Post New

Growth Trend Timing exits stocks when the S&P500 is below the 10 month (43 week) SMA, but *only* if neither of the two FRED indicators is down YOY. If both are up YOY then you are to ignore the SMA sell signal.

The S&P500 has been above the 43 week SMA for quite some time now, for 28 weeks. But the FRED signal has been to obey an SMA sell signal.

Two days ago FRED did the monthly update, and now both are up YOY, so any SMA sell signal should be ignored.

Ray

Would you mind posting the dates for the FRED signal or point me where to look.

Thanks

Craig

Links: https://fred.stlouisfed.org/data/RRSFS.txt and https://fred.stlouisfed.org/data/INDPRO.txt
You have to look at the "Last Updated:" line.

For bash users, here are the two key lines in my script:

curl -sL "https://fred.stlouisfed.org/data/INDPRO.txt" >z1
sed -n '/^[12][0-9][0-9][0-9]-/p' z1 | awk ' $1 >= "1950" {print $0}' | sort -r -k1,1 -u - ind_db.txt | tac >ind_db.txt

(Although I now see that there is a potential race condition on the file ind_db.txt. Not safe to read and write the same file in a pipe.)

My script runs every day and downloads & keeps the most recent file from FRED. Below is a list of the last handful. You can see that the updates come around the middle of the month and sometimes also get an update a few days or weeks later.

rrsfs-2021-12-15.txt
rrsfs-2022-01-14.txt
rrsfs-2022-02-10.txt
rrsfs-2022-02-16.txt
rrsfs-2022-03-16.txt
rrsfs-2022-04-14.txt
rrsfs-2022-05-11.txt
rrsfs-2022-05-17.txt
rrsfs-2022-06-15.txt
rrsfs-2022-07-15.txt
rrsfs-2022-08-17.txt
rrsfs-2022-09-15.txt
rrsfs-2022-10-14.txt
rrsfs-2022-11-16.txt
rrsfs-2022-12-15.txt
rrsfs-2023-01-18.txt
rrsfs-2023-02-14.txt
rrsfs-2023-02-15.txt
rrsfs-2023-03-15.txt
rrsfs-2023-04-14.txt
rrsfs-2023-04-24.txt
rrsfs-2023-05-16.txt
rrsfs-2023-06-15.txt
rrsfs-2023-07-18.txt
rrsfs-2023-08-15.txt
rrsfs-2023-09-14.txt
rrsfs-2023-10-17.txt

More about the FRED data...... Sometimes the FRED data gets readjusted long after the first report and multiple times. Here, for example is the Real Retail and Food Services Sales (RRSFS.txt) value for one particular month (Feb 2022) from different file dates:

rrsfs-2022-03-16: 2022-02-01   231587
rrsfs-2022-04-14: 2022-02-01   233096
rrsfs-2022-05-11: 2022-02-01   233216
rrsfs-2022-05-17: 2022-02-01   233062
rrsfs-2022-06-15: 2022-02-01   233062
rrsfs-2022-07-15: 2022-02-01   233062
rrsfs-2022-08-17: 2022-02-01   233062
rrsfs-2022-09-15: 2022-02-01   233062
rrsfs-2022-10-14: 2022-02-01   233062
rrsfs-2022-11-16: 2022-02-01   233062
rrsfs-2022-12-15: 2022-02-01   233062
rrsfs-2023-01-18: 2022-02-01   233062
rrsfs-2023-02-14: 2022-02-01   232712
rrsfs-2023-02-15: 2022-02-01   232712
rrsfs-2023-03-15: 2022-02-01   232656
rrsfs-2023-04-14: 2022-02-01   231820
rrsfs-2023-04-24: 2022-02-01   229631
rrsfs-2023-05-16: 2022-02-01   229631
rrsfs-2023-06-15: 2022-02-01   229631
rrsfs-2023-07-18: 2022-02-01   229631
rrsfs-2023-08-15: 2022-02-01   229631
rrsfs-2023-09-14: 2022-02-01   229631
rrsfs-2023-10-17: 2022-02-01   229631

Maybe my script is overkill and you'd be fine with just the most recent file. I dunno, I just keep all the unique files according to "Last Updated". But you do need to grab the file right before you make any decisions based on the data.
Of course, if you are doing backtesting you need to use the data that was current at that time, not the most recent data, to avoid look-ahead bias.

And because I don't know when to stop, here is all the changes to the Feb'2015 value:
(The 1st column is the number of months that had the same value.)

      2 rrsfs-2016-01-20: 2015-02-01 183651
      1 rrsfs-2016-02-19: 2015-02-01 183472
      2 rrsfs-2016-03-15: 2015-02-01 184318
      2 rrsfs-2016-04-13: 2015-02-01 183697
     17 rrsfs-2016-05-13: 2015-02-01 184176
      5 rrsfs-2017-02-15: 2015-02-01 184119
      9 rrsfs-2017-05-12: 2015-02-01 184362
      6 rrsfs-2018-02-14: 2015-02-01 184531
      8 rrsfs-2018-06-12: 2015-02-01 184592
      7 rrsfs-2019-02-14: 2015-02-01 184621
      8 rrsfs-2019-07-11: 2015-02-01 185061
      3 rrsfs-2020-02-14: 2015-02-01 184978
     12 rrsfs-2020-05-15: 2015-02-01 184982
     14 rrsfs-2021-04-26: 2015-02-01 185197
     14 rrsfs-2022-05-11: 2015-02-01 185154
      7 rrsfs-2023-04-24: 2015-02-01 185007

Ugh! Unless you have all the historical data this will play merry hell on backtests. I had no idea that FRED updated old values so much. They last updated the Feb 2015 value 7 (seven!) years after the initial report. Probably worse than that -- I only started collecting the FRED data in 2016.

Ray

Thanks for posting all the data. I am wondering if the adjustments are being made because there is some adjustment to current data which affects all previous data. Similar to a dividend being issued makes an adjustment to all previous data. If this is the case, then year over year comparisons should not change significantly. Is it much trouble to ask to see the values change for 2016-02-01?

Thanks

Craig

Is it much trouble to ask to see the values change for 2016-02-01?

Not a problem.

      1 rrsfs-2016-03-16: 2016-02-01 188176
      2 rrsfs-2016-04-13: 2016-02-01 188570
      2 rrsfs-2016-05-13: 2016-02-01 189012
      1 rrsfs-2016-06-14: 2016-02-01 188983
     14 rrsfs-2016-06-16: 2016-02-01 189012
      3 rrsfs-2017-02-15: 2016-02-01 188931
      1 rrsfs-2017-03-15: 2016-02-01 188636
      1 rrsfs-2017-04-14: 2016-02-01 188764
      9 rrsfs-2017-05-12: 2016-02-01 189737
      6 rrsfs-2018-02-14: 2016-02-01 189957
      8 rrsfs-2018-06-12: 2016-02-01 189912
      7 rrsfs-2019-02-14: 2016-02-01 189962
      8 rrsfs-2019-07-11: 2016-02-01 190515
      3 rrsfs-2020-02-14: 2016-02-01 190478
      9 rrsfs-2020-05-15: 2016-02-01 190538
      3 rrsfs-2021-02-17: 2016-02-01 190681
     14 rrsfs-2021-04-26: 2016-02-01 190591
     14 rrsfs-2022-05-11: 2016-02-01 190613
      7 rrsfs-2023-04-24: 2016-02-01 190372

      1 indfs-2016-03-16: 2016-02-01 106.3275
      1 indfs-2016-04-01: 2016-02-01 104.1893
      1 indfs-2016-04-15: 2016-02-01 103.9943
      1 indfs-2016-05-17: 2016-02-01 104.3721
      1 indfs-2016-06-15: 2016-02-01 104.4165
      1 indfs-2016-07-15: 2016-02-01 104.3717
      8 indfs-2016-08-16: 2016-02-01 104.4140
     13 indfs-2017-03-31: 2016-02-01 103.2685
     13 indfs-2018-03-28: 2016-02-01 102.3429
     27 indfs-2019-03-27: 2016-02-01 102.2225
     14 indfs-2021-05-28: 2016-02-01 98.9583
     10 indfs-2022-06-28: 2016-02-01 98.8946
      8 indfs-2023-03-28: 2016-02-01 98.9136

It struck me that, as far as backtesting goes, the changes might not matter much. GTT looks at the YOY value. So even though the values may change, what matters is the ratio of this year to previous year. So let's look at the ratios of 2016-02-01 vs 2017-02-01.

Aaaand ... for RRS (retail sales) the ratio varied between +1.64% and +2.25%. Always up YOY in the database reports 2016 to 2023, for February.


But that's not the case for Industrial Production: Total Index (INDPRO)
The ratios for 2016-02-01 vs 2017-02-01 varied between +0.46% and -0.49%.

Ex: The report of 6/15/2017 gave:
 2/1/2016 103.2685      2/1/2017  103.7416    +0.00458 (up YOY)
 
The latest report (10/17/2023) shows:
  2/1/2016  98.9136     2/1/2017   98.4322    -0.00487 (down YOY)

So in mid-2017 the signal would have been INDPRO showing no recession.
But if you did a backtest today, you would think that the mid-2017 signal was possible recession.
Ugh.

Since the goal of GTT timing is to identify recessions, it's worth knowing about another indicator that works pretty well in real time (recessions are officially identifiable only retroactively). It is called the Sahm Rule, and it uses unemployment as its metric.

Here is a link for those who may be interested.

https://fred.stlouisfed.org/series/SAHMREALTIME

Here is a recent discussion by the inventor herself:

https://www.bloomberg.com/opinion/articles/2023-11...

Baltassar

Nice short perspective of the current environment 'Why My Recession Rule Could Go Wrong This Time' by Sahm. Love her warning not to make too much (or too little) of her indicator.

But again, a recession is not inevitable. Indicators of economic downturns like the Sahm rule are empirical regularities from the past, not laws of nature.

Allocate Smartly compared a backtest of GTT vs a modified GTT that used the unemployment rate compared to asset prices alone. The results were very similar agreeing on allocation 96% of the time.
Philosophical Economics' Growth-Trend Timing (Redux) - Allocate Smartly
https://allocatesmartly.com/philosophical-economic...

RAM

I recently re-read the Growth-Trend Timing(GTT) paper. Now that several years have passed, the details of GTT are now more clear to me.


"When the 10-month SMA strategy switches inside of recessions, it tends to be profitable. When it switches outside of recessions, it tends to be unprofitable."

"GTT directs the moving average strategy to turn itself off during periods when those signals are unanimously disconfirming recession, i.e., periods where they are all confirming a positive fundamental economic backdrop."

This post was prompted because the RRS signal just turned negative, so a SMA sell signal should be obeyed. The recent months have been interesting, as the SMA and Recession Disconfirming indicators have been seesawing with the net result so far being to stay in stocks. SMA was up when Growth Trend was down and now SMA is up while Growth Trend is down.

S&P500 is +6.0% above the SMA.

SMA was up when Growth Trend was down and now SMA is up while Growth Trend is down.

Please clarify.

Please clarify.

For daily, Oct 16 showed potential recession, but SMA was +1.9%, so no sell signal.

Oct 17 - Oct 24, SMA was +, so don't pay attention to recession indicators.

Oct 25 - Oct 31, SMA was negative (sell signal), but recession indicators were both up (Disconfirm recession), so ignore SMA sell signal and stay IN stocks.

Nov 1 - Nov 14 SMA was +, so ignore recession indicators. Which were both up anyway.

Nov 15, SMA was +, so ignore recession indicators. Even though it said potential recession.

The only recent signals to sell were 4/17/22 (for 3 weeks), 12/11/22 (for 5 weeks) and 3/12/23 (for 2 weeks). Previous was 3/1/2020 to 5/31/2020.

Note that I use 2% below the SMA, because that backtests better than just below the SMA. Going by just below SMA, you would have been out from 4/17/22 to 1/15/23.

Post New

Berkshire Hathaway FAQ

---

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