Tuesday, 24 December 2013

Walk Forward Optimization

In his book, THE EVALUATION AND OPTIMIZATION OF TRADING STRATEGIES (2nd Edition), Bob Pardo wrote: "I have always viewed Walk-Forward Analysis as an idiot-proof wayof optimizing a trading strategy."

With such a tool available, I'd be an idiot not to use it.

Walk Forward optimization has been available for a while in most of the platforms I use: NinjaTrader, MultiCharts and TradeStation. All allow me to optimize a strategy and then show the strategy unseen data and see how it would have performed. All of the above 3 platforms performs WFO adequately. However, the one that currently provides the most statistical information about a WFO is TradeStation. The WFO in TradeStation is The Grail which TS purchased some years ago now and integrated into TradeStation. The  developer of The Grail working for TS was part of the deal.

A key issue in WFO is the data. Probably the best quality historical data currently available to retail traders at a reasonable price is the IQ Feed and its twin, Kinetick if you are a NinjaTrader user.

As long as I can get the data issue resolved, I can use one platform to do my research an another to do the trading. Why would I want to do this? For a number of reasons including:
1. Preference
2. Running multiple algos on the same underlying and needing different broker accounts for each algo of the same symbol
3. Spreading risk between brokers (This is an issue for me since MF Global and PFG Best - I dodged both those bullets but am nevertheless gun shy).

Optimization is meaningless unless the WFO test is passed.

A good walked forward algo should have stats such as the ones below from one of the longer term flobots I run currently run on the ES:

Again, passing all these tests is no guarantee of future profitability bt it is the best way I know of getting a great flobot. The repeated profitability and good stats in the Out Of Sample data as the WFO rolls forward gives me the confidence of robustness.

Lastly, Happy Holidays to everyone. Snow has finally arrived here in the Haute Savoie. The house is full of family although not everyone is here this year - 4 are in Oz - I am jealous that they can go to the beach.


  1. Hi EL, I'm almost done catching up your archived posts. I've 2 more questions if you don't mind.

    1. In April 2012, a reader asked:"Do you think there's any edge to be gained from using bid/ask data aka footprint chart? do you use it or not really? You answer was "not really." I was having the same question as you dropped CVD a long time ago. Do you still think it's lost its edge in market?

    2. How do you determine your Fav Fibs? Is it based on today's open, yesterday's close or some other significant level?

    Thank you!

    1. Sarah, glad to hear the blog's useful.
      1. I think that the proliferation of HFT has clouded CVD, at least for me. I get tons more info from what price is doing and from MP.
      2. I take any significant swing hi and swing lo points and project a FavFib. Those that are taken out show continuation.

  2. and by now Pardo must be the richest man in the world!

    1. I did a little testing for Bob in about 1984 and he was a good guy. He's just sharing and I think that his book is very worthwhile if you want to learn bout WFO.