The Power of Multi-Model Forecast in Stock Market

A multi-model forecast provides a significant improvement over the best individual forecast. It can be explained by existence of many different independent factors contributing to the error in each forecast which distributed around actual value.

Today, airplanes are equipped with a few different altimeters: barometric altimeter, radar altimeter, and GPS. Not to mention, also pilots use a visual estimation of altitude. Why do we need to use so many measuring tools? Firstly, any of them can fail. Secondly, there is another reason why multiple measurements are used everywhere. It is accuracy. For example, meteorologists always use many different methods to improve the quality of weather forecast.

The phenomenon of multi-model forecast improvement can be compared with the power of Expert Method. This method can be illustrated by the following. As example, an experimentalist shows a pen and asks a group of several people to write down their estimate of the length. Then he collects notes and calculates the average number. Normally it is almost a precise result. Why does it work? Because everyone makes error in different direction so that the averaging is able to self-compensate erroneous deviations.

As experiments show, in case of the stock market, a multi-model forecast provides a significant (10-25%) improvement over the best individual forecast. Also tests show the advantage of using forecasts from different models even if they are of different quality. The explanation is that there are many different independent factors. These factors contribute to the error in each forecast and the results from different models are normally distributed around actual value.

Evidently, to make multi-model idea more efficient, the methods should be different by their nature. Traditionally, fundamental factors and technical analysis are the major stock market tools. As well as, within technical analysis, there are several different models: technical indicators, chart pattern analysis, Elliott Wave theory, cycle analysis, candlesticks model, trend lines analysis, regression models, etc. Most of these methods are statistically proven and widely used that often create self-fulfilling results.

As a rule, learning and also correctly using many of technical analysis methods may require a lot of time, especially, in a modern dynamic trading environment. Fortunately, the forecast methods combined with computer power have become a good solution to make the works less time-consuming and more effective. These days, except different linear and non-linear solvers and analytical methods, Neural Network (NN) can help to automate a lot of computational tasks. A properly trained NN may enable predictions to the highest accuracy.

However, implementing NN application can be difficult for non-experts. Besides, one of the big obstacles of implementing NN predicting system is a formalization of inputs. Luckily, there are some software tools that already well-developed and do not require a deep technical understanding. These tools are optimized for each method and users might not notice even all computational power behind the buttons, windows, and charts.

Useful resources:

Introduction to Candlesticks
A Basic Introduction To Neural Networks
The download page for computer programs that enable using Neural Network to automate many methods of technical analysis and predict future prices - stock market software tools