Neural Networks in Investments
Investment managers often find themselves overwhelmed with the large amount of data obtained from the financial markets. Most of the data available is numeric and noisy in nature, making the decision-making process harder. These decisions usually rely on the integration of statistical measures that attempt to compress much of the data and qualitative depictions such as graphs and bar charts with news events and other pertinent information. Investment decisions usually involve non-linear relationships among the various components of the data. Computers in general, are very adept at dealing with large amounts of numeric information. However, some algorithms are crucial in analyzing and combining disparate information that can impact security prices. Artificial Intelligence based methods uses clever algorithms and rules of thumb (heuristics) in the decision-making process. Neural Network and expert systems applications have been successfully deployed in the domain of Finance, and in the area of investment management.
This paper discusses the basics and the theory behind neural networks and provides an introduction to an application area of neural networks in the domain of Finance. The application areas of Neural Networks discussed in the paper are corporate finance, financial institutions, and the professional investor. The purpose of the second paper will be to discuss the specifics of each of these applications.
Neural network computing is an information processing method that was developed from research to make computers that could imitate the way people learned. The field initially grew from 1930s ideas about how biological systems like the human brain works. Today neural network systems are being used in business, government, and academic research because of their power to model data quickly and to produce better results than other more traditional data analysis techniques. At the simplest level, neural networks are a new way of analyzing data. The revolutionary aspect of neural networks is their ability to learn and trace the complex patterns and trends in data. Neural networks are made up of neurons and behave like the human brain, and has the ability to apply knowledge from past experience to new problems. Neural networks acquire this knowledge by training on a set of data. After the network has been trained and validated, the model may be applied to data it has not seen previously for prediction, classification, time series analysis or data segmentation.
Unlike traditional statistical methods, neural networks do not require assumptions about the model form. A statistical analysis requires a certain form to be assumed such as linearity, which characterizes relationships between variables. Neural networks are more tolerant of imperfect data, such as the presence of missing values or other data quality problems. Neural networks perform better than traditional...