Breadth Bottoms

In my historical and recent reading here, bottom detectors are discussed a lot. Mungofitch seems to have a good one, which is highly respected. Others have shared their versions. I wondered if the component in the weekly published BCC that uses breadth showed anything useful on its own. It’s probably much simpler than the above tools, but I was curious. I realize this has probably been looked at before, so sorry if this is a duplication.

Downloading the signal values shows the 9 day WMA levels (see link below). I tied these to the one-year (252 day) forward GSPC returns, investing at the close one day after the WMA value hits below a certain threshold. Given the descriptions of the other bottom detectors, might this simple indicator be useful to identify a long term bottom?

Based on what I saw for this 9 day WMA calculated by the gtr link above, I looked at two thresholds: -25 and -15.

The first period examined is 12/30/1926, when the WMA is first calculated, through 9/22/2021, so we can measure the future one year return of GSPC. I also looked at two other shorter periods.

For the full period of 94 plus years, here is a summary of the days when the WMA was below -25:

Occurred on 46 days, or about 0.2% of the time.

These occurred in 13 “clusters”, where each cluster ranged from 1 day to 8 consecutive days.

The one year returns were negative for three clusters (6 days), which started on:

10/29/1929
12/16/1930
10/1/1931

The one year returns were positive for the other 10 clusters (40 days), which started on:

6/1/1932
10/13/1937
3/29/1938
10/22/1957
5/28/1962
5/21/1970
10/9/2008
11/20/2008
12/21/2018
3/12/2020

I also looked at a -15 threshold, but with mixed results. I did this because it looks like the WMA measure just dropped below -15 on 9/23, last Friday.

For the full 94 year period, this occurred on 227 days, or 0.95% of the time, This was in 37 clusters. 15 of the clusters had negative one year returns; 22 were positive.

If you exclude the nasty early to mid 1930s and start on 1/1/1938, there were 25 clusters with the WMA < -15 (148 days, or 0.7% of the time). 5 clusters had negative one year returns (2 in 1969 and 3 in 1973); 20 clusters had positive one year returns.

If you look at only the period starting 1/1/1974, there were 9 clusters with the WMA < -15 (62 days, or 0.5% of the time), and all showed positive one year returns.

So, no big conclusions, but thought I’d share what I saw.


Here’s the gtr link to the NHNL component, drawn from the weekly post:

GTR1 Backtester),0,3,excd.a)et3:styp.a:et10!11!18!48:dspo(1)al252:rank(class.a,permco.a,step3)et1:NHNLDiff:et-999999:StockCount:sum(1,1,step4):PcntNHC252:linear(100,ratio(sum(ifgt(ratio(gprc(1),hgprc(2,251)),1,1,0),1,step4),StockCount)):PcntNLC252:linear(100,ratio(sum(iflt(ratio(gprc(1),lgprc(2,251)),1,1,0),1,step4),StockCount)):PcntNHC252WMA9:ratio(sgwsum(PcntNHC252,0,9,8,7,6,5,4,3,2,1),sgwsum(linear(1,1),0,9,8,7,6,5,4,3,2,1)):PcntNLC252WMA9:ratio(sgwsum(PcntNLC252,0,9,8,7,6,5,4,3,2,1),sgwsum(linear(1,1),0,9,8,7,6,5,4,3,2,1)):NHNLDiff:linear(1,PcntNHC252WMA9,-1,PcntNLC252WMA9)

That doesn’t seem to look right in the right panel as I draft this, so here it is in two parts:

First part of the link: GTR1 Backtester

The rest of the link: ?lf-1lp-1h1::iflt(linear(1,ord(1),-1,date2ord(19731217)),0,3,excd.a)et3:styp.a:et10!11!18!48:dspo(1)al252:rank(class.a,permco.a,step3)et1:NHNLDiff:et-999999:StockCount:sum(1,1,step4):PcntNHC252:linear(100,ratio(sum(ifgt(ratio(gprc(1),hgprc(2,251)),1,1,0),1,step4),StockCount)):PcntNLC252:linear(100,ratio(sum(iflt(ratio(gprc(1),lgprc(2,251)),1,1,0),1,step4),StockCount)):PcntNHC252WMA9:ratio(sgwsum(PcntNHC252,0,9,8,7,6,5,4,3,2,1),sgwsum(linear(1,1),0,9,8,7,6,5,4,3,2,1)):PcntNLC252WMA9:ratio(sgwsum(PcntNLC252,0,9,8,7,6,5,4,3,2,1),sgwsum(linear(1,1),0,9,8,7,6,5,4,3,2,1)):NHNLDiff:linear(1,PcntNHC252WMA9,-1,PcntNLC252WMA9)

11 Likes