12.2) Using 'Walk Forward Optimization' to Improve Trading Results | Walk Forward Analysis

This video shows you what Walk Forward Analysis is. Sometimes called Walk Forward Optimization, it also shows how it can solve a number of issues encountered by algorithmic traders, and how the technique can be used to improve the effectiveness of your trading system.
Brought to you by Darwinex: UK FCA Regulated Broker, Asset Manager & Trader Exchange where Traders can legally attract Investor Capital and charge Performance Fees:
www.darwinex.com/?...
OTHER PARTS TO THIS EPISODE:
Easy access to all parts of this episode:
Part 1: • 12.1) What is 'Walk Fo...
Part 2: • 12.2) Using 'Walk Forw... (This part)
VIEW THE WHOLE 'BACKTESTING & OPTIMIZATION' SERIES IN ONE PLACE:
• Algorithmic Backtestin...
-----------------------
IMPORTANT REQUEST: Please please please.. if you find this content useful, please do consider liking and sharing it on TH-cam, Twitter, Facebook, LinkedIn, and whatever other social networks you have circles in.
Darwinex relies almost exclusively on organic growth, primarily through recommendation via informative content.
TH-cam’s algorithms measure the quality of Darwinex content on the basis of:
- Reach
- Engagement
- and several other related variables
With seemingly small actions such as:
- Clicking the Like button
- Clicking the Subscribe button
- Clicking the Share button (on TH-cam) and distributing our content
- etc
… YOU inform TH-cam’s algorithms of your sentiment towards Darwinex, thereby directly helping Darwinex MASSIVELY in achieving organic growth.
Thank you very much for your kind consideration!
-----------------------
Risk disclosure:
www.darwinex.com/legal/risk-d...
** Fancy joining a vibrant community of algorithmic traders, quants and data scientists focused on financial hacking? Join the Darwinex Collective Slack Workspace:
join.slack.com/t/darwinex-col...
#WalkForwardOptimization, #WalkForwardAnalysis, #Backtesting, #AlgoTrading, #MultiStageOptimization, #Pardo, #Kaufman

[@julesbarnabas3542]

All this channel is pure gold. Thank you.

[@eliremy6979]

Walk forward analysis 💎 big eye opener for me Thank you 🙏🏾

[@objflex9075]

Just finished my first optimization using an algorithm I designed. Going to do my first walk-forward validation. It was very confusing at first but you make it easy to understand after spending some time watching your videos on walk-forward. Unfortunately, I had to learn the hard way at first because I took the parameters from the first optimization and used those for live trading. Lesson learned. Thanks again.

[@romassdzindalieta2647]

Great content Darwinex team!

[@martyntinsley]

Thanks for the feedback Romass. Appreciated :)

[@harryokonkwo3419]

Beautiful explanation. Love this

[@randomdude79404]

Great content Darwinex , would it possible if you could show how this can be done in MT4 if you don't have a video on that already ?

[@nigelmdletsshe8397]

Easy to understand explanations, thank you so much

[@dpc107]

very well explained thanks

[@GanovAlex]

thanks, good video and WFA explanation

[@seanjohnson7145]

Love your videos!! thanks so much :D

[@Darwinexchange]

Thanks Sean. You are welcome!

[@martyntinsley]

Hi Sean - Always great to get feedback like this. Thank you.

[@leonjbr]

Clear and useful.

[@martyntinsley]

As always, thanks Leon

[@kersyduru119]

Thanks a lot for this. I am always thrilled by your lessons. I am a newbie (learning forex but have experince with programming). I was wondering if you could guide me on how to start off and get to the point of developing my own EA. I will appreciate it a lot.

[@FranciscusRoorda]

Thanks, it is explained clearly but still I am puzzled. What is the relationship between Optimalisation 1 and 2 ? (etc.). If Optimalisation2 starts anew and results in different prameters, then what is the use op Optimalisaton 1 etc.. ? Then you could do only the last pre-live optimalisation. So, it seems obvious that somehow the results of stage 1 are used in stage 2 etc - but how?

[@urib24]

Great videos!
A question: what if my strategy is made of 20 different parameters that can and should be optimized (some of the parameters act as a couple, meaning they are related (for example - one parameter sets the high limit and its couple sets the low limit), what is the best practice you suggest in order to obtain overall optimization? And could that be achieved by using the WFA optimization method?

[@OneCreator87]

I am having so much difficulty with confidence in my optimization/selection process, I hope this helps.

[@gadeichhorn]

Interesting, how would you merge results from the different optimization phases to make one set of parameters? thanks for putting this together!

[@martyntinsley]

Hi Gadi. You wouldn't. That's the whole point of WFA really. You would only use parameters from the pre-live optimization for live trading. This means that these parameters are in tune with current market conditions. The purpose of the previous stages is to prove that the model works. Each stage produces parameters in tune with the market conditions 'at that time' which then get validated in the walk forward 'at that time'. So the Pre-live opt is merely an extension of that process to produce the best parameters to use now in your live account. Obviously, you would only actually trade using these if the cumulative walk forward results were to your satisfaction. Hope that clarifies.

[@gadeichhorn]

@@martyntinsley thanks! yes, that makes sense. this is a process you will repeat periodically to improve your live setup over time after going live?

[@martyntinsley]

@@gadeichhorn You got it!

[@qamer88]

@@martyntinsley So if WF does not validate optimization results in any phase it mean either strategy does not have edge, or indicator that was used is not proper right?

[@kewtheii6764]

@@martyntinsley exactly the answer to the question i had, amazing! Thank you!

[@nehemaialord2653]

But what happens if the first stage of optimisation and walk forward passes? Why do you have to re optimise? Is the reason for repoptimsing to find the common parameters that were successful through out the whole process?

[@Zwinzor]

Great video but I still have some questions about it. Like what settings do you need to use in the end for your pre-live OPT?
For example I test 2 indicators:
OPT1 gives me best settings 20 for the first indi and 10 for the second indi -> WF1 gives me an ROI of 2000;
OPT2 gives me best settings 15 for the first indi and 15 for the second indi - > WF2 gives me an ROI of 3000;
OPT3 gives me best settings 15 for the first indi and 20 for the second indi - > WF3 gives me an ROI of 2500;
OPT4 gives me best settings 10 for the first indi and 20 for the second indi - > WF4 gives me an ROI of 1500;
OPT5 gives me best settings 20 for the first indi and 10 for the second indi - > WF5 gives me an ROI of 2500;
Pre-live OPT gives me best settings 15 for the first indi and 20 for the second indi - > WF6 settings to use as robust system?
Thx for clarifying this for me when (I use MT4).

[@antoniomachado8782]

If I understood the video correcty, as well as some of the good comments above, where Martin gave good feedback, I would say to you:
- the best settings for you to go live are the settings of your Pre-live OPT.
- Not necessarily because those settings are, some how, more 'similar' to the settings used in some of the previous validated Optimizatins (although, if that happens, I personaly think it adds some more confidence),
but because they are the most recent settings - therefore, representing the current conditions of the market;
- as to the role or the usefulness of the previous 5 Opts and corresponding WFs, it is huge, their role was to validate (!!) (or not) your overall system (concept) of using the two indicators you decided to put together, with the rules you set.
- If the different WF phases curves of results were good, then, most likely than not, your system has an edge. But, if the WF phases didn't overall produce a relatively good enough curve and results, then, the
concept for that strategy was probably not good and there wasn't a real edge behind it.
Hope this helps and my interpretation is correct.

[@tomtownsend4210]

This is helpful on a number of fronts. I still don't understand the "how-to" of this and hopefully someone can help. After gathering multiple rounds of optimized parameters in steps 1-4 and their accompanying walk forward test results, what goes into the "pre-live optimization" section that informs this test based on the prior 4 walk forward tests?

[@tomtownsend4210]

Ok, after watching this over and the 13 videos as well, I think I may understand better: the 1-4 or 5 optimization and walk forward periods should each show a relatively good return and curve for their "recent market activity" period, and thus, we hope we can trust that the final pre-live optimization can be relied on to produce solid results as well when it is turned live. Is that essentially how this works?

[@martyntinsley]

@@tomtownsend4210 Yes perfect! Basically, the walk-forward validation phases do TWO things. Each "individual" walk forward validates that the parameters obtained in the previous optimization work on out-of-sample data - i.e. that they have not suffered from over-fitting. This is the same as any standard WF phase. But the "Cumulative" walk forward (all WF equities combined), proves whether or not your "WFA MODEL" works. If the resulting cumulative equity curve is considered good enough, then as you rightly said, this then gives you the confidence that when you now perform a pre-live optimization, it will also produce parameters that produce a similar result in your live account. But basically, yes it looks like you came to this conclusion yourself anyway which is great! It's always better to do that than someone telling you because you will understand it better. Good job.
And thanks for the feedback :)

[@kylemaharaj9563]

Is there a way to automate the re-optimizations on MQL5? Thank you!

[@simonhuber6859]

So which metrics of which year do I take to let the algo trade live then? The ones that performed the best in the out of sample periods and the most recent year therefore or the in sample improved metrics?

[@Darwinexchange]

Hi. With WFO, firstly you are looking to get consistently good results across all 'out-of-sample' stages. When this is the case it informs you that you are not over-fitting and that your process as a whole is generalizing well. Then, secondly with that confidence in place (if it is), then you should use the parameters from the 'Pre-Live Optimization' phase to trade in live. However, if your overall out-of-sample (for all stages) is not good, then it informs you that you are probably over-fitting and so you shouldn't start trading with any parameters. Instead you need to go back to the drawing board and simplify things (to reduce over-fitting), and/or work on improving the edge of the strategy, until you do get good results in out-of-sample across all phases. Once you've done this successfully, you can then use the 'pre-live optimization' params. Hope that clarifies, and makes sense :)

[@eliremy6979]

🦉🦉🦉🦉

[@marcusl8688]

We run pre-live opt to produce parameter for live account which suits the current market condition. However, the market condition varies from time to time, the covid-19 impact on the market condition as an example. What can we do to solve the abrupt changes in market condition?

[@antoniomachado8782]

If the changes in market conditions are relatively abrupt, there is nothing you can do. Just use solid money management rules, don't risk too much and try to have strategies (uncorrelated) that perform well in different market conditions.
Example: when Covid started, a good trend following strategy would most likely produce amazing results... (offseting eventual not so good results from other strategies).