Different Algos for Different Currency Pairs

afeudale · Post by **afeudale** » Tue Jan 30, 2018 5:10 pm

One thing I don't see discussed much on this forum or others is about the different reactions that different currency pairs have.

To elaborate, we anecdotally hear things like "USDCAD has a slow reaction time to news" or "EURUSD reacts quickly". However, when we're designing trading systems, we somehow overlook this and try to design a "one-size fits-all" approach. Intuitively this doesn't make sense.

Indeed, my testing (backtesting in Empty4...I know...I know...) confirms exactly this:

My programs are mostly using variants of CSS for trend determination. EURUSD for example, needs an hourly CSS of >0.8 and a weekly CSS of >0 before it's a good idea to jump on the trend. Indicating hourly trend is much more important than weekly.

On the other hand, USDJPY needs only a positive (>0) hourly CSS, but weekly has to be stronger at >0.2 before jumping on the trend.

Has anyone else encountered/used this type of pair-specific info? And if so, how did you determine it if not through backtesting?

afeudale · Post by **afeudale** » Thu Feb 01, 2018 5:18 am

Here is a screengrab from today's trading based on this latest EA.

Pair is GBPJPY and the EA took about 60 trades (0.01 lots) throughout the day.

Currently sitting at $573 profit on LIVE account (I only run EA's live...demo is not as exciting....live on the edge...hahaha)

No manual intervention. I'm tempted to pull profits, but we'll see what happens overnight

Post by **renexxxx** » Thu Feb 01, 2018 8:36 am

afeudale » Thu Feb 01, 2018 3:18 pm wrote:Here is a screengrab from today's trading based on this latest EA.

I am not sure if/where "this latest EA" has been introduced before and how the second post relates to the first post. I am interested to learn what your "RangeExpert_v32" does to make this $573 profit on a live account. It looks like it is sending a gazillion of trades, very close together. So, this profit can also quickly turn negative.

afeudale · Post by **afeudale** » Thu Feb 01, 2018 1:05 pm

Hi Rene, Hope all is well

I'll try to answer your questions below:

1) No, I have not posted my EA's on here before. I posted often when I first started trading about 2 years ago, but was disappointed in the closed-mindedness of some members, so my posting dwindled. As an example, I posted about Empty4 backtesting a few weeks ago (which has been successful for me) and a bunch of users jumped on it, essentially calling me an idiot for believing so. Range_Expert_32 is just the 32nd iteration of a particular EA I've been working on the last two years. I probably coded 200 other ones as well and abandoned them. It's funny, as initially it was a "Range Trading EA" (thus the name), but now it is a trend trading EA (but kept the same name...lol)

2) Loosely describing the EA, it takes advantage of two proven market inefficiencies: trend and seasonality. There are only a handful of inefficiencies that can be exploited, listed here: http://www.financial-hacker.com/build-b ... d-systems/ and there are a ton that don't work, listed here: http://www.financial-hacker.com/sevente ... nderstand/ Trend and seasonality were chosen because they were the easiest to program.

3) Regarding the "gazillion trades", you may be too quick to judge. If I showed you a screencap that had only 2 or 3 0.5 lot-size trades, you would think it is more sound trading? Not so. This EA took 50-60 0.01 lot size trades through a concept know as "dollar cost averaging". There is no way to predict exactly when to pull the trigger on a trade (when conditions are met), so you buy small lots (in this case every 15 mins), while the conditions of trend and seasonality are still valid. This way, your buy-in price is the average of price action over that time period. Those gazillion trades you see are actually more efficient over time than if you only had a couple of trades. It also has the effect of pyramiding (another proven concept), where additional trades are added to a winning position.

Back to the original question:

Are you and other users, using the same conditions to enter trades for all currencies, or are they tailored for each currency? For example, I believe it's the Baluda Super-CSS that uses a 0.8 CSS crossover to trigger trades. From my testing, this is bound to fail, as only a couple of currencies obey this rule. Others may need a 0.4, 0.2 or 0.0 crossover and all at different time scales (hourly, daily and weekly).

Thanks

renexxxx » Thu Feb 01, 2018 3:36 am wrote:
I am not sure if/where "this latest EA" has been introduced before and how the second post relates to the first post. I am interested to learn what your "RangeExpert_v32" does to make this $573 profit on a live account. It looks like it is sending a gazillion of trades, very close together. So, this profit can also quickly turn negative.

afeudale · Post by **afeudale** » Thu Feb 01, 2018 1:12 pm

For anyone who's curious, here is the trade result after the overnight period (same trades...EA is still holding...SL is trailing).

growfx · Post by **growfx** » Thu Feb 01, 2018 1:49 pm

afeudale » Tue Jan 30, 2018 5:10 pm wrote:One thing I don't see discussed much on this forum or others is about the different reactions that different currency pairs have.

To elaborate, we anecdotally hear things like "USDCAD has a slow reaction time to news" or "EURUSD reacts quickly". However, when we're designing trading systems, we somehow overlook this and try to design a "one-size fits-all" approach. Intuitively this doesn't make sense.

Indeed, my testing (backtesting in Empty4...I know...I know...) confirms exactly this:

My programs are mostly using variants of CSS for trend determination. EURUSD for example, needs an hourly CSS of >0.8 and a weekly CSS of >0 before it's a good idea to jump on the trend. Indicating hourly trend is much more important than weekly.

On the other hand, USDJPY needs only a positive (>0) hourly CSS, but weekly has to be stronger at >0.2 before jumping on the trend.

Has anyone else encountered/used this type of pair-specific info? And if so, how did you determine it if not through backtesting?

Interesting topic!
I have seen similar things regarding CSS and trending.
I have just started Reading about CCFp and it seems to be similar to CSS.
Have you evaluated CCFp in the same way?
Do you know if that indicator suffer in the same way as CSS when it comes to levels and trending?

BostonEJ · Post by **BostonEJ** » Thu Feb 01, 2018 3:39 pm

growfx,

I haven't done a deep dive into the differences but CSS uses TMA and CCFp has selections for different moving average types as well as settings for the fast and slow averages.

I cant speak to issues around levels and trending.

EJ

afeudale · Post by **afeudale** » Thu Feb 01, 2018 5:41 pm

I haven't heard about CCFp before, but I checked that link briefly and it seems to be a currency strength indicator, similar to CSS.

In that case, I would maintain that CCFp requires the same theory as per my research - that different currencies will require different strengths on different timescales before a real trend is established.

I really find it interesting that those threads (i.e. CCFp) talk about how well their algorithm works but make no mention of which currency pairs that applies to.

In my opinion, this should be the first question. I have found only 5 pairs that trend with enough inefficiency to be exploitable for profit - all others are useless long-term - no matter what algo or system you throw at it.

So, while CSS and CCFp (and likely many, many other indicators) may work fantastically on EURUSD as an example, they will fail miserably on CAD or AUD pairs.

Here's my two years worth of research in a nutshell - and enough info to design your own system based on it:

1) Find which pairs exhibit true trends and are inefficient (I used 15 year Empty4 backtests with full tickdata)
2) Use an indicator of choice to determine what constitutes a trend for this period (CSS, CCFp, MA, etc)
3) Use seasonality to determine from past behaviour (backtests) if jumping on a trend indicated by #2 above is likely to be profitable
4) Press "GO" and count the money (time will tell)

growfx » Thu Feb 01, 2018 8:49 am wrote:
Interesting topic!
I have seen similar things regarding CSS and trending.
I have just started Reading about CCFp and it seems to be similar to CSS.
Have you evaluated CCFp in the same way?
Do you know if that indicator suffer in the same way as CSS when it comes to levels and trending?

afeudale · Post by **afeudale** » Thu Feb 01, 2018 5:57 pm

Here's today's EURJPY trade...with a couple lingering GBPUSD trades from yesterday. Yesterday's GPBJPY basket of trades closed out at $900 profit. It hit $1500 at peak, but I restrained myself from profit taking. Design rules and stick to them, even when tempted not to!

Post by **Wavegarrick** » Thu Feb 01, 2018 6:02 pm

Afeudale

I believe it's the Baluda Super-CSS that uses a 0.8 CSS crossover to trigger trades. From my testing, this is bound to fail, as only a couple of currencies obey this rule.

You are continuously referring to the level cross value of 0.8 or 0.2 etc. Baluda's superslope has always been set with a level cross value of 2.00 and at times we have used 0.0 but I have never seen anyone use a difference of 0.80 in all my years of testing. Can you please be specific or are you referring to the threshold. Right now you are confusing the heck out of me.

Cheers
Leon