Artificial intelligence in trading
Previous sections introduced the basic principles and algorithms for building neural networks. However, our primary interest lies in the practical application of the presented technologies, and, I am certainly not the first to consider it.
Computer technologies have long been integrated and successfully applied in trading. It's difficult to imagine trading without the use of computer technologies nowadays. Primarily, thanks to the internet and computers, traders no longer need to be physically present on the trading floor. The trading terminal software can be easily installed on any computer and even on mobile devices (smartphones, tablets). This enables traders to analyze the market and execute trading operations from virtually any location on our planet.
The aforementioned trading terminals not only facilitate trade execution but also provide all the necessary tools for detailed real-time market analysis. They include features for constructing graphical objects on price charts and a variety of indicators that can dynamically update values and display them on the chart according to the current market situation.
Another direction of applying computer technologies in trading is algorithmic trading. Algorithmic trading involves creating computer programs (robots) that execute trading operations without human intervention, following a predefined trading strategy. This method has its own advantages and disadvantages compared to manual human trading.
A created program can work tirelessly 24 hours a day, 7 days a week, which is impossible for a human. Accordingly, the program will not miss any signal to enter or exit a position. The robot will strictly follow the specified algorithm. In contrast, a human, while evaluating the market situation, may consider personal past experiences and subjective feelings, which can vary.
First and foremost, deviating from the trading strategy disrupts the balance between profitable and losing trades, and over a long time frame, it's likely to have a negative impact on the trading account balance.
On the other hand, it can be quite challenging to precisely describe all aspects of a trading strategy in mathematical terms. In this case, the trader's personal experience and their personal feeling of the market will play a significant role. The program does not have these features, and the tolerances built in by the programmer may not be ideal.
Among the benefits of algorithmic trading, we can also include the absence of psychological factors in programs. Meanwhile, the psychological barrier often causes traders, especially newcomers, to deviate from their trading strategies.
On the other hand, time series are variable. Therefore, any trading strategy has a limited lifespan. As a consequence, over time, there's a need to adapt trading systems to current market conditions, and a classical robot can't evaluate its performance or make changes to its trading algorithm or parameters without human assistance.
So what do we expect from the application of artificial intelligence and neural networks in particular?
When building a mathematical model using a neural network, we do not prescribe the entire trading algorithm, as in classical algorithmic trading. We simply provide a training dataset and let the neural network itself discover patterns and correlations between the input data and the final outcome. In doing so, we expect the neural network to capture not only the obvious patterns but also the subtle fluctuations that can enhance the effectiveness of the trading system.
When creating a training dataset for the neural network, we should not limit ourselves to the input data of a single strategy. There may be much more input data than a human is capable of processing. However, the final mathematical model might produce signals that don't fit neatly into any of the expected strategies. As a result, we expect to obtain performance higher than that of robots built according to the classical algorithmic trading scheme.
And, of course, the learning ability of neural networks enables the creation of methods for assessing the performance of a strategy and initiating the training process of the neural network in a timely manner for adaptation to current market conditions.
Thus, we anticipate a reduction in the negative aspects of algorithmic trading while retaining its positive aspects.