Backtesting

[userque]

Continued from:

How do you use TSP center to your advantage?
http://tspcenter.com/forums/viewtopic.php?f=14&t=17181

Few of us here are actuaries or some type of statistical wizard that you may be--or that we need to be one in order to pick a TSP destiny successfully---but to my untrained eye, you make it seem that the TSPCALC dataset and the resultant analyses results derived from it are somehow flawed--enough to be discredited from our use. I hope not, because my recent results since I started using it have been in line with my expectations.

Here's my question: do you or do you not agree that buying and holding a *fixed* (i.e. 100% in C) allocation over a 10 year term *has* returned markedly less than SPECIFIC TSPCalc-driven seasonal strategies for that same period (data optimization vs. back-testing methodologies be darned)?

If the answer is "yeah buddy", then that was the ONLY point I was trying to make.

Was working on a reply to you...it was pretty long (of course). Lost it somehow when I opened a second tspcenter window to read another post. But fordest is correct, I didn't realize we were hijacking this thread. I'll respond again to a new thread later today or late tonight.

I actually found that reply I was working on ... had too many tabs open, and it wasn't near where I thought it should be.

Again, I just wanted to clarify, if not for you, then for others that may get the wrong understanding from your characterizations. One doesn't have to be an actuary or wizard to have certain understandings. I'm far from both.

I never ever said TSPCalc or it's data was flawed. I've posted that I like the site. I've posted that I would use the site if I weren't working on something different. I don't know how to be more clear about how I feel about the site, but I had to (for now) address your misunderstanding about that as well. (Research silent acquiescence.)

Regarding your question:

March 2009 to March 2019 returned a total of 316% You can let me know what the best seasonal did during that 10 year time frame. Or I guess you could just optimize a new strategy over those years and beat any other strategy in the trading universe.

(Points: Either of us can cherry pick any time period we want. Additionally, you can ex post facto produce a strategy 'perfectly' adapted to any past time period. This exercise is not meaningful in a real life trading scenario.

You brought up your strategy. You said that it is recently performing as expected. Every strategy out there goes through periods where they are totally in tune with the markets. That's not hard to do. [You know the cliché... a broke clock will be correct twice a day ...] Consistency is.)

What number is that strategy? How did that same exact strategy do last year?

I'm only asking because you brought it up. I'm not really that concerned--. I believe everyone should do whatever they do...especially when it comes to their own money.

Whether you believe me or not is not my issue. I respect free will and am not here to win friends and influence people (look that latter phrase up ); I just didn't want those that may read your posts one day in the future to walk away with a misunderstanding ... at least not before hearing a counter-argument.

I get it. Some take offense to being 'challenged,' even if the challenge is justified. Unfortunately, I can't help that. But I do try to be as gentle as possible. I hope you do the same when you eventually correct me on some point I make in the future.

[12squared]

userque,

If you knew a strategy's creation date, would its performance from that point forward be valid "backtesting"?

[userque]

userque,

If you knew a strategy's creation date, would its performance from that point forward be valid "backtesting"?

Not necessarily.

Suppose 100 strategies exist, all created in the past. Further suppose that you know the creation dates of each--thus, satisfying your above criterion.

You examine the performance of all of them over the past year. And you pick the best. ... This would not be proper backtesting.

In this case, you made a decision with future data (the year you examined). That future data was used to decide which strategy to pick and validate for that same year. (You must simulate being in the past ... before that year 'actually' happened. Then pick a strategy--knowing only what you knew then.

In real life, we can't pick a strategy based on how it will do in the next upcoming year. So this shouldn't be done with proper backtesting.)

Proper backtesting will simulate 'going back in time and doing things with no knowledge of the future.'

The above was my example. If you offer a more detailed example of a method you believe is proper backtesting, feedback from that example may be clearer and more helpful to you.

Backtesting and Forward Testing: The Importance of Correlation
https://www.investopedia.com/articles/t ... lation.asp

[SnareMV17]

These are strategies based off the daily returns of the TSP. It is all hindsight, yes, so technical backtesting has not occurred in the traditional sense. However, the point of these strategies is to prescribe a trading recipe for an every day of the year. Whether one selects a strategy this year and follows for X years or 5 years ago and follows for X years, the results would be identical for every year going forward ASSUMING you follow the strategy. These aren’t strategies that require any real subjective analysis that should in any way alter ones thinking or placement of funds. Therefor backtesting isn’t a necessary step. However, if your plan is to Kissel follow these strategies and allow your own emotional “analysis” to inform you when you feel you are smarter than the strategy, that bias should be back tested.