CONSTRUCTING, TRADING, AND EVALUATING PORTFOLIOS

To maximize implementation of the model's insights, the portfolio construction process should consider exactly the same dimensions found relevant by the stock selection model. Failure to do so can lead to mismatches between model insights and portfolio exposures.

Consider a commercially available portfolio optimizer that recognizes only a subset of the variables in the valuation model. Risk reduction using such an optimizer will reduce the portfolio's exposures only along the dimensions the optimizer recognizes. As a result, the portfolio is likely to wind up more exposed to those variables recognized by the model, but not the optimizer, and less exposed to those variables common to both the model and the optimizer.

Imagine an investor who seeks low-P/E stocks that analysts are recommending for purchase, but who uses a commercial optimizer that incorporates a P/E factor but not analyst recommendations. The investor is likely to wind up with a portfolio that has a less-than-optimal level of exposure to low P/E and a greater-than-optimal level of exposure to analyst purchase recommendations. Optimization using all relevant variables ensures a portfolio whose risk and return opportunities are balanced in accordance with the model's insights. Furthermore, the use of more numerous variables allows portfolio risk to be more finely tuned.

Insofar as the investment process, both stock selection and portfolio construction, is model-driven, it is more adaptable to electronic trading venues. This should benefit the investor in several ways. First, electronic trading is generally less costly, with lower commissions, market impact, and opportunity costs. Second, it allows real-time monitoring, which can further reduce trading costs. Third, an automated trading system can take account of more factors, including the urgency of a particular trade and market conditions, than individual traders can be expected to bear in mind.

Finally, the performance attribution process should be congruent with the dimensions of the selection model (and portfolio optimizer). Insofar as performance attribution identifies sources of return, a process that considers all the sources identified by the selection model will be more insightful than a commercial performance attribution system applied in a one-size-fits-all manner. Our investor who has sought exposure to low P/E and positive analyst recommendations, for example, will want to know how each of these factors has paid off and will be less interested in the returns to factors that are not a part of the stock selection process.

A performance evaluation process tailored to the model also functions as a monitor of the model's reliability. Has portfolio performance supported the model's insights? Should some be reexamined? Equally important, does the model's reliability hold up over time? A model that performs well in today's economic and market environments may not necessarily perform well in the future. A feedback loop between the evaluation and the research or modeling processes can help ensure that the model retains robustness over time.