Update: Oct 29 numbers in the post changed to reflect SPY. the prior post reported another ETF
Since holding SPY for 4 hours on a Sunday afternoon isn’t as risky as holding SPY on a weekday afternoon, such as Wednesday afternoon when the Fed is releasing notes, I was embarrassingly (since I've worked with volatility) alerted by the ex-CEO of a financial firm that time, used in Post 2, is the wrong variable in comparing my model’s risk to the risk of buy-and-hold SPY.
On reflection SPY variance is the correct variable. I use variances because they can be added, then convert to volatility. SPY’s 5-year daily variance has been:
.165 between 9:30 and noon, .198 between noon and 4pm, and .256 overnight.
So 5-year variance for all days is 1260 * (.165+.198+.256)=780; Volatility = 27.9
Variance for 129 afternoons in the original model is 129 *.198; Volatility = 5.1
Variance for 208 afternoons in the relaxed model, described below, is 208 *.198; Volatility = 6.4
Therefore Buy&Hold SPY produces about 3 times the profit but at 6 times the risk of my model.
So from a risk-return viewpoint my model beats B&H but much less than Post 2 incorrectly indicated.
This corrected risk analysis helps when we relax the model’s triggers. I presented the most restrictive model which yields positive returns on 72.1% of the afternoons averaging .203% daily trading about 1 in 10 days. (These numbers differ a bit from early numbers because they include 3 months of 2012.)
By relaxing triggers on momentum, premium/discount, country interaction and other factors the model produces buy signals about once every six days. The win ratio, however, drops to 70% and daily return drops to .143% but with more trading days the net return jumps from 20.5% to 25.9%. Yet as shown in the last column below this model's return/risk ratio stays about constant with the original.
So viewed formally statistics show the models' return-to-risk ratio exceeds Buy&Hold's by a almost two. Viewed casually though, it seems much more: I prefer holding SPY for 182 afternoons and earning 26% rather than holding for 5 years and earning 65%. In future e-mail buy alerts I'll note whether the buy comes from the original or the relaxed model.
* Columns 7 and 8 aren't Sharpe ratios because values aren't annualized and risk-free-rate isn't included so only last column matters. Subtracting the risk free rate from cols 7 and 8 numerators would increase last column edge even more