I've been experimenting with various lengths of WFO training cycles and am getting some interesting results. I read in Pardo's book that the length of the training cycle should be considered one of the strategy's optimization parameters. This makes intuitive sense, since different strategies exploit different inefficiencies that change in different ways.

The interesting thing about the results I'm getting is that there is no clear winner in terms of all performance metrics. For example, the cycle length that results in the highest annual return is not the one that results in the highest Sharpe ratio. The one that results in the lowest capital requirement is different again.

I should also note that I am confident in the statistical significance of all the Walk Forwards I have tested, based on the number of trades in each test cycle.

I'm wondering which performance metric others would choose? I am tempted to go for a Sharpe ratio of 3.17 over 3.01, even though the annual return is about 40% less for the former. Then again, I am not sure about Sharpe ratio, given that it assumes a normal distribution and it is simple to show using Excel that the strategy's trade results are not normally distributed. In this case, it would seem to make more sense to consider a trade off between the highest annual return and the lowest capital requirement.

What would others do in this situation???

Hey Boatman!

Great with a Sharp Ratio of 3.17, really good job! What type of strategy is it? I'm not asking you to give me the code or so, just curios what type of indicators/filters/approach you use. My best Sharp Ratio so far is around 1.5 and that strategy is based on a Normalize indicator and LowPass filters.

Take care

Hey ibra

Sorry for the tardy reply.

Its a pretty simple range breakout strategy that uses volatility filters and the higher time frame trend for entries. It uses some of John Ehlers' trend indicators for exits. It trades a portfolio of six assets - EURUSD, AUDUSD, EURJPY, USDJPY, GBPJPY and XAUUSD.

I've found that Ehlers' trend indicators are very useful for trend following systems. I haven't been able to build a successful system using his cycle indicators, but am working on it.

That's great. When selecting which assets to trade, do you usually go through them one by one?
I'm still working (when spare time allows me) on various strategies, haha, but it seems that I can't really "finish" a strategy before coding another one. I'm not trading the built-in strategies though since I really can't see the sport in it. How many strategies does your portfolio consist of?

Yes, I went through each asset one by one in this case. Although I did read about an approach that someone else was using (DdLV I think) where he ran a walk-forward test on multiple assets and selected those to trade via the calculated OptimalF. I think its a good approach and makes sense, and I do intend to adopt it at some point! I believe the thread was called "design approach" or something similar if you want to search for it.

I know what you mean about not finishing off strategies! I have a couple of others that I am testing, but not quite satisfied with. I find it hard to know when to keep developing the strategy, and when to abandon it and focus on something else.

At the moment, my portfolio only consists of the one strategy. I hope to add another one or possibly two in the next few weeks, but I'm not quite satisfied with their testing yet. We'll see.

I am also not trading the built in strategies. I ran Z12 for a while, but had disastrous results that were clearly not aligned with the backtest. Often, even the backtest itself fails, indicating that something is clearly wrong. It seems that others have had problems too, judging by some of the posts here. I couldn't get a very satisfactory answer from support, so I decided that it is better to build my own strategies. At least then I can only blame myself when something goes wrong!