KJ Trading Systems Kevin Davey - Ask Me Anything (AMA)

November 14th, 2018, 08:13 AM

Hi Kevin,

If you were developing on ES, would you consider the in-sample dates as in-sample, or OOS when you came to test that strategy on other US indices?

As always, thanks for your time.

kevinkdog · November 14th, 2018, 08:34 AM

sixtyseven

Hi Kevin,

If you were developing on ES, would you consider the in-sample dates as in-sample, or OOS when you came to test that strategy on other US indices?

As always, thanks for your time.

Since I do walkforward, parameters usually change enough that I can consider the test valid for each.

So if I run an ES strategy with walkforward, and then an RTY strategy with walkforward, I'd consider them 2 separate tests.

Without walkforward, that second market test is not really OOS, but it is not really IS either - it is somewhere in between...

November 19th, 2018, 01:46 PM

Hey Kevin,

I was hoping you might be able to shed some light on the correct way to initially test a strategy in your view.

Would it make sense to take a 4 year historical period as in sample data, optimize to seek out the best parameters during this period, without over-optimizing of course, and then move forward to Walk Forward Optimization?

How do you ensure you aren't over optimizing, is it more of a feel thing rather than exact science?

How would you go about the Walk Foward Optimization in terms of size of data set it's being tested on?

Any fedback highly appreciated as always!

Thanks.

kevinkdog · November 19th, 2018, 02:09 PM

WhatIfGod

Hey Kevin,

I was hoping you might be able to shed some light on the correct way to initially test a strategy in your view.

Would it make sense to take a 4 year historical period as in sample data, optimize to seek out the best parameters during this period, without over-optimizing of course, and then move forward to Walk Forward Optimization?

How do you ensure you aren't over optimizing, is it more of a feel thing rather than exact science?

How would you go about the Walk Foward Optimization in terms of size of data set it's being tested on?

Any fedback highly appreciated as always!

Thanks.

Thanks for the questions. These are complicated questions, way too much to put in a thread post.

A few years ago, I did an Elite member webinar where I discussed a lot of this.

Also, my first book "Building Winning Algorithmic Trading Systems" also covers your questions in much more detail. You probably can find a copy at a library, or maybe online with a free trial at overdrive.com

Sorry I can't give short and sweet answers here. You have asked some great questions!

kevinkdog · November 20th, 2018, 01:19 PM

You may have heard about a big option seller going belly up last week because of the rise in Natural Gas.

Regardless of your style of trading, there are lessons for all of us in that disaster.

I wrote a short article about it for my friend and trader Adam Grimes: https://adamhgrimes.com/guest-post-ohhh-is-that-a-shiny-penny-on-the-ground/

Please feel free to leave comments about it!

November 24th, 2018, 07:27 PM

WhatIfGod

Hey Kevin,

I was hoping you might be able to shed some light on the correct way to initially test a strategy in your view.

Would it make sense to take a 4 year historical period as in sample data, optimize to seek out the best parameters during this period, without over-optimizing of course, and then move forward to Walk Forward Optimization?

How do you ensure you aren't over optimizing, is it more of a feel thing rather than exact science?

How would you go about the Walk Foward Optimization in terms of size of data set it's being tested on?

Any fedback highly appreciated as always!

Thanks.

This paper on System Parameter Permutation might give some insight in regards to Optimization .

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2423187

November 25th, 2018, 10:00 AM

kevinkdog

Thanks for the question. Not sure if my algo trading day is typical of what others do, but here it is:

1. Make sure all strategies are running, and are in sync - live accounts versus strategy positions. I usually check this multiple time a day, and evening.

If you don't mind me asking, is this precautionary or did you have to deal with strategies not running or doing crazy stuff in the past?

kevinkdog · November 25th, 2018, 12:22 PM

skoa

If you don't mind me asking, is this precautionary or did you have to deal with strategies not running or doing crazy stuff in the past?

Thanks for the question.

Strategies can and do go out of sync, so you have to be prepared to fix and adjust. Sometimes it is a code issue, sometimes a data issue, sometimes exchange issue, sometimes internet issue, etc. The list is long. One thing algo trading is NOT: it is NOT unattended trading. Even hedge funds, etc have people monitoring positions.

December 2nd, 2018, 02:51 PM

Hi Kevin,

I had a quick question from your book; for the Euro strategy example, you had allowed yourself one month to develop a strategy based on the defined goals.

I was wondering how much time it took you to come to the final strategy and if you went through other ideas before the final published version?

Cheers!
Michael

kevinkdog · December 2nd, 2018, 06:48 PM

amoeba

Hi Kevin,

I had a quick question from your book; for the Euro strategy example, you had allowed yourself one month to develop a strategy based on the defined goals.

I was wondering how much time it took you to come to the final strategy and if you went through other ideas before the final published version?

Cheers!
Michael

Thanks for the question, and thanks for reading the book. I developed that strategy more than 5 years ago, so unfortunately I do not have exact numbers for that. I am pretty sure I went through multiple ideas and variations during preliminary testing, but I don't recall what or how much. Trying multiple ideas etc. can be dangerous if done incorrectly, as I describe in the book.

Thanks!

KJ Trading Systems Kevin Davey - Ask Me Anything (AMA)

Discussion in Trading Reviews and Vendors

KJ Trading Systems Kevin Davey - Ask Me Anything (AMA)