Treynor-Black portfolio weights are sensitive to large alpha values, which can result in practically infeasible long/short portfolio positions.

Benchmark portfolio risk, the variance of the return difference between the portfolio and the benchmark, can be constrained to keep the TB portfolio within reasonable weights.

Alpha forecasts must be shrunk (adjusted toward zero) to account for less-than-perfect forecasting quality. Compiling past analyst forecasts and subsequent realizations allows one to estimate the correlation between realizations and forecasts. Regression analysis can be used to measure the forecast quality and guide the proper adjustment of future forecasts. When alpha forecasts are scaled back to account for forecast imprecision, the resulting portfolio positions become far more moderate.

The Black-Litterman model allows the private views of the portfolio manager to be incorporated with market data in the optimization procedure.

The Treynor-Black and Black-Litterman models are complementary tools. Both should be used: the TB model is more geared toward security analysis while the BL model more naturally fits asset allocation problems.

Even low-quality forecasts are valuable. Imperceptible R-squares of only .001 in regressions of realizations on analysts' forecasts can be used to substantially improve portfolio performance.