Daily Strategy: High Mean, How High the StdDev?

[Rothwell]

There has been quite a bit of effort to find the best “high mean to low standard deviation” daily strategy with some surprising low StdDev to high Mean performers (here’s looking at you 22435).

What about looking in the opposite direction? How high of a standard deviation vs high mean would still be “acceptable”? At a quick look 17900 has a Mean of (2004 – 2017) 37.46 with a standard deviation of 17.95. That puts the standard deviation just below 50%.

Best year 2009: 88.46
Worst year 2014: 19.12

[TopComm]

This would depend on one's risk tolerance.

Divide the Mean by the Standard Deviation to get the Sigma (σ)
For 17900 it is about 2σ. This means that 1 out 21 years could result in a negative return.
For 22435 it is about 16σ. 7σ+ would be 1 out of 390 billion years could result in a negative return.

1σ = 1 out of 3 could be negative
2σ = 1 out of 21
3σ = 1 out of 370
4σ = 1 out of 15,787
5σ = 1 out of 1.7 Million
6σ = 1 out of 506 Million
7σ = 1 out of 390 Billion

For me, 4σ+ would give a reasonable assurance. Like you pointed out above, I also look at worst year. So a 4σ+ strategy with a higher worst year than my current strategy would warrant a new selection.

[ewok55]

Has anybody tried to see how one of these daily seasonal strategies would do at predicting an actual year? For example, a seasonal strategy based on 2004 to 2016 predicting 2017? I'm wary of using a strategy based on 2004 to 2017 "predicting" performance for a year that is already included in the formulation (e.g., 2008). I'm cautiously optimistic, but am just wondering if this was figured in with all the high CAGRs I'm seeing.

[Octjan2]

This would depend on one's risk tolerance.

Divide the Mean by the Standard Deviation to get the Sigma (σ)
For 17900 it is about 2σ. This means that 1 out 21 years could result in a negative return.
For 22435 it is about 16σ. 7σ+ would be 1 out of 390 billion years could result in a negative return.

1σ = 1 out of 3 could be negative
2σ = 1 out of 21
3σ = 1 out of 370
4σ = 1 out of 15,787
5σ = 1 out of 1.7 Million
6σ = 1 out of 506 Million
7σ = 1 out of 390 Billion

For me, 4σ+ would give a reasonable assurance. Like you pointed out above, I also look at worst year. So a 4σ+ strategy with a higher worst year than my current strategy would warrant a new selection.

So with these odds, If I am following 17842 I am 1,344 times more likely to win the powerball than having a negative yearly return? Powerball odds are 1 in 292 million.

[12squared]

Has anybody tried to see how one of these daily seasonal strategies would do at predicting an actual year? For example, a seasonal strategy based on 2004 to 2016 predicting 2017? I'm wary of using a strategy based on 2004 to 2017 "predicting" performance for a year that is already included in the formulation (e.g., 2008). I'm cautiously optimistic, but am just wondering if this was figured in with all the high CAGRs I'm seeing.

Try this:
Set the range to only four years, 2004-2007, which covers only the time period before the big drop. Set the mean to a min. of 30. Sort by Std Dev ascending. Choose any strategy with a SD < 10. Look at the performance from 2008-2017. Set the min. Mean to 29 and you'll get 100+ choices with SDs less than 7.
What do you observe?

[12squared]

There has been quite a bit of effort to find the best “high mean to low standard deviation” daily strategy with some surprising low StdDev to high Mean performers (here’s looking at you 22435).

What about looking in the opposite direction? How high of a standard deviation vs high mean would still be “acceptable”? At a quick look 17900 has a Mean of (2004 – 2017) 37.46 with a standard deviation of 17.95. That puts the standard deviation just below 50%.

Two things to consider:

If you try to optimize Sigma on a monthly basis (i.e. Mean / StdDev for each month) you will find it impossible that to achieve an annual Sigma >10. Conversely, most high (>10) Sigma strategies I've seen have relatively low Sigma for January (<0.4) compared with the highest (~0.99).

The Mean and StdDev we've been discussing for months is based on the Jan-Dec (aka annual performance. Given a 10+ Sigma strategy, if you calculate the 12 month return on any of the other 11 month intervals, Sigma decreases to single digits. This is because all of the high sigma strategies have been optimized for the Jan -Dec interval. The interval ending in June is typically the lowest. Many high annual Sigma strategies have had PIPs < 10% over the course of any year.

12 month PIP table 1-6-18b.png

Averaging sigma over all 12 intervals would provide a better measure of consistency throughout the year. Any strategy whose lowest 12 month Sigma is >2.0 has a low risk of negative returns.

[Rothwell]

12sqaured, I am currently following 19046. Why? Six Sigma, which I mentioned elsewhere. What I have is a colleague who has less than 10 years until retirement and would like to invest the first half of that time a bit more aggressively.

When you load the TSPcalc website, it loads highest overall mean without any filters, which as of 1/6/2018 is:

21980
Mean: 39.38
StdDev: 29.28
CAGR: 36.84
Sigma: 1.34σ

Not great at all and the data is skewed because of the (really) outlying years 2008 & 2009.
But when I look at:

17900
Mean: 37.46
StdDev: 17.95
CAGR: 36.41
Sigma: 2.08σ

What popped out at me was that the CAGR was effectively “identical” with a significantly lower StdDev.

[Aitrus]

Roth,

Keep in mind that it also loads 2018, which as of right now is a real drag on the stats because it counts 2018 as a full year in the Mean / StdDev / CAGR stats. Dial it back to 2017 to get the stats for full years only.

[Rothwell]

Aitrus, (big fan by the way, I've been in Jahbulon’s since 2015 up until this last November) if you click through the linked strategies you will see that I did only include complete years

[Aitrus]

Thanks, Roth. Much appreciated.

I was responding to your statement about the basic load of the TSPCalc. I hadn't clicked on your links. Sorry for the confusion.

[man00dude]

My strategy was all over the place for 2017. Ended up over 21%.

[Octjan2]

This would depend on one's risk tolerance.

Divide the Mean by the Standard Deviation to get the Sigma (σ)
For 17900 it is about 2σ. This means that 1 out 21 years could result in a negative return.
For 22435 it is about 16σ. 7σ+ would be 1 out of 390 billion years could result in a negative return.

1σ = 1 out of 3 could be negative
2σ = 1 out of 21
3σ = 1 out of 370
4σ = 1 out of 15,787
5σ = 1 out of 1.7 Million
6σ = 1 out of 506 Million
7σ = 1 out of 390 Billion

For me, 4σ+ would give a reasonable assurance. Like you pointed out above, I also look at worst year. So a 4σ+ strategy with a higher worst year than my current strategy would warrant a new selection.

So with these odds, If I am following 17842 I am 1,344 times more likely to win the powerball than having a negative yearly return? Powerball odds are 1 in 292 million.

Let the countdown begin! 17842 is currently -1.99 for the year. Either faulty math or one heck of an odd year. With 3 months left it may go down to the wire. Since I am very loosely following 17842 I am hoping it storms to the finish line and gets some positive gains.

[michigande]

Powerball must be looking like a sure thing at this point! I lost money following #27777 for a while before I flinched and pulled the plug. Best of luck to you, hopefully 17842 has a banner october through december and you make money there!

"If I am following 17842 I am 1,344 times more likely to win the powerball than having a negative yearly return? Powerball odds are 1 in 292 million."