Analysts should be aware of several key considerations when using historical data:
Stationarity: Prefer the longest dataset with stable statistical properties (stationarity), as longer data histories give more precise estimates.
Frequency trade-offs:
Higher-frequency data (e.g., daily/monthly) improve estimates of variances, covariances, and correlations, but not the mean.
More frequent data can introduce asynchronicity, where variables don't align in time (e.g., due to time zones), distorting correlations.
Dimensionality problem:
When analyzing many variables (assets), the number of observations must exceed the number of variables to avoid spurious zero-volatility portfolios.
Solution: use factor models, assuming returns are driven by common factors plus specific noise.
Normality assumptions:
Asset returns often show skewness and fat tails, violating normal distribution assumptions.
While acknowledging non-normality is important, accounting for it increases complexity—often not worth the cost unless precision is critical.