banjanxed : "any backtest analysis on the ACCLAIM variant vs. standard AIM, or any other AIM variant?"
A variant of ACCLAIM might be to use LD-AIM with a variable amount of virtual and real stock according to the vWave level.
If for instance you were happy to be all in virtual stock (all in cash) at the 60 high vWave levels and zero virtual stock (all in stock) at the vWave=40 lows, you might establish a AIM with all virtual stock, but then adjust that to a blend of virtual and real stock using the stochastic indicated amount where the stochastic top and bottom were aligned to those 40 to 60 vWave levels.
i.e. 1 - (( current - low ) / ( top - low ))
If for instance the current vWave was 55 then
1 - ( ( 55 - 40 ) / ( 60 - 40 ) ) = 0.25
so 25% of the virtual stock might be held in real stock.
If later the vWave declined to 45 then
1 - ( ( 45 - 40 ) / ( 60 - 40 ) ) = 0.75
so 75% of virtual stock might be held in real stock.
Beneath that the LD-AIM would run as per normal AIM, just that you'd periodically adjust the virtual/real stock levels over time to reflect the vWave indicated amounts.
AFAICT Glenn's ACCLAIM works along somewhat similar lines but using different maths and using the VIX instead of the vWave.
To test that would require actual historic figures for the vWave, which I haven't got, so as an alternative if you stochastic the VIX level each month to identify an appropriate amount of stock exposure using extremes of VIX=10 up to VIX=90 then you get
In this VIX's case there is some marginal improvement on a risk adjusted basis of the VIX over the index e.g. if I descale the 3.53 index gain (100% exposure) to the same 87% average VIX indicated stock exposure then 3.56 * 0.87 = 3.1, whereas the VIX timed approach actually achieved a 3.37 gain over the Jan 1990 to present period.
These figures are based on monthly realigning to whatever the indicated virtual/real amounts were at the time. If you delay the realignments perhaps until some other trigger factor occurred then the results may be improved, but equally similar predictive measures might be made against any other investment with potentially as good additional benefits (or loss if the predictor turns out to be bad).
Looking at the monthly open, high, low, close VIX prices I obtained from yahoo, the average is 20.5 with a 8.73 standard deviation. If we set for example a top of the average + 1 stdev and a bottom of the average - 1 stdev = 11.77 to 29.23 range.
Normal distributions imply that the VIX will be within that range for 65% of time, so if we used a timing method that went all-in at the upper value and all-out at the lower value, and in both cases stayed in that state until the opposite case was encountered, then a 2.235 gain is evident against average stock exposure time of 38.23%. In this case the timing added significant risk-adjusted benefit, however that is largely because we used forward time known values i.e. the average and stdev across the whole set. For a real world case you'd be limited to only knowing the average and stdev up to the present date. Correcting the method to use current average and stdev values results in 2.36 gain with 45.8% average stock exposure time - still pretty good (if scaled to 100% exposure = 5.15 gain factor).
To put some more meaningful figures onto those gain factors, the test period was 19.9 years, so the above 2.36 gain factor = 4.4% p.a. annualised whilst the index gain of 3.53 = 6.54% p.a. annualised. These values exclude income (cash interest and/or dividends). Achieving a 2.1% lower p.a. whilst only averaging 46% stock exposure is pretty good as if you could scale that to 100% average exposure it would be the equivalent of 9.6% p.a. on 100% stock exposure (excluding income).
Here's the same chart with that extra detail included (I'm writing this as I'm running the tests)
Here we see the timing based on these measures had us out for extended periods of time (flat green line periods), possibly providing opportunities to have the funds invested elsewhere where maybe a 'in' signal was evident.
Some time back Tom was going to forward iWave data to me to do analysis along the lines of the above, but that never happened before the iWave was taken private. It would be interesting to know how well the iWave did/does fair and/or the same for the vWave.
As ever, better predictors provide better rewards.
News 
Market Data 
Discover 
AI
