Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
KJ Trading Systems Kevin Davey - Ask Me Anything (AMA)
I know of a similar story. I had friends working at an institution when a known hot-shot trader opened an account with them. He was known to have made a lot of money for some of his clients and there was a lot of competition for his services. In any case he opened an account with this institution to trade his personal account.
The guys at the trading desk all executed his orders, so had real time access to his trades. When they saw the trades come in, no one thought to trade alongside him as they regarded the trades too risky and stupid. Once his account was up close to 1,000% in less than a year (pyramiding into a single trend), everyone admired his guts, but everyone still considered his strategy too risky.
After that, he had a couple of losing trades and the account was only up about 500% (down approx 50% from its peak) when he closed it. Funnily enough all of the traders considered themselves right, i.e. the guy was just lucky, he was taking on too much risk, etc. So even though they had access to the data to analyse the trades and see what made him rich, none of them considered it worth their time. They all thought they were smarter...
Firstly, wanted to thank you for all the insightful information presented in your book. Its truly a great read.
I really liked the practical & simple step by step methodology for creating trading strategies compared to all the hi-tech overcomplicated quant stuff we come across these days
I wanted to ask you 3 things
1) Tradestation has 2 options of WFO analysis
a) Single WFA - &
b) Cluster WFA - this uses anywhere between 5 - 20(In-Sample) runs of data and applies it to anywhere between
10-30%(Out-of Sample) Data.
Does a strategy have to PASS all the combinations of Cluster analysis or just the SINGLE WFA analysis to be considered to be taken to the Next step of Monte-Carlo analysis / incubation?
2) Tradestation has added a monte-carlo feature to their WFO, is this the same as the one you recommended in your book ?
3) If the only optimisable variables in a strategy are the exits, and if the exits itself are not fixed dollar value
( except for a fixed stop loss to minimise unforseen losses) but variable exits such as trailing ATR exits, Parabolic Exits, or exit based of a bollinger band then would one need to run the to optimise these exits in a WFO ? Wouldn't that be curve fitting ?
Thanks for the kind words, and for the questions. Here are some answers:
1. I do not use TS Walkforward tool. I use an inexpensive 3rd party piece of software that does the analysis differently
2. I use my own Monte Carlo simulator, written in Excel
3. Any time you are doing any sort of optimization you are curve fitting. I do not see your particular setup of variables being any better or worse than any other set (say with just entry parameters you optimize). The key is really to keep the # of variables low.
I use WFO more like a robustness test. If the strategy will work similar with different settings it would be a sign of robustness. However I also find that if leaving say 1 year of OOS to test the WFO strategy on 50 % of them do not improve from original and can be even worse wile the other 50 % benefited from it. I usually pick my strategies to trade from a batch that have performed recent months and find that 8 times of 10 they will perform in the near future as well so one way to do WFO could be to test it against last 1-6 months data to see if it's performance is such that you want to trade it.
I've read your book and it's a great piece of work. However, I am having a hard time understanding why using Walk Forward is so important. Usually what I do is optimize on my in-sample and use Monte Carlo Annual Return and Draw Down (95% confidence) to see how well my strategy works.
I then test my system on out of sample data and again use Monte Carlo so see whether I am happy with the results. Shouldn't this provide enough proof of robustness before taking it to the incubation phase?
Regarding walkforward, if you use a different process and verify it works well with real money, I say stick to it. Next month, I am hanging out with 3 great trader friends of mine in New York, and I'm the only one who uses walkforward. So, you definitely don't need it to succeed, but it is what I use.
That being said, here is why I like walkforward compared to traditional out of sample testing. For me, it is about maximizing the amount of out of sample results:
Traditional out of sample 80/20, Test on 10 years of data >>> optimize first 8 years, last 2 years is out of sample
Walkforward, Test on same 10 years of data, with 2 year IN period >>> 8 years out of sample performance.
So, over the same data, I get 4x the amount of out of sample data with walkforward.
Thanks for the reply. I got mixed feelings about WFO. I want it to work but at the same time, I'm worried it creates another layer of curve fitting possibilities and potential mistakes.
Understood. If you do walkforward wrong, it can easily be curvefit. Many common software packages allow you to curvefit walkforward, with certain settings/features...
Trading: Primarily Energy but also a little Equities, Fixed Income, Metals and Crypto.
Frequency: Many times daily
Duration: Never
Posts: 5,049 since Dec 2013
Thanks Given: 4,386
Thanks Received: 10,206
Walkforward (the way Kevin teaches it) is the bane of my existance, but also probably a life saver. I have so many things that look good at first but never can get past walkforward. That tells me they wouldn't work in real life either.
Do you have any examples (settings/features not packages) where this is the case. If done correctly I can't see how your walkforwards could be increasing curve fitting.
