Portfolio optimization is an important aspect of investment management, aiming to construct portfolios that offer the best risk-return trade-off based on an investor’s objectives and constraints.
-H depends on the time fame, so you use it in your usual time frame that you trade (daily, intraday etc.)
-true that they swap covar matrix for H, it's just a different objective function and the article claimed that it's more effective.
The way I look at this optimization scheme: in pairs or basket (stat arb) trading, you use cointegration to construct pairs that are mean reverting. Here, similarly, you use H to construct a basket of stocks that is trending (or mean reverting if you optimize for it).
I have not looked in the details. At first sight
- H is noisy indeed
- actually, rather than H, there is changing fractal dimensions at different resolutions
But, complexities apart, have they not just swapped the noise of the covariance matrix for the noise of the Hurst exponent? What do you think?
The attempt to maximize a SNR is laudable in any case
-H depends on the time fame, so you use it in your usual time frame that you trade (daily, intraday etc.)
-true that they swap covar matrix for H, it's just a different objective function and the article claimed that it's more effective.
The way I look at this optimization scheme: in pairs or basket (stat arb) trading, you use cointegration to construct pairs that are mean reverting. Here, similarly, you use H to construct a basket of stocks that is trending (or mean reverting if you optimize for it).