Can we predict the financial stock market using data mining techniques?

Can we predict the financial stock market using data mining techniques?

In this report 2 types of options are discussed for forecasting the financial stock market. Both of them seem convincing in their own way, but can we make a wise decision using either of the two approaches.

Type 1 - Computer automated programs are developed using data mining and predictive technologies for the trade markets. Here the historic data is the key and is used to develop prediction models. This helps investors to understand hidden patterns from the historic data and in turn enables them to do a prediction or forecast of their future investments/trade. Data analysis and record of various global events are usually used to develop the prediction model which supports numerically and graphically.

I read a paper and couple of articles on this subject. In the paper states that it is possible to predict but the article gives a complete opposite picture. In the paper the researchers collected and combined six months data. They generated a prediction model of whether or not the stock prices go up or down. Further predictions were made and the model was tested the following day’s closing prices. If the following day’s price changes as per the prediction the model developed it is considered valid. As per the literature if the closing price can be predicted to increase or decrease 70% of the time at the 0.07 confidence level, then this analysis would be an easy and useful aid in financial investing. Further, it would show that the results are better than random choices at a reasonable level of significance.

 Type 2 – The hypothetical results of data mining always crumple when they collide with real world investing scenario. There are lots of theories about how to predict future stock market performance. The most popular source of market prediction is data mining investing strategies. Every year people pour lot of money to develop data mining strategies and come up with models which use back test or simulation techniques.

We must understand that even if stocks have rallied every second for Thursday in May for the past 50 years, it doesn’t mean they will rally this year on the second Thursday in May. If history could determine what the stock market is going to do next, the world’s richest investors would be historians, data-processors and librarians which is not the case.

Both the arguments sounds convincing but which of the two would be a better choice.

http://www.iaeng.org/publication/IMECS2010/IMECS2010_pp555-559.pdf

http://mat.tepper.cmu.edu/blog/?p=804

http://online.wsj.com/article/SB124967937642715417.html