Firstly, one or two missed trades are nothing unusual in any trading strategy. Even the best traders typically have a 60-percent win ratio. So I think it's far too early to write the commercials off based only on a couple of days of a single signal. I've found the most robust signals based on the Commitments of Traders data typically last three or four weeks on average.
Secondly, I just did a little study and found that relying on the commercial traders in the S&P 500 the week immediately following big shifts in their positioning isn't a very useful strategy, either. My trading setup for this market works with a three-week delay for the commercial traders. Making your move on the open of trading the Monday following big shifts in the commercial net derivatives positioning would have gained 38 percent in 2008 - not bad, considering how the market tanked! - but during the 2003-07 bull, such a strategy would have gained less than half the return of the market.
Far more importantly, to my mind, even those gains would have occurred in a highly unreliable fashion. My backtesting shows a volatility-adjusted out-of-sample efficiency of just 48 percent for the compound annual return since 1995. That's a completely untradable score. It means the out-of-sample period scored just 48 percent of the in-sample backtested period. You want to see 70 percent or higher for the setup to have a chance of being reliable in real life.
But add a three-week delay before executing signals and the out-of-sample efficiency jumps to 120 percent. And throw in a second signal based on fading the small traders, again with a three-week delay, and that score jumps to 200 percent. As it happens, the small traders were already giving me a bullish signal for the S&P 500. So my own bullish signal for this setup takes effect on the open of July 20. Until then, that setup remains in cash.
Incidentally, out-of-sample testing is just one of many measures of robustness I think you've got to use in backtesting. (See my backtesting results table for some additional ones I use.) But it gives you an idea of how much of the common wisdom out there of how to use the COT data is based on little more than eyeballing some very complex data - not any actual real research. Good luck the rest of this week.