Post New

Back in the following post - https://www.shrewdm.com/MB?pid=430752207 - musselmant wrote / quoted the following:

"When we see days like these, where 90% of the volume on the NYSE is higher, and 90% of all stocks are in the green, it tends to be bullish for the next 12 months of gains....

The last 90/90 day was April 9th of this year, when we had a 90-day pause on Tariffs. The S&P 500 is up 29.8% since then; the Nasdaq 100 has gained 37.5%.

Since 1982, we have seen one negative, one flattish, and 12 positive sets of returns over the 12 months that followed a 90/90 day. It’s not a guarantee, but it suggests favorable odds for remaining constructive." Barry Ritholz

Within the data collected for calculating the Extremely Simple Bottom Detector (ESBD) is [NYSE advn], [NYSE decln] & [NYSE unchn] and so could [%advn] (of stocks) be used to help determine re-entries into the market following corrections & bear markets?

In general it's difficult to determine when to re-enter. Frequently a bottom detector can fire multiple times at various levels of distress as the market makes its way towards the ultimate low. And in a fully fledged 'full panic' bear market this may exhibit itself in multiple waves of bottom signals (of various strengths) between which the market takes a bit of a 'breather' before further concern / panic sets in leading to further declines. As a result I've frequently suggested that one looks for an 'up day on increasing volume'. But how strong an 'up day'? musselmant's quote may suggest a possible measure of 'up day' strength.

After some additional work I propose that a good potential 'pay attention as the decline may be over' indicator is …

[NYSE advn] / [NYSE total] > 86.48% [AND] [^GSPC volume today] > [^GSPC volume yesterday]

… where the 86.48% is the mean - 2 standard deviations of the [NYSE %advn]. This combines a representation of the quoted idea that '90% of all stocks are in the green' combined with an 'up day on increasing volume' (that is a 86.5% green day will result in the S&P500 being positive and so combining that with increasing volume).

It seems to work pretty well. If you combine this with a MajorBot or ExtrmBot from the ESBD the only time this re-entry signal failed to fire is during the 2011 correction (the advance criteria was met but not the increasing volume) - so you might still want some sort of back-up entry criteria. Remove the 'up volume' criteria & there are far more interim signals and the re-entry indicator will fire during 'breather' periods. However generally speaking if one sees an 86.5%-90% advance day on increased volume (from the prior day) the worst is likely over for the current (bear market) decline.

For the visually inclined, this is how it looks when combined with the ESBD - https://ibb.co/V004nBYG - through to the end of the prior week.

Platy

[NYSE advn] / [NYSE total] > 86.48% [AND] [^GSPC volume today] > [^GSPC volume yesterday]

Is that 86.48% value from the regression calculation, or from tuning, or...?

Is there a mound of toast range around that value? If so, how broad?

Eric Hines

Is that 86.48% value from the regression calculation, or from tuning, or...?

It's the outcome of the

mean of the [NYSE %advn] PLUS 2x standard deviation of [NYSE %advn]

where the 2 was chosen to approximate the 90% (as 2.5x - 3x - 95.6%-104% - would pretty much 'never' fire due to the normal distribution being only an approximation of the actual distribution) where the mean (49.97%) & sd (18.26%) are both calculated since 2005-10-27 (the complete data series I retain). So *somewhat* tuned - but no real effort to optimize it - and in practice 85%-87% works better than 90% when combined with the 'improved volume'. 90% is just too restrictive when combined with the 2nd criteria, but works a better than 86.48% when used it's own single criteria. I expect that a one could find a better figure - the 'mound of toast' range appears quite broad, but I've made no real effort to explore it widely preferring to obtain something that works for the data I have & doesn't seem too tightly tuned.

(Interestingly using 85%-86% provides a re-entry point for 2011 on 2011-10-27 and so offers further improved performance across the data set at the potential expense of further curve fitting - YMMV. It's tricky however 85% or 86% may well be what folks prefer to roll with. Part of the reason with posting all this is to let other folks express an opinion on the tuning or otherwise.)

One of the potential 'builds' I'm considering is just to measure the mean & standard deviation across only a year or two to allow for it to flex a bit, but I've not really explored that yet.

Platy

(I note that despite extensive checking of the first post I wrote "… where the 86.48% is the mean - 2 standard deviations of the [NYSE %advn]" this should be "… where the 86.48% is the mean + 2 standard deviations of the [NYSE %advn]". Apologies.)

Platykurtic

For the NSYE advn and total, would you mind sharing your data source. As discussed in other threads, different data sources have different values.

Aussi

Platykurtic

For the NSYE advn and total, would you mind sharing your data source. As discussed in other threads, different data sources have different values.

Aussi

I meant to say, name your data source, I didn't mean to share the data.

Regards

Aussi

For the NSYE advn and total, would you mind sharing your data source. As discussed in other threads, different data sources have different values.
It's a bit manual, however I key the data in from the WSJ Markets Diary. I should, but have not bothered to automate the collection, as there's some value in keying the data manually & getting a 'feel' for the breadth of the market along the way. For the volume data I use Yahoo! for ^GSPC also keyed manually.

https://www.wsj.com/market-data/stocks/marketsdiar...

For the historical data I (believe I) used http://unicorn.us.com/advdec/ (or similar) and switched to the WSJ source as that most closely matched the historical collection at the time (it also looks like I can push back much further in-time - I may do so as I recently discovered Libreoffice Calc isn't limited to the apparent 10k rows that I thought it was). As a general comment there's been significant drop in freely available data for small individuals. It's interesting as larger operations can likely scrape the data anyway - it's just some programming for them - if they don't buy it outright.

With a free account at Barcharts, one can do the same scraping on 20 pages of data a day, and their data is extensive - proprietary tracking for Advs/Decs/pama 5/20/50/100/150/200, etc etc. FWIW I happen to scrape those pages weekly to drive and derive some short and intermediate term breadth signals.

A paid account is pretty cheap and gets you api access to non-proprietary indicators and a good number of actual downloads. I could afford it but I'm not sure about investing even more time in more data, given the apparent statistical irrelevance of my signal tracking.

FC

With a free account at Barcharts, one can do the same scraping on 20 pages of data a day, and their data is extensive - proprietary tracking for Advs/Decs/pama 5/20/50/100/150/200, etc etc.

You can also download data, limited to 1000 rows. Limited to 1 download per day. Much easier than scraping one page at a time.

Per account. Account = email address.

Nothing says that you can only have 1 email address. There are plenty of places you can sign up for a free email address.


Do 2 sorts, one ascending and one descending to get the 1st & last 1000 rows. Which nicely handles the Russell 2000.

Post New

Mechanical Investing FAQ

---

Contact Shrewd'm
Contact the developer of these message boards.