Just a thought - is there possibly a secondary criterion (orthogonal to the current sector ranking) that would assist in selecting the "better" sector (or maybe determine position sizing) for weighting the top 2 sectors?
Since the sector change to XLU took place on July 25, XLU has been a poor performer while XLB was the star (assuming it was held since that date while in reality it was swapped in/out vs XLK which was more or less flat). See the S&P sectors performance chart (http://stockcharts.com/freecharts/perf.php?[SECT] and set it to 49 days).
It's interesting to review the commentary at the time (http://market-mousetrap.blogspot.com/2013/07/07262013-sector-update.html) relative to the actual performance. XLI looks like the 2nd best performer after XLB, tho according to the commentary, it was ranked dead last at the time the switch to XLU was made... hindsight is 20/20?
In the background I've had a friend ask me every question under the sun. "Why 1 sector instead of 2?" "Why this measurement instead of another range?" I finally worked out a way to test this questions and measured EVERY number of sectors on EVERY measurement point.
The first thing I found was that 2 sectors outperformed 1. That was bizarre, but I made the switch.
But that "bizarre" result led me to keep trying to find out why and how 2 sectors would outperform 1. Turns out there is a momentum weakness that often overlaps the first position. Over any three year period it would outperform the S&P, but there were some 1 and 2 year periods it would underperform.
I had noted the same problem when I developed the full model: out of 98 industries, the 1st industry performed the WORST, even though the top 10 outperformed. I've been treating that 1st of 98 position as a stop loss for the past two and a half years, but had no way to do so on the sector model (since it only has 9 positions instead of 98).
I've been suspecting that this has been part of the reason the full model outperformed the sector model -- but for two and a half years I had no way to correct it.
I finally figured out how to correct it, and the improvement will be rather substantial.
Not really "hindsight" as much as "helplessness" until I could figure out a way to solve the stop loss problem that would overtake the single sector model on occasion.
Just a hunch - I suppose it's reasonable to expect that a more concentrated portfolio (comprised of a single sector) would entail more "risk" (namely: less consistency over a given period of time). In other words, you'd have to wait longer for your single-sector portfolio to beat the benchmark compared with a 2-sector portfolio, all other things being equal. The point about the sector model not being "fine grain" enough to diversify the positions relative to the 98-industry full model makes a lot of sense to me. You simply can't bet on a single flaky horse (however promising the long-term track record) with so few races.
"Risk" goes both ways, though. "Risk" just means that you can't trust it to follow the full S&P benchmark -- but if you are trying to outperform, you don't WANT to follow the full S&P benchmark.
The problem had to do with the juxtaposition of mean reversion and momentum:
1) Mean reversion simply says something is a certain deviation from the mean. 2) Momentum simply says something is going in a particular direction.
The top 10% on my metrics is a certain deviation from the mean, but could STILL be going in that direction for a while. The key is to find a knife that has stopped falling and THEN pick it up.
When I originally tested my metrics on 98 industry groups, positions 2-11 were the best, and position 1 was the worst of all 98.
When I tested on the DOW I had the same effect -- positions 2-4 outperformed positions 1-3.
However, positions 1-10 out of 98 and positions 1-3 out of 30 still outperformed the S&P. With only 9 sectors I couldn't figure out a way to shave off that final bit of negative momentum, and figured I'd just need to ride it out.
I was intending to expand the sector model to all 40 that SPDRS.com offered, but I couldn't really do a lot of testing with that number and couldn't do real time trading either. Fortunately I found a different way to solve the problem:
A) I pick the top 2 on a mean reversion metric, and B) I eliminate the 1 with the worst momentum metric.
Of course, I had to work out a momentum metric, but when I studied ALL sector positions on ALL metrics and graphed it out, it was rather obvious. :-)
Determining whether a falling knife is equipped with a rocket ready to launch in the other direction is tricky business. You mentioned before that aligning timeframes is an important aspect of making joint criteria work. Both mean reversion and momentum measurements are dependent on a timeframe. Is it simply a matter of choosing the same timeframe or is there possibly another relationship that governs how they relate to each other for the sake of "divide and conquer"?
You hit the nail on the head: momentum is a shorter time frame than mean reversion (about half). It spends half of the time going in one direction and then half coming back. If the metric on the shorter time frame is more extreme than on the longer one, it's still going the wrong way.
Just a thought - is there possibly a secondary criterion (orthogonal to the current sector ranking) that would assist in selecting the "better" sector (or maybe determine position sizing) for weighting the top 2 sectors?
ReplyDeleteThat's the purpose of the sector changes. The current two sector configuration is XLU & XLB. Of those two, XLB is the better short term play.
ReplyDeleteWhen I do change to the single sector configuration (very soon), the performance gain should be between 5%-10% per year.
Since the sector change to XLU took place on July 25, XLU has been a poor performer while XLB was the star (assuming it was held since that date while in reality it was swapped in/out vs XLK which was more or less flat). See the S&P sectors performance chart (http://stockcharts.com/freecharts/perf.php?[SECT] and set it to 49 days).
ReplyDeleteIt's interesting to review the commentary at the time (http://market-mousetrap.blogspot.com/2013/07/07262013-sector-update.html) relative to the actual performance. XLI looks like the 2nd best performer after XLB, tho according to the commentary, it was ranked dead last at the time the switch to XLU was made... hindsight is 20/20?
In the background I've had a friend ask me every question under the sun. "Why 1 sector instead of 2?" "Why this measurement instead of another range?" I finally worked out a way to test this questions and measured EVERY number of sectors on EVERY measurement point.
ReplyDeleteThe first thing I found was that 2 sectors outperformed 1. That was bizarre, but I made the switch.
But that "bizarre" result led me to keep trying to find out why and how 2 sectors would outperform 1. Turns out there is a momentum weakness that often overlaps the first position. Over any three year period it would outperform the S&P, but there were some 1 and 2 year periods it would underperform.
I had noted the same problem when I developed the full model: out of 98 industries, the 1st industry performed the WORST, even though the top 10 outperformed. I've been treating that 1st of 98 position as a stop loss for the past two and a half years, but had no way to do so on the sector model (since it only has 9 positions instead of 98).
I've been suspecting that this has been part of the reason the full model outperformed the sector model -- but for two and a half years I had no way to correct it.
I finally figured out how to correct it, and the improvement will be rather substantial.
Not really "hindsight" as much as "helplessness" until I could figure out a way to solve the stop loss problem that would overtake the single sector model on occasion.
Just a hunch - I suppose it's reasonable to expect that a more concentrated portfolio (comprised of a single sector) would entail more "risk" (namely: less consistency over a given period of time). In other words, you'd have to wait longer for your single-sector portfolio to beat the benchmark compared with a 2-sector portfolio, all other things being equal. The point about the sector model not being "fine grain" enough to diversify the positions relative to the 98-industry full model makes a lot of sense to me. You simply can't bet on a single flaky horse (however promising the long-term track record) with so few races.
ReplyDelete"Risk" goes both ways, though. "Risk" just means that you can't trust it to follow the full S&P benchmark -- but if you are trying to outperform, you don't WANT to follow the full S&P benchmark.
ReplyDeleteThe problem had to do with the juxtaposition of mean reversion and momentum:
1) Mean reversion simply says something is a certain deviation from the mean.
2) Momentum simply says something is going in a particular direction.
The top 10% on my metrics is a certain deviation from the mean, but could STILL be going in that direction for a while. The key is to find a knife that has stopped falling and THEN pick it up.
When I originally tested my metrics on 98 industry groups, positions 2-11 were the best, and position 1 was the worst of all 98.
When I tested on the DOW I had the same effect -- positions 2-4 outperformed positions 1-3.
However, positions 1-10 out of 98 and positions 1-3 out of 30 still outperformed the S&P. With only 9 sectors I couldn't figure out a way to shave off that final bit of negative momentum, and figured I'd just need to ride it out.
I was intending to expand the sector model to all 40 that SPDRS.com offered, but I couldn't really do a lot of testing with that number and couldn't do real time trading either. Fortunately I found a different way to solve the problem:
A) I pick the top 2 on a mean reversion metric, and
B) I eliminate the 1 with the worst momentum metric.
Of course, I had to work out a momentum metric, but when I studied ALL sector positions on ALL metrics and graphed it out, it was rather obvious. :-)
Determining whether a falling knife is equipped with a rocket ready to launch in the other direction is tricky business. You mentioned before that aligning timeframes is an important aspect of making joint criteria work. Both mean reversion and momentum measurements are dependent on a timeframe. Is it simply a matter of choosing the same timeframe or is there possibly another relationship that governs how they relate to each other for the sake of "divide and conquer"?
ReplyDeleteYou hit the nail on the head: momentum is a shorter time frame than mean reversion (about half). It spends half of the time going in one direction and then half coming back. If the metric on the shorter time frame is more extreme than on the longer one, it's still going the wrong way.
ReplyDelete