Curve Fitting vs. Forward Looking

[bamafamily]

Hey all,
I am pretty new around here but have been really intrigued by the seasonality method of trading the TSP. I come from a background that has always dealt in utilizing TA Indicators to try and make my trades. That along with the fundamentals going on in the world. I am a member of a couple of other sites that use mostly momentum based algorithms to produces their trades. In deriving these algorithms, someone applies the formulas to a time in history that does not include the most recent 'x' years so as not to curve fit the entire data period. (I usually don't include the last 4-5 years when creating my algorithms) When I feel I have a good solution for the historical period, I will then apply those same TA variables to the 4-5 years that were not included. Sort of a forward looking test if you will to see how robust my algorithm is against data that did not help create it.

Fast forward to seasonals. Since we only have data that goes back to 2004, and can only see the top 100 daily strategies, I have attempted to record some of the higher ranked strategies from the 2004-2012 period (std dev<10%) and then record the returns of those same strategies from the 2013-2017 period.. in a perfect world, the top 5 strategies from 2004-2012 would also be the top 5 strategies from 2013-present....but we don't live in a perfect world...so...here is what I came up with...
Comments welcome..

forward Looking.jpg

Bama

[mmmmmbeer]

Wow.. thats really interesting. I can't tell you I totally understand the details of what you're talking about, but I think I'm picking up the gist.

Your chart seems pretty on though. Showing 7980 at a 28% mean just utilizing your algorithm. Of course 2017 isn't done yet, but with 18.48% thus far for 2017, is at 29.962% mean for 2013-2017. Thats pretty danged close to the actual numbers just utilizing a predictor.

Soooo... what is it showing using the real numbers for 7980 from 2013-2017 for the next 5 years mean in 2018-2023?

[evilanne]

So Bama, What are you going to do? try seasonal or not?

[bamafamily]

mmmmmbeer....I did not utilize any algorithm for this...I just took the strategies and saw what they produced from the years 2004-2012. I then took those same strategies and saw how they did between 2013-2017. (Instead of looking at the mean/std dev across 1 big time period (2004-2017)) This tells me if the strategies that were robust during 2004-2012 were also robust during 2013-2017....they did not do that bad...I am still analyzing....

Wow.. thats really interesting. I can't tell you I totally understand the details of what you're talking about, but I think I'm picking up the gist.

Your chart seems pretty on though. Showing 7980 at a 28% mean just utilizing your algorithm. Of course 2017 isn't done yet, but with 18.48% thus far for 2017, is at 29.962% mean for 2013-2017. Thats pretty danged close to the actual numbers just utilizing a predictor.

Soooo... what is it showing using the real numbers for 7980 from 2013-2017 for the next 5 years mean in 2018-2023?

[bamafamily]

EA.....
I am still analyzing the seasonal against the daily. If I do seasonal, it looks like I need am already where I need to be (F) for tmj100...
If I do the daily, it looks like most of the strategies people are looking at wont be changing until 9/18...If things look ok, I might dip my toe in the S fund at the beginning of the month to get in sync (currently in F), then hit 'I' on 9/18 and 'G' on 9/21 (since I will be out of IFT's and G is the only place I can go....)
but I will probably only go this way if if the 'S' price on 9/1 is close to the price you guys paid for it on 8/23...

What about you???

B

So Bama, What are you going to do? try seasonal or not?

[evilanne]

I'm not planning on going 100% into anything but I have been using the monthly & daily to allocate a certain percentage based on the seasonal strategies among other inputs. I'm more concerned with capital preservation than with trying to make the largest possible gains at this point. I could see using it in a separate IRA account at some point with ETFs at discount brokerage when I do a partial transfer out of TSP.

[bamafamily]

I am thinking about allocating a small part of my IRA(s) with the Fidelity/Vanguard equivalent ETF's as well....I have 2 other diversification strategies I use for my non-TSP accounts, but I have always been intrigued by seasonal tendencies.....

As for me...I am trying to gain the most I can before retirement with the least amount of fluctuation/std. deviations....

While strategy 7980 looks very good, the std deviation is a tad high for me.....
I have been looking at the Dailies from 04-06 since that is the last time we had a rise in interest rates...I would be happy with 15% and a Std Dev of 5%....

btw...what daily might you be following?

B

I'm not planning on going 100% into anything but I have been using the monthly & daily to allocate a certain percentage based on the seasonal strategies among other inputs. I'm more concerned with capital preservation than with trying to make the largest possible gains at this point. I could see using it in a separate IRA account at some point with ETFs at discount brokerage when I do a partial transfer out of TSP.

[mjedlin66]

This is excellent. Good work.

[Sad Al]

OMG I had to refresh my high school statistics vocabulary. It meant what I thought it meant, but I had to make sure. Thanks for all the work on the seasonal and daily strategies mjedlin66, bamafamily, and all others! I wish I had it back in the good old days.

[12squared]

As for me...I am trying to gain the most I can before retirement with the least amount of fluctuation/std. deviations....

While strategy 7980 looks very good, the std deviation is a tad high for me.....
I have been looking at the Dailies from 04-06 since that is the last time we had a rise in interest rates...I would be happy with 15% and a Std Dev of 5%....

Although standard deviation is a good way to characterize the range variation around the mean, the number of samples (e.g. years) which contribute to the StdDev should also be considered. The greater the sample size, the higher the confidence that can be placed in estimating the range of the data.

For a 15% return and a StdDev of 5 percentage points, we should be 90% confident that the margin of error is +/- 2.3 for data spanning 13 years (e.g 2004-2016), or a . However, the margin of error increases to +/- 4.7 when looking only at a 3 year window (e.g 2004-2006).
Assuming a mean return of 15%, we should expect the returns in 9 out of 10 years to range from 12.7% to 17.3%, based on 2004-2016 data. Using 2004-2006 data (or any other 3 year span) the expected rate of return would be 10.3% to 19.7%

https://www.mathsisfun.com/data/confide ... lator.html

[bamafamily]

Interesting...thanks for the insight....The only thing I would question is the variability of the confidence factor?? How do we know we are 90% confident....??? I suppose you could swag the confidence factor based on how many samples you have??? e.g. 100 samples and one is 95% confident...5 samples and one is only 60% confident.
In reality, i was using the small sample factor of 04-06 just to see how the TSP funds did across a rising interest rate timeframe....
thx again for the comments....
Bama

Although standard deviation is a good way to characterize the range variation around the mean, the number of samples (e.g. years) which contribute to the StdDev should also be considered. The greater the sample size, the higher the confidence that can be placed in estimating the range of the data.

For a 15% return and a StdDev of 5 percentage points, we should be 90% confident that the margin of error is +/- 2.3 for data spanning 13 years (e.g 2004-2016), or a . However, the margin of error increases to +/- 4.7 when looking only at a 3 year window (e.g 2004-2006).
Assuming a mean return of 15%, we should expect the returns in 9 out of 10 years to range from 12.7% to 17.3%, based on 2004-2016 data. Using 2004-2006 data (or any other 3 year span) the expected rate of return would be 10.3% to 19.7%

https://www.mathsisfun.com/data/confide ... lator.html

[bamafamily]

After today's drubbing, if tomorrow is down as well at 11am, I might pull some from F and throw it in S...Probably following one of the dailies from the first post of this thread.....

Bama

So Bama, What are you going to do? try seasonal or not?

[bamafamily]

I dipped my toe in both the S and I fund today....we shall see how it goes...

After today's drubbing, if tomorrow is down as well at 11am, I might pull some from F and throw it in S...Probably following one of the dailies from the first post of this thread.....

Bama

So Bama, What are you going to do? try seasonal or not?