Equity Valuation and Portfolio Management

BACKTESTING OUR FACTOR TRADING STRATEGY

Using the model scores from the trading strategy example, we build two optimized portfolios and evaluate their performance. Unlike the five equally weighted portfolios built only from model scores, the models we now discuss were built to mirror as close as possible tradable portfolios a portfolio manager would build in real time. Our investable universe is the Russell 1000. We assign alphas for all stock in the Russell 1000 with our dynamic factor model. The portfolios are long only and benchmarked to the S&P 500. The difference between the portfolios is in their benchmark tracking error. For the low-tracking error portfolio the risk aversion in the optimizer is set to a high value, sectors are constrained to plus or minus 10% of the sector weightings in the benchmark, and portfolio beta is constrained to 1.00. For the high-tracking error portfolio, the risk aversion is set to a low value, the sectors are constrained to plus or minus 25% of the sector weightings in the benchmark, and portfolio beta is constrained to 1.00. Rebalancing is performed once a month. Monthly turnover is limited to 10% of the portfolio value for the low-tracking error portfolio and 15% of the portfolio value for the high-tracking error portfolio.

Exhibit 12.14 presents the results of our backtest. The performance numbers are gross of fees and transaction costs. Performance over the entire period is good and consistent throughout. The portfolios outperform the benchmark ...