A genuine edge based on price movements will work anywhere, but not equally well everywhere at the same time. The forex pairs are not traded by the big banks in the same way. This means for example that EUR/USD trends are short-lived and reverse without warning, while GBP/JPY is known for monster dramatic moves, while EUR/CHF just limps along. That said, characteristics like volatility rotate between different pairs, so even EUR/CHF has some exciting times over the course of a year. All it boils down to for us private retail traders is that pairs can move strongly, some weakly. This will necessarily affect the outcomes of a TA-based strategy, but that alone doesn't prove absence of edge.
If by trading data you mean price behaviour, isn't trading data the core of TA? Isn't TA simply the pictorialisation of trading data? Sorry, but I take the orthodox view, that analysis is either Technical or Fundamental and that if it isn't either, then it isn't analysis. But I'd be happy to be shown how wrong that is.
It doesn't really matter what approach you take, so long as you make money in the long term. I am of the view that is important to prove a statistical edge before committing any reasonable capital to trading activity. It is quite hard to do this with lines subjectively placed on a chart, arguably impossible. I don't have anything new to add to the age old debate about the relative efficacy of TA as compared to a quantitative approach.
Well, I agree, making money in the long term is the goal. And also the proof of a strategy. You kicked the thread off reporting a strategy which doesn't display a universal edge. I can only suggest there isn't any way round this, and in the long term it won't matter.
Thank you for this post. It might be an interesting idea to include a tool to monitor the properties of the current market versus the market conditions over which the models were created. I could look at how the back test performed without the application of my filters, monitor the win rates and drawdown in chunks of time and then compare them to the performance of the back test with the filters applied, over the same periods of time to find a relationship. There might be a relationship to exploit as an early warning of the likely future reliability of the signal. Or it might simpler to monitor the performance of the signals at the pair level, to know when it is underperforming and is time to switch off, at least that would minimize the losses.
I trade 20+ currency pairs and I found that no matter what strategy I use, my filter settings would have to be different due to the different dynamics of each instrument. Also, if you are doing back testing you would need to know what happened during certain periods of time so that you can decide if the event which caused your test to fail was a one off event (eg Brexit) or if it is likely to occur again (GFC?). Depending on the strategy you use, you might want to consider excluding certain currency pairs which would not result in many trading opportunities or perhaps exclude those currency pairs which could increase a trader's risk (EURCHF?, EURAUD?) You mentioned that you have a problem with new data. My guess is that the dataset you've used for setting up your filters is incomplete or that it has errors. If you are using metatrader historical data, it is broker dependent and often has errors or is incomplete and therefore its not reliable for fine tuning your strategy. Ideally a strategy should be forward tested for >1 year before we could say that our edge is working.