| Advances in telecommunication and software technologies have changed the way that securities are traded on the stock market. Algorithmic trading, which is also referred to as automated or black box trading, accounts for a large percentage of orders placed in the market, especially after the year 2000. It provides investors with many benefits such as reduced transaction costs, higher accuracy and speed, anonymity, transparency, and also access to different markets. Yet, it has a few limitations including lack of intelligence and lack of adaptability to the market conditions. Algorithms execute blindly what they are trained without having the capability to distinguish different conditions in the market. Such weaknesses make it vulnerable to unforeseen events like market crises, which may result in a large amount of loss. For example, on May 6th, 2010, the Dow Jones Industrial Average fell 600 points in about five minutes that led to a loss of $600 billion in the market value of US corporate stocks. A large number of researchers attribute that crash to algorithmic trading orders, which were not intelligent enough to find out the financial crisis. Algorithms should be able to determine when to place different orders such as buy and sell, and more importantly algorithms need to identify situations where staying out of the market is more beneficial than placing buy or sell orders. As a result, protecting algorithms through supervised learning processes is of great importance to algorithmic trading. In this thesis, a new algorithmic based on value trading is proposed to identify when to place buy, sell, or stop orders. After classifying the orders into those three categories, an Artificial Neural Network (ANN) with three layers, input, hidden, and the output is used to learn from previous trades. The ANN has five neurons in the input layer, ten neurons in the hidden layer, and three neurons in the output layer and is used to learn from the past patterns and make predictions for the future. In the last phase, the learning performance measures including accuracy, precision, recall, and F-score are measured. |
