticipated signal in respect to traditional ones, but also on market entrance at the best moment, avoiding market negative variations.

In the second part, it is proposed an improvement of the trading system by reinforcement learning algorithm. Throughout a reward mechanism, the algorithm learns itself and autonomously a trading strategy. It combines with different weights all trading signals indicators created before to optimize investment returns from the trading strategy.

In conclusion, further investigations in this field can take many directions. Such as an improvement of deep learning models by find the best algorithm with the best number of parameters or Choose a detailed trading system and improve its performance by have the advantage of a predicted time series. These could be two simple suggestions, by analysis on this field are still wide ranging and it still leaves further room for deeper investigations.