Currency Scan

Nice thread. You should not give up on trading.

good grief, it seems like forever since I've been in here. Still grinding along with Numerai comp. They have a revamped data-set out and one of the targets now is "consistency" which measures how well a model does across different time periods. For this round I jumped off the logloss boat for a minute and really tried to focus on that metric. As a former trader (yeah I'm still on the sidelines) having a model that does well over time is more important to me than one that gains .00xx in accuracy. Currently sitting in 19th position out of 378. Puts me in the top 5%, I'll take that any day on that measurement.

It takes courage to admit that forex retail trading doesn't take you anywhere. I think it is the right decision.

Retail brokers do have an awful lot of possibilities to shift the odds against you by increasing trading costs through slippage, spread widening, delaying the platform, disconnect the platform etc.

Good luck.

Back in March of 2016 I started competing at Numerai as one of their anonymous Data Scientists (ObjectScience). Over the last several months I've managed to hold a ranking in the top 1% (.007 if you're counting) for average live logloss (tracked by OmniAnalytics an independent consulting firm). It has been a really interested experience and something I've enjoyed working on a great deal. It was also really gratifying to see the work and research that I started seventeen years ago with my CTA still have value and merit. It makes me a little nostalgic and wonder what I could have accomplished if Refco hadn't tanked.

As much as I've enjoyed that project it's time for me to move on. When I put the Currency Scan on hold I was at a point where I needed to integrate the Machine Learning models into MS Excel but at the time my coding skills weren't broad enough. The last 18 month's have seen my Python skills improve dramatically and I think before I move on to something else I want to finish the scan, see it fully functional they way it was intended.

Not sure if I'll ever return to trading, but the scan is like my baby, it's been with me in one form or another since 2003 and I want to see it finished and fully functional. So, you may see me sneaking in and out of here a little bit.

Without further ado, a shot of my most recent ranking: 19th out of +2700 Data Scientists

Congratulations on your efforts!

Thank you. It was really satisfying to see the work still had value.

And so the work begins, resumes, something. Gotten quite a bit done the last couple of days. With Numerai out of my system and a good camping trip behind me it's time to start grinding on the scan again. Right now I'm making some adjustments in the pairs covered (to reflect the current data-source) and am finally documenting the processes. I've been sort of slack on that and I find myself "wondering" a lot when I look at the back-end after long breaks. When it's done I'll (finally) have a decent work-flow for updates and core changes.

Hopefully in a couple of days it's fully functional again then I can move on to the process building the data-sets for the machine learning side of things. Anxious to get there and start seeing those processes, and the eventual integration into the scan itself, take place.

with mapping errors and old crusty data...

Making progress. Dashboard/Reporting is properly pointing back to the calculation pages which are grabbing from the data pages which are pointing back to source files... need to double check everything but the basic scan and charts should be dropping back in here next week. Then it's time to start the ML grind, that's gonna be a huge project , but I'm pretty excited TBH. Then it's back to dreaming up, creating and testing new data-sets. And then, who knows.

I think we're about 95% functional again. Tomorrow I'll double check everything, rewrite the keys and notations (better descriptions etc) then it should be business as usual and on to the ML expansion, finally. That's been a long time coming.

A quick follow up to last nights signal in the EURCHF. We've seen the pair touch last nights S1 which accounts for nearly 100 pips. The pair is signaling short term bearish momentum as a result of that pa. There could be some more left on the table as this is a pretty long term signal. Even in a demo account I'm not brave enough to hang on to anything CHF...

The "I Can't Sleep, Midnight Edition"

This currency scan is brought to you by, "I drank coffee at 8PM and what was I thinking?"

Markets are pretty quiet right now. Smattering of momentum here and there in the short term price action. We see a small break lower in the AUD/CAD and it has some room to run (not much) before it get's tangled up with S1.

The EUR/CAD is also interesting and might warrant some additional research. After a break lower and decent bearish momentum Monday, the pair is beginning to signal a little "over-sold" in the intra-day time frames (APL, the short-term asymmetrical forecast is getting bullish on the pair).

AUDCAD has for the moment lost its momentum. There is a bearish cycle developing in the pair so it possible it could continue as we get into the US session, given the fact it's also consolidating after a break lower, it could be poised for a move back up. With signals being mixed and the risk profile neutral, this would be no-mans land for me. If I wasn't in it at this point not sure I'd give it any further consideration.

EURCAD still looks a little over-sold in the intraday time-series and we're beginning to see a bullish forecast develop in the broader index. These can take a while to develop and the pair is still at risk for macro news impacts. Barring that, I wouldn't be surprised to see the pair bet back into the neutral zone of it's current trend channel via prolonged consolidation at current levels or an outright correction higher.

EURCAD Pushing Higher

We're seeing some lift in the EURCAD as the broader Index comes out of it's bearish cycle. The pair itself should find initial resistance around 1.4844/54. This will also get it back inside its channel which may suck some of the wind out of the bullish momentum.

Attached Image (click to enlarge)

Potential short-term buy signals in the EURCAD when bearish momentum subsides.

The Euro Index is pushing lower again this morning and it's dragging the EURCAD with it (it might be the other way around, bear with me, it's early here in AZ.) The pair is beginning to show some asymmetry in the intraday risk profiles and long signals could emerge once this sell-off slows. When the dust settles it might be worth reviewing the "why's" and "what's" of the move to see if a reversal back up is possible.

Additionally, if the Euro Index doesn't take out its most recent low as this bear cycle runs its course, we could see bullish pressure across the board in the EUR pairs. The next couple of days could determine if we cycle back up or continue this longer bear trend.

Both the AUDNZD and the EURCAD may warrant some research.

If either of those pairs fall into your technical or fundamental work it may not hurt to run them through your analytical processes. The AUDNZD is still displaying some downward momentum but it may have found some short term technical support. It's not signaling (for me) a bullish trigger but it is getting a little asymmetrical in my risk profiles.

The EURCAD is finding support along the bottom of its trend channel and the bearish momentum we saw earlier has waned. This pair is signaling for a move higher in my analysis so I'd give precedence to this pair first when you run your analysis/models.

Attached Image (click to enlarge)

Euro Index cycles higher takes EURUSD and EURCAD with it.

The Euro Index moved higher in early sessions, all but the EURCHF benefited. Models initially signaled potential longs in the EURCAD at 1.4759 (see Nov. 7th post). The pair is 1.4837 now with R1 between 1.4844 and 1.4863. Those levels also coincide with trend-line and oscillator resistance and could prove difficult to push through in the US session.

Attached Image (click to enlarge)

EURCHF Starting to look over-bought

With the run-up in the Euro Index in the last couple of trading days the EURCHF has been dragged into a potentially over-bought territory. It may warrant some additional research. If the fundamentals don't justify current levels we could be looking at some asymmetrical risk in the pricing, technical easing could follow once the momentum in the pair and the Index subside.

I'll be spending the better part of this week setting up a dedicated platform for the scan. I need the thing to run itself in the background so I can focus on development. Things may be a little sparse this week. With some luck next week though, I'm getting text messages that let me know when the price action is worth a look and I can then post up.

I'll be uploading Sunday's scan in just a bit. Currently setting up another development environment on the workstation. One of the cool things about the comp at Numerai is they allow you to have three separate accounts to compete with. This is going to allow me to benchmark a best in class modeling approach for the scan. Currently I use a XGBoost ensemble (that's the one that got me in the top 1% for nearly 20 weeks). I've also setup a GPU environment for TensorFlow (Google's deep learning algorithm). I thought as long as I was at it I might as well benchmark Pytorch against the other two processes. One of these or an ensemble of the outputs will eventually become the Machine Learning model in the scan. Pytorch doesn't really do Windows so I've enabled WSL and linux. Python and Pytorch will follow right after they download. Pretty interesting stuff. Looking forward to the testing.

Change is in the air
Just a quick note on some changes I'll be making. Going forward I'll be posting the scan and the corresponding charts in their respective threads, i.e. EUR/USD and Euro Index outlooks will be posted in Ata's thread: https://www.forexfactory.com/showthread.php?t=699860 and so on.

I'm going to keep this thread strictly for announcements regarding the scan (changes, bugs reports, updates etc.) and will try to keep those to a minimum so as not to bore you to death. I will ONLY be posting the scan when I feel there is price action that warrants additional research. I don't want to add "look-at-me" noise to any of the threads and want to make sure the scan is meaningful, so be sure to subscribe to my feed.

If you are for some reason interested in the guts of the thing (modeling, research and such) you can follow me on Twitter. That's where I moan and groan about the data-science and programming side of it all throughout the day.

Documentation
I've had requests from Twitter followers for more detail on how to use the scan. It's in the works I promise. When the scan is finished I'm going to follow up with a companion website that will document in excruciating detail (in separate sections of course) the back end of the thing and how to use it to quickly spot price action that warrants additional research and analysis.

Hopefully the website and my continued participation here and at Numerai will also add credibility to my work. I want to get back into the workforce, either as a retail analyst, prop-trader or currency quant, so a lot of this is something akin to a dynamic resume that will hopefully catch the attention of the right people.

A quick note on planned updates to the scan
In addition to the planned Machine Learning models, I'm also going to add one and possibly two Deep Learning models to the scan. One of those processes will be trained for longer-term forecasts (i.e. a week from today) to compliment the Macro Technical Analysis, and the other will be trained to forecast 24 hours from today to compliment intraday TA.

Development and implementation are planned as follows:
- Restructuring the scan for better integration of future modules. (80%, the last 20 is going to be a grind)
- Machine Learning integration (intraday) (80%, code in use at Numerai, extensive track-record)
- Deep Learning (TensorFlow Intraday) (40%, code in use at Numerai)
- Deep Learning (Pytorch Macro) (10%, WSL/Linux Environment in place, code in development)
- Fundamental forecasts (dev on these will be outsourced due to time restraints, looking for a good math/python person if you know one)
- Behavioral models (dev on these will also be outsourced as the math here is more complex and I don't want to get it wrong)
- Documentation/Website (I can do this in my sleep, unfortunately that's when I'd have to do it as I just don't have time, so that will also get outsourced. if you know anyone who has worked with Porto, let me know.)

That's all for now. Thank you for the continued interest in the Currency Scan. Best of luck trading. - JP