If more history means a better estimate, why can't you just use fifty years of GM's returns?
Loading…
The historical covariance estimator
Write and record which quantities are estimated from data and which are chosen.
Estimation error versus non-stationarity
Record the trade-off in sample length, using GM's 2009 bankruptcy and Motorola's 2011 split as examples, and the problem with choosing a full cycle such as Kondratiev's.
Stable parameters, not forecasts
State the narrow claim that expected returns and covariances are stable, and contrast portfolio construction with Buffett's approach of picking undervalued stocks to beat the market.