Too long ago, we promised a follow-up entry describing Stracia and our analytical process. This entry is the second in that series, and we hope not to let ourselves be so distracted by “optimizing our optimizer” that the third entry (which will discuss our modeling methodology) is as long in coming.

At its foundation, Stracia is an investment philosophy: Its purpose is to pursue both beta (yes) and alpha returns by articulating buy and sell decisions based on critical analyses of how market and macroeconomic conditions may effect the performance of various asset classes. It stresses fundamental research supported by rigorous quantitative methods. On another, nuts-and-bolts level, it is a collection of optimization algorithms, statistical techniques, and valuation methodologies — software — designed to measure and express the credence of this philosophy in quantitative terms. So it is both an intellectual system for approaching trading decisions and the proprietary algorithms — the process — that supports the approach.

It may also be useful to stress what Stracia is not. It is not an automated investment system or “some kind of robot” that seeks to remove the investment manager from the decision-making process, or to replace his own judgment with merely mechanized trading rules or signals. (Most especially, it is not some “elite” investment product or black-box that will help you “make money now!,” “discover cheap stocks!,” etc., etc. All such pitches are for snake oil and beware.)

Rather, the emphasis is on using quantitative models to add substantive understanding to what the manager already knows (or to challenge his convictions) about the nature of cross-category asset-class performance, to provide a framework for understanding how asset classes outside of a manager’s specific investment mandate may impact his holdings universe, and to add to his total, qualitative knowledge of these relationships and their expected impact on portfolio performance.

Stracia is designed to support competitive decision-making and as such (and as mentioned), could be used either as the foundation for a global, multi-strategy hedge-fund approach, or to support a fund-of-funds or separately managed accounts (SMA) strategy. More generally, it could be used to support active-indexing or passive-rotation (sector, high-low quality) strategies — hence the willing exposure to beta as well as to alpha.

We have endeavored to make Stracia as fully articulated as possible through the creation of several dozen proprietary models. These models consist of three kinds:

• number-crunching subroutines that either compare time series to one another or conduct stochastic simulations — such as on asset returns, economic or other trend data;
• risk/return optimizers that incorporate the subroutines’ results at the decision level, supporting our own buy/sell decisions (we do not make investment recommendations); and
  
• artificial neural networks.

Models of the first type are generally bivariate, yielding results independent of other subroutines and thereby allowing the manager to evaluate the expected impact of market-moving economic events as they develop and (if desired) in isolation; models of the second type are backward-looking optimizers, translating the output of individual subroutines into a coördinated execution strategy; and the neural networks are generally of the supervised, backward-propagation, or backprop, variety.

Much of this Web site is concerned with describing the use of these models and the interpretation of their results in detail. Since they are the core of the Stracia framework, they deserve their own entry. Stay tuned for an in-depth discussion of our modeling methodology. ♦