Since starting this “Challenging the 2013 Nobel Prize” series VIX, the market's accepted measure of volatility, has averaged 10.85 - the lowest 2.5 month average in history. My model was developed with 5 years of history so it's useful to measure performance versus VIX during this entire period. The plot of model returns vs VIX levels shows a visual difference (wins more prevalent) only at high VIX levels. Having good returns at any VIX level says the model is robust with respect to VIX.
(Another pleasant fact. The plot shows only 3 losses exceeded -0.5% whereas 15 gains were greater than 0.5% .)
From a statistician's view the model's win ratios of 73.6%, 68.2% 75% and 79.5% are almost statistically equal so VIX level doesn’t matter. But from the a data-miner's view, like myself, the results differ and one should expect to win more often when VIX is above 13 ( Win ratios of 75.0% or 79.5%) than at levels under 13 (Win Ratios of 73.7% or 68.2%.) Because of this I'll include VIX level in my future e-mails on signal days and you should keep this table in mind -- even thought statisticians will say 74.4 is the best Win Ratio estimate.
Follow-up, Post 6 showed that the model could predict asset classes correlated with SPY (plus surprisingly, highly-rated corporate bonds.) An economist asked if the model could predict volatility?
It's well known that SPY and VIX are inversely correlated so he specifically asked: Does the model predict afternoon VIX in excess of what existing SPY-to-VIX negative correlation implies? The answer is "No." It predicts volatility as it predicts other correlated assets reported in Post 6, i.e., XIV (a tradable ETF tied to VIX short futures because of negative correlation) rose 70% of the days as implied by its correlation to SPY.
So LQD, highly-rated corporate bonds, remains the only asset not correlated with SPY that the model can predict. Again, any opinions on this will be appreciated.