50% returns selling options with max 50% drawdown per trade!

[nooby_mcnoob]

Found this code floating around, adapted it for my own data and fucking thing works?
Basically, with DTE=30 and choosing ABS(S-K) carefully, you get decent returns. 50% with my data.

So, what am I missing besides the obvious that I have been up since 3am and shouldn't be thinking?

Thank you sirs/adams.

 

[gaussian]

So, what am I missing besides the obvious that I have been up since 3am and shouldn't be thinking?

You're curve fitting.

By choosing your parameters in this way you are optimizing on historical data, rather than out of sample runs.

Split your sample set into a training set and n-1 sequential test sets. I guarantee you will not find the same results.

 

[nooby_mcnoob]

You're curve fitting.

By choosing your parameters in this way you are optimizing on historical data, rather than out of sample runs.

Split your sample set into a training set and n-1 sequential test sets. I guarantee you will not find the same results.

I did split it. But I only have 6 months of options data so I guess I'll have to wait another 6 months.

But yes you're of course right.

 

[gaussian]

I did split it. But I only have 6 months of options data so I guess I'll have to wait another 6 months.

But yes you're of course right.

You can buy historical data from a number of cheap sources. I think historical option data has a black friday sale.

It comes down to not going back to changing parameters. An alternative for small data is to use block-based resampling to generate more data based on some underlying distribution. You can split the set into n/2 training sets and n/2 test sets as well (sequential) and see how it runs. You can also attempt to walk-forward test it to remove any lookback bias.

 

[nooby_mcnoob]

You can buy historical data from a number of cheap sources. I think historical option data has a black friday sale.

It comes down to not going back to changing parameters. An alternative for small data is to use block-based resampling to generate more data based on some underlying distribution. You can split the set into n/2 training sets and n/2 test sets as well (sequential) and see how it runs. You can also attempt to walk-forward test it to remove any lookback bias.

Yeah that's annoying I'll just assume it doesn't work since if it did, it would be stupid.

 

[nooby_mcnoob]

You can buy historical data from a number of cheap sources. I think historical option data has a black friday sale.

It comes down to not going back to changing parameters. An alternative for small data is to use block-based resampling to generate more data based on some underlying distribution. You can split the set into n/2 training sets and n/2 test sets as well (sequential) and see how it runs. You can also attempt to walk-forward test it to remove any lookback bias.

Actually, is this over fitting thesis true? This came up while I was on a walk with the dog, so it is possible the dog was channeling her thoughts through me.

Say you are selling options 30 days out on SPY on either side of the current price. So you'll sell puts and calls that are ITM by more than (say) 1/2 ATR.

There are only a few possibilities:

1. Price doesn't move at all (can't happen on something like SPY, but OK at least you still get the premium)
2. Price moves up and keeps going
3. Price moves down and keeps going
4. Price moves up/down, reverts

In any of these cases except the first, you will have a drawdown of one side and a profit on the other side. If you are using 1/2 ATR as your base, then it is entirely possible that you will get a 100% movement (i.e., another 1/2 ATR) on one side.

So logically, it makes sense.

I reserve the right to 1) blame the dog, 2) blame my lack of sleep.

 

[taowave]

So you are selling the "guts" of a box??

 

[dozu888]

eat like bird poop like elephant...

every few years there will be some hedge funds running some strategy like this and put up a string of all-win months and investors who don't know any better rush in, only to be wiped out by 1 black swan..

if it's just a straight sell then there is no edge, correct?

 

[taowave]

Looks like a straight strangle sell to me,or should I say the risk equivalent.

Nothing new here

eat like bird op like elephant...

every few years there will be some hedge funds running some strategy like this and put up a string of all-win months and investors who don't know any better rush in, only to be wiped out by 1 black swan..

if it's just a straight sell then there is no edge, correct?

 

[nooby_mcnoob]

Looks like a straight strangle sell to me,or should I say the risk equivalent.

Nothing new here

Yep that's what I realized.