reading 2.7

how to estimate a constant VCV matrix

getting var and covar from sample stats

VCV matrix measures

variance of each asset

covariance between every pair of assets

what is a critical step in usuing sample stats for VCV matrix

selecting sample size (recommended # of obvs is 10x the portfolio size)

advantage of usuing multifactor models for VCV matrices

lot less observations required

Instead of asking:

"How does Asset A move with Asset B?"

we ask:

"How do both assets respond to common risk factors?"

why use factor models for VCV matrices

reduce number of estiamtes

improve forecasting effiency

reduce estimation errorwher

where does var come from

common factor risk

assets specific risk

where does covar come from

shared factor exposures

not— company specific risk

shortcomings of factor based VCV matries

Biased and inconsistent

what does it mean matris is biased

inputs are estimated so they gonna be misspecified

the matrix will not a be a predictor of true returns

what does it mean the matrix is inconsistent

as sample size increases model doesnt go to true matrix

is the sample VCV matrix consistent and unbiased

yes!

shrinkage estimats

what; weighted average of sample and target matrix

Results: more efficient wiht smaller error terms

may be biased by more precise

what does data smoothing lead to

underestimate risk

overstate returns/ diversification

lead to distorted portfolio analysis and bad asset allocation decisions

how to unsmooth the data:

take weighted avg of the current “true” returns and previously observed returns

shortcoming of unsmoothing data model

true current return is not directly observatble

how are vol clustering addresed

through arch models