Oct

3

Remember trend days? Recently trade has been, I wouldn't say choppy, but with a lot of reversals. The legs are better than trend days in the past, point wise at least. Which led me to study trend days over the years. See how they cluster, and there have been none this year, at least how I defined them. Interesting to say the least. What does it mean? Algos? …that's my theory.

Adam Grimes asks:

What definition are you using?

Zubin Al Genubi answers:

Pick one, any one will do. The moral of the story is the same. Hint: These were up days only.

Adam Grimes responds:

I think the definition matters quite a bit. I would typically work with a definition like:
- opens near the low, closes near the high
- range > X * ATR(20) (starting with values of something like 3 for X)

Key concept is a larger-than-average day with open and close at opposite ends of the day. The reason I asked is I count far more than you do so I wondered if you were looking at "truly exceptional" days. I'll take a look at distribution by year later today.

I don't see any value in thinking in points, either. Saying the legs are longer point-wise when the price level is so much higher doesn't say anything at all, right?

I don't replicate your numbers at all. These are using open and close in the bottom/top X of the range. One table for day's range > 1.5 * ATR, the second for > 2 * ATR. I can't imagine the usefulness of extremely restrictive criteria for trend days, which is why I asked how you were defining them. At any rate, ES futures do not bear out your claim. What am I missing?

[Click to view tables.]

Bullish days: high-low range > 1.5 * ATR(20) table.
Bullish days: high-low range > 2 * ATR(20) table.

And why only look at bullish days? The thesis collapses further when we consider both directions (which I would argue we should do).

Bearish days: high-low range > 1.5 * ATR(20) table
Bearish days: high-low range > 2 * ATR(20) table

Is my math off? (Seriously asking, working with a relatively new and somewhat unvetted system here.) Or what am I missing?

Zubin Al Genubi writes:

Adam, My definition was pretty tight so your math is not off, my def is not too good. I subsequently did down days, which were larger than up days, and fewer of them. I am trying to find conditions that precede trend days. The only thing I am seeing is other trend days, like those 3 big up days 2 months ago ( which my system did not catch) Still playing with it as well.

Adam Grimes offers:

Interesting timing. I'm redoing some old work on the same subject:

A new look at measuring trend days
Catching the wave: ideas for finding trend days
Finding trend days in index futures

Ralph Vince offers:

Isn't it vital as well that the open be nearly the same as the low or the high!

Zubin Al Genubi responds:

There are a lot of interesting ways to define it, and I like your percent of new bar highs one. The range constriction break out is a good precursor. I've found there's not that many trend days so it doesn't seem like something to count on. I have a few more definitions I will be looking at: range efficiency |C-O|/(H-L), Kaufman efficiency ratio: |C-O|/\sum|\Delta C|, Choppiness-type ratio: total path divided by daily range, e.g. \sum|\Delta C|/(H-L). I just used retracement for the post.

I found an average of 28 trading days between trend days, but with wide dispersion over the years. Some years more, some less. That's what led me to the study, my observation of few trend days, lots of Z days this year, with long too straight un-choppy, overly efficient legs, caused by (in my theory) by bots.

Looking at your low vol triggers (for example NR5,NR7) over that last 5 years, I'm finding it persists to the next day. Not what I would have expected. Also, I found 0 instances of 3-consecutive-decreasing inside days (Russian Doll set up) over the last 10yrs!

Adam Grimes writes:

Yes, that no longer surprises me (failure of low vol triggers). Technical analysis literature and marketing has a lot to say about compression as a trigger. I don't think it's any kind of trigger. What's the best predictor of tomorrow's volatility? In models, the answer is today's volatility, by a wide margin. (Consider ARCH and GARCH models.) So low vol tends to lead to further low vol. (Also the observation one of our list members misses when he's concerned about low VIX = impending market crash.)

But if and when you DO get a move out of low vol it tends to be more directional and probably leads to a higher probability of trend days. One issue with all of this work is that good trend days are sparse enough you can't really generate reliable stats.

Zubin Al Genubi observes:

There were only 7 bottom tick/top tick day in the last 15 years and no perfect down days. So the definition needs to be wider. I used 10/90.

Adam Grimes responds:

Yes it has to be something like this. There's also a trend day subset where one part of the day might be a powerful and large trend and the other a range or a trend in the opposite direction. Intraday data can obviously untangle this, but one of the limitations of using daily data is that these types of days will be missed in the count.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search