Equity Strategies

Description

Proteom creates high performance securities portfolios with a rare combination of algorithms, hardware, and software tools derived from computational biology.  Merging the techniques of computational biology with more conventional market analysis creates a powerful approach to understanding financial markets.  This allows the Investment Manager to custom design portfolio strategies to meet performance criteria specified by the Investment Manager.

The investment methodology uses several proprietary technologies to identify complex symmetries in equity pricing.  Equity pricing data are exposed to a combination of decryption and pattern recognition processes.  These processes are related to systems used in the encryption of information, or in the study of proteins and genes, and are expedited on parallel computer clusters that provide access to very large data structures with great speed.  These early solutions are then used to inform simulated market conditions so that competing strategies can be assessed for their relative strengths and weaknesses.  Although individual behavior may be impossible to predict with any accuracy, group behavior presents a very different opportunity.  As strategies compete in simulated markets, again on parallel clusters, more information is obtained about meta-relationships to time and market conditions.  Proteom maintains a flexible and secure laboratory for incubating parallel computer cluster based algorithms designed to help clients better understand complex events.  Finally, several selection processes are used to impute convergence toward predefined goals.  This approach should maximize out of sample demonstrations and a continuity of security and accuracy. 

Tools

• Genetic algorithms simulate evolutionary processes observed in nature.  To do so, they are designed to solve virtually any type of “optimization” problem.  Re-tooled for financial markets, genetic algorithms can show how to optimize financial portfolios. 

•  Neural networks loosely simulate human brain processes to solve a wide range of problems.  Neural networks are particularly effective at recognizing subtle patterns amidst seemingly chaotic movements. 

•  Computer clusters are networked computer workstations with an additional layer in the operating system allowing them to work together on the same problem.  Because individual processors in the cluster can skip ahead to next steps as soon as they detect progress or failure, computer clusters are capable of processing large volumes of data very quickly.  Crucial analyses that might take months through conventional computing take only minutes using the cluster.

The cluster was designed to address the high-volume computation and data management challenges unique to financial modeling.  The cluster provides more than three times the computing power of IBM’s famed Deep Blue supercomputer.

The following is a general description of the development and testing of a typical Proteom product, an enhanced index strategy that confines the portfolio and has defined benchmarks.  To design such a product, we go through the following steps:

In the decryption phase, the tools of computational biology are used to develop pattern recognition systems whereby each equity price history is decrypted and profitable patterns are identified.  This phase of development has its origins in protein searching and matching.  A distance profile is a summary statistic that qualifies the content of a protein.  Simple modifications of these statistics characterize the architecture as well.  Such an approach was used to take a mutant gene from the worm, and find it in another species, the fly, allowing geneticists to move their research toward human applications.  The figure on the left shows targeted proteins from both species, while the figure on the right shows the daily profit / loss stream from a typical equity, 3M, a member of the S&P 500 Index.

Applying modified protein search algorithms to equities provides a method for assessing cycles and their strengths.  This information is then used to assess the timing of holding various positions.  Once patterns are identified, other procedures are used to assess their stability as well. These cycles and stabilities are then weighted for simulation.  These holding methods can then compete with each other in virtual (simulated) financial markets.  For instance, the hold period in 3M, below, was predicted and held during the bold (blue) phase in 2003.

Simulation allows strategies to compete with each other and then to evolve into complex sets of optimized strategies.  These optimized strategies are then selected for investment based in the predicted performance.  This process is a modification of neural networking processes used to build digital versions of biological nervous systems.  The neural network is determined by (1) the architecture of the network, and (2) the communication rules.   While this might seem trivial, the lack of spatially accurate network architecture has been a primary barrier to biological modeling, and disease modeling.  Disease modeling, especially developmental diseases, but also pathological processes that involve any time course (including trauma recovery). depend upon spatial modeling.  A neuron, once turned off, damaged, or removed, not only affects the neurons to which it is “synapsed”, but also neourons that it is near.  Accurately predicting and building neural networks from electron micrographs of the worm (left) is much like designing out of sample investment strategies (right).