TSPCalc Math Tutorial

[mjedlin66]

Hey guys, my math tutorial has been downloaded over 350 times in the last two days, but I haven't received any feedback.

I added a CAGR tutorial to it yesterday.

I have never written a math tutorial before. This is stuff I take for granted, so I want to be sure I do a good job explaining it. Please provide detailed feedback.

1. Where is it confusing?
2. What parts of it need clarification?
3. Do you now understand the concepts of Compound Gains and CAGR, and what they mean to you?
4. Do you still have any questions?

I will add a standard deviation page soon.

Thank you,
-Matt

https://tspcalc.com/bin/tutorial.xlsx

[evilanne]

It looked very thorough. I only viewed it briefly because I trust that the numbers in the website are correct since I know it was looked at, tested & double checked last year and don't plan on doing any calculations myself

[TopNotch]

This question isn't directed to the tutorial, but the "Seasonal Strategies" window on tspcalc.com that says, for example, "There are 20205 strategies in the database". But some threads here discuss higher numbered strategies, like #20241, for example.

So I'm wondering why a strategy can be numbered higher than the total number of strategies in the database.

I appreciate all of the excellent work that went into TSPCalc.

[robika]

This question isn't directed to the tutorial, but the "Seasonal Strategies" window on tspcalc.com that says, for example, "There are 20205 strategies in the database". But some threads here discuss higher numbered strategies, like #20241, for example.

So I'm wondering why a strategy can be numbered higher than the total number of strategies in the database.

I appreciate all of the excellent work that went into TSPCalc.

Yeah, I don't get it either. I just created strategy 24059 but in the seasonal strategies filter it says "There are 20213 strategies in the database." Why is there a difference of 3,846?

[ewok55]

I have a question ... If TSPCalc uses all possible combinations of IFTs, why are there only 5 strategies (14038, 15430, 1873, 4496, 5715) that might have had a negative return in 2017? Am I missing something? I can surely think of more strategies than that having negative returns, so then TSPCalc must not be calculating all possible combinations?

To replicate: Searched 2017 to 2017, sorted by mean in ascending order. No other filters.

[fordest]

This question isn't directed to the tutorial, but the "Seasonal Strategies" window on tspcalc.com that says, for example, "There are 20205 strategies in the database". But some threads here discuss higher numbered strategies, like #20241, for example.

So I'm wondering why a strategy can be numbered higher than the total number of strategies in the database.

I appreciate all of the excellent work that went into TSPCalc.

Yeah, I don't get it either. I just created strategy 24059 but in the seasonal strategies filter it says "There are 20213 strategies in the database." Why is there a difference of 3,846?

I don't know the exact reason. But I know I have entered fund numbers that either don't load or load an alternate number. That would account for some of the missing numbers but not the why.

[fordest]

I have a question ... If TSPCalc uses all possible combinations of IFTs, why are there only 5 strategies (14038, 15430, 1873, 4496, 5715) that might have had a negative return in 2017? Am I missing something? I can surely think of more strategies than that having negative returns, so then TSPCalc must not be calculating all possible combinations?

To replicate: Searched 2017 to 2017, sorted by mean in ascending order. No other filters.

It does not contain every possible combination. It only has the ones that people have entered through searching and creating and modifying. When something new comes along it will get a new number and be saved in the database.

[evilanne]

I don't think all the lower numbers are used
Strategy ID# 68 is G Fund. Strategy ID#s 80-83 - F/S/C/I Funds

I think Matt's original strategy was #24 but I think when everyone started using it was in 3500 range for numbering purposes.

[ewok55]

I have a question ... If TSPCalc uses all possible combinations of IFTs, why are there only 5 strategies (14038, 15430, 1873, 4496, 5715) that might have had a negative return in 2017? Am I missing something? I can surely think of more strategies than that having negative returns, so then TSPCalc must not be calculating all possible combinations?

To replicate: Searched 2017 to 2017, sorted by mean in ascending order. No other filters.

It does not contain every possible combination. It only has the ones that people have entered through searching and creating and modifying. When something new comes along it will get a new number and be saved in the database.

If it only contains patterns made by other people, I find this potentially troubling, as it is no longer "a priori," but made in hindsight. That is, when I search for the best pattern from 2004 to 2004, I thought it was actually calculating it, not just showing a pattern that somebody else might have made (based on data AFTER 2004). Is there a way to have it create patterns randomly?

[mjedlin66]

1. None of this has anything to do with the subject matter. God bless the internet.

2. The database does not contain every possible strategy. There are over 100 million possible strategies, and my database has something like 20k. A strategy run for the first time by anyone will automatically be added. It is what it is, should be no cause for concern.

3. There are strategies that have been removed over time. Most of them were duplicates. When the website was in its infancy, the site was not very good at catching duplicates. At the moment, there are no known duplicates. However, those old strategy #s remain so that if users have saved the strategy# or the link somewhere, it will still work. The website cross-references all old duplicate numbers and automatically forwards the user to the one that was kept in the database.

[mindofmush]

All seasonal strategies use historical data or "hindsight" to find patterns that reoccur every year in the same season. The monthly seasonal strategies divide the data into 12 parts and then choose the fund that performs best (most of the time) for each of the months. The daily strategies divide the data into daily segments and the calculator allows you to create the best pattern you can within the limits of 2 IFTs (+G) per month. The TSPcalc Filter lets you test your creation to maximize your results or to choose someone else's work.

If you don't like the idea of using historical data patterns to guess the most probable choice, then the Seasonal strategy is not for you.

[ewok55]

All seasonal strategies use historical data or "hindsight" to find patterns that reoccur every year in the same season. The monthly seasonal strategies divide the data into 12 parts and then choose the fund that performs best (most of the time) for each of the months. The daily strategies divide the data into daily segments and the calculator allows you to create the best pattern you can within the limits of 2 IFTs (+G) per month. The TSPcalc Filter lets you test your creation to maximize your results or to choose someone else's work.

I'm all about using historical data. I'm even more interested in seeing how historical data, using an a priori method, could have predicted data that would be "future" to the data being used in the method (i.e., how a method developed using solely 2004 to 2004 data could predict 2005 data). One of the fundamental assumptions of correlation is that the dependent variables (i.e., results) are independent of the independent variables (i.e., models being formulated).

If you don't like the idea of using historical data patterns to guess the most probable choice, then the Seasonal strategy is not for you.

I love using historical data patterns, and see great promise in this. My post merely voiced my surprise to learn that some "historical" data was not independent of "future" data. That is, a pattern developed and discovered from doing a search (2014 to 2014) might actually include a user's knowledge of what happened in 2015.

Again, I see great promise in the work being done here, but am merely trying to add my own thoughts/concerns/hopes. My initial thoughts were that it is "too good to be true;" however, I am so hopeful that I am actually trying out one of the methods I found using tspcalc with my real funds. The statistician in me, however, makes me think that (because these patterns were not created using solely a priori data), returns will be lower than those expected based on the website.

In the meantime, here's hoping I'm wrong, but merely bringing up another consideration (and maybe a way to consider?) data presented in the current tspcalc website. For improvement, I could see it being helpful to enter every possible combination of IFTs (if such computing power were available), which would then allow for a priori use of models from former years to "predict" data available from years following. Again, I'm not discounting the use of historical data, merely voicing my concern that there are different ways to use historical data (i.e., with and/or without use of historical data following the predictive data).

[mjedlin66]

What is priori? Google doesn't seem to know.

[mmmmmbeer]

What is priori? Google doesn't seem to know.

A Priori Knowledge. A Priori is a philosophical term that is used in several different ways. The term is suppose to mean knowledge that is gained through deduction, and not through empirical evidence. ... One example is inborn knowledge. Another example often used is mathematics.

[evilanne]

https://www.merriam-webster.com/dictionary/a%20priori

Your tutorial was fine...appreciate you taking the time to explain everything in detail to educate everyone. If there are no follow up questions/comments it is all good If you shut down TSPcalc for a period of time, I'm sure you would get plenty of feedback because it would be dearly missed.