12/10/10 Updated Update with 3k Returns Distribution Available
Okay, so just keeping everyone updated on recent progress. Re-calibration has indeed brought about some positive results as expected. So far the test run has brought it down to 40% of the previous long-run setting. Still some optimisation and analysis potential out there so will keep you updated as this comes through.
Distribution with old 10k Calibration
Distribution with 4k Calibration set
Distribution with 3k Calibration set
The graphs above detail the effects of the calibration shift. The Algo remains normally distributed throughout the period and across all calibration sets (will try and include chi-tests with the final release).
Another thing you might notice (already better reported elsewhere – will try and hunt some background literature for the main article) is the more ‘beta’ styled distribution of the benchmark over the 90s onwards as opposed to the more ‘normal’ distribution from 1954 to 1990 (see distribution graph here).
Then again, the previous time period, ’54 to ’90, is still 36 years worth of data compared to the twenty years since ’91 so this might also be adding to the skew. In the long run, things should normalize, which would be good news for everyone…