3 ms·

Thanks for sharing my post here. There's actually one more thing I tried but didn't make it into the post - if we treat the 12 trading windows as distinct asset

Thanks for sharing my post here. There's actually one more thing I tried but didn't make it into the post - if we treat the 12 trading windows as distinct asset classes characterized by their own expected return/volatility plus the pair-wise correlations, it's possible to construct tangent portfolio/efficient frontier and get the optimal weights. The negative expected return on a few of them certainly makes the exercise less meaningful though.

ndr 6y ago
Thanks for writing this! What time zone are your dates/times in?
- hueyduck 6y ago
  Thanks for the support! I am in EDT (GMT-4).
shocks 6y ago
> There's actually one more thing I tried but didn't make it into the post - if we treat the 12 trading windows as distinct asset classes characterized by their own expected return/volatility plus the pair-wise correlations, it's possible to construct tangent portfolio/efficient frontier and get the optimal weights. This was recently explored here [1] and here [2] 1: https://insignificantbits.com/2020/05/02/turnips/ https://insignificantbits.com/2020/05/02/turnips/ 2: https://news.ycombinator.com/item?id=23058276 https://news.ycombinator.com/item?id=23058276
- hueyduck 6y ago
  Excellent point - I've actually read [1] but not [2]. It's worth noting that neither of them suffers from the same negative weights problem that I am seeing, which is probably due to the fact that 1) an explicit bound between 0 and 1 is imposed on the weights 2) (to a much lesser extent) the analysis is conducted conditional on a specific realized path as of Wednesday.