My brand new COTs Timer trading setup for gold will end its five-week bullish signal on next week's open of trading. It's been bullish since Nov. 24. I've just posted the results and parameter values for the new setup on my latest signal table. On Dec. 29, the setup will either go to cash or bearish. This setup is the second one I've developed based on my new testing that includes detrended price data (in which the long-term bullish or bearish bias of the prices is stripped away) and Monte Carlo testing (which sees how the setup would have done in 6,000-plus randomized market runs). My first new setup using these tests was for the S&P 500. See more on that one here.
My gold setup uses two groups of traders to come up with long, short or cash calls. It's been in the market only 44 percent of the time since 1995, yet beat gold handily. If the two traders don't agree, the setup goes to cash. The first signal fades (trades opposite to) the large speculators when their net percentage-of-open-interest position hits extremes of positioning. It has no trade delay. This setup remains bullish this week. The second signal fades the large specs when their total open interest hits positioning extremes. This setup gave a bearish signal the week of Nov. 4 and stayed bearish three weeks before going back to bullish. So with the trade delay of seven weeks it will put the overall setup either in cash or bearish next week, depending on what the first setup does. See all the details in the notes to that table. I'll upload a sample spreadsheet for this setup at some point soon.
Portfolio Update: Also, just updated my portfolio page with current positioning based on my COTs Timer trading system and a recently closed trade for my Nikkei trading setup. Hope you have a great holiday and good luck in the New Year.
TAGS: gold, S&P 500, SPX, Nikkei, COT, Commitments of Traders, derivatives, Black Swans, market timing, trading system development, CFTC, Commodity Futures Trading Commission, COTs Timer, out-of-sample testing, Monte Carlo, walk-around testing
5 comments:
Alex,
Tim here. On your stop loss value you use the most recent maximum drawdown. On the SP500s it was around 10% I believe. My question is, has this been tested also? OR is there perhaps a better metric - a combination of largest most recent drawdown and say a resistance support level or something like that?
Thanks
Alex,
A followup to my previous question - how do you define the time period for "recent" largest drawdown? How long?
Thanks,
Tim
Hi Tim,
The stop is based on the largest entry-to-intratrade low hit during the backtested period, which starts in 1995 for most markets. I didn't do any testing on this stop level, in part because it's determined a posteriori. It couldn't possibly ever be exceeded during testing.
There are probably many other possible stop methods, including some based on price. As I've just mentioned in a recent comment, I've tested various price filters, but they significantly degraded the results. Not that there aren't any that could improve results. I haven't done an exhaustive search either. Your input on specific parameters for this is welcome.
Regards,
Alex
Alex,
Ok, I will start testing some of the metrics we were using on the fund. One that might be useful is this:
Since we know what the holding period is up to 3 weeks anyway due to the lag in the signal, so, I take the 1 std dev value of the past 50 days (typical time period) and any negative move beyond it would cause me to sell it. I don't want to stay in a down move if it is out of 66% of the moves. Heck, I can even look at a looser stop using 2 std dev.
I think that using the time period back to 1995 is a bit too long since there are many things that have caused the "speed" of the market to change: decimalization, internet, etc. I found that using semi-comparable time frames for buy and sell signals is vital.
Simply, if the SP has moved, say, 20 points in the past 50 days, and all of a sudden it moves beyond 20, than one of 2 things has occurred: a black swan event or a significant erosion/build of the underlying market. Either way, the market would then be outside of the parameters we would be comfortable with. If the system works with a particular construct of of parameters, then the time to sell is when the parameters move outside of those constructs. It is less about the market and more about the system's variables.
Any thoughts?
Hi Tim,
Makes sense, but the timeframe is the issue - and how to decide we've moved into a different kind of market. The reason I like the 1995-today dataset is because it includes at least one major bullish, bearish and trading-range market. It's hard to gauge until after the fact that we've entered a new market condition, so it's good for a setup to operate robustly in all conditions, if possible. Otherwise, you need a filter to gauge the type of market and enough data within each market type to get a robust setup. With the weekly data of the COT reports, that last issue is a problem.
Regards,
Alex
Post a Comment