A neural network is a mathematical modeling tool that has the capacity to learn by example. This is an extraordinarily useful ability, especially in financial modeling, where the predictive inputs are usually known and there are countless example forecasts. Networks first need to be trained by being presented with hundreds of facts, each fact consisting of inputs and corresponding outputs. Through a unique feedback process, the network learns how those inputs are related to the outputs, and develops a general model that describes the relationship.
Neural networks use a different technique from standard analysis that is better suited to real world problems. Mathematical models are normally built by making a priori assumptions about the functional form of the solution. These are called parametric models, and are solved by regression methods to determine a number of coefficients. This is sufficient if you know that the solution must be a second order polynomial, or some other simple, well-known function. But in the real world, relationships are not necessarily simple. Inputs and outputs could even be related in a non-linear fashion. If you do not have to guess the functional form of the answer, you have a big advantage.
There are a number of funds using neural networks to improve stock selection, and neural networks have been the subject of articles in Barron's, The Economist, and Futures Magazine. It is a new cutting-edge technology. Parallax currently uses two professional neural network programs: BrainMaker and NGO.
The Wizard equity pricing model uses neural networks to learn how a stock's fundamental balance sheet data has been translated to a market price in the past, within the context of its industry group and economic sector. The models created for each industry group are then used to estimate current fair market values for these stocks. The application of neural networks to this problem is a natural use of the technology since there are many factors making up a stock price. The analyst must juggle many factors at the same time to estimate value. This is difficult for one stock, let alone all the stocks in an industry group. Neural nets will do this easily and in an unbiased manner.
Nine factors such as earnings per share, earnings growth, return on assets, return on equity, current ratio, sales per share, book value per share, operating profit margin, and cash per share are fed to the network for each stock, for up to 40 back quarters, within a single industry group. The network is presented with the actual market price at the time the earnings were announced. It then tries to learn how these fundamentals were interpreted by the market into a price.
Once trained, the latest quarter's information is fed into the model, and the expected "fair value" is determined for each stock. In order to be able to estimate valuation error, seven prices are actually determined by the net by repeatedly splitting the per share data, retrieving the estimate, and then unsplitting it. The final average price and standard deviation are calculated from these estimates. A low standard error means the estimate is well determined.
These values can then be used to rank stocks within the group, find under-valued purchase candidates, find short sale candidates, help form spread trades, or be used together to rank the industry group.
Find out how to contact Parallax.
Find out about product trials.
Go to the Parallax home page.