I've just posted details of my newly revised trading setup for the Russell 2000 index on my latest signals page table. This setup is in cash. My previous setup for this index has been on a bullish signal, and I'm presently long RUT myself, so I will be looking at the technicals in coming days to sell. My initial floor is any drop below the July 7 low, minus a little cushion so I don't get knocked out by the stop-runners. (I usually set my cushion at the recent largest Average True Range on the 15-minute chart.)
I'm looking forward to trading this new RUT setup. Its backtesting results are sweet, including a Sharpe ratio of 5.1 and Robust Sharpe of 4.5. Its Compound Annual Growth Rate beat the underlying index since 2003 by over 60 percent (including trade friction of 0.2 percent per trade). This again lays to rest any notion that the Commitments of Traders data has lost its utility in recent years. The setup works by combining the signals of two of my best setups from the "smart money" commercial traders when they hit specific extremes in their futures and options positioning as a percentage of the total open interest. The two setups effectively look at the data from different time frames and use different moving averages and standard deviation values. When they don't agree, which happened 54 percent of the time since 1995, the setup goes to cash. Despite this, the setup had compound annual growth of 20.3 percent (25.5 percent since 2003). In out-of-sample testing, it scored a CAGR efficiency of 93 percent and regressed annual return efficiency of 107 percent - very nice numbers.
I also filtered possible setups with a new test I've devised for robustness, which I call the "walk-around test." This exercize varies each of the parameter values in 16 different permutations. The idea is to help see to what extent the setup may have been the result of data-mining. If "neighbouring" setups with slightly differing parameter values were to produce much worse results, that would be a bad sign. It would mean the setup chosen likely wouldn't perform anywhere nearly the same in real-life conditions. In fact, the RUT setup I've chosen did very well in this test as well, with an average walk-around efficiency of 91 percent. This means that in all 16 tests - and for 16 different measures like CAGR, Sharpe score, regressed annual return and out-of-sample results - the "neighbouring" setup's result was 91 percent of the chosen setup's result on average. A promising sign. Good luck this week!
TAGS: COT, Commitments of Traders, market timing, trading system development, CFTC, Commodity Futures Trading Commission, COTs Timer, Russell 2000, CAGR, RAR, Sharpe, out of sample testing, walk-around testing
No comments:
Post a Comment