Over the past 2 months, I've been paper-trading and some real trading in two new (for me) markets, the Euro and crude oil. The strategy I'm using was originally developed for the ES and continues to do just fine in it, but I was getting fewer signals than usual and I've always wanted to see if this approach was as universal as it appeared to be, so the time appeared to be right to try branching out. What's been interesting, beyond the fact that, on paper at least, this approach appears to work just fine in both new markets, with some slight tweaking of the optimal parameter values, is that it has actually worked better than the ES strategy over the same time period. And not only in terms of getting more trades, which has definitely been true, but also in terms of profit factor (ES pf ~2, Euro ~3.2, crude ~3.9). I know that among some of the profitable traders here are a subset who trade multiple markets, so I'm curious if any of you have ever taken an approach developed in one market and then had it sustainably work even better in a different market. Not that I'd stop trading the ES, but it would be great if I could do even better over the long haul in the Euro and/or crude.
A robust strategy must work in multiple markets, although it might require different parameter sets. If a strategy only works in a single market, there is likely something wrong with it. Only exception are strategies that exploit inefficiencies specific to a certain market, such as seasonal strategies.
Well, I want to believe that, too, but I can't help but wonder if there are crucial differences between equity indices, currencies and commodities that make this statement false. I am already seeing the potential for a relatively minor difference in crude's dynamics, which would make one of my parameters unnecessary.
From what I found so far, the differences between time frames within the same market are normally higher than the differences between markets. From the ~20 markets that I use for testing, most strategies work with about 15 of them, but only within a certain timeframe range.
I agree that strategies should generally work across many markets. They don't have to all be knock-your-socks off profitable but I tend to like to see that a strategy produces positive results in most markets. I do have certain exceptions to the rules though: 1. I don't care if a strategy works in most markets but not the ES. The ES tends to be a completely different animal. And vice versa - strategies that work for the ES tend not to be so cooperative in the other futures markets. 2. I don't care if a strategy works in most markets but not the currencies. My experience has been that the currencies are their own beasts. Or, when the work they produce profits but increase the drawdown to the point that it's not worth trading them in the portfolio. 3. I don't care if a strategy works on on most markets but not a particular group. So, if it fails to work in most of the grain complex for example, I am ok with that - I'll just eliminate the entire grain complex from trading. I tend not to like to tweak parameters across markets though. I'm very wary of that because it comes a bit too close to curve fitting for my liking but I know that many people have success doing this. Thats my two cents....
I can see how any tweaking could turn into curve-fitting. That's one reason why I'm gathering more data and sample trades. I realize that it could be the case that I just happen to be in a "lucky" period for these new markets for values of the parameters outside of the optimal range I've determined for the ES, but that over time, the parameter's optimal values in these markets will converge to the same values as in the ES. On the other hand, since one of the parameters I'm looking at tweaking is directly correlated to the asset's volatility, I think it makes sense that it would not have the same value for ES, Euro and crude.
I see this attitude often, so I can't resist to comment. Most strategies have large profit differences from one year to the other. Trading a strategy real for one year is therefore completely meaningless. When you think that this proves a strategy, you are severely mistaken, and up for nasty surprises. Only back testing over more than 5 years can give useful insight in the profitability of a strategy, and only when done correctly, i.e. with enough trades, out of sample data, and a realistic slippage/fill simulation. When your strategies all work nice on paper, but fail otherwise, there were likely some fundamental mistakes in your paper tests.
Isn't it possible for you to run a backtest on these markets for the last 3 to 5 years ? That would prove some more than just 2 months of paper testing (which seem to include some parameters tweaking). IMHO.