Financial Signal Processing Inc.

Where do financial signals come from and how do they improve your investment performance ?

We build our own web crawlers, and we crawl data on the Internet. Our tools differ from many other data collection tools built for the financial industry in the scope of the data collected. Often, data collection tools for the financial industry (like, Reuters, Bloomberg, Thomson etc) crawl data from limited sources, such as the Wall Street Journal, New York Times, CNN, CNBC, Financial Times, and the Economists, and gather data from financial or news-related sites. We have built a general web crawler similar to Google's web crawler, and we crawl data all over the Internet. For example, we crawl shopping sites to see what people are buying, we crawl fashion sites to see what fashion is going to be popular, we crawl bulletin boards, forums, and social networks sites to see what people are talking about. We gather data from many different sources to find out the true demand for commodities. Since all the information about human life today is reflected in some form on the Internet, why use secondary sources ? We believe that all the information needed for financial analysis can be found on the web, although it involves a great effort to obtain it.

We crawl a great amount of data, but it is not the amount that makes us special. If it were only the amount of collected data that mattered, Google would be the world’s largest financial institution (which is not necessarily what they are interested in, as they already have a lot of cash, and it requires special exemptions to avoid to be regulated as a mutual fund). Furthermore, due to the availability of open-source tools such as Cloudera and Lucene, it is not too difficult nowadays to build your own web crawlers. However, it should be emphasized that it is still very hard to build a high-performance web crawler. For example, Microsoft's Bing and Yahoo web Crawler are still not as good as Google's, in spite of the huge funds that have been invested in these projects. Another crawler with excellent speed performance is Baidu, although it has a smaller data index.

Our financial signal crawlers do not store the crawled data, and this is one of the major differences between our crawlers and the ordinary web crawlers. If we store the data that we crawled, we need to build data centers of the same scale as Google's data centers. It is when our crawlers bring back raw data that the financial signal processing engine starts to operate, and to extract signals from these data. Unfortunately, the signal noise ratio(SNR) is EXTREMELY low in the web data, and there are a lot of noises. Therefore we have to “squeeze” raw data really hard in order to find small amounts of signals in it, and these are the signals that reveal the real trends. We use combined techniques of signal processing and natural language processing. Our systems convert these signals into numbers, which are multi-dimension vectors. We only need to save these vectors and use them later on to form different signals and signal indicators. These can be analyzed using common signal processing techniques in order to identify patterns and cope with noises. For example, one can apply MIMO analysis to these vectors. Because our datasets encompass almost the entire Internet, we can provide an unbiased, high fidelity, high sampling rate signal that surpasses many of the alternatives, and would avoid stories such as this.

We provide these "hard-sought" signal vectors to our customers, and they can integrate them into their analysis, or into their trading systems. Our customers may be interested either in raw signals or in processed signals which include forward predictions.

About Us:

Financial Signal Processing (FSP) is located in the Silicon Valley, California. We have many years of experience in building web crawlers and applying signal processing techniques to wireless and wire-line communications, such as mobile phones, wireless home networks and DSL technologies. Then we were told that our techniques to handle extremely low signal noise ratio (SNR) systems can be applied to financial and investment applications. That is how it got start.

Financial Signal Processing Online Resources

Signal Processing Basics

Kelly Criterion

Universal Portfolio Theory

Natural Language Processing

Maximum Entropy

People with Cryptography background interested in investments

James Simons

Elwyn Berlekamp

Information theorists interested in investments

Claude Shannon

Edward Thorp

Open source tools to build a web crawler

Cloudera

Lucene

Advanced Signal Processing

EE379C (Stanford Univ)

EE479 (Stanford Univ)