Hi all,

curious about a few things here and wanted to start a discussion :) , I noticed one of the algorithms in the strat library → 

which refers to this paper by Avallaneda → 

In QC's implementation (attached backtest, the PCA approach), the reversion was based on just the OLS' residuals and its mean/std, but the residuals themselves was never actually modeled using the O-U process to try getting the speed of reversion, ‘K’, nor incorporated the drift estimate to get the modified s-score →

96178_1629391465.jpg96178_1629391698.jpg

 

The std that was used in the QC implementation for the s-score was hence the sample's residual std instead of the theoretical equilibrium std, The algorithm performs considerably okay though, so I'm wondering if anyone have any insight on trying to incorporate the full model, and if it would make much difference (while I'm implementing it myself :D :) 

 

Author