Pair Divergence Method

The E/U buy yesterday was a win, and looking over the charts today, I found this E/U sell, which would still be in progress.

edit: would have hit TP level

E/U short - win
U/C short - loss

So I'm not 100% sure about fills during NFP today, but these popped up. A few still in progress. No more than two trades at a time would have been open.

edit: open euro trade was a win, u/c trade was a loss.

I finished back testing the last half of 2008 with these expectancy numbers (note this is NOT % return!):

Trading June-August would have been rough on your account if you were trading 3%/trade. Cumulative 28% draw down! However, if you had kept on trading, you would have ended the 6 month period +33%. Not to mention the stellar performance since then.

I hope to get the first half of 2008 tested ASAP.

U/C 30m sell today would have been a win:

edit: and one from yesterday I missed

Two more wins. They were both close calls, peaks differing by one or two pips on each.

I really don't want it to look like I'm posting winners only, but that's what it seems like recently =x

U/C trade hit profit overnight, while I would not take the second trade posted here because it's so close to when I would quit on a Friday.

I've recorded 12 wins and 3 losses so far this month (average win being 93% of the average loss).

It seems silly to just keep posting charts like it's a trading journal or something. Maybe I will start one of those?

U/C loss a bit ago.

Finished my testing of May '08 with 16 wins, 20 losses, average loss being 0.96 of the average win. This resulted in an expectancy of -0.09

So far for 2008:
Losers: may, june, july, august
Winners: september, october, november, december
Untested: january, february, march, april

2009: No losing months.

Really makes me

I'll be double checking results soon.

It's possible that the necessary correlation began in September, 2008 and still continues. It looks like that is the case.

I don't know how to prove it. I don't know of any figures that will give past correlation data.

If that is the case, it would be easy to avoid such times with a correlation indicator. There are several out there I have seen.

Because of the "shifting" problem with the overlay indicator, I had to go back and make sure E/U and U/C had all matching bars on 15/30/60m time frames throughout 2008 and 2009 for backtesting purposes. I can post the data if anyone is interested.

Because of having that data in excel, you can do the CORREL() function on X bars of past data to find the correlation. Then to kick it up a notch, add in conditional formatting color gradient, and it starts to look really cool. I suppose this is what MT4 correlation indicators already do for you as new bars are added to the chart.

I've attached an example for anyone unfamiliar with the concept. I used typical price HLC/3 15m data from May 08; times are MT4 broker times. The question I have: We can spot where divergences appear in real time...we don't need a lagging indicator to do it for us; therefore, we should apply this to a higher TF as a filter, right? But is that an arbitrary filter, or one that actually means something?

The whole idea behind this method is that the two pairs we're trading are highly correlated on a grand scale most of the time. We are just diving into the smaller time frames with a sniper rifle to pick out the times when they sway from the norm, assuming they will always go back to the norm.

You wouldn't trade G/U and U/J with this method because their correlation isn't anywhere near as consistent with other available pairs such as E/U and U/C. So why trade E/U and U/C when they're having an off day? But how do we clarify off day? Won't it be off X-days-period? I won't jump to conclusions until I can pull some data.

I'll try to pull daily TF data for as long as I can reliably go back (I imagine until 2005 at least?) and throw it into excel with the correl() function.

edit: probably give it 20 days worth of data for the correlation coefficient? That's a month. Maybe compare with 5 for a week?

Hopefully this gives you an idea of the correlation during the period where your expectancy was down. It uses the same 50 period look back that Mataf uses.

Thanks for your post! Will be great to compare notes with.

I graphed daily correlation for periods of 20 and 50 across 2008 and 2009, and I didn't see much you could take away from it. I will post the charts when I have time to clean them up. I think if you're going to use 20 or 50 data points, it is silly to use daily data. Going to look at 12 and 24 period hourly correlations next. If there are obvious trends there, maybe we can take something away from that.

February ended with 27 paper trades for me: 17 wins 10 losses. Expectancy of 0.23 (avg win = 96% of average loss).

I'm spending this week trading live in preparation of going live! So I may post some more charts.

As far as those correlation numbers go, I don't like the idea of making a "should I trade this setup?" decision based on lagging indicators. I would rather base the decision on the pattern being traded and market hints. One thing I've found fairly consistently: if a trade goes immediately into loss and never sees much profit by the time it's 80-90% of the way to SL, it will rarely get back to TP. I think I've also spotted a few "types" of setups that have higher win rates than average, but that might just be confirmation bias rearing its ugly head.

All results posted so far are still unfiltered except for time of day, no more than two trades at a time (don't run into this very often at all), and no open positions before interest rate announcements

Has anyone ever noticed the overlay indicator repainting while on a live feed? I've been scared for this all along, and I think it might have happened this morning, but I'm not 100% sure...

Yes, I've seen it repaint many times.
Any time a bar of data is missing from the overlaid pair, the alignment shifts all the way back.
This is what makes backtesting difficult.

When a new bar begins, if one pair receives it's price tick and the other doesn't, the alignment shifts and then shifts back when the tick is finally received.

Sorry Bill, that's not what I meant. Have you ever had two peaks at say, 1.0770, then you double check them again later for whatever reason, and one is now at 1.0770 with the other at 1.0771?

I hope that is more clear

No, I've never had that happen.