Artificial intelligence (AI) is used in the financial services industry to automate, enhance, and optimize processes; make more accurate predictions; and autonomously learn from experience.
AI in finance includes machine learning, deep learning, natural language processing, graph algorithms, evolutionary learning, and other techniques. You can apply these techniques using MATLAB®.
Machine learning approaches are categorized as unsupervised learning, supervised learning, or reinforcement learning:
- Unsupervised learning is useful for grouping unlabeled historical data sets and finding patterns in data using clustering. For example, investors use cluster analysis to build diversified portfolios.
- Supervised learning is useful when you have historical inputs along with labeled output. It infers a function that can predict future outputs and falls into two categories: classification and regression. Classification identifies a discrete label of which a new observation belongs, such as trade or do not trade, while regression is used to predict a continuous value, such as price or salary.
Deep learning, a subset of machine learning, utilizes neural networks and is applied to machine learning problems simultaneously perform feature extraction and prediction within the neural network architecture. This approach eliminates the need to perform feature extraction prior to developing a predictive model. Moreover, deep learning requires a substantial historical training data set to build a robust and accurate predictive model. For example, nonlinearities in oil price distribution such as volatility are captured by neural network models.
Reinforcement learning helps alleviate this challenge by generating the needed data. It does this through repeated simulations (via trial and error) with a reward structure for good outcomes. Its aim is to learn a “behavior” as opposed to fitting a model with the highest possible accuracy. The goal of reinforcement learning is to train a model to take actions or make decisions in order to maximize the cumulative reward. One financial application is to train an agent to hedge a European call option contract and save on transaction costs.
Natural language processing, another AI in finance technique, employs algorithms to retrieve essential data from textual data representations of natural language. Its key applications are text generation, text classification, sentiment analysis, and topic modeling.
Sentiment analysis is an application of natural language processing in which analysis of news and monitoring of social media are used to build sentiment measures of the financial market, which can be used to drive real-time trading decisions. Other applications include assessing counterparty credit risk and analyzing surveys to understand why customers are satisfied or dissatisfied.
Practitioners of AI in finance often use graphs to make visual representations of data structures involving complex interrelationships. Graphs are composed of nodes and edges that can be weighted. Because data is connected in a smart way, one application is to construct a diversified portfolio by identifying correlated assets.
Inspired by biological evolution, AI in finance used evolutionary or genetic algorithms to develop new forecasting techniques and sophisticated trading systems. The iterative process consists of crossovers, mutation, and selections to get a satisfactory level of convergence, resulting in better trading parameters over time.