A reader asks, “Is a system more robust if it is tailored to a specific market condition?” This is response to a point in Blogging Cross-Talk: Psychology Observations where I said,
Systems can be optimized to work in one market regime, with a signal or overlay allowing the system to switch when regimes change, or the systems can be optimized to work “overall” regardless of the particular regime in place at the moment.
The first thing I would need, to answer a question about “robustness,” is a working definition of the term. I would call a system “robust” if it met my predetermined absolute benchmarks, or performed better than a competing system in terms of risk-adjusted metrics.
Evaluation of a system designed or tailored to a specific market regime is done in terms of the entirety or totality of the system performance. I will use two public-domain systems as examples, CANSLIM and “Thermostat.”
CANSLIM is a popular method for individual stocks, which focuses on those with high earnings growth, institutional accumulation of shares, and other metrics (including low float size in the original version). CANSLIM includes a market-timing element, or regime-switching measure, whereby the practitioner refrains from trading stocks when overall market action is unfavorable. So in order to evaluate CANSLIM performance, I would not just look at the performance of trades taken when the market was favorable, but I would include the return on alternate investments (i.e. “cash”) when the market was unfavorable, and evaluate that return stream against my absolute benchmarks or competing system.
“Thermostat” is less well-known, and is a futures-trading system designed to run across several markets simultaneously. “Thermostat” has three parts, a trend-following system, a mean-reverting or “countertrend” system, and a “regime detector” that decides, for each market, which system should be applied at the moment. You could consider it a blend of two systems, traded across a variety of relatively uncorrelated markets. This is an idea that has definite scale requirements, that is, it needs a certain large (for retail chumps like me) amount of capital to implement, but it’s a good idea, nonetheless.
I think it’s possible to build very good systems that are tailored to specific market regimes, as long as the regime-detecting element works well enough, and I think that the robustness of the tailored system has to be measured across its entire return stream – including alternate investments when the system is “out of market.” It’s a fruitful area for analysis, and one that I’ve investigated and will keep investigating, even though it’s not an idea I trade with today.
If you liked this post, you might be interested in subscribing to my RSS feed. If you prefer, you can get a nightly RSS email update sent on the days that I post! There are convenient “Subscription” icons near the top of the right sidebar.
To view my actual trades and model portfolios for the different systems I track, visit The Rempel Report. If you’d like to become of member of The Rempel Report, you can register here. At The Rempel Report, I track model portfolios for four different mechanical trading systems, disclosing all results (good and bad) at regular intervals. I also track my personal portfolio, and disclose all trades before I make them. Members receive email notification of new posts and can contribute to the site through comments. Registration is still free!
2 Comments
Bill,
Do you use a system that combines trend-following and “countertrend” indicators? Also, I would like to read your ideas on “regime detecting” elements. I remember seeing a chart by Ned Davis Research that adjusted OB/OS oscillator levels based on his monetary model. Interesting stuff.
The Timing system has a longer-term trend-following element combined with the Fear/Greed portion, which tends to be countertrend. I don’t use anything that does this for individual stocks or groups of ETFs right now.
I think the TF/CT method works in one of two situations: either a “regime detector” is used to switch the model, or the two systems are used simultaneously but on different timeframes, i.e. long-term TF but taking advantage of short-term CT opportunities.
The little bit of playing I’ve personally done with “regime detecting” has been in the concept of “efficiency of movement,” i.e. how much of the issue’s movement in the past x units of time has been in the direction of trend? Obviously nothing ever gets 100% efficient (unless it’s “limit lock” several days in a row), but comparing the efficiency in recent times to some historical (or longer term) norm appears to be a promising technique there.