Optimization

[Mr Smith]

I'm new to system development and have two questions which I wanted to put to the community. I have a mean reversion strategy which makes use of a SMA.

1. I back tested the data for 25 years and saw that for security A, a 5d SMA worked best and for security B, a 20d SMA worked best. Should I use these results to trade or are they completely spurious as all I'm doing is fitting any strategy to the data. My view is that 25 years is a long time and the difference between 5 and 20d is not statistically significant over the time period.

2. For the mean reversion I use an oscillator. The oversold level is 30 while I found that better results were obtained when using a overbought level of 90. Have a fallen into another optimization trap? Or can the overbought and oversold levels on an oscillator not be symmetrical I.E. -100 and 100 or 30 and 70.

Any help would be much appreciated

 

[stepandfetchit]

IMHO: However, before putting too much confidence in my statement, know that I am no good in directional trading, the basis for your post! (others here likely have more relevant feedback)
1) Yes, you are curve fitting! However, that does NOT necessarily mean your thinking is flawed. Consider, using the "data" you extracted, then applying it in history to good periods, then bad periods, to observe how it would perform. -- You may need to tweak your criteria with some general market conditions.
2) I have no suggestions for #2.
good luck.

 

[lindq]

My view is that 25 years is a long time and the difference between 5 and 20d is not statistically significant over the time period.

The difference between a 5 or 20 day MA is the difference between a week and a month.

That's a very significant difference if you're trying to use an MA to generate trading signals.

 

[tommcginnis]

DO NOT be afraid to tailor the SMA (and/or, the choice of oscillator!) to the underlying.

THERE IS NO rule that says the look-back parameter needs to be the same for every underlying OR that it needs to be the same over history. (The latter being a formula for spectacular failure.)

USE WHAT WORKS, test it again. Doubt, review, and test some more. Cross-check!

NEVER be sleeping; the market won't.

 

[ironchef]

I'm new to system development and have two questions which I wanted to put to the community. I have a mean reversion strategy which makes use of a SMA.

1. I back tested the data for 25 years and saw that for security A, a 5d SMA worked best and for security B, a 20d SMA worked best. Should I use these results to trade or are they completely spurious as all I'm doing is fitting any strategy to the data. My view is that 25 years is a long time and the difference between 5 and 20d is not statistically significant over the time period.

2. For the mean reversion I use an oscillator. The oversold level is 30 while I found that better results were obtained when using a overbought level of 90. Have a fallen into another optimization trap? Or can the overbought and oversold levels on an oscillator not be symmetrical I.E. -100 and 100 or 30 and 70.

Any help would be much appreciated

IMHO: However, before putting too much confidence in my statement, know that I am no good in directional trading, the basis for your post! (others here likely have more relevant feedback)
1) Yes, you are curve fitting! However, that does NOT necessarily mean your thinking is flawed. Consider, using the "data" you extracted, then applying it in history to good periods, then bad periods, to observe how it would perform. -- You may need to tweak your criteria with some general market conditions.
2) I have no suggestions for #2.
good luck.

You, I and thousands of other traders tried using SMA, EMA, oscillators.... Why? They are so logical, simple, elegant and make sense. Just look at them superimposed on a price chart, it is easy to see why they should work well! Besides, hundreds of books were written on them, by trading gurus and experts, they must work!

I back tested 5, 10, 20, years data and found they worked well!

Encouraged, I tried demo accounts, well, they worked but now not as well. OK, I went live. Surprise, they didn't work. On cases that worked, I realized that buy and hold gave me better returns.

Back testing and then go live is like using the rearview mirror to drive your car.

So, what is a trader to do? If you back test and it works, try demo and go live like I did. If it works for you live, great, you found your method, keep going. If it does not work, OK you are not alone, go back to the drawing board and try something different. There is no one size fits all in trading.

Good luck.