Be nice to people. This changes the whole environment.
- Manlobbi
Stocks A to Z / Stocks B / Berkshire Hathaway (BRK.A)
No. of Recommendations: 18
Chart: timing with Nas100 RS screen
Just looking at statistics for applying timing to a screen doesn't give you a feel for the effect. Nor does it help much to decide which of several timing schemes works best for you.
So I created a graph of the top 10 Nasdaq 100 stocks with the greatest momentum, monthly.
The chart shows:
Untimed.
timing with the 52 week SMA of NASDAQ Composite (^IXIC)
Timing with the 52 & 65 week (252 days and 325 days) SMA of the S&P 500.
The sell signal is 1% below the SMA.
Link to chart:
https://docs.google.com/spreadsheets/d/1GUanGFQB7w...
No. of Recommendations: 4
The sell signal is 1% below the SMA.
RayVt
Thanks for posting the graphs. Is the sell next day close or something else?
Aussi
No. of Recommendations: 2
So I created a graph of the top 10 Nasdaq 100 stocks with the greatest momentum, monthly.
The chart shows:
1-Untimed.
2-timing with the 52 week SMA of NASDAQ Composite (^IXIC)
3-Timing with the 52 & 65 week (252 days and 325 days) SMA of the S&P 500.
The sell signal is 1% below the SMA.
I like it.
If I had to make a choice from that chart 2-timing would win for the long-term.
Worth noting that the three major time outs were during the periods as expected but held us out for longer periods(years) than expected.
GD_
No. of Recommendations: 3
Is the sell next day close or something else?
Something else.
Using the daily closing values that come out of GTR1 ...
Then stripping that down to the value on the first day of the month ...
Compute the return for that month. ((next_month_value / this_month_value) -1)
(It is an interesting contortion in Excel to isolate the SOM values from the daily values and put them in a table with no holes.)
For each month, look down the table of signals and find the signal (IN/OUT) that was computed just before this month.
If that signal was OUT, then the return for that month is forced to 1.0000.
The IN/OUT signal at the end of month N determines the return to be used for month N+1, actual or 1.0000.
The signals are computed in a separate operation.
There is some granularity involved, because the signals are computed on a daily basis but reported on a weekly basis. The signal computed for a week is to be applied to the next week. In this case, the last week reported in a month is applied to the next month.
(It is a PITA to ensure that you are applying the signals to the correct month instead of effectively applying the NEXT month's signal, which would be lookahead error.)
No. of Recommendations: 4
Worth noting that the three major time outs were during the periods as expected but held us out for longer periods(years) than expected.
Yes.
What you can't see in the chart is the many periods that you were OUT for only 1 or 2 months. Also the periods where you were IN for one month in the middle of a long period of being OUT.
Those whipsaws could be exhausting and get you to throw in the towel.
For this screen you need both nerves of steel and the patience of a saint. Tough combination.
In actuality you would not sit in cash during the out periods, but switch to a short-term income thing like SGOV or BIL or NEAR or ICSH.
Hard to explain to the spousal unit that you are investing in a great thing that doubles your money every 3-4 years but for the last 2 years you're sitting in a MMF earning 2%.
No. of Recommendations: 4
Even harder to explain that you have a 40% drawdown after having said we will take out some money later as we are really doing well.
Aussi
No. of Recommendations: 5
I have said this before.
I got started in serious investing in 1997 when I discovered the Motley Fool.
A year and a half later mortgage rates has dropped and we went to refinance our mortgage. From 14% (Thanks, Jimmy Carter) to 8%.
I ran the numbers and told my wife, "We can either do a straight refi and reduce the monthly payment, or we can take $20,000 cash back and the payment will be the same as we pay now. It looks like if I had invested serious money in Motley Fool Foolish Four method we would have made a large gain and I think that this will continue."
She said "We are not having any trouble making the current payment, so even if you lose the entire $20,000 we would still be okay. Anyway, we were flat broke when we got married so we would not be any worse off than when we started."
I said, "Okay, we'll take the $20,000 and I'll invest it. I will keep track of what the mortgage balance would be if we didn't take the money, and compare that to the investment value. In a few years we will know if we made a mistake or not.
And whenever the investment account gets as big as the remaining mortgage balance, I will tell you and we can decide if we want to sell the stocks and pay off the house."
A few years later the stocks _had_ grown to be more than what we owed on the house, so I presented it to her and said, "Do we want to pay off the house now?" Fully expecting as a typical woman she would want to pay off the house.
She said, "Are you nuts? Keep investing. Just tell me when we have enough that we can retire."
When a bear market hit, and when I got a call from the broker demanding me to wire money and meet the margin call, and the times we were down 50%, she just said "Just tell me when I should stop spending money."
Unusual woman.
My friend at work, also doing Motley Fool, his wife insisted on paying off their house ASAP. After I retired (early) he was still prioritizing the mortgage, still working on getting it paid off when he got caught in a layoff a few years later.
No. of Recommendations: 1
Am I reading the graph correctly? If your start was anytime other that ~2H00 to ~2H08, Untimed gives the better return?
So, basically, this would have helped one miss the dot com and financial crisis busts (which ain't nothing), but other than that, worse?
No. of Recommendations: 6
Am I reading the graph correctly? If your start was anytime other that ~2H00 to ~2H08, Untimed gives the better return?
So, basically, this would have helped one miss the dot com and financial crisis busts (which ain't nothing), but other than that, worse?Yes.
People think the purpose of timing is to improve the return. When they find it does not, they say "timing the market does not work".
But that is NOT the purpose of timing.
Going back at least to the 2013 Mel Faber paper, the purpose of timing is to reduce the standard deviation (volatility) at the cost of reducing the overall return. The win is if timing reduces the stdev percentage-wise more than it reduces the return percentage-wise.
The other win is "avoiding the full brunt of a large decline."
2/1985-12/2025
Untimed:
CAGR 22.9%
Stdev 30.2%
MaxDD(12) -66%
Sortino 1.35
Timed (65 week SMA of SPX):
CAGR 22.6%
Stdev 26.1%
MaxDD(12) -33%
Sortino 1.63
No. of Recommendations: 4
Timing, based on backtests, reduces the sequence of returns risk, so the safe withdrawal can be increased even though overall return may be smaller.
Aussi
No. of Recommendations: 0
RayVt
How do you post your tables so they stay as tables?
Thanks
Aussi
No. of Recommendations: 5
My experience is as rayvt has said: it is dodging the big bears that makes all the difference over an investing lifetime. It's partly the math (big losses take forever to recover), but mainly the psychology: it's a lot easier to be aggressive if you're reasonably confident you won't fall off a cliff. You will fall into a few ditches, of course ...
Baltassar
No. of Recommendations: 2
How do you post your tables so they stay as tables?
1. Detabify
2. Enclosed with < pre > tag. (Without the spaces of course.)
No. of Recommendations: 1
Aussi,
I don't know what ratvt means by "detabify". When I post the screen picks, the table has tabs between all the screen names and picks while each line is followed my a return. At the very beginning of the table I use < pre > (no space) and at the end of the table I use < /pre > ( again no space).
Larry
No. of Recommendations: 3
If the field widths are regular and small, then tabs are probably ok after being passed through shrewdm processing.
But that's not the case when the fields are varying widths. Or when there are blank columns (fields).
Copy/pasted from Excel:
Not untabified:
Rows 2/1/1985 12/1/2025 Rows 9/1/2006 12/1/2025
2 CAGR 23.5% 261 CAGR 20.9%
492 Stdev 25.4% 492 Stdev 21.1%
MaxDD(12) -36% MaxDD(12) -24%
Sortino 1.78 Sortino 1.77
Untabified:
Rows 2/1/1985 12/1/2025 Rows 9/1/2006 12/1/2025
2 CAGR 23.5% 261 CAGR 20.9%
492 Stdev 25.4% 492 Stdev 21.1%
MaxDD(12) -36% MaxDD(12) -24%
Sortino 1.78 Sortino 1.77
When pasted into notepad, these look identical. And they look like the untabified version.
No. of Recommendations: 2
I still use the Export Text Macro that was part of the Radiscreen set of spreadsheets.
No. of Recommendations: 4
Another important caveat: at the top of a market (you tell me ... ) timing **always** looks its worst relative to simple b&h.
No. of Recommendations: 8
I tested on ChatGPT today to see if getting out of a stock when its voume falls compared to its norm, and you can add 1.4% CAGR by doing so when
ADV10 / ADV90 < 0.50
Effects versus baseline:
CAGR: 19.06 → 20.48% (+1.42%)
Sharpe: 0.702 → 0.734
MDD: −54.7 → −54.0%
Beta: 0.700 → 0.688
Recovery: 1892d → 1716d
The interesting thing is that essentially everything improved simultaneously:
CAGR ↑
Sharpe ↑
beta ↓
recovery time ↓
drawdown slightly ↓
That is unusual. But it only improved results if you checked daily and got out when daily dollar volume showed the drop-off, not if you waited to check on your normal new trading date.
By your normal next trade date: 82% were no longer top 5
65% were no longer even top 10.
So the dollar volume deterioration predicted you were going to be getting out of those stocks anyway, and would get you out early if you check daily.
For stocks triggering the volume rule:
Return period Avg return if you ignored signal and held
Next month -1.8%
Next 3 months -3.6%
Next 6 months -2.9%
For normal held stocks:
Return period Avg return
Next month +2.1%
Next 3 months +5.2%
Next 6 months +8.4%
The interpretation matters.
This looks much more like:
early warning of rank deterioration
than:
a separate volume factor with independent predictive power
The sequence appears to be:
stock becomes a momentum winner
investor attention cools (dollar volume fades)
returns flatten
momentum rank falls
monthly system eventually removes stock
The daily rule mainly gained by doing:
sell several weeks earlier.
No. of Recommendations: 10
I tested on ChatGPT today to see if getting out of a stock when its [dollar] volume falls compared to its norm, and you can add 1.4% CAGR by doing so when
ADV10 / ADV90 < 0.50
A few questions & comments:
1) I would not take ChatGPT as an authoritative source. Just maybe a pointer to something to check.
2) Have you backtested this? How?
3) Does ADV# mean "average dollar volume over the last # trading days"?
4) Where can you get ADV10 & ADV90? Daily.
5) I guess that you could do it yourself by keeping the daily data on every potential stock, so that you'd have this information by the time it (might) become one of the topN picks. Doable with Nas100 stocks, difficult on S&P500 stocks.
6) What do you do when you have sold a stock early but it is still in the topN picks on the next portfolio trading date? You cannot get in and get out simultaneously, so what do you do?
7) Hmmmm. Do you or should you bypass buying a new topN stock that has this declining ADV10/ADV90?
8) At least for this monthly screen, the nominal holding period is one month, so there is no reason to look at the 3 or 6 month difference in returns. In my backtests on GTR1, 2 month period is 10% less than 1 month. 3 month is quite a bit less.
9) "By your normal next trade date: 65% were no longer even top 10." So is this even a problem that we need to solve? On average we'd only be selling a stock 2 weeks early.
10) How often does it happen, that a top 10 stock that you hold has a sharp dropoff in daily dollar volume before the next normal trading date?
11) My big question is "Is this actionable?"
No. of Recommendations: 7
I tested on ChatGPT today to see if getting out of a stock when its [dollar] volume falls compared to its norm, and you can add 1.4% CAGR by doing so when
ADV10 / ADV90 < 0.50
A few questions & comments":
I fed Chatgpt the price and volume history of all Nasdaq100 members that I already gave the group, so that was in fact a backtest, better than gtr since we can actually look at each stock's identity which gtr did not allow.
Yes ADV# mean "average dollar volume over the last # trading days"
Where can you get ADV10 & ADV90? Daily. I paid Norgate for the data.
Yahoo has average trading volumes. If you only owned 5 stocks it would not be that burdensome to keep track.
What do you do when you have sold a stock early but it is still in the topN picks on the next portfolio trading date? You cannot get in and get out simultaneously, so what do you do?
I pretended you re-purchased it the next month if it were still in your top 5 list then (82% of the time it isn't going to be).
Do you or should you bypass buying a new topN stock that has this declining ADV10/ADV90?
o for example:
Rank Nasdaq-100 stocks by 9-month momentum.
Start from rank #1.
If rank #1 has bad dollar volume, skip it.
Check rank #2, #3, etc.
Continue until five acceptable stocks are selected.
The key result was:
ADV10 / ADV90 < 0.50, which helped in the daily mid-month sell test, did not help in monthly-skip form.
It reduced CAGR versus baseline.
The only monthly-skip variant that improved was:
skip if ADV10 / ADV30 < 0.60
But you could use that .6 standarad.
"By your normal next trade date: 65% were no longer even top 10." So is this even a problem that we need to solve? On average we'd only be selling a stock 2 weeks early."
It would take extra work for 1.4% added CAGR. Probably not worth the effort for me but for people doing automated trading it would be.
10) How often does it happen, that a top 10 stock that you hold has a sharp dropoff in daily dollar volume before the next normal trading date?
Results:
Starting rank when held Frequency of early-sale trigger before next rebalance
#1 stock 6.8%
Top 5 holdings overall 11.9%
Top 10 holdings overall 14.8%
So best ranked stocks it is not very common.
My big question is "Is this actionable?"
Depends on how greedy you are and if you can automate or want to check daily. Over time an added 1.4% compounds to a big number; Wall Street heads have rolled for less than 1.4%
No. of Recommendations: 6
It would take extra work for 1.4% added CAGR. Probably not worth the effort for me but for people doing automated trading it would be.
Over time an added 1.4% compounds to a big number; Wall Street heads have rolled for less than 1.4%If the GTR1 backtests are right, we are talking about a screen that has long-term CAGR ranging from 22.9% to 26.8%.
Right there, that's a lot more than the S&P500's ~11.5%. So bumping that by 1.4% doesn't seem that big of a deal.
GTR1 for equal-weight SP500, 252 day hold is average 12.3% CAGR.
I am
highly doubtful that this 22%+ CAGR is realistic going forward, because if true why hasn't it been arbitraged away? If Wall Street would kill for an extra 1.4% what would they do for an extra 10%?
"How often does this happen?"
Starting rank when held Frequency of early-sale trigger before next rebalance
#1 stock 6.8%
Top 5 holdings overall 11.9%
Top 10 holdings overall 14.8%
So best ranked stocks it is not very common.Probably not worth it if it takes a lot of daily effort. It's just gilding the lily.
Top 10HTD12 shows avg CAGR 24.5% with AT 2.5
Top 5HTD12 shows 27.4%, with AT 1.8
Top 5HTD10 shows 27.1%, with AT 2.0
Top 5HTD15 shows 26.2%, with AT 1.6
Top 5HTD7 shows 25.8%, with AT 2.6
(I'm gonna have to look closely at that 5HTD12.)
==========================================
Here are a couple of top 10 lists, about a year apart.
Two things stand out.
There is a huge turnover in the top 10. There is a quick dropoff in the top ~5 and the dropoff flattens out after ~#6.
The first shows why this needs to be a monthly or bi-monthly screen.
The seconds shows why a large(ish) HTD works okay.
5/11/26 Ticker Perf Year
1 SNDK 4048.26%
2 WDC 1073.30%
3 MU 811.36%
4 STX 768.04%
5 AMD 348.16%
6 LRCX 295.24%
7 AMAT 185.64%
8 MRVL 181.05%
9 KLAC 168.82%
10 GOOGL 158.31%
6/19/25 Ticker Perf Year
1 APP 344.23%
2 AXON 163.67%
3 MSTR 146.75%
4 DASH 96.86%
5 NFLX 82.60%
6 GILD 66.00%
7 FTNT 65.88%
8 ZS 65.79%
9 SHOP 56.62%
10 MELI 51.87%
No. of Recommendations: 2
I don't recall when but earlier I hunted for funds that publicly claimed to be using momentum on Nasdaq100 stocks and none were doing this simple version (and the ones with more complication were under-performing what we are doing).
No. of Recommendations: 5
I tested trailing stop losses today on the startegy; only at the 35% loss level was it an improvement.
The winner is:
35% trailing stop
Compared with baseline:
CAGR: 19.06 → 19.18
Sharpe: 0.702 → 0.714
MDD: −54.7 → −46.1
Beta: 0.700 → 0.669
Recovery: 1892 → 1629 days
And the trigger frequency is surprisingly low:
about 0.27 positions/month
No. of Recommendations: 3
I checked the "Turn-of-the-month" effect on the screen I announced with the timing rule (sell and stay out of the market if SPY<325 day SMA), buy 5 or 10 Nasdaq100 stocks by the best 9 month price momentum.
Using "buy on the first day of a month at the close using the previous day's closing price to calculate which stocks to buy, and if the timing rule was triggered"= trade day -1 using day -2's data.
I looked at buying later or earlier.
Buying earlier, on the last day of the month based on the 2nd-to-last day's closing prices,
to the 5th-to-last day using the 6th to last day's closing prices,
was better for the collective period of the backtest 10/93 forward.
But it has weakened since 2000.
I ran the breakdown on the Top-5 strategy, comparing:
Current workflow: signal on −1 (data from the last trading day of month; trade next day i.e. trade the first day of the next month).
Late-month workflow: signal on −6 (6th trading day before month-end; trade next day i.e. the 5th-to-last day of the month instead)
Decade robustness
Period −1 current −6 late-month Difference
1994–1999 73.9% 84.7% +10.8%
2000–2009 10.6% 14.5% +3.9%
2010–2019 16.7% 21.2% +4.5%
2020–2026 28.7% 28.5% −0.2%
Interesting point:
It did not reverse recently.
It mostly just disappeared recently.
So:
pre-2020: persistent advantage in trading earlier, i.e. in the final week of a month rather than the 1st of the next month.
post-2020: essentially flat difference
during major stress periods
Dot-com collapse (2000–2002)
Workflow CAGR MDD
−1 4.4% -63%
−6 8.7% -60%
Financial crisis (2007–2009)
Workflow CAGR MDD
−1 7.9% -58%
−6 10.8% -54%
COVID crash (2020)
Workflow CAGR MDD
−1 31.1% -35%
−6 30.4% -36%
Inflation / tech bear (2022)
Workflow CAGR MDD
−1 17.3% -40%
−6 16.9% -42%
SPY regime split
When SPY >325d SMA:
Workflow CAGR
−1 28.2%
−6 31.6%
When SPY <325d SMA:
Workflow CAGR
−1 0.4%
−6 0.6%
So almost all of the difference occurs during favorable market environments.
My conclusion after the robustness check:
Originally I thought:
"switch everyone to late-month"
Now I would soften that.
The result looks more like:
a historically useful enhancement whose advantage has weakened after 2020.
So use as you see fit. In a way it makes it easy: during the last week of a month do your trading whenever you have time but get it done on or before the 1st of the next month.
No. of Recommendations: 3
And looking at the data, it really was only trading 5 days before the end of the month (using the 6th-to-last day close to calculate momentum and timing trigger) was why it looked like the new possibility of trading earlier had "weakened" since 2000.
Trading on the last day of the month, or the 2nd to last day, still were excellent and were improvements to the 1st of the month.
So my conclusion now is: trade on either of the last 2 days of the month using 1 day earlier from the day you trade to find closing prices to do calculations of 9 month momentum and (if you use my timing rule) the timing rule.
No. of Recommendations: 3
I checked the "Turn-of-the-month" effect ... a historically useful enhancement whose advantage has weakened after 2020.
I have read studies that said the best day to trade is at the open on the last business day of the month.
Now I have come to think that any such findings are just statistical noise.
When looking at the trading-day-of-month for any historical period, there will be ONE that's best. That's just how randomness works. One will be best and one will be worst.
Also, there is a lot of randomness just in the trading day. Our backtests don't capture that.
Case in point: The other day I was looking to buy LITE. It opened at 870 with a pretty wide bid-ask spread. I put in a buy limit order at 880. Which did not fill, so I bumped my limit up a few times to try to get a fill inside the spread. My last bid was 903.
LITE continued going up to the day's high of 977!
Given all those observation, I now think that that we are just seeing a pattern in randomness.
Just pick a trading day that is easy.
For most people the easiest is probably:
* Run your screen picks during the weekend after the last Friday of the month,
* Do your trades at or near the open on the upcoming Monday.
* Of note, market (not limit) trade orders placed during market close will execute at the open. Similar to market-at-close, all buy & sell orders fill at the same price. You can place your sell orders during the weekend and then do your buys after the market opens.
No. of Recommendations: 5
So my conclusion now is: trade on either of the last 2 days of the month using 1 day earlier from the day you trade to find closing prices to do calculations of 9 month momentum and (if you use my timing rule) the timing rule.
I should add this:
For a couple of years I was doing this -- run the screen(s) in the evening of the next-to-last trading day of the month and do the trades in the morning of the last trading day of the month.
It was a bit hectic; a lot of work had to be done that evening, especially if you are running more than one screen. Given the extreme volatility of this type of screen I think any day-of-month effect is tiny compared to the daily volatility of the stocks.
LITE being a case in point, 10.9% gain in that one day, 13.7% swing between high and low.
AMD had a similar story that same day, just not as big.
No. of Recommendations: 6
It would seem that with a high volatility screen like this that a shorter moving average timing would be better than a longer SMA period.
It's hard to compare the results of different timing periods by just looking at the overall long-term returns or even by looking at a few chosen time periods. A great overall return might be hiding a 5 or 10 year period with absolutely TERRIBLE losses. Losses that would cause anyone to throw in the towel.
A good bit of information would be to look at all the rolling N-year periods in the long-term run. This backtest, 1985 to 2025, is 40 years. The rolling periods start every month. I examined several rolling periods, so we can get an idea of what we might encounter.
There are 480 rolling 1-year periods, 432 rolling 5-year periods, 372 10-year periods, and 312 15-year periods.
Each month buy/hold the top 10 Nasdaq-100 stocks with the highest 52 week price momentum. Keep any existing holding that is still in the top 12.
First, CAGRs without timing.
# Yrs --> 1 2 3 5 10 15
# >0 387 406 406 416 361 312
# <0 93 62 50 16 11 0
Win % 81% 87% 89% 96% 97% 100%
Yrs --> 1 2 3 5 10 15
Avg 28.9% 25.2% 23.4% 22.9% 21.8% 20.2%
Med 25.7% 23.5% 22.6% 23.6% 20.8% 18.8%
Min -63.1% -48.5% -32.5% -15.8% -4.6% 4.2%
Max 230.9% 142.9% 114.5% 78.1% 53.2% 46.0%
2/3 range 10.6% 15.5% 16.5% 15.5% 18.1% 16.8% <<--- 1/3rd are worse & 2/3 are better.
95% range -30.9% -16.5% -5.1% 0.8% 3.7% 10.4% <<--- 5% are worse & 95% are better.
The thing that is quite important is the worst return (read: loss) of a multi-year period. A large loss over 2 or 3 years will cause people to abandon a winning screen.
Timing is applied at the trade day of each month (month = 21 trading days), using the most recent N week SMA of the S&P500 index. Sell when the current price is 1% below the SMA, buy when at or above the SMA.
CAGR of 43 week (215 day) SMA
# Yrs --> 1 2 3 5 10 15
# >0 379 415 433 424 372 312
# <0 86 53 23 8 0 0
Win % 82% 89% 95% 98% 100% 100%
Yrs --> 1 2 3 5 10 15
Avg 26.1% 23.7% 22.4% 22.2% 22.4% 21.8%
Med 18.8% 20.0% 20.8% 21.2% 21.2% 19.8%
Min -32.6% -16.8% -8.2% -4.6% 3.7% 9.7%
Max 230.9% 134.2% 109.4% 75.5% 46.8% 41.2%
2/3 range 9.4% 13.3% 14.1% 15.6% 16.7% 17.7% <<--- 1/3rd are worse & 2/3 are better.
95% range -14.4% -5.8% 0.1% 2.2% 8.2% 12.1% <<--- 5% are worse & 95% are better.
CAGR 65 week (215 day) SMA
# Yrs --> 1 2 3 5 10 15
# >0 389 431 447 430 372 312
# <0 60 28 9 2 0 0
Win % 87% 94% 98% 100% 100% 100%
Yrs --> 1 2 3 5 10 15
Avg 27.8% 25.6% 24.3% 24.3% 24.7% 24.6%
Med 20.7% 21.8% 23.2% 22.6% 23.2% 24.3%
Min -32.6% -17.4% -3.5% -1.1% 7.2% 13.6%
Max 230.9% 142.9% 114.5% 78.1% 47.9% 41.9%
2/3 range 10.0% 14.0% 15.0% 17.6% 20.2% 20.4% <<--- 1/3rd are worse & 2/3 are better.
95% range -14.3% -3.1% 2.7% 4.9% 12.8% 15.9% <<--- 5% are worse & 95% are better.
Compare the number of multi-year periods with losses for the different timing schemes.
For 3-year periods 50 had a loss without timing, 23 had a loss with 43 week timing, and 9 had a loss with 65 week timing.
For 5-year periods the respective numbers with a loss were 16, 8, and 2.
No. of Recommendations: 6
NAS100 TOP 12
52W Performance
2026-05-26
1 SNDK 3866%
2 LITE 1150%
3 WDC 865%
4 MU 704%
5 STX 621%
6 INTC 498%
7 AMC 326%
8 LRCX 277%
9 MRVL 224%
10 GOOGL 127%
11 WBD 200%
12 BKR 80%
What a ride.
GD_
No. of Recommendations: 4
NAS100 TOP 12
52W Performance
2026-05-26
Some more of nas100 top 12 stock data info and replace AMC with AMD
Stock Data Ticker symbol Name Industry MarketCap VolumeAve P/E 52WK high 52WK low LastPrice LastTradeTime
1 SNDK SANDISK CORPORATION Computers, Phones & Household Electronics 2.35E+11 1.46E+07 56.20 $ 1,641.74 $ 36.21 $ 1,589.55 5/26/2026 21:48
2 LITE LUMENTUM HOLDINGS INC. Communications & Networking 7.09E+10 5.88E+06 169.04 $ 1,085.68 $ 71.04 $ 910.81 5/26/2026 21:48
3 WDC WESTERN DIGITAL CORPORATION Computers, Phones & Household Electronics 1.81E+11 7.64E+06 34.19 $ 536.12 $ 50.62 $ 524.65 5/26/2026 21:48
4 MU MICRON TECHNOLOGY, INC. Semiconductors & Semiconductor Equipment 1.01E+12 4.79E+07 42.30 $ 916.80 $ 92.22 $ 895.88 5/26/2026 21:48
5 STX SEAGATE TECHNOLOGY HOLDINGS PUBLIC LIMITED COMPANY Computers, Phones & Household Electronics 1.90E+11 4.46E+06 80.31 $ 862.65 $ 113.20 $ 845.76 5/26/2026 21:48
6 INTC INTEL CORPORATION Semiconductors & Semiconductor Equipment 6.21E+11 1.52E+08 0.00 $ 132.75 $ 18.97 $ 123.52 5/26/2026 21:48
7 AMD ADVANCED MICRO DEVICES, INC. Semiconductors & Semiconductor Equipment 8.22E+11 4.39E+07 167.97 $ 506.96 $ 108.62 $ 503.89 5/26/2026 21:48
8 LRCX LAM RESEARCH CORPORATION Semiconductors & Semiconductor Equipment 4.04E+11 9.33E+06 60.90 $ 323.98 $ 79.49 $ 322.68 5/26/2026 21:47
9 MRVL MARVELL TECHNOLOGY, INC. Semiconductors & Semiconductor Equipment 1.82E+11 2.59E+07 67.83 $ 217.45 $ 58.61 $ 208.26 5/26/2026 21:48
10 GOOGL ALPHABET INC. Software & IT Services 4.71E+12 2.79E+07 29.64 $ 408.61 $ 162.00 $ 388.88 5/26/2026 21:48
11 WBD WARNER BROS. DISCOVERY, INC. Media & Publishing 6.77E+10 1.64E+07 0.00 $ 30.00 $ 9.11 $ 27.00 5/26/2026 21:41
12 BKR BAKER HUGHES COMPANY Oil & Gas Related Equipment and Services 6.62E+10 8.26E+06 21.29 $ 70.41 $ 36.36 $ 66.73 5/26/2026 21:29
GD/
No. of Recommendations: 6
NAS100 TOP 12
52W Performance
2026-05-26
What a ride.I recall Micron (MU) from way back in the early Motley Fool days. As I recall, it went in and out and in and out of some of those early MI screens. Guess we should have just held on to it.
Although ... it went from $83 in 2000 to $2 by 2008.
So I guess not. Buying the top N momentum stocks and selling them when they drop out of the top would appear to be a good way.
MU and INTC and GOOG are the only ones from that era that stick to my mind.
I checked finance.yahoo this morning to see what the market was doing -- which turned out to be "Meh, not much". Except for the news item "Micron up 20% today." Daaaang!
Those stocks you listed that have gained 200%+ (tripled or more) in 52 weeks will likely stay in the top 12 for a long time. Not that you can't lose big money on them, though. SNDK could lose 75% and it would *still* be in the top 12.
I have gotten a bit nervous about the preponderance of such similar tech stocks in the top N of the Nasdaq 100 so I have created a similar screen from a different universe of stocks.
Here is that list:
1 BE 1452.36%
2 MU 859.49%
3 VICR 699.02%
4 CIEN 650.92%
5 TTMI 588.88%
6 DOCN 474.21%
7 TER 405.51%
8 POWL 401.38%
9 COHR 386.04%
10 AMD 356.79%
11 FORM 349.39%
12 NESR 342.56%
Yeah, they are still mostly technology. But not in the Nasdaq 100.
No. of Recommendations: 5
today's ChatGPT study:
Your timed Top-5 25.6%
After tax (hypothetical model)
Strategy After-tax CAGR Tax drag
SPY 9.8% -0.8%
QQQ 12.8% -0.9%
Your timed
Top-5 20.8% -4.8%
So short-term cap gain taxes hurt your strategy much more compared to holding SPY and QQQ and only selling them at the very end of the test period —as expected—but they do not remotely destroy the edge in this approximation.
Compounding impact from $100k over 25 years:
Strategy Pretax ending value After-tax ending value
SPY ~$1.24M ~$1.03M
QQQ ~$2.49M ~$2.02M
Timed
Top-5 ~$29.9M ~$11.3M
The key observation:
Even though your strategy lost almost 5 percentage points of CAGR to annual taxation, the remaining after-tax return still substantially exceeded passive alternatives.
Also note something unintuitive:
Your strategy did not lose:
25.6%×.25
because:
loss years generated tax assets,
taxes were paid only on net annual gains,
compounding still continued on remaining capital.
So the tax drag was significant but not proportional.
And trading on either of the last 2 days of the month added 4% CAGR to the strategy beyond what I usually use in the report (which pretended trading the 1st day of the month). The last 5 days of the month were good most periods but the last 2 in all periods, i.e. their advantage did not degrade post-2000.
If you use 9 mo momentum on the market's top 100 stocks not in the Nasdaq100, 5 stocks monthly, the result is NOT GOOD.
https://gtr1.net/2013/?nas100.a:etnull:MktCap:tn10...Today I made what use to take me 6 months working; thank you momentum strategy!
No. of Recommendations: 4
If you use 9 mo momentum on the market's top 100 stocks not in the Nasdaq100, 5 stocks monthly, the result is NOT GOOD. https://gtr1.net/2013/?nas100.a:etnull:MktCap:tn10...
Why does it default to a start date of 1997/09/02 ? It must be something about the MktCap, because when you delete that it goes back to 1985/02/01 which is the start date it uses for "nas100.a". Weird.
When I look at these screens with a lot of short-term trades in a taxable account, I assume that the tax reduces the CAGR by 25%. That is, 25% of the gain gets whisked off by the IRS each year.
Of course, these screens are best done in a Roth IRA or even a regular IRA.
The 52 week momentum top 5 HTD 12 has a CAGR range, min & max, of 24.1% & 29.6%. Avg 27.4%.
When you pull "Portfolio Values: Download daily portfolio values", the initial value for all cycles is $5. The final values range from $33,780 to $196,136, with median $108,192.
(For comparison, VFINX (S&P500 index) final value is $382. BRK-A is $2,718.)
I load the *lowest* cycle into the spreadsheet which computes the statistics with & without timing. So presumably you can expect to get a better return in real life.
Today I made what use to take me 6 months working; thank you momentum strategy!Amen.
Not so fun on the days when you lose the same.
But remember the days just a few weeks ago when people were running around with their hair on fire proclaiming "the crash is here!!!!!!!" ? Good times, good times.
No. of Recommendations: 5
Strategy CAGR Beta Sharpe Sortino Ulcer Max DD MDD recovery Longest other recovery
Timed Top 1 16.5% 0.88 0.54 0.63 45.7 -73.6% 4,945 days 1,596 days
Timed Top 5 23.8% 0.84 0.78 0.92 20.5 -55.1% 176 days 1,782 days
Timed Top 10 19.2% 0.79 0.73 0.85 18.2 -50.9% 175 days 1,532 days
QQQ 10.2% 1.19 0.50 0.66 43.7 -83.0% 5,446 days 716 days
SPY 8.2% 1.00 0.51 0.65 16.0 -55.2% 1,773 days 2,407 days
So 5 stocks bought the last day of the month and held 1 month, paying .1% spreads, and on your next trade date exiting the market if SPY<325day SMA, is the best of the tested #1, or #1-5, or #1-10 9 month momentum stocks in the Nasdaq100 as it changed over time, within the database I have (gtr goes back further).
You would have doubled the indices with lower beta, higher Sharpe ratio, smaller max drawdown, and faster recovery from the max drawdown.
No. of Recommendations: 2
1,782 days Longest other recovery
That would be really tough to trade through.
No. of Recommendations: 4
1,782 days Longest other recovery
That would be really tough to trade through.
That's why most people would not be able to run this strategy.
Here's what I get for the longest bad periods.
10/2/2000 to 11/3/2003 THREE years
Values:
10/2/2000 - $1925 Out on 11/1/2000 at $1731
In on 7/1/2003 at $1731
Timing took you out for 31 months.
That's a long time sitting in a 3% MM fund when your backtest says you should be getting 20+%.
Untimed:
11/1/2000 - $2652
7/1/2003 - $1181
Next date above $2652 was 2/1/2011 at $2634.
An entire decade!
BTW, timed on 2/1/2011 - $5481
---
Next one , 2/1/2008
Timed, $3048 to get back in on 9/1/2009 at $3048.
Almost 1.5 years. Timing took you out for 19 months.
Untimed
2/1/2008 - $2080 The high was on 6/2/2008 at $2289
9/1/2009 - $1465
Next date above $2289 was 11/1/2010 at $2396
---
Next one, timing took you out 5/2/2022 to 4/3/2023
Untimed, the high was 11/1/2021 and the next date above that was 7/1/2024.
33 month recovery time.
---
Most people would not be able to run this strategy.
No. of Recommendations: 2
Just consider how long the S&P and Nasdaq100 took to make their money back from losses;
most humans would have given up rather than face that.
That is why we are using data not human feelings: I am greedy and want the most money. I long ago decided to tolerate losses and long recovery periods if it meant making more money. The evidence is what it is.
No. of Recommendations: 5
Just consider how long the S&P and Nasdaq100 took to make their money back from losses;
most humans would have given up rather than face that.
That is why we are using data not human feelings: I am greedy and want the most money. I long ago decided to tolerate losses and long recovery periods if it meant making more money. The evidence is what it is.
It occurred to me that in this type of screen the most important thing for timing is to reduce the drawdowns and have the discipline to stay the course during long periods where you are out of the stocks and sitting in an ultra-short bond ETF. It would be really nice if your out periods were shorter than the un-timed recovery of the bear markets.
"The moving average timing strategy makes the majority of its money by avoiding large, sustained market downturns. To be able to avoid those downturns, it has to accept a large number of small losses associated with switches that prove to be unnecessary."
"Avoid big losses. That’s the way to really make money over the years.
If you can be somewhat close on the way up and avoid the full brunt of large declines on the way down, you have a decent shot of outperforming over the long term and an even better chance of smoothing out your ride." -Julian Robertson
I have tried several timing schemes on this backtest. The best ones have MaxDD around -25%. Which is not actually too bad. The exact parameters of the SMA timing don't much matter. Every reasonable set of parameters worked pretty much the same; just minor differences in CAGR and MaxDD and number of out periods and total months out.
The problem is that they have you out and sitting in cash for 30 months and 18 months at a time. Pretty hard to not get itchy fingers somewhere around the 9'th month.
Of course, that's better than untimed with -66% MaxDD and 10+ years to recover.
I've experimented with a few simple ways of trying to recognize being past the bottom while the SMA is still saying to sell -- and everything I tried failed. I guess this screen is just too volatile.
No. of Recommendations: 5
The problem is that they have you out and sitting in cash for 30 months and 18 months at a time. Pretty hard to not get itchy fingers somewhere around the 9'th month.
Of course, that's better than untimed with -66% MaxDD and 10+ years to recover.
I've experimented with a few simple ways of trying to recognize being past the bottom while the SMA is still saying to sell -- and everything I tried failed. I guess this screen is just too volatile.
How about an arbitrary rule for getting back in. When you’re still out 12 months after your sell signal, you get back in no matter what.
No. of Recommendations: 4
How about 32% CAGR for 5 stocks monthly trading since the Nasdaq100 was invented. Possible per gtr (am testing in Chatgpt now and will report back)
https://gtr1.net/2013/?~Liquid_consolidators::nas1... nas100 stocks
1 year top 10
then
top 5 stocks by ave dollar volume in the last month / return in the last 700 market days
No. of Recommendations: 4
in oct 1993 forward Norgate data using chatgpt:
Untimed comparison
Metric Top5 12M Mome Top10→Top5 ADV20/ADV700 Difference
CAGR 23.09% 20.39% -2.70%
Max DD -67.86% -51.11% +16.75 percentage points better
Recovery 1190 days 1155 days 35 days better
Sharpe 0.774 0.871 Much better
Ulcer Index 25.39 17.30 Much better
Timed comparison (SPY > 325d SMA)
Metric Top5 12M Momentum Top10→Top5 ADV20/ADV700 Difference
CAGR 23.25% 18.23% -5.02%
Max DD -43.26% -33.06% +10.20 percentage points better
Recovery 1703 days 154 days 1549 days better!
Sharpe 0.819 0.921 Better
Ulcer Index 19.15 11.30 Much better
So if someone cared primarily about:
sleep-at-night factor
drawdown
time spent underwater
the ADV20/ADV700 version is actually impressive.
The most surprising number is not the drawdown; it's the recovery time:
Timed 12M momentum: 1703 days
Timed ADV20/ADV700: 154 days!
That's roughly 4.7 years faster recovery from the worst drawdown.
If I saw these numbers without knowing the strategy, I'd conclude:
ADV20/ADV700 is acting like a quality filter. It is sacrificing some of the explosive winners but avoiding a lot of the catastrophic collapses.
So far people who are willing to sacrifice 5% cagr, to get recovery in 6 months from their worst, drawdown, this is the way to go.
No. of Recommendations: 4
How about an arbitrary rule for getting back in. When you’re still out 12 months after your sell signal, you get back in no matter what.I don't think that would work.
Most of the OUT periods are only a few months long.
Here is the number of months OUT in a row and how many times it happened.
# months # of occurrances
1 11
2 6
3 2
4 4
7 1
8 1
14 1
17 1
31 1
Lots of whipsaws, get out just to get back in the very next month or the month after that.
Here are the longest OUT periods along with the untimed values.
Date Value signal # months out
10/2/2000 2949.850 out 1
11/1/2000 2651.888 out 2
12/1/2000 1836.939 out 3
1/2/2001 1598.507 out 4
2/1/2001 1704.723 out 5
3/1/2001 1213.951 out 6
4/2/2001 1115.060 out 7
5/1/2001 1312.679 out 8
6/1/2001 1389.376 out 9
7/2/2001 1425.945 out 10
8/1/2001 1402.360 out 11
9/4/2001 1289.362 out 12
10/1/2001 1087.102 out 13
11/1/2001 1221.470 out 14
12/3/2001 1288.974 out 15
1/2/2002 1332.662 out 16
2/1/2002 1279.423 out 17
3/1/2002 1288.866 out 18
4/1/2002 1258.799 out 19
5/1/2002 1077.335 out 20
6/3/2002 1014.651 out 21
7/1/2002 999.307 out 22
8/1/2002 913.733 out 23
9/3/2002 863.804 out 24
10/1/2002 898.274 out 25
11/1/2002 993.539 out 26
12/2/2002 998.539 out 27
1/2/2003 931.837 out 28
2/3/2003 926.671 out 29
3/3/2003 901.603 out 30
4/1/2003 971.167 out 31
5/1/2003 1015.460 IN
2/1/2008 2080.525 out 1
3/3/2008 1941.838 out 2
4/1/2008 2102.354 out 3
5/1/2008 2213.794 out 4
6/2/2008 2298.746 out 5
7/1/2008 2168.512 out 6
8/1/2008 2098.480 out 7
9/2/2008 1960.124 out 8
10/1/2008 1591.297 out 9
11/3/2008 1433.314 out 10
12/1/2008 1235.188 out 11
1/2/2009 1387.917 out 12
2/2/2009 1337.783 out 13
3/2/2009 1195.046 out 14
4/1/2009 1275.693 out 15
5/1/2009 1303.381 out 16
6/1/2009 1337.018 out 17
7/1/2009 1369.713 IN
2/1/2022 26543.570 out 1
3/1/2022 25322.720 out 2
4/1/2022 26922.390 out 3
5/2/2022 23670.210 out 4
6/1/2022 22343.580 out 5
7/1/2022 21212.650 out 6
8/1/2022 22622.680 out 7
9/1/2022 22138.430 out 8
10/3/2022 21009.090 out 9
11/1/2022 22894.530 out 10
12/1/2022 23328.740 out 11
1/3/2023 21491.030 out 12
2/1/2023 21881.610 out 13
3/1/2023 20617.680 out 14
4/3/2023 20922.880 IN
11/2/1987 24.792 out 1
12/1/1987 21.301 out 2
1/4/1988 26.893 out 3
2/1/1988 26.023 out 4
3/1/1988 27.271 out 5
4/4/1988 26.560 out 6
5/2/1988 27.511 out 7
6/1/1988 27.295 out 8
7/1/1988 28.879 IN
No. of Recommendations: 3
I've experimented with a few simple ways of trying to recognize being past the bottom while the SMA is still saying to sell....
How about an arbitrary rule for getting back in. When you’re still out 12 months after your sell signal, you get back in no matter what.
Alternatively, how about the SMA rule for getting out, one of Mungofitch's bottom catchers for getting back in, and then ignore the SMA rule until back above the SMA curve?
Eric Hines
No. of Recommendations: 4
Alternatively, how about the SMA rule for getting out, one of Mungofitch's bottom catchers for getting back in, and then ignore the SMA rule until back above the SMA curve?Tried that, too.
non funzionaToo bad we can't post charts here. But go to yahoo and look at the chart for "^IXIC" (NASDAQ Composite) for 10/2/2000 to 4/21/2003.
Those are the dates encompassing the "OUT" period. 4/21/2003 was the week where the SMA signal said to buy.
There were a few potential bottoms along the way, each one lower than the previous one. I would call the first three head-fakes. By the time you could call the bottom you would have gotten in right about the next peak.
The last bottom was 9/30/2002 at 1140. The "IN" signal was 11/21/2003 at 1434. If it took 1 month to decide that 9/30/02 was the bottom, you'd have gotten in at 1360. BFD.
The first major bottom was 4/2/21. Say you got back in a month later 5/7/01 at 2235 and stayed in ignoring everything until the SMA buy rule kicked in on 4/21/03 at 1134. Not good.
So, one right bottom call out of 4-5 candidates. The first ones were fails, the last (true) one got you in a year earlyfor little effect -- 1360 rather than 1434.
But here is the real problem:
# of months out Count
1 11
2 6
3 2
4 4
7 1
8 1
14 1
17 1
31 1
The longer the OUT period was, the fewer there were. The long periods -- 7, 8, 14. 17, 31 -- only occurred one time each. There is no way you can come up with a robust rule for getting back in early.
I think it's like musselmant said, in this screen you just have to take your lumps and soldier on.
"Successful investors need the emotional discipline to execute their planned strategy faithfully, come hell, high water, or the apparent end of capitalism as we know it." --William Bernstein
No. of Recommendations: 0
So is this using ADV20/ADV700<0.5 to get out of a stock mid-cycle, as well as a pre-selection check?
Or is ADV10/ADV90 better for the former and ADV20/700 for the latter?
Learnng lots lately; thanks for piqueing my curiosity!
----
Referring to:
I tested on ChatGPT today to see if getting out of a stock when its voume falls compared to its norm, and you can add 1.4% CAGR by doing so when
ADV10 / ADV90 < 0.50
--- and ---
ADV20/ADV700 is acting like a quality filter. It is sacrificing some of the explosive winners but avoiding a lot of the catastrophic collapses.
So far people who are willing to sacrifice 5% cagr, to get recovery in 6 months from their worst, drawdown, this is the way to go.
No. of Recommendations: 2
Also, as a reality check, when SPY falls below it's 325 day SMA, the idea is to sell all picks at the next monthly rebalance and not make any purchases, correct? Not to sell all immediately, right?
No. of Recommendations: 3
Absent testing the benefit of waiting or not can't be known but yes the assumption was you look only at your usual trading date. My theory was that monthly traders are not going to look daily in practice.
Just as an example I just tested a separate timing trigger and compared waiting to trade day or selling a stock on the trigger (share volume falling below 60 day average share volume) and there was not much difference in CAGR; better to act now but not a high price to pay to wait.
No. of Recommendations: 2
getting out mid-month gives you a small added CAGR boost but nothing much
No. of Recommendations: 3
getting out mid-month gives you a small added CAGR boost but nothing much
Then operationally it's probably best to do whichever is most convenient.
For someone who isn't deep into this stuff, just make the timing decision at the normal once-a-month recast day.
This would be people, like son/daughter/SIL etc. that you passed this simple-to-do strategy to.
Some people who _are_ deeply into this stuff, probably have it automated and can get the sell/buy signals as they occur, it is probably okay to sell when the sell is signaled and then wait for the next monthly recast date for a buy signal.
Personally, I have scripts that run automatically every day. The signals are then emailed to me every day or every weekend.
If you have a gap between the sell & buy thresholds, it is unlikely for a buy signal to come quickly after a sell signal.
FWIW, here is the most recent signal (on 6/5/2026):
"NAS100 timing: Timing: IN stocks (+12.9% vs. SMA) (# 56 consecutive weeks)"
We have a long way to fall before we get a sell signal.
No. of Recommendations: 1
getting out mid-month gives you a small added CAGR boost but nothing much
Then operationally it's probably best to do whichever is most convenient.
For someone who isn't deep into this stuff, just make the timing decision at the normal once-a-month recast day.
This would be people, like son/daughter/SIL etc. that you passed this simple-to-do strategy to.
Some people who _are_ deeply into this stuff, probably have it automated and can get the sell/buy signals as they occur, it is probably okay to sell when the sell is signaled and then wait for the next monthly recast date for a buy signal.
Personally, I have scripts that run automatically every day. The signals are then emailed to me every day or every weekend.
If you have a gap between the sell & buy thresholds, it is unlikely for a buy signal to come quickly after a sell signal.
FWIW, here is the most recent signal (on 6/5/2026):
"NAS100 timing: Timing: IN stocks (+12.9% vs. SMA) (# 56 consecutive weeks)"
We have a long way to fall before we get a sell signal.
No. of Recommendations: 3
if you can automate, apply the volume rule to the overlap strategy to determine whether to purchase, and then whether to sell mid-month, also.
The SPY>325day SMA timing signal for all purchases (or go to cash), then also the volume rule on buy date, generated 29.1% cagr post 1999.
If you also apply the volume rule mid-month to sell early (even if you have to get back in to the same stock the next month) it ups cagr to 31.8%
loss episodes longest days to recover
strategy CAGR Sharpe Beta UI worse than-9.99% recovery % Slower than SPY Slower than NDX
Overlap 28.4% 0.87 0.82 15.61 46 757days 67.4% 58.7%
Entry filter 29.3% 0.89 0.81 15.42 46 757d 63.0% 56.5%
Entry + exit 31.8% 0.93 0.81 15.04 47 800d 68.1% 59.6%
9m Momentum 5 18.3% 0.67 0.82 25.61 36 606d 75.0% 58.3%
PHL180day 5 23.4% 0.85 0.74 16.89 34 1131d 82.4% 70.6%
SPY 7.9% 0.49 1.00 16.22 8 1020d — 62.5%
NDX 7.5% 0.40 1.21 44.61 15 3292d 60.0% —
No. of Recommendations: 5
here is latest data from the data I have after follow-up with chatgpt on various errors:
Strategy Offset CAGR Shrpe Beta UI MDD times drawdowns
-10% or more
max days ever to
recover #days to recover from MDD only
Momentum5_-1
Momentum5 -1 19.8% 0.712 0.811 26.951 -58.9% 36 651 651
Momentum5_+1
Momentum5 1 20.5% 0.731 0.824 22.246 -57.2% 34 1373 1373
NDX 7.5% 0.403 1.206 44.611 -82.9% 15 3292 3292
Overlap_entry
_exit_-5 27.8% 0.882 0.706 17.400 -62.8% 38 491 67
Overlap_entry
_exit_-4 24.9% 0.807 0.724 18.873 -59.4% 35 484 43
Overlap_entry
_exit_-3 19.8% 0.707 0.684 19.069 -54.3% 38 483 58
Overlap_entry
_exit_-2 25.9% 0.863 0.683 18.857 -45.4% 34 465 123
Overlap_entry
_exit_-1 25.9% 0.830 0.734 16.185 -58.6% 42 489 44
Overlap_entry
_exit_+1 21.5% 0.727 0.757 17.608 -48.8% 48 529 43
PHL180_5_-1 22.0% 0.815 0.722 19.235 -54.7% 38 1214 1214
PHL180_5_+1 20.1% 0.758 0.744 20.829 -51.7% 33 1403 1403
SPY 7.9% 0.491 1.000 16.215 -55.2% 8 1020 869
highest CAGR was to a) stay ouf the market on trade day if SPY<325day SMA, b) buy the overlaps only of the two possible 5 stock strategies (1 is 9 month momentum winners in the Nasdaq100, the other the highest 5 of the ratio price/(sum of 180 day high price and 180 day low price), c) trade on the last day of each month, d) don't enter any position on trade day if that particular stock's 5 day share volume (not dollar volume) is less than its 60 day share volume; e) during the month after purchase, exist any stock no longer meeting "d" criteria i.e. make sure daily, f) despite getting out of a stock, if on the list and meets buy criteria on the last day of the month for holding the next month, you re-buy even if you got out early.
This strategy recovered from its worst drawdown in only 44 trading days, and its worst-ever recovery period for any -10% or worse drawdown was much less than SPY or NDQ's.
However, I’ll quantify each volume-triggered early sale: what happened if sold at the trigger versus held to the normal next rebalance, then summarize win rate and whether a few names explain most of the benefit.
Thought for 2m 29s
Computed.
For Overlap + volume entry filter + mid-month exit, trade day -1:
Item Result
early-sale events 68
Measurable events 67
Unique stocks sold early 45
Beneficial exits 34
Harmful exits 33
Beneficial rate 50.7%
Median stock return after exit to next rebalance -0.03%
Average stock return after exit to next rebalance +0.99%
So the rule is not consistently predictive. It is basically a coin flip by count.
The portfolio improvement came from a few large helpful exits offsetting many small or harmful exits. Top positive contributors:
Stock Approx portfolio benefit
TSLA +10.8%
VRTX +8.3%
SHLDQ +6.5%
PTON +5.7%
ISRG +5.4%
Top 5 helpful stocks explained about 62.6% of all positive exit benefit.
Marginal effect of adding the mid-month exit on top of entry filter:
Metric Change
CAGR +0.26 pts
Sharpe +0.015
Beta -0.022
UI -0.87
MDD +1.26 pts better
MDD recovery 18 days faster
Longest ≥10% recovery 140 days worse
My read: the entry filter is cleaner than the mid-month exit. The mid-month exit helps some headline metrics slightly, but its edge is not broad; it depends on a small number of exits.
No. of Recommendations: 2
Can someone post the yearly returns for this strategy or a version like it (by year)?
I'm curious if astronomical returns in this wave and prior waves drove all or most of it ...
In other words, how consistently does it produce positive returns by year.
Thanks if able!
No. of Recommendations: 3
d) don't enter any position on trade day if that particular stock's 5 day share volume (not dollar volume) is less than its 60 day share volume;
Market days, right? Not calendar days? And average N-day volume, right?
Ha! Turns out you don't need to pay for this, or to try to scrape it from Yahoo or other web sites.
GoogleFinance spreadsheet will grab daily volumes, all you need to do is calculate the averages.
Finviz shows "RelVolume" which is the ratio of a stock's current DAY trading volume to its 3-month average volume. That might be good enough. At this moment (market is open) the only fails are #17 & 18, CSCO & ADI.
Not sure how they handle "day volume", as RelVolume is live when the market is open. Obviously the volume 2 hours after market open is not comparable to an entire day's volume.
Barchart has all the volume averages you would ever need.
e) during the month after purchase, exist any stock no longer meeting "d" criteria i.e. make sure daily,
Ah well, can't automate checking the google spreadsheet (I think), have to check manually.
How often does this happen? Often enough so it isn't just a few rare but big events?
"Unique stocks sold early 45" Out of how many stock-months? I think you said from 1999, so 27 years * 12 mo/yr * 5 stocks/mo = 1620.
45 occurrences out of 1620 candidates is just noise.
Oh: "So the rule is not consistently predictive. It is basically a coin flip by count."
So no need to bother. Simpler is better, so just trade at end of month and go to sleep until the next month's trade date.
The more rules there are, the more likely that the rules are (over)tuned for that one specific history.
My read: the entry filter is cleaner than the mid-month exit. The mid-month exit helps some headline metrics slightly, but its edge is not broad; it depends on a small number of exits.
This screen is extremely volatile. Probably what we are seeing here is just a natural outcome of that volatility. "CAGR +0.26 pts" is well within the range of CAGRs we see just between the trading cycles of one day apart. It's just random noise. You cannot predict random noise.
That simplifies the "exit" rules to use, viz. don't try to be clever, just look once a month.
... recovered from its worst drawdown in only 44 trading days ...
I'm not sure that "days to recover" is useful or meaningful when using a good timing method.
Example:
Sell at 2% below 52 week SMA of ^IXIC (Nasdaq 100 index), monthly check.
peak 11/2007 @ $2012
OUT 2/2008 @ $1663 (-17% drop)
IN 9/2009 @ $1623
recover 1/2010 @ $2025
recovery time 26 months.
Untimed:
peak 11/2007 @ $2517
recover 1/2011 @ $2524
recovery time 39 months.
Bottom: 3/2009 @ $1195 (-50% drop)
The goal of timing is to sidestep the large drawdowns. Which it did. How long it takes to recover is up to the stock, which timing has no control over. Unless you have a superb bottom detector.
No. of Recommendations: 8
Can someone post the yearly returns for this strategy or a version like it (by year)?
I'm curious if astronomical returns in this wave and prior waves drove all or most of it ...
In other words, how consistently does it produce positive returns by year.Asking for a firehose to take a sip of water from. ;-)
Here is what you really want to see. The rolling N-year CAGR returns beginning every month and not just January. For a few numbers of years. There are a lot of bad 1-year periods but fewer bad 3-year periods. And the bad years are
really bad.
No timing, 52 week momentum Nasdaq 100 stocks, top 10 HTD 12, 21 day hold, worst cycle. Data from GTR1.
**** 1985 - 2026 ****
# Yrs --> 1 2 3 5 10 15
# >0 387 406 406 416 361 312
# <0 93 62 50 16 11 0
Win % 81% 87% 89% 96% 97% 100%
Yrs --> 1 2 3 5 10 15
Avg 28.9% 25.2% 23.4% 22.9% 21.8% 20.2%
Med 25.7% 23.5% 22.6% 23.6% 20.8% 18.8%
Min -63.1% -48.5% -32.5% -15.8% -4.6% 4.2%
Max 230.9% 142.9% 114.5% 78.1% 53.2% 46.0%
2/3 range 10.6% 15.5% 16.5% 15.5% 18.1% 16.8% <<--- 2/3 are better.
95% range -30.9% -16.5% -5.1% 0.8% 3.7% 10.4% <<--- 95% are better.
--------------
**** 2006 - 2026 ****
# Yrs --> 1 2 3 5 10 15
# >0 188 179 182 181 121 61
# <0 41 38 23 0 0 0
Win % 82% 82% 89% 100% 100% 100%
Yrs --> 1 2 3 5 10 15
Avg 22.2% 19.7% 18.6% 20.6% 21.7% 20.4%
Med 21.7% 21.5% 22.2% 20.8% 21.0% 21.2%
Min -49.6% -22.3% -10.1% 4.1% 13.6% 15.6%
Max 117.5% 55.8% 38.2% 38.8% 30.5% 24.4%
2/3 range 9.6% 13.8% 14.9% 15.5% 19.6% 19.7% <<--- 2/3 are better.
95% range -33.6% -15.9% -4.5% 8.1% 15.2% 16.2% <<--- 95% are better.
-----------------------------------------------------------------------
With a simple & reasonable timing scheme
**** 1985 - 2026 ****
# Yrs --> 1 2 3 5 10 15
# >0 375 421 437 428 372 312
# <0 90 47 19 4 0 0
Win % 81% 90% 96% 99% 100% 100%
Yrs --> 1 2 3 5 10 15
Avg 27.4% 24.9% 23.5% 23.4% 23.6% 23.5%
Med 20.2% 21.9% 21.8% 21.5% 22.1% 22.9%
Min -32.6% -19.2% -8.2% -2.7% 6.0% 12.3%
Max 230.9% 142.9% 114.5% 78.1% 47.6% 41.6%
2/3 range 10.1% 14.0% 14.9% 17.0% 19.2% 19.3% <<--- 2/3 are better.
95% range -15.7% -7.5% 0.4% 4.0% 11.7% 14.8% <<--- 95% are better.
--------------
**** 2006 - 2026 ****
# Yrs --> 1 2 3 5 10 15
# >0 176 190 197 180 121 61
# <0 46 27 8 1 0 0
Win % 79% 88% 96% 99% 100% 100%
Yrs --> 1 2 3 5 10 15
Avg 19.7% 18.2% 17.3% 18.3% 18.7% 18.1%
Med 17.8% 18.7% 18.2% 18.7% 19.5% 18.2%
Min -32.6% -19.2% -2.6% -1.1% 11.2% 14.8%
Max 66.3% 49.6% 32.4% 32.2% 25.9% 22.0%
2/3 range 9.2% 13.5% 12.5% 16.2% 15.8% 17.1% <<--- 2/3 are better.
95% range -17.1% -9.7% 2.2% 2.6% 12.8% 15.6% <<--- 95% are better.
No. of Recommendations: 4
Series CAGR Sharpe Beta UI MDD Longest more than-10% crash recovery period (MDD recovery might be shorter)
Momentum5 21.75% 0.765 0.814 20.22 -55.51% 528d
PHL180_5 22.21% 0.819 0.726 18.77 -55.51% 1162d
Overlap 26.89% 0.849 0.773 17.65 -60.56% 533d
Entry filter 27.87% 0.861 0.775 17.08 -60.56% 488d
SPY 7.90% 0.491 1.000 16.22 -55.19% 1020d
NDX 7.55% 0.403 1.206 44.61 -82.90% 3292d
This is the 10/93 to 4/2026 data period I have.
It looks at the simplest strategy: buy top 5 by 9 month momentum ("Momentum5") but goes to cash on next month purchase date if SPY<325 dah SMA.
"PHL180 5" is top 5 by price/(180 day price high +180 day price low), with same timing rule
"Overlap" is the overlapping stocs in those 2 screens (up to 5)
"Entry filter" applies another criteria to a purchases: last 5 day average share volume (not dollars)must be above last 60 day average share volume; if not, don't buy that stock. Using this criteria mid-month adds a little CAGR but is due to very few stocks so I'm not using it.
In tests of best day to buy, 2nd to last day of month was better than the last day of month, and last day was better than the 1st day of the next month.
Monthly trading.
No. of Recommendations: 1
"d) don't enter any position on trade day if that particular stock's 5 day share volume (not dollar volume) is less than its 60 day share volume;
Market days, right? Not calendar days? And average N-day volume, right?" Yes market not calendar days.
"e) during the month after purchase, exist any stock no longer meeting "d" criteria i.e. make sure daily," This effect was due to ony a small number of trades and when I looked at ignoring it during the month there was only a tiny effect. Interpretation is that demand faing this much precedes drop or lack of a future up period. I pretended, however that you bought back in if no longer true on trade date. ’ll quantify each ADV-triggered early sale: what happened if sold at the trigger versus held to the normal next rebalance, then summarize win rate and whether a few names explain most of the benefit.
Computed.
For Overlap + volume entry filter + mid-month ADV exit, trade day -1:
Item Result
early-sale events 68
Measurable events 67
Unique stocks sold early 45
Beneficial exits 34
Harmful exits 33
Beneficial rate 50.7%
Median stock return after exit to next rebalance -0.03%
Average stock return after exit to next rebalance +0.99%
So the rule is not consistently predictive. It is basically a coin flip by count.
The portfolio improvement came from a few large helpful exits offsetting many small or harmful exits. Top positive contributors:
Stock Approx portfolio benefit
TSLA +10.8%
VRTX +8.3%
SHLDQ +6.5%
PTON +5.7%
ISRG +5.4%
Top 5 helpful stocks explained about 62.6% of all positive exit benefit.
Marginal effect of adding the mid-month exit on top of entry filter:
Metric Change
CAGR +0.26 pts
Sharpe +0.015
Beta -0.022
UI -0.87
MDD +1.26 pts better
MDD recovery 18 days faster
Longest ≥10% recovery 140 days worse
My read: the entry filter is cleaner than the mid-month exit
My read: the entry filter is cleaner than the mid-month exit. The mid-month exit helps some headline metrics slightly, but its edge is not broad; it depends on a small number of exits.
More <-10% drops but massively more up gain between them and faster recovery (see above).
A matter of preference on what is most important to you. For me in last 28 years recovery times has mattered more: if I get 3X market I am wiling to have drops as long as I recover faster. Sharpe, Beta, and UI all better.
No. of Recommendations: 2
The only warning I would give is post-1999 has worst performance than 185-forward tests from gtr . I made sure to test full on gtr and 1993 forward with my data, and post 1999 since post-1999 of 9 month momentum showed 19% cagr when it was higher 1985 forward.
No. of Recommendations: 6
mo wrote: "Can someone post the yearly returns for this strategy or a version like it (by year)? I'm curious if astronomical returns in this wave and prior waves drove all or most of it ... In other words, how consistently does it produce positive returns by year."
41 years of annual returns for {Nas100Momentum} follow a log-normal distribution with fat tails.
The abnormal bad years were: 2008 and 2022.
The abnormal good years were: 1991, 1998, 1999, and 2020.
Total return by year:
Year S5T N1T N1Mo SPY QQQ myN1Mo
2026ytd 8.5 14.9 91%
2025 17.8 23.1 51.0 17.7 20.8 68%
2024 25.4 32.4 15.3 24.9 25.6 25%
2023 26.5 54.9 21.0 26.2 54.9 28%
2022 -18.1 -33.1 -43.5 -18.2 -32.6 -35%
2021 28.3 28.6 13.1 28.8 27.4 11%
2020 18.9 50.1 118.9 18.4 48.6
2019 31.3 40.6 33.2 31.2 39.0
2018 -4.3 -0.3 0.9 -4.6 -0.1
2017 22.0 33.1 39.0 21.7 32.7
2016 11.8 7.3 25.4 12.0 7.1
2015 1.3 9.9 22.3
2014 13.7 19.0 38.1
2013 32.4 37.3 54.7
2012 16.2 18.7 23.1
2011 1.7 2.2 -1.5
2010 15.0 14.1 32.3
2009 26.2 52.4 9.5
2008 -36.6 -41.0 -41.8
2007 5.6 16.0 44.2
2006 15.8 8.3 5.3
2005 5.0 -0.7 35.7
2004 10.9 6.3 25.0
2003 28.7 46.5 65.6
2002 -22.0 -35.3 -24.7
2001 -11.6 -26.0 -19.4
2000 -9.4 -40.2 -17.4
1999 20.8 90.0 185.9
1998 29.2 82.5 104.5
1997 33.5 22.0 31.8
1996 23.0 41.0 16.1
1995 37.6 42.8 31.1
1994 1.2 1.8 16.0
1993 10.0 11.1 19.9
1992 7.7 8.7 7.7
1991 30.6 66.3 156.6
1990 -3.3 -7.2 9.2
1989 31.4 29.6 72.0
1988 17.1 15.4 42.1
1987 5.1 13.0 49.5
1986 18.3 8.6 27.5
1985 23.2 22.7 39.9
S5T is the GTR1 index (close to SPY).
N1T is the GTR1 index (close to QQQ).
N1Mo is the GTR1 screen {Nas100Momentum} traded monthly, 0.1% friction, depth 5 stocks.
GTR1 results are from 19850201 to 20251128.
SPY and QQQ ETF returns are from M*.
myN1Mo is my backtester results for {Nas100Momentum} (not gtr1).
2026ytd is to Friday June 5th.
http://gtr1.net/2013/?~Nas100Momentum_20191030_rge...https://gtr1.net/2013/?~N1T:h1::trp%281,1%29ne-999...https://gtr1.net/2013/?~S5T:s19850201h1::trp%281,1...
No. of Recommendations: 4
By the timing that I like best, there are only a few periods where the timing has you out for more than a few months.
Here is the OUT statistics for 2/1985 to 12/2025:
# of months out # of times
1 5
2 3
3 5
4 1
6 1
8 1
12 1
13 1
17 1
18 1
The 18 & 12 is really a 31 month bear market with one whiplash IN month in the middle. 10/2000 to 4/2003. The dot-com bust.
The 17 is the bear market 2/2008 to 6/2009. The financial/bank bust.
The 13 is the 2002 bear market, 2/2022 to 2/2023.
Really, just 3 significant out-of-market periods in 41 years. Hard to make predictions on something that only happened 3 times in 41 years.
You don't need to be perfect, you just need to be Good Enough™
I don't mind not catching the absolute bottom, since I'm sitting cozy in short-term T-bills while waiting for the storm to pass and the sun to come out.
I have a theory why the "overlaps" works so well.
Here's the CAGRs for the number of stocks from 10 to 1:
10 24.5%
9 24.6%
8 25.2%
7 25.8%
6 26.7%
5 27.4%
4 28.6%
3 29.8%
2 31.7%
1 28.7%
Taking the overlaps from the top 5, you are essentially doubling up on some of those top 5 stocks.
No. of Recommendations: 5
2000 - 2002 would have probably caused me and others to throw in the towel (unfortunately).
No. of Recommendations: 1
Sorry, convos been long and hard to nail down exactly: which timing do you prefer, Ray (gtt?)?
No. of Recommendations: 6
which timing do you prefer, Ray (gtt?)?
Once I put my mind to it, it was easy to try many different timing schemes. The benefit of doing it all in a spreadsheet, from the GTR1 "daily value" results.
I figured that this high-volatility screen in a specialized market segment would probably not be best suited to my standard timing of GTT on the S&P500.
I tested about 15 different ones.
The best overall was:
52 week SMA of ^IXIC (Nasdaq 100 Index), sell at 4% below the SMA, buy at 0% above the SMA, in between stand pat (no action).
The two periods of this timed screen tested were 1985-2025 (since NASDAQ 100 inception) and 2006-2025 (since we retired.)
The SMA periods tested were 43 , 52, and 64 weeks.
The indexes tested were ^IXIC and ^GSPC (S&P500).
The sell signal % tested were 0%, -1%, and -4% below SMA.
The timing signal obeyed was the most recent weekly signal at the trading day.
I was looking for the best combination of MaxDD and Sortino Ratio and low whiplash count. Regardless of any change in CAGR.
Some were best in one date period, others were best in the other date period.
The one I settled on was the best combination in both date periods.
Look, for the entire 1985-2025 the (untimed) growth of $10,000 was $46,000,000. Of course, with huge volatility & huge and long drawdown.
For comparison, VFINX (S&P 500) grew $10,000 to $766,000.
The growth of all the tested timings ranged from $17,000,000 to $50,000,000.
Heck, nobody is going to turn down growing $10K to even $17 million.
No need to mess around with declining volumes or exiting mid-month or any other tweaks.
It doesn't need to be the best, it just needs to be good enough.
You wouldn't even need to mess with timing except for the -67% drawdown that took 10 years to recover.
No. of Recommendations: 1
Appreciated. Yeah, you just have to time things I think ... Praying for always up doesn't seem like an adequate "plan".
No. of Recommendations: 2
Just a note, ^IXIC is the Nasdaq, ^NDX is the Nasdaq 100
No. of Recommendations: 2
Just a note, ^IXIC is the Nasdaq, ^NDX is the Nasdaq 100
You are correct. I mis-called it.
FWIW, timing with ^NDX has worse results than ^IXIC.
No. of Recommendations: 3
recomputed cleanly for -2, -1, +1.trading days. +1=first day of each month, -1 means last day of a month. Using Norgate data and Chatgpt to cacuate.
Signal data = prior trading day before the trading day, so prior day's close and volume to do the math
Trade = selected offset day close
Timing rule = SPY > 325-day SMA; if it isn't, you go to cash on trade day SMA is simple moving average
Other criteria for entry:
Overlap of picks between two different 5 stock strategies are bought, but only if 5 day share volume is greater than 50% of the 60 day trading share volume ave
Full available data period: Oct 1993–Apr 2026 mdd=max drawdown UI=Ulcer Index CAGR=compound annual growth rate "Longest recovery" is greatest # of days to make back any loss of -10% or greater in that strategy SPY=S&P500 NDX=Nasdaq100 "Best day" is the day with best CAGR to do your buys (and sells) for that strategy
Strategy
Best day CAGR Sharpe Beta UI MDD Longest recovery from any -10% or worse loss
SPY +1 10.65% 0.635 1.00 14.66 -55.2% 1020d
NDX +1 13.72% 0.611 1.21 40.16 -82.9% 3292d
9m Momentum
Top 5 +1 27.94% 0.890 0.90 19.82 -56.4% 1199d
PHL180
Top 5 -1 29.96% 0.991 0.81 17.16 -55.5% 1162d
Overlap +
volume
entry -2 37.97% 1.061 0.82 17.72 -45.2% 462d
Post-1999
Strategy Best day CAGR Sharpe Beta UI MDD Longest recovery
SPY -2 7.98% 0.495 1.00 16.21 -55.2% 1020d
NDX -2 7.54% 0.403 1.21 44.60 -82.9% 3292d
9m Momentum
Top 5 +1 21.24% 0.750 0.83 21.38 -56.4% 1199d
PHL180
Top 5 -1 22.36% 0.825 0.72 18.73 -55.5% 1162d
Overlap +
volume
entry -2 27.61% 0.878 0.73 19.18 -45.2% 462d
So for the practical version people might actually use monthly, the best tested day was -2, not +1. 6% cagr improvement from 9 month top 5 with SPY timing rule. Almost 4X index fund return despite owning only 5 stocks
No. of Recommendations: 3
I've been trying to boil this thread down to something I understand and can calculate on my own using BarChart and Stockcharts websites.
Could I get a reality check? - Does the below seem reasonable or am I missing something important?
Many thanks in advance!
---
Nasdaq-100 stocks:
* Each month, trade on the 2nd to last day of the month, using the previous day's Nasdaq-100 closing prices
* On Trading Day (at or near open), check to make sure that SPY is above its 325 trading day MA. If not, sit in Ultra-Short Bond ETF while out of market
* Keep any existing holdings that are still in the top-12 9-month performance rankings (let winners run)
* Pick/Hold the Top 9-month performance Nasdaq-100 stocks, filling to five positions (Note: 9-12 months performance all seem to work)
-- Don't enter a position on trade day if that particular stock's 5-day share volume (not dollar volume) is less than its 60-day share volume;
-- Alternative to improve recovery time with small loss of CAGR: Skip a pick if its dollar volume ADV20/ADV700 is declining (Question: 20-day ADV < 700-day, right?)
* Buy each month (and continue to own unchanged stocks) as long as SPY is above its 325 MA on trading day.
-- TS Option: Place a 35% trailing stop on each stock
No. of Recommendations: 4
I am constantly trying to improve the strategy. Here is the latest, which generated a 36% compound annual growth rate 12/31/99-4/28/2026 (it doesn't include the monster run-up we had since April).
1) you determine if SPY is above its 325 day SMA (it is right now). If not you go to cash on trade day.
2) you trade the 2nd-to-last TRADE day of each month.
3) you get the list of Nasdaq100 stocks and make sure Nasdaq didn't make changes since you last looked.
4) you make two lists: a list of the top 9 months price change stocks, top 5 in the Nasdaq100, through yesterday's close (i.e. close on the 3rd-to-last day of the month).
5) you make another, separate list of the top 5 in the Nasdaq100 measured by the next ratio: price / (sum of highest price in last 1 year + lowest price in the last year)
6) on the 2nd-to-last day of every month you buy the stocks that overlap from those 2 lists i.e. in #4 and #5
Often there will be fewer than 5 total stocks. I tested going back to the two lists to buy more so you own 5 or 10 stocks, and it massively reduced return. So my advice is to buy much more of the overlap list you create and if you still want some more stocks use less money to buy other stocks from the 2 lists.
No. of Recommendations: 12
I've been trying to boil this thread down to something I understand and can calculate on my own using BarChart and Stockcharts websites.
Could I get a reality check? - Does the below seem reasonable or am I missing something important?
{8 rules}
IMHO, this is way over-tuned. I became sensitized to overtuning in the early years of TMF and Mechanical Investing, when many of the screens that backtested great failed miserably after the were published. The more rules there are and the more fine-tuned the rules were, the more likely that the rules were tailored to fit just the data they were trained on.
One thing that we tried to avoid this was to divide the historical data in half and apply the rules to each half, to see if there any major difference. Or, as I did, apply the rules to separate 5-year and 10-year data.
This comment here:
"I am constantly trying to improve the strategy." makes me fear that this overtuning is going on.
"Simple is better, because simple is more robust."
"In uncertain environments, simple heuristics tend to be more robust than complex decision rules. And markets are very, very uncertain."
"The best systems are simple, and for good reason. Complicated systems have more opportunities for failure." -- Scott Adams
Okay, my thoughts:
Timing should not be a major part of the screen. It is well known that timing is to reduce the volatility (standard deviation) and avoid the huge drawdowns. Timing almost always also reduces the CAGR. A good timing scheme lowers the stdev more than it lowers the CAGR (percentage wise). Reducing the stdev by 10% while the CAGR is reduced by only 5% is a win.
I don't see 9 month momentum being better than 12 month momentum.
GTR1 shows 9 month as 18.0% CAGR and 12 month as 24.6% CAGR.
12 mo:
https://gtr1.net/2013/?~Nasdaq100%20top%2010:s1985...9 mo:
https://gtr1.net/2013/?~Nasdaq100%20top%2010:s1985...Of course, this depends on the GTR1 database being accurate. I have been assured that it is.
Also, the data for 12 month momentum is widely available. 9 month, not so much. AFAICT, you'd have to compute it yourself. You can using Yahoo data but it is a lot of work.
Here is one source, and you don't even have to log in.
https://www.barchart.com/stocks/indices/nasdaq/nas...The two screens have a large overlap, which makes sense because they are just two different ways of measuring essentially the same thing -- the momentum over the last 9 or 12 months. Owning the overlaps is basically just doubling down on a few of the highest ranked stocks. You wouldn't want to do that as the only strategy -- too risky.
For timing, I think that SPY is not the best index to use. It is quite a different universe than the Nasdaq stocks. My timing backtest says that ^IXIC (ETF: ONEQ) works better than SPY.
My timing backtests with SMAs of 10, 12, and 15 months showed that 12 months ws best overall.
REMEMBER: The purpose of timing is *not* to improve the CAGR, it is to mitigate the downturns.
I don't think the exact day-of-month to trade makes a significant difference. Likewise the short-term vs. long-term average trading volume.
If you look at GTR1's detailed CAGRs, you will see that it varies greatly depending on the exact start date in the cycle. For the 9-month momentum the CAGR ranges from 16.0% to 19.4%. Any tweak that raises the average CAGR from, say, 18% to 18.5% is just random noise. The CAGR you would actually get could be anywhere from 16% to 19%.
"Using stops is 10th grade risk management." The timing on a simple monthly basis is all you need. Fine-tuning it has almost no effect, it just adds complexity.
This screen has a very large turnover. Most of the realized gains will be short term, so it is best to do it in an IRA or Roth.
----------------------
What's left?
* Once a month, somewhere near the end of the month, rank the Nasdaq 100 stocks by the 12 month momentum (either straight RS or price/(52wk HI + 52wk LO).
* Sell the ones that have fallen out of the top N.
* Use that money to buy equal shares of the ones that you don't have.
* But first, check the timing rule. If it says to get OUT, sell everything and switch to a short-term T-bill ETF.
* If you want, run one portfolio in two accounts, each using a different momentum type.
* Once those portfolios have done good, gild the lily and run an "overlap" portfolio in another account.
No. of Recommendations: 5
* But first, check the timing rule. If it says to get OUT, sell everything and switch to a short-term T-bill ETF.
This advise has always been the hardest part of MI for me to execute timely but is good advise. May be why my results are just average.
Instead I Sell/Buy using a Variable allocation dependant on the Trend analysis.
GD_
No. of Recommendations: 1
Thank you, Musselmant!
A couple follow-up questions:
* Are you letting winners run? If so, how does that integrate with buying overlaps?
* Any thoughts on Cash vs an Ultra-Short Bond ETF if the timing kicks one out?
No. of Recommendations: 2
And thank you Ray!
A follow-up question for you:
How would one decide on the 12 month momentum calculation, choosing between either straight RS or price/(52wk HI + 52wk LO)?
Straight RS would seem a lot simpler; is there value to be gained by performing a further calculation?
Also what are your thoughts on Cash vs ST T-Bill ETF when not in?
Thanks again to all involved here - this board seems to have maintained the spirit of the old Foolish MI board, which is grand!
No. of Recommendations: 2
Follow-ups follow-ups:
Are you using ^IXIC for the timing step? SMA325?
No. of Recommendations: 6
Once a month, somewhere near the end of the month, rank the Nasdaq 100 stocks by the 12 month momentumFor those who are interested, this is easily done on Barcharts:
https://www.barchart.com/stocks/quotes/$...Baltassar
No. of Recommendations: 6
How would one decide on the 12 month momentum calculation, choosing between either straight RS or price/(52wk HI + 52wk LO)?Flip a coin.
Of course, straight RS is simpler and extremely easy to find. No computation, just look at the figures.
Actually the 52 week Hi & lo are also easy to find. You just have to load the data into a spreadsheet and do the calculations and ranking yourself.
I have all this stuff automated with bash scripts (running with Cygwin on Windows 10) so it was easy for me to clone the RS script into the HL-52 script.
I ran only the top 10 RS for over a year then branched out to the HL-52 after reading Jim's post #4412
I ran an older Jim screen for several years. It did great and I split half of it off to the RS screen in Jan 2025. Sadly, that screen did absolutely *nothing* in 2025 while the RS screen almost doubled.
Also what are your thoughts on Cash vs ST T-Bill ETF when not in?I've been spoiled by Fidelity automatically sweeping your cash into SPAXX.
Here's the deal. Most of the timing OUT periods are only a couple of months long. The amount of interest you'll get in 1 or 2 months is trivial, so it doesn't matter if it is pure cash or bonds.
But *some* of the out periods are very long, so you would want to put the money into some interest-bearing asset.
But interest-bearing assets that have good yields are very much subject to capital loss if/when rates change. It's dumb to open yourself to a loss in your supposedly "safe, waiting for the bear to leave" money, so you should put in in very short-term income assets like a 1-3 month MMF or T-bill.
FWIW. The last OUT period with my current timing scheme was the one month of May 2025. I was using a different timing scheme then which said to stay in. (You would have missed out on a 12% gain. Bummer.)
The OUT period prior to that was 3/2022 until 3/2022, out for 12 months. (You avoided a -19% loss)
"The moving average timing strategy makes the majority of its money by avoiding large, sustained market downturns. To be able to avoid those downturns, it has to accept a large number of small losses associated with switches that prove to be unnecessary. Numerically, more than 75% of all of MMA’s trades turn out to be losing trades." -
http://www.philosophicaleconomics.comBut also, "A strategy that makes comparisons to a single prior level (i.e., momentum) is vulnerable to single-point anomalies in the data, whereas a strategy that makes comparison to an average of prior levels will smooth those anomalies out." -
http://www.philosophicaleconomics.comHmmm, I may have to look into that. Jim has mentioned that a number of times.
No. of Recommendations: 5
Are you using ^IXIC for the timing step? SMA325?
Sell on -4% below the 52 week SMA of ^IXIC.
Buy on at or above the same SMA.
In between, stand pat, take no action.
No. of Recommendations: 1
How would one decide on the 12 month momentum calculation, choosing between either straight RS or price/(52wk HI + 52wk LO)?
Flip a coin.
---
Well that seems clear! ;-)
So was there a perceived advantage to the Hi/Low method that didn't pan out or was it just for love of numbers?
No. of Recommendations: 4
So was there a perceived advantage to the Hi/Low method that didn't pan out or was it just for love of numbers?
There's probably an 80% overlap between the two screens, so not much difference in expected return.
Frankly, the first portfolio (52 week RS) did so well in 2 years that I was getting freaked out. So I decided to diversify a bit and spin off a HL-52 portfolio.
That one is up 88% in 11 1/2 months.
The RS portfolio is up 150% in 18 months.
Scary. There's no way that things could continue like that.
And it's not even MU that is the best winner.
BTW, when a stock in the screen goes bad, it goes bad really fast. Like -20% in 2 weeks.
The saving grace is that a bad stock usually drops off the screen soon.
No. of Recommendations: 4
I have let winners run since using real money and won big (MU up over 500% since October). I am testing on Norgate data on Chatgpt now; so far not only do tests show that wise, but even using sales proceeds to buy more if still on the list adds another 4.4% cagr! All major fronts, sharpe, beta, UI, recovery time better than SPY, NDQ, better than 9-12 month momentum, better than price/(hi price 52 weeks + low price 52 weeks).
I don't agree doing what is simple to look up is best. 9-12 month mo works but 9 is best in backtests; there is nothing special about 200 days or 1 year. Googlesheets can do your calculation instantly with no effort beyond checking additions to Nasdaq100 since you last looked.
And in my tests using SPY as the timing mechanism, and 325 days beat using NDQ/QQQ and beat 200 days. 300-350 days all were good and prevented whipsaws that shorter SMAs caused.
And while reducing drawdowns is #1 goal it is not suspicious if it also improved other metrics, just a good thing. Why not have your cake and eat it too, Marie be damned.
The overlaps are simply better than either list, and 5 beats 10 stocks by quite a lot.
The recovery is better than with SPY or NDQ.
And tests work post 1999, just not as well as 1985-forward. I haven't even included post April 2026 yet! I am making a killing this year.
No. of Recommendations: 4
using sales proceeds to buy more if still on the list
What does this mean?
Say you own 5 stocks A, B, C, D, E.
New month picks are C, B, W, X, E.
So you should sell A & D and buy W & X and keep B, C, E.
Normal operation would be to take the money raised from A & D and buy W & X in equal weight.
A complete rebalance would mean you might sell some of B, C, E so that all 5 stocks start the month at equal weight.
Are you saying that you would buy MORE of B, C, and E? Just because they are still on the list? That would mean shortchanging the new picks W & X.
If you are retired and the account is an IRA, no money can be added to the account. You can only work with what's there.
9-12 month mo works but 9 is best in backtests
...
SPY as the timing mechanism, and 325 days beat using NDQ/QQQ and beat 200 days
I am reminded of the joke about the guy who installed 5 "gas saving" devices on his car, each promising to save 25%.
When he drove around the block his gas tank overflowed.
CAGR 20% is so much better than CAGR 10% that an additional 1%-2% barely matters. Anyway, just a one day difference in start date makes a +/- 4% difference in CAGR.
I do not know why we are getting different results on our choice of timing criteria.
Maybe because you are using daily prices for the SMA and I am using weekly prices?
Maybe because I use 4% below the SMA for the sell signal and you use 0%?
Maybe because I am using historical data from Yahoo and you are using Chatgpt AI?
(I assume we both only consider the timing signal on the trading day.)
No. of Recommendations: 4
9-12 month mo works but 9 is best in backtestsGTR1 shows 12 is better than 9. Not much though. 0.3% to 0.6%
Again, I do not know why we are getting different results. I'm thinking that it must be because we are using different data sources. But the differences are so small that it's probably just minor differences in the exact figures. Round off to zero decimal places and we get the same results.
GTR1 reports data out to 6 decimal places -- which is just ridiculous.
From GTR1:
19850201 to 20251128
Top 10, 252 days (12 months)
Avg Min Max
CAGR: 24.625883 23.190943 26.638947
Top 10, 189 days (9 months)
Avg Min Max
CAGR: 23.300272 21.831718 26.605984
Top 10 HTD 12, 252 days (12 months)
Avg Min Max
CAGR: 24.519554 22.958212 26.837484
Top 10 HTD 12, 189 days (9 months)
Avg Min Max
CAGR: 23.603947 22.298716 26.066774
Ah, I'm using 0.25% friction on all sales. That surely makes a difference.
Maybe it depends on the start date?
Maybe because I require initial price to be more than $10? To avoid penny stocks, but have there ever been penny stocks in the Nas100? IDK.
20010201 to 20251128
Top 10 HTD 12, 252 days (12 months)
Avg Min Max
CAGR: 15.312686 13.922186 16.894257
Top 10 HTD 12, 189 days (9 months)
Avg Min Max
CAGR: 15.393602 13.852866 16.735155
9 better than 12 by 0.08% on average. But 12 better than 9 for min & max.
These differences are so small it is just random noise. There is no reason to prefer one over the other.
OTOH, 6 months is not good.
Top 10 HTD 12, 126 days (6 months)
Avg Min Max
CAGR: 12.848204 10.670835 14.536878
As is 15 months.
Top 10 HTD 12, 315 days (15 months)
CAGR: 12.625591 11.064657 14.364594
13 months not so great either
CAGR: 13.054716 10.581337 14.954453
Looks like 12 month lookback is the upper limit. Looks like anywhere 8-12 months is good.
No. of Recommendations: 3
No. of Recommendations: 0
I haven't done this yet; I have just let winners ride.
---
By "winner", you mean, a stock that is still in the Top 12 sort, right?
No. of Recommendations: 0
Rayvt wrote:
"I do not know why we are getting different results on our choice of timing criteria...
Maybe because I use 4% below the SMA for the sell signal and you use 0%?"
-----------------------
Ray - to clarify, do you mean ^IXIC needs to be 4% above its 325-day SMA to pass the timing test?
No. of Recommendations: 0
Sorry!
- Ray had already answered this question: "Sell on -4% below the 52 week SMA of ^IXIC."
And that "Sell If" is at the end of the month, right - not mid-month or with daily checks?
And then, before buying, confirm ^IXIC is above its 325 day SMA, right?
No. of Recommendations: 4
"I haven't done this yet; I have just let winners ride.
---
By "winner", you mean, a stock that is still in the Top 12 sort, right? "
Until now I was using simply top 10 by 9 month momentum (with Spy required to be above its 325 day SMA, which has been for a long time).
But my research showed top 5 were much better so I am putting much more in them than #6-10 (by chance INTC was # 10 before I did this so I doubled my money on INTC in one month).
And now the overlap strategy that overlaps the top 5 with the top 5 of the related Nasdaq100 screen (price/ (highest price in 1 year + lowest price in 1 year) ) tested even better (but you won't have 5 stocks; 20% of the time there are no overlaps and 3 stocks is typical) is better yet, so I have to decide how concentrated I can tolerate. MU is up over 500%.
The testing showed buying more of existing holdings as long as they keep showing up on the overlap list was even better (using proceeds of sales of stocks leaving the list); that will be scarier yet but is what the testing showed. With MU I would have made even more doing this i.e. when I was up 300% buying yet more as it went up over 500%.
Top 10 momentum by 9 months to 12 months is good. 9 better than 12. Timed with SPY rule better yet. Price/high+low top stocks also good, and when combined as an overlap even better.
Version 10/29/1993–4/9/2026 CAGR Sharpe UI MDD
Monthly equal rebalance 41.8% 1.14 16.9 -42.4%
Hold until removed, invest cash only in new names 43.2% 1.13 18.9 -43.7%
Post-1999 check:
Version Post-1999 CAGR
Monthly equal rebalance 31.1%
Hold until removed 35.8%
No. of Recommendations: 1
what I meant was if you sell stocks leaving the list, you use the sales proceeds to buy equal dollars of the things on the list with that new money, even if you already own some of them.
Ah, that is interesting. That would have you moving to a larger and larger allocation to the stocks that stayed on the screen a long time. Even if they cut barely made the cutoff while the new stocks were ranked higher. Sounds risky. If the same stock stayed on the screen for a year or two it would be near 100% of the portfolio. Also, I don't know how yuo could backtest this. GTR1 doesn't have that option.
No. of Recommendations: 1
By "winner", you mean, a stock that is still in the Top 12 sort, right?
Yes.
No. of Recommendations: 3
to clarify, do you mean ^IXIC needs to be 4% above its 325-day SMA to pass the timing test?
No, the opposite. And not 325 days. 52 weeks.
(BTW, I hate it when people say days -- never clear if it means market days or calendar days. In GTR1 it means market days, in google sheets it means calendar days. Weeks is unambiguous.)
Here is the formula: =if(C6>0%,"BUY",if(C6<-4%,"SELL","No action"))
C6 is: =A6/B6 -1 Which is the percentage of current price vs the average of the closing prices of the last 52 weeks.
A6 is =googlefinance(".ixic")
B6 is the average of =GOOGLEFINANCE(".ixic", "price", today()-365, today(), "WEEKLY")
That is, the average of the last 52 weeks of IXIC,
No. of Recommendations: 4
And that "Sell If" is at the end of the month, right - not mid-month or with daily checks?
And then, before buying, confirm ^IXIC is above its 325 day SMA, right?
Mid-month, daily, or weekly, don't matter. Once a month at the regular trade date is good enough.
Most of the sell signals have a buy signal soon afterward, so you aren't gaining anything by selling early. It's just a small whiplash and doesn't have any big effect.
What *does* have a big effect is the times when timing takes you out for 9 or 18 or 38 months. And for those times, getting out a week or two earlier doesn't make any difference.
Timing is essentially a "see lightning, hear thunder" type of thing. (Indian joke, to pick braves qualified to go hunting pony soldiers. You don't have to be eagle-eyed and bat-eared, you just need to be able to tell there's something there.)
BTW, you can't have different periods for the sell signal and buy signal. Otherwise you could have one saying to BUY while the other says to SELL.
No. of Recommendations: 2
The 2nd link you posted
https://gtr1.net/2013/?s20080131h21f0.10000::nas10...throws a GTR1 error: "Variant 0: Corrupt or obsolete cache file (c1/dba26c96a57534bd10807d172e6480a1 is in state -99) [5373]."
Maybe it'll straighten itself out in a few hours.
The 1st link runs ok.
https://gtr1.net/2013/?h21f0.10000::nas100.a:et1:a...I downloaded the daily values and could not tell when the BCC had it going to cash. I tested that by seeing if two successive days had the same value. But of course that won't work if GTR1 imputes a dividend on the cash.
Loading the cycle 0 data into my spreadsheet (note, the MaxDD is the 12-month, not alltime):
Cycle 0, no timing
CAGR 25.0%
Stdev 35.0%
MaxDD(12) -63%
Sortino 1.29
Cycle 0, with timing 52 wk SMA, -4% below ^IXIC
CAGR 22.9%
Stdev 30.1%
MaxDD(12) -37%
Sortino 1.41
Cycle 0, with timing 64 wk SMA, -4% below ^GSPC
CAGR 23.1%
Stdev 31.2%
MaxDD(12) -50%
Sortino 1.33
For comparison, with the BCC data:
CAGR 23.6%
Stdev 30.5%
MaxDD(12) -44%
Sortino 1.46
These are all pretty darn close. Just choose whichever combination of CAGR, Stdev, MaxDD, and Sortino you prefer.
The CAGRs for BCC range from 23.349739 to 28.411659 and
with no BCC from 24.967045 to 31.041727.
Hmmm, I wonder if this BCC calculation doesn't have an inadvertent lookahead bias. That is a very common problem and very easy to make. I made this mistake myself in the first iteration of my spreadsheet.
When changed from "ratio(gprc(1)" to "ratio(gprc(2)" the CAGR range becomes 22.235964 to 26.787621.
"ratio(gprc(3)" goes to 22.135351 to 26.537304
Big difference in outcome between 1 and 2, very little between 2 and 3.
A one day change in the SMA comparison should not have a significant change in outcomes, so there is reason to suspect a lookahead error.
With the BCC data using "ratio(gprc(2)"
CAGR 21.1%
Stdev 30.7%
MaxDD(12) -46%
Sortino 1.27
Big difference between 23.6% and 21.1%.
Oh, "Close all liquid positions on
any market dates where [BCC] == 0"
This BCC does the timing on a *daily* basic. That is unrealistic and would cause a lot of in and out trading. A real-world portfolio would only check timing signals on the regular trade day, or perhaps weekly. Not every day.
No. of Recommendations: 1
"Here is the formula: =if(C6>0%,"BUY",if(C6<-4%,"SELL","No action"))"
---
Thank you Ray - that formula was very helpful in groking the nested IFs!
I see the utility of using weeks rather than days to avoid confusion - but that raises a related question - for use in Google Sheets, should the formula indeed specify 365 days or something like 252 which is what I thought was the equivalent of 52 weeks in Google Sheet-speak?
"B6 is the average of =GOOGLEFINANCE(".ixic", "price", today()-365, today(), "WEEKLY")
That is, the average of the last 52 weeks of IXIC,"
Sorry if this is a dumb question.
No. of Recommendations: 3
I downloaded the daily values and could not tell when the BCC had it going to cash.
Easiest way is to count stocks and download the file and see when there are zero stocks.
Aussi
No. of Recommendations: 1
In Google Sheets ...
The formula "=GOOGLEFINANCE(J4, "price", today()-365, today(), "daily")" delivers 250 daily closing prices, dated 6/16/2025 through 6/12/2026.
Today being 6/14/2026.
So, data for all the market days in the last 365 calendar days. How's that for doubly confusing?
For weekly it delivers 53 weeks of prices, dates 6/13/2025 through 6/12/2026.
p.s., I run my timings daily in a Linux bash script. For my kids, I gave them a google sheet link that they can look at over the weekend.
No. of Recommendations: 0
Easiest way is to count stocks and download the file and see when there are zero stocks.
Doesn't work. It only counts the number of stocks passing each step of the screen.
Doesn't show the "[BCC] > 0; Cash When None" step.
But it's funny, there were two days when 6 stocks instead of 5 were selected.
No. of Recommendations: 0
Can you post the link you are using. I haven't seen that before where Cash when none doesn't show zero stocks.
Aussi
No. of Recommendations: 1
Rayvt wrote:
"The formula "=GOOGLEFINANCE(J4, "price", today()-365, today(), "daily")" delivers 250 daily closing prices, dated 6/16/2025 through 6/12/2026.
Today being 6/14/2026.
So, data for all the market days in the last 365 calendar days. How's that for doubly confusing?
For weekly it delivers 53 weeks of prices, dates 6/13/2025 through 6/12/2026".
---
Oh.. my.
All I can say, is thank you for your patience.
No. of Recommendations: 0
No. of Recommendations: 2
I just ran the two "overlaps" GTR1 links, did SHOW COUNTS, downloaded the reports, and processed to show the number of times each stock count was picked.
Number of days: 10,286
without BCC:
Count # of stocks
56 0
481 1
1993 2
3998 3
3226 4
532 5
With BCC:
Count # of stocks
56 0
481 1
1993 2
3998 3
3226 4
532 5
Exactly the same.
With BCC
http://gtr1.net/2013/?~Nas100_Mom9_PHL252_OverlapT...No BCC
http://gtr1.net/2013/?~Nas100_Mom9_PHL252_OverlapT...------------------
But this overlaps data is actually meaningless. This shows the number of picks EVERY DAY, and not the number of picks in each holding period (month).
Ex:
19871016 4
19871019 4
19871020 4
19871021 4
19871022 0
19871023 2
19871026 2
19871027 3
19871028 4
No. of Recommendations: 1
Well, that was surprising to me. It doesn't show zero for BCC=0
There is a work around. You can check the "Signal Values" and run backtest and download the report. Something I have not checked, but I think is how it works, the signal value goes to zero, and the trade is made the next day.
Also, could someone advise what the ""2" and "-1" do in the command imports(0,BCC,2,-1)? I use imports(0,BCC,0,0) but I use gprc(0) in the imports field.
Aussi
No. of Recommendations: 3
2 is lag days
-1 is default value
Check the GTR1 Results for the Command Translation:
Create [BCC]: Import [BCC] as signal (screen_number=0, lag_days=2, default_value=-1)
No. of Recommendations: 2
You can check the "Signal Values" and run backtest and download the report.
That worked.
I made a new post on the GTR1 timing findings, so it wouldn't get buried in this thread.
No. of Recommendations: 1
I believe the "overlaps" backtest linked in post #5663 is wrong.
Details in a new post. Mentioned here so that it won't get lost in the long thread.
No. of Recommendations: 1
Rayvt wrote:
* Once a month, somewhere near the end of the month, Just as a reality check - is there a significant advantage beyond it being something of a convention to trade at the end/beginning of a month?
--
Out of curiousity, does the timing of trades generally impact the logic of screens? I've meant to ask before, but it's the holding period that is more important, right? Trading for example, mid-month should not be inherently different than on the last day of the month, correct? (beyond any minor effects of certain timing periods that may or may not be signficant)
I ask as a reality check - is there an actual significant benefit to trading at the end of the month or is it just a convention to more easily be able to crunch data sets and talk about them?
No. of Recommendations: 3
If you look at the historical daily returns of the 88 ETF trading simulations on Allocate Smartly there is evidence that trading on the last day or couple of days have done slightly better over the long term history. But Allocate Smartly also states that this effect has gone away in the last 5 years.
No. of Recommendations: 4
I've been doing my trades on the last day of the month for quite a while. It's a PITA most of the time. Especially if you are running multiple monthly screens.
You have to run all your screens overnight in the middle of the week, with no time to think about things and it consumes your leisure time for that night.
For bookkeeping all the monthly account/portfolio balances I've always treated a month to end on the last Friday of the month. For a couple you can easily have 6-8 accounts between the two of you.
It's just a lot simpler to align the trading month the same way.
If you are trading with a GTR1 backtested screen, its 21 day cycle does not line up with calendar months anyway. Close, but not exact.
No. of Recommendations: 0
Thank you for the reality check!
No. of Recommendations: 6
Pandrea,
Thanks. I find that getting questions often helps me, by getting me to verify my findings claims and assumptions. And sometimes discovering that there was a mistake or something overlooked or misunderstood.
In this instance I decided to create the timing signal from scratch and make the parameters to be variables so that many combinations could be tested. Instead of the previous ad-hoc hardcoded values.
The CAGRs with timing are suspiciously high, so I triple checked that there was not inadvertent look-ahead bias. The statistics held even with lags of 1, 2, and 3 weeks. Astonishing.
(All done in Excel spreadsheets.)
[Ah, I get it. Not panda. P and Rea.]
-------------
I found the best timing scheme for these related momentum screens to be sell at -4% below the 64 week SMA of S&P500.
It was best in Sortino, STDEV, and MaxDD(12).
Much less important, also marginally best CAGR of the timings tested. But all the CAGRs were in the same ballpark.
There were 12 OUT periods.
Of those, 6 were whipsaws. Defined as switching between IN and OUT within 2-3 months.
The longest OUT period was 30 months.
The second longest OUT period was 18 months.
The portfolios:
RS52 5HTD12, P-HL52 5HTD12, Overlaps of top 5 in RS52 & P-HL52
Rankings by Sortino ratio:
P-HL52, Overlap, RS52
Ranking by MaxDD (by 12 month drawdown of the end-of-month values):
Overlap, P-HL52, RS52
All three had about the same STDEV, close to 25%.
The portfolios ranking by timed CAGR for 2006-2026:
Overlap , P-HL52, RS52
No. of Recommendations: 3
Testing on chatgpt suggests the 2nd-to-last trading day of the month beats the last as well as the first of the month.
No. of Recommendations: 4
I suspect we are just seeing the randomness. And perhaps observation bias. And surely false accuracy.
One of the days of the month is going to be best. Statistically there is a 3/21 chance that one of the last 3 trading days will show as the best. We may be just seeing what we want to see.
Here's the false accuracy bit.
From GTR1 for the RS52 5 HTD12 screen 21 day holding period.
The CAGRs for the 21 cycles 2/1/1985 - 11/28/2025 range from 24.1% to 29.6%. That is a 5.5% range.
For 9/1/2006 - 11/28/2025 the CAGRs range from 19.5% to 26.1%. That's a 6.6% range.
I bet that any day-of-month benefit is lost in this range of returns.
Heck, any of us would be ecstatic to get the lowest of these -- 19.5% CAGR for ten years.
What does Chatgpt say about the best improvement you'd get by trading the optimal day of the month? Is it just 2 numbers that are in the above range?
If so, we are probably just looking at a blip in the random noise.
I have never purposely used Chatgpt, from what google AI (yes, I appreciate the irony) says, it "is trained by fed massive amounts of data from the internet (books, articles, websites, and code)."
So what is the chance that it is simply echoing what it has read from internet posts, articles, and papers?
No. of Recommendations: 1
CGPT is only using my Norgate data, nothing else.
End-of-month actions by big boys can explain end-of-month effects rather than randomness.
No. of Recommendations: 1
I gave cgpt Norgate data to 10/1993 and ask it to give results to then as well as post 1999 for the 2000s plus when I care individual years. I ask for cagr, sharpe, beta, UI, mdd, recovery from mdd, max recovery days from all -10% or worse drops, to examine if a few stocks warp the results, if mound-of-toast type results show criteria to be good (thus I use 325 days for the SMA of SPY because 300-350 were all good and better than 200 or 253 days), to see if works out-of-sample when i find a good outcome in a more limited period, to see it during SPY and QQQ downturns, to compare with them on the above stats, if you can think of it I have asked. Hundreds of hours of conversations with it by now. I updated Norgate data to a few days ago recently (so results went up since this year has been great for the strategy). I am still working on improving it. Generally Gtr has equal results but not always and I have used it to draft gtr tests, too but cgpt has the benefit of giving me the identity of the picks over time to see how robust. Ave gains, htd (price/(hgprc(1,252) plus lgprc(1,253) was the best htd so far), weighted choices (buying extra 1 and 2 stocks), 1 stock strategies, 5, 10, combining (overlap list then phl for more to choose 5 stocks every month). Worst single start results, average across starts. I am testing a 1 stock criteria as final sort that is very tuned to see how overtuned now since it gave 45% cagr. In general result suggests tripling SPY and QQQ with higher sharpe and lower beta and mdd and quicker recovery time is possible with only negative more -10% plus drops (but up periods more than making up for them).
No. of Recommendations: 3
Alternatively, how about the SMA rule for getting out, one of Mungofitch's bottom catchers for getting back in, and then ignore the SMA rule until back above the SMA curve?
---
Tried that, too.
non funziona
Has anyone looked at using the 99-day method? Use the momentum method as along as the last change in the 99-day channel was up. When the channel moves down, step aside. Mungofitch found that applying the 99-day rule to the S&P500 index not only reduced volatility but also increased the CAGR by a couple of percentage points.
DB2
No. of Recommendations: 4
overlap list then phl for more to choose 5 stocks every month
Now that is an interesting twist.
Looking at the overlap counts I have been concerned about the number of months where the number of overlaps is just 0 or 1. The count also swings wildly from month to month.
But maybe the times of low overlap are the times when timing has you out of the market, so it's a non-issue?
I ran one recent check where the overlap count was 4. The next check 2 weeks later the count was 2. I would be scared going from, say, 5 stocks at $10K each to 1 stock at $50K.
But maybe the (slight) higher returns of Overlap come from these times you put it all on red? Dunno.
This is effectively doubling up on the PHL-top5 stocks that are also in the RS-top5. You don't need to run a seperate portfolio for overlap, you could just do the overweighting in the PHL portfolio.
Hundreds of hours of conversations with it by now.
Be wary of the possibility of curve-fitting.
"Curve-fitting bias refers to the systemic error that occurs when a mathematical model or trading strategy is excessively tuned to historical data. Instead of capturing the true, underlying trend, the model memorizes specific past data points and noise, losing its ability to generalize or perform well on new, unseen data."
I don't have Norgate data nor have I used Chapgpt, but the biggest value I see here is the evidently confirmation that the GTR1 backtest is valid.
(thus I use 325 days for the SMA of SPY because 300-350 were all good and better than 200 or 253 days), to see if works out-of-sample
Better in what regard? The risk metrics or the CAGR? Better be risk metrics, because that's the benfit of timing.
(BTW, Sortino Ratio is a much better measure than Sharpe Ratio. Sharpe penalizes upside volatility as well as downside volatility.)
Being a simple man, for timing I just use the Yahoo historical price data in my own spreadsheet to get the timing signals and then apply them to the GTR1 daily prices.
We have 40.8 years of prices and 21 overlapped cycles of prices. I have only run the timing backtest on 2 periods, 1985-2026 and 2006-2026. There was no set of timing criteria that worked best for both periods. A better test would be to run the timing backtest on 2 distinct non-overlapping periods, 1985-2006 and 2006-2026.
But.....I don't think it makes any real-world difference.
I tested several lookback periods, various percentages below the SMA for the sell signal, waiting for 2 or 3 or 4 successive weeks of being below the SMA sell cutoff before taking the sell signal, with and without the FRED indexes used by GTT.
Nothing stood out. And none of them were outstanding. What they all did was improve the maximum drawdown, standard deviation, and Sortino Ratio.
Even then, they all failed to reduce the maximum drawdown to a tolerable level. Untimed MaxDD(12 month) was -55%. Timings had -35% to -39%.
That's because of the high volatility. The 3 worst months were -31%, -31% and, -29%. No timing scheme is going to get you out fast enough to avoid those. Of the 490 months, 48 of them had a loss of more than -10%. Timing cut that down to 34.
Lay out the IN/OUT signals by date for a few of the timing schemes. What you see is not major differences. What you see is a one or two month difference in the start of an IN or OUT period. Every OUT period of several months were the same in all the timings, the only difference was slight difference on the start of the period.
Short OUT periods don't do much for you. The value of timing is being OUT during most of the long bear markets.
And, no, checking the timing weekly instead of monthly did not improve things. The week-to-week volatility is also high.
Of the 2128 week, 39 had loss of more than -10%. Weekly timing just had you missing out on the extraordinary high return weeks.
No. of Recommendations: 2
Has anyone looked at using the 99-day method? Use the momentum method as along as the last change in the 99-day channel was up.
I haven't.
First, it's a lot of extra data to look at 99 daily prices to detect a 99 day high, and then counting those to see if one occurred in the last 99 days.
Then you've got the problem of aligning that with the once-a-month trade date.
Mungofitch found that ... increased the CAGR by a couple of percentage points.
We're already looking at beating the S&P500 by 15 points, with or without timing.
These screens have such high volatility that it's virtually impossible to reduce the volatility down to anywhere even near the untimed volatility of SPY.
With timing we are getting that down to only about TWICE that of the S&P500.
Unlikely that the 99-day method would do much better.
No. of Recommendations: 3
After making that observation that the main difference in the timing parameters is just a 1 month lead or lag in the change of the signal, I started to wonder just what the actual difference were.
And once I started I kept thinking up more ways to look at it. Kept me busy for a couple of hours of Excel work. ;-)
So here's the differences applied to one particular cycle of the P-HL52 screen, from 2/1/1985 to 12/1/2025.
The timings all used the weekly price of ^GSPC (S&P 500).
Each line is the month which has a different signal of the two timings, the (untimed) return of that month, and which timing scheme turned out better for that month.
Better means being OUT for a loss month or being IN for a gain month.
Whiplash is when an IN month is between 2 OUT months, or vice versa, so you are going IN or OUT for just one month.
Also is the number of months IN and OUT, and the number of sells. These numbers are almost identical for all the timing parameters I tested. Demonstrating that, yes, the main difference is the 1 month lead or lag of a signal.
There are 556 month in total.
# in # out # sells
1st --> 64 wk SMA for 1 wks -1% below GSPC 397 100 18
2nd --> 52 wk SMA for 1 wks -1% below GSPC 396 101 19
Date 1st 2nd ---whiplash--- Mo Rtn #1 best #2 best
8/7/1988 out IN WL2 -5.7% 1st
10/2/1988 out IN -0.7% 1st
8/7/1994 IN out 9.3% 1st
11/5/2000 IN out -29.3% 2nd
5/4/2003 out IN 11.8% 2nd
8/2/2009 out IN -5.2% 1st
11/6/2011 IN out WL1 2.1% 1st
1/1/2012 IN out 6.1% 1st
6/3/2012 out IN WL1 6.0% 2nd
1/3/2016 IN out -7.8% 2nd
11/4/2018 IN out WL2 1.7% 1st
3/1/2020 IN out -13.2% 2nd
3/6/2022 IN out WL2 9.6% 1st
2/5/2023 out IN -8.5% 1st
3/5/2023 out IN -2.8% 1st
# in # out # sells
1st --> 64 wk SMA for 1 wks -4% below GSPC 408 89 12
2nd --> 52 wk SMA for 1 wks -4% below GSPC 410 87 15
Date 1st 2nd ---whiplash--- Mo Rtn #1 best #2 best
8/7/1988 out IN -5.7% 1st
9/4/1988 out IN 2.2% 2nd
10/2/1988 out IN -0.7% 1st
9/6/1998 IN out WL2 9.8% 1st
11/5/2000 IN out -29.3% 2nd
4/7/2002 out IN WL2 -11.6% 1st
5/4/2003 out IN 11.8% 2nd
1/6/2008 IN out -13.6% 2nd
8/2/2009 out IN -5.2% 1st
7/4/2010 IN out WL2 13.6% 1st
8/7/2011 IN out -4.9% 2nd
1/1/2012 IN out 6.1% 1st
2/5/2023 out IN -8.5% 1st
3/5/2023 out IN -2.8% 1st
# in # out # sells
1st --> 64 wk SMA for 1 wks -4% below GSPC 408 89 12
2nd --> 58 wk SMA for 1 wks -4% below GSPC 410 87 14
Date 1st 2nd ---whiplash--- Mo Rtn #1 best #2 best
10/2/1988 out IN -0.7% 1st
9/6/1998 IN out WL2 9.8% 1st
8/2/2009 out IN -5.2% 1st
7/4/2010 IN out WL2 13.6% 1st
2/5/2023 out IN -8.5% 1st
3/5/2023 out IN -2.8% 1st
And finally,
# in # out # sells
1st --> 64 wk SMA for 1 wks -4% below GSPC 408 89 12
2nd --> 64 wk SMA for 1 wks -1% below GSPC 397 100 18
Date 1st 2nd ---whiplash--- Mo Rtn #1 best #2 best
4/3/1994 IN out -2.1% 2nd
5/1/1994 IN out -10.3% 2nd
7/3/1994 IN out WL2 -4.9% 2nd
12/4/1994 IN out 0.8% 1st
1/1/1995 IN out -2.5% 2nd
9/6/1998 IN out 9.8% 1st
10/4/1998 IN out 8.6% 1st
1/6/2008 IN out -13.6% 2nd
7/4/2010 IN out WL2 13.6% 1st
8/7/2011 IN out -4.9% 2nd
6/3/2012 IN out WL2 6.0% 1st
No. of Recommendations: 2
AVE 3.1 stocks
uses SPY>325SMA as timing rule
cagr 32.2 since nasdaq100 started
beta .88
sharpe .88
mdd -57.7
2 losing years:
-02.3% 2008
-22.5% 2022
I used Chatgpt to add back stocks from the 2 overlapped screens to test how to reach 5 stocks every month but don't know how to do that in gtr so the above is just overlaps with timing rule.
In chat gpt using hold-til-drop up to ran 15 of price/(high in last 1 year + low in last 1 year) screen (1 of the 2 in the overlap) was best overall way to reduce turnover but maintain stats on performance.
https://gtr1.net/2013/?~Nas100_Mom9_PHL252_Overlap...Using 200 days for the timing rule reduces cagr to 27.19, sharpe to .788, beta to .83, mdd -58.4%.
You can stick in what days you want into the screen.
No. of Recommendations: 3
musselmant asked how "to test how to reach 5 stocks every month" with 2 overlapped screens.
Try combining the 2 screens into one screen. Something like:
https://gtr1.net/2013/?~Mom9Top5_PHL252Top5rankSum...Define {Mom9Top5_PHL252Top5rankSum}
Create [Mom9m]: [Total Return Multiplier over 189 days]
Create [rankMom9]: [Rank by [Mom9m] (Desc) at step2]
Create [PHL252]: [[Closing g-price; quote_lag=3 days]/[[Lowest closing g-price over 252 days; lag=1 days] + [Highest closing g-price over 252 days; lag=1 days]]]
Create [rankPHL]: [Rank by [PHL252] (Desc) at step2]
Create [rankSum]: [[rankPHL] + [rankMom9]]
step0: [NASDAQ 100 Member; lag=1 days] == 1
step1: [Mkt Days Since Security Opened; lag=1 days] >= 252
step2: [rankPHL] <= 10
step3: [rankMom9] <= 10
step4: [rankSum] Bottom 5; Cash When Fewer
Author: lizgdal
Subject: Re: Overlaps GTR1 backtest is flawed.
Date: 06/16/26
"The screen in question {Nas100_Mom9_PHL252_OverlapTest} finds the overlap between the 2 screens {Mom9Top5} and {PHL252Top5}. Similar results can be obtained through a single screen that combines the criteria..."
https://www.shrewdm.com/MB?pid=787818796
No. of Recommendations: 3
In chat gpt using hold-til-drop up to ran 15 of price/(high in last 1 year + low in last 1 year) screen
Yeah, I discovered that, too.
Conceptually the HTD keeps stocks that are still good, but the dropoff isn't very sharp as you go a bit further down the list, so good but just not one of the top N. It mainly reduces the turnover with little impact on the CAGR.
Lower turnover equals less work and less trading friction.
Based on 10 HTD 12 I initially backested 5 HTD 6.
Then accidently ran 5 HTD12.
Surprise!! That was superior to 5 HTD 6. Tried other HTD numbers and 12 seemed good enough.
Wierd.
Using 200 days for the timing rule reduces cagr to 27.19
Oh no!! "Only" 27%.
LOL.
No. of Recommendations: 2
"to test how to reach 5 stocks every month" with 2 overlapped screens.
I'm not sure the sum of the rankings gets us what we want, which was to start off with the stocks that are in both screens, regardless of their ranking in their screen, then fill in with the stocks that are not common.
The overlap screen that musselmant posted does not always have 5 stocks, sometimes it is heavily weighted to just 1 or 2 stocks. Makes it impossible to compare that with your rankSum screen. We are already having a small number of stocks. In Jim's various sum-of-ranks screens he has 20 or 25 stocks, so much less risky, only 4%-5% each instead of 20%.
The ranking criteria in the two screens are different, so there is no reason to prefer the #2 ranked stock in one screen to the #4 ranked stock in the other.
The problem is how to decide to fill in after the common stocks.
First, obviously, take the PHL stocks that are also in RS. That gets you the common stocks regardless of their rankings in their own screen. The #1 in one screen may be the #5 in the other. If both screens say a stock is good, then it is probably very good.
You could fill in with the PHL stocks that are *not* in RS, or vice versa. But that feels unsatisfying, as it would fill in with the non-overlap stocks exclusively from one screen or the other.
It seems reasonable -- and easy -- to just pick the non-common stocks alternately from each screen (in rank order), until you've got 5 stocks total.
I have no idea how to backtest any of this in GTR1.
Frankly, I am almost inclined to run an overlap screen with 10 stocks selected from the top 10 of the PHL and RS screens, while keeping those screens at 5. Just because in total we will not be so concentrated in a small number of stocks. That will lessen the pain of the days when, for example, SNDK loses -14% and MU loses -10%. Because those two stocks are in all 3 screens.
No. of Recommendations: 4
A field can find the overlaps:
Overlap: if(rankPHL<=5,if(rankMom9<=5,0,100),100)
Symbol [Mom9m] [PHL252] [rankMom9] [rankPHL] [Overlap] [rankSum]
MU 2.53 0.71 1 2 0 3
WBD 2.09 0.72 3 1 0 4
AVGO 2.04 0.70 5 3 0 8
GOOGL 1.89 0.69 7 4 100 111
LRCX 2.05 0.67 4 7 100 111
GOOG 1.86 0.69 8 5 100 113
IDXX 1.72 0.68 10 6 100 116
AMD 2.18 0.60 2 18 100 120
AMAT 1.61 0.64 13 8 100 121
KLAC 1.67 0.63 12 9 100 121
EA 1.57 0.63 14 10 100 124
PLTR 1.98 0.60 6 19 100 125
INTC 1.71 0.60 11 20 100 131
ASML 1.51 0.60 15 17 100 132
SHOP 1.42 0.61 19 14 100 133
CEG 1.46 0.61 17 16 100 133
https://gtr1.net/2013/?~Mom9Top5_PHL252Top5overlap...
No. of Recommendations: 3
This version combines the overlap field with a minimum function:
Create [rankSum]: [[Overlap] + [Min([rankPHL], [rankMom9])]]
Ticker Symbol [rankMom9] [rankPHL] [Overlap] [rankSum]
MU 1 2 0 1
WBD 3 1 0 1
AVGO 5 3 0 3
AMD 2 18 100 102
GOOGL 7 4 100 104
LRCX 4 7 100 104
GOOG 8 5 100 105
IDXX 10 6 100 106
PLTR 6 19 100 106
AMAT 13 8 100 108
KLAC 12 9 100 109
APP 9 23 100 109
EA 14 10 100 110
INTC 11 20 100 111
MNST 21 12 100 112
https://gtr1.net/2013/?~Mom9Top5_PHL252Top5min:s19...
No. of Recommendations: 5
in cgpt, using low combined ranks of the 2 lists to buy 4 stocks, then next PHL low rank of stocks you didn't have in your 4, was the best way to reach 5 stocks. You can use googlesheets to do the 2 lists on all nasdaq100 stocks and calculate lowest combined rank to find the 4, then manually read down the PHL list to find your 5th stock.
No. of Recommendations: 3
in cgpt, using low combined ranks of the 2 lists to buy 4 stocks, then next PHL low rank of stocks you didn't have in your 4, was the best way to reach 5 stocks. You can use googlesheets to do the 2 lists on all nasdaq100 stocks and calculate lowest combined rank to find the 4, then manually read down the PHL list to find your 5th stock.
Doesn't this just say that PHL is better than RS?
This selects the PHL stocks that are also in RS, then add enough PHL stocks to get to 5.
No. of Recommendations: 2
"in cgpt, using low combined ranks of the 2 lists to buy 4 stocks, then next PHL low rank of stocks you didn't have in your 4, was the best way to reach 5 stocks. You can use googlesheets to do the 2 lists on all nasdaq100 stocks and calculate lowest combined rank to find the 4, then manually read down the PHL list to find your 5th stock."
Doesn't this just say that PHL is better than RS?
This selects the PHL stocks that are also in RS, then add enough PHL stocks to get to 5.
Waitaminute. This is just taking the top 5 PHL stocks, albeit in a different order. But it's the same 5 stocks.
Unless I'm missing something in your description.
No. of Recommendations: 4
I tested just PHL and no, the overlaps were better, the sum of ranks at some point but not others.
one version:
The best one-stock HTD variant we found was approximately:
CAGR: 45.4%
Sharpe: 1.03
Beta: ~1.0
MDD: -51.1%
Compare that with SPY:
Metric 1-Stock HTD SPY
CAGR 45.4% ~9–10% (2000+)
MDD -51.1% -50.8%
Recovery ~41 months ~75 months
10%+ drawdowns More frequent Fewer
Typical recovery Faster Slower
The maximum drawdown differs by only about 0.3 percentage points, essentially the same historical worst-case loss, while the strategy's CAGR is several times higher over the audited post-1999 period.
What impresses me even more is the recovery time. Even though the worst drawdown is nearly identical to SPY's, the strategy recovered in about 41 months versus roughly 75 months for SPY. For many investors, that difference matters as much as the depth of the decline.
You got up to 48% cagr if you accept a 61% drop. A tuned result.
But nice.
Strategy Post-1999 CAGR Relative rank
Sum of Ranks (9-month Momentum + PHL) 27.5–28.1% 🥇 Best
PHL Top 5 25–26% 🥈 Second
9-month Momentum Top 5 18–20% 🥉 Third
Overlap (Momentum Top 5 ∩ PHL Top 5, hold up to 5 names/cash otherwise) 17–18% Fourth
No. of Recommendations: 5
Has anyone looked at using the 99-day method?
---
I haven't. First, it's a lot of extra data to look at 99 daily prices to detect a 99 day high, and then counting those to see if one occurred in the last 99 days. Then you've got the problem of aligning that with the once-a-month trade date.
Understood that it is more work.
Mungofitch found that ... increased the CAGR by a couple of percentage points.
---
We're already looking at beating the S&P500 by 15 points, with or without timing.
Yes, but there are two separate facets. One is stock picking, the momentum part. Clearly these Naz screamers are going to outperform the S&P500 index. The other facet is timing, the topic of this thread. It is well recognized that it is hard to get a higher CAGR with any timing. Mungo's 99-day rule is the only method that comes to mind that improves return in addition to reducing volatility.
I could be wrong, but it seems that a method the improves CAGR with an index would be likely to improve the performance after being applied to a momentum screen.
DB2
No. of Recommendations: 4
but there are two separate facets. One is stock picking, the momentum part. Clearly these Naz screamers are going to outperform the S&P500 index.
Yes. Although I wouldn't say "clearly these stocks...." The only reason we are looking at these are BECAUSE they are screamers.
Do the same backtest using Sp500 (VFINX) instead of Nas100. Over the same time period it beat B&H by 3.4 points. (16.6% vs. 11.2%).
The other facet is timing, the topic of this thread. It is well recognized that it is hard to get a higher CAGR with any timing. Mungo's 99-day rule is the only method that comes to mind that improves return in addition to reducing volatility.
If a timing scheme improves the return, that is just a happy coincidence. The true purpose of timing is to avoid the brunt of a huge downturn in deep bear markets. Anything beyond that is just icing on the cake.
There are others. 43 week SMA using GTT is one. For these NAS screens almost all the timing variations I've tried improved the return -- albeit just a little -- while greatly reducing the drawdowns from sky-high to just massive-nosebleed high.
However, I will rummage around and look for the 99-day spreadsheet I made a while back and see to trying that.
GTT works on normal screens because it prevents you from selling when the general economy is doing good. I think it doesn't work here because these stocks are in a special sector and not part of the general economy.
I could be wrong, but it seems that a method the improves CAGR with an index would be likely to improve the performance after being applied to a momentum screen.
Indeed. I've thought about this a lot. It lends itself to special pleading, which I try to be wary of.
I think it works with high volatility investments when it cuts off the large downturns more than it cuts off the high positive excursions.
No. of Recommendations: 3
Here are the statistics for RS52 using the cycle which has the 25th percentile return.
For a few timing schemes. The no new 99-day high in the last 99 days is #4.
Timing is computed weekly, except daily for the 99-day, and id applied at the monthly trading day.
For 1985-2026, timing had little effect on the return. #4 was worst on risk metrics.
For 2006-2026, #4 had the slightly best return, slightly worse risk metrics except for Sortino Ratio.
For me, the only reason to use #4 was if I believed that the future was going to look more like 2006-2026 instead of 1985-2006.
#1 25th-1985 Untimed
2/1/1985 12/1/2025 9/1/2006 12/1/2025
CAGR 26.7% CAGR 22.2%
Stdev 37.0% Stdev 31.2%
MaxDD(12) -73% MaxDD(12) -52%
Sortino 1.30 Sortino 1.21
Grth $10K $155,966,040 Grth $10K $476,424
# Sells 0 # Sells 0
Max mo out 0 Max mo out 0
#2 25th-1985 64 wk SMA for 1 wks -4% below GSPC
2/1/1985 12/1/2025 9/1/2006 12/1/2025
CAGR 26.6% CAGR 22.7%
Stdev 33.6% Stdev 27.9%
MaxDD(12) -50% MaxDD(12) -43%
Sortino 1.48 Sortino 1.43
Grth $10K $151,100,781 Grth $10K $512,211
# Sells 12 # Sells 9
Max mo out 30 Max mo out 19
#3 25th-1985 64 wk SMA for 1 wks -1% below GSPC
2/1/1985 12/1/2025 9/1/2006 12/1/2025
CAGR 27.6% CAGR 22.3%
Stdev 33.1% Stdev 27.9%
MaxDD(12) -43% MaxDD(12) -43%
Sortino 1.61 Sortino 1.41
Grth $10K $209,475,557 Grth $10K $481,191
# Sells 16 # Sells 10
Max mo out 31 Max mo out 19
#4 25th-1985 No new 99day high GSPC
2/1/1985 12/1/2025 9/1/2006 12/1/2025
CAGR 26.1% CAGR 24.2%
Stdev 34.6% Stdev 29.0%
MaxDD(12) -54% MaxDD(12) -47%
Sortino 1.41 Sortino 1.45
Grth $10K $131,804,070 Grth $10K $652,635
# Sells 17 # Sells 7
Max mo out 28 Max mo out 15
