Re: Backtesting for the Optimum HBPP Allocations
Posted: Thu Aug 15, 2019 7:26 am
Interesting question & reporting - thanks guys!
Applying what I know about statistical analysis to this question...I think you'd have run a permutation test, also known as a bootstrap. I would do this: Take each year's gain and average it with the year before and year after (since a given year's performance is not independent of adjacent years). Randomly shuffle each year since 1972 (or 1974 if you're afraid of the artificial gold spikes in 1972-73), then string together the first 15 years that come up to produce your outcome statistics. Do this several hundred times, or as long as it takes to get a stable distribution. Hope one of you has some time to do this ? (Sadly I don't.)
Going further than that, I'd label each year according to the economic condition represented i.e. prosperity, deflation, inflation, recession. Make sure each of these is represented in the random sample according to historical distributions.
You'd then have to run this for each combination of asset %'s to determine the optimal set. I believe this is what HB did at some point, which is how he arrived at the 25x4 percentages. I'm not sure if he balanced the economic conditions though, so he may have overweighted inflation & recession. That's why the Golden Butterfly is an interesting variation, to me: it recognizes that prosperity has dominated long enough that is not an unreasonable expectation that it will continue to do so going forward.
Applying what I know about statistical analysis to this question...I think you'd have run a permutation test, also known as a bootstrap. I would do this: Take each year's gain and average it with the year before and year after (since a given year's performance is not independent of adjacent years). Randomly shuffle each year since 1972 (or 1974 if you're afraid of the artificial gold spikes in 1972-73), then string together the first 15 years that come up to produce your outcome statistics. Do this several hundred times, or as long as it takes to get a stable distribution. Hope one of you has some time to do this ? (Sadly I don't.)
Going further than that, I'd label each year according to the economic condition represented i.e. prosperity, deflation, inflation, recession. Make sure each of these is represented in the random sample according to historical distributions.
You'd then have to run this for each combination of asset %'s to determine the optimal set. I believe this is what HB did at some point, which is how he arrived at the 25x4 percentages. I'm not sure if he balanced the economic conditions though, so he may have overweighted inflation & recession. That's why the Golden Butterfly is an interesting variation, to me: it recognizes that prosperity has dominated long enough that is not an unreasonable expectation that it will continue to do so going forward.