Currency Scan

A quick look inside the scan... this is how you make MS Excel rip:
- adopt an "object oriented" approach (lock all objects down so users can't inadvertently tank your work)
- use built in functions first (they will always be faster)
- use optimized VBA for light custom functions
- pass off more demanding stuff to python
- if you have to, use c/c++ (if python wrapped around c++ isn't available)

There isn't much you can't do with excel and it allows you to build dashboards on top of an application almost every business has and most people have some familiarity with. It does have some pitfalls as does everything.

My day in a nutshell...

A quick look at the AUD/USD

Just a quick update on some stuff happening with the scan. This past week I setup a stand-alone box for the TriggerScan (kind of a Currency Scan Lite version). The configuration of the box and the TS are both +90% finished. Monday or Tuesday they should go into testing and I'm guessing by the end of next week I can start using them in full-time production. The TS is currently restricted to just a couple of signals on the Weekly and Daily Asymmetrical Pricing Forecasts (APF) but expanding to the other intra-day and price-action signals won't be an issue. It's going to allow me to generate a lot of insight/content while also keeping up with other projects like the Machine Learning stuff. Pretty excited by it all.

added a poll to the thread. thinking about adding 4H models the scan, curious to know users would find that useful.

https://www.forexfactory.com/showthread.php?t=576234

I've done a little in house-cleaning this week on the scan, dropped some things I'm not using and am adding things I think will add some value (4hr data for Ata and position scaling). The latter isn't meant to be a recommendation to trade or a hard and fast rule to building positions. It's meant to be a piece of analysis that recognizes that forecasts can and do often initiate research/insight, very early in a move, and if that's not accounted for, it can have a detrimental affect on a traders timing. By presenting this timing/scaling factor, I can hopefully encourage readers to look deeper, not just at the "what" of their trades, but also the "when", and how much risk they take on at any given time.

For my own stress testing I'm also going to be tracking this on myfxbook in demo mode. I want to give myself some peace of mind that the analysis has some level of merit. In a perfect world I'd throw 200k at the thing and work it in real time, but I just can't risk it (too old for that and I lost too much on the Refco BK!).

I'm not sure if the trade explorer here supports demo trading, if it does and it's not too much work to set it up each month I'll add that as well. I'm just working with what I know right now to keep production flowing.

Now it's time for some documentation...

I'm finally going to start working on the companion website, the "how to use the scan" stuff, and that means putting things to the test. It's been a while since I've traded and while I don't think of the scan as a stand alone recommendation/signal I've always believed that forecasts need to be actionable (even if it's triggering additional research) and for those forecasts to be valuable, they have to be profitable.

So before I can really construct a meaningful "how" I've got to spend some time back in the demo-trenches observing the what. With that in mind I'll probably be holding off on posting much analysis. I want to make sure the scan is useful and what I'm saying and writing is actionable in a profitable way. Not sure how long that will be. I'll keep you all up-to-date as I grind along. - JP

Fast forward a month or so... Demo trading was interesting. Net results were positive, right in the range I would want if I were trading live (managing funds). Returns were a tick over 1% for the month and would have been a little better if i wasn't testing so many different things. Trades were broken up across different styles with different parameters. Taking trades from the short-term side with no regard for the longer-term signals was pretty disastrous. The net loss on that was about half a percent and never felt comfortable for me, I'm not really an intraday kind of trader. Taking small positions on what I felt was asymmetrical pricing in the market and letting them run for days, sometimes weeks, paid off. This was basically the way I did it back in the fund. That of course also had the benefit of the machine learning models (they're coming I promise) but the underlying premise was the same, asymmetrical risk.

One of the things on the short-term trading that became really clear was how much market timing plays a part. If there was a short-term bullish signal in the scan but the EU session had established bearish momentum, getting long was a terrible idea. Having said that, trying to ride that momentum never really brought any joy either as much of it had already run its course. In those scenarios you were much more likely to hit stops but never hit TP targets. I think in a lot of ways the EU traders have an advantage over those of us Stateside as they are in markets when the initial short-term momentum begins. Getting in on those waves hours after they start doesn't really work.

So I think going forward, especially given the time of day I write from, I've got to be super careful about any inference regarding short-term moves. I also think it's really important to understand that the most profitable trades I took from the scan made money the same way I did as a trader. No stops, small positions relative to holdings and just let the signal play out until it was neutral. Most retail traders can't (and don't want to) trade that way. Large amounts of leverage forces the use of stops or your risk-of-ruin goes through the roof. Not that, IMHO, stops prevent ruin, they just delay its arrival, but that is a debate for another day.

Now I've got to think about how to present the analysis. If the short-term information in the scan is only meaningful in improving entry into long-term trades that traders aren't able to take... Well, you see the problem. The first part of the solution is to make sure I'm clearly labeling the analysis as long-term (days to weeks). The second portion of that may be to completely hide the scan from view to prevent anyone trying to use it in a manner where it really isn't useful. I'm not sure. All this has to be kicked around a little. What's most important to me at the end of the day is people don't lose money based on my work.

On a different note, until I know what I'm doing with it, I've buried the website. I can't exactly write a "how to use the scan..." When I'm not convinced the majority of people reading it can or will be able to use it in the right way.

For now I'm going to keep marching on with the ML models and I'll get back to posting here. I think at the core, the information contained in the scan is very good, especially for seasoned traders with substantial capital, but presentation is going be everything so that newer trader with smaller accounts don't get torpedoed.


Probably worth mentioning again, these results are DEMO and there was a bunch of different time-frame testing going on here. This is not how I would trade my personal account. It's just to show that I did indeed spend the last month stress testing some of the signals and that the information below has shaped the presentation of the information going forward. The returns, losses and stats are completely meaningless outside of that.

I've been giving this a lot of thought the last few days, I think it's time to move back into the professional side of things. Trading the last few weeks have driven home the strengths and weaknesses of the research and modeling. It's just not something that will fit into the retail sector well but I do think it has utility at the real money and professionally leveraged/managed levels. I'm going to start trading live again and slowly assemble a team that can move the project forward. It's been a really interesting journey this far and I appreciate all the interest and support in the work. I wish you all the best with your trading. - John

Fast forward a bunch of months: I tried to push the CS off into the institutional side of things, but I just can't muster the manpower to make a proper go of it. There is so much demand for talent it's difficult to come up with a package that will actually keep people. So I'm not sure what the path forward is. I think the CS has a lot of good info in it, but it might just be poor timing. I think for now I'm going to get my stuff reorganized over the next couple of weeks, get everything working "locally" again and will start posting the CS here again (on the site). I've made some cool improvements over the last few months, actually did a complete rewrite of the backend (again) and automated some sub-reporting. So you'll start seeing my ugly mug around these parts before too long. On another note, I've continued to participate at Numerai (the AI Hedge Fund project), the last few months have been crazy tough but I have recently found myself back in the top ten (objectscience) on the leader-board: I'll accept that. See you all soon!

Just realized I need to change my signature, not looking for work/partnerships at this time. Just want to focus on research, analysis and modeling. Maybe do a little writing as well. All of that can be accomplished here.

Man, this update has been a long time coming. First things first, a few weeks ago I submitted my final models to the Numerai Hedge Fund. All those models have resolved and I leave their data science competition in the top 20, out of several hundred participants and a claimed 40,000 total. Pretty happy with that, but it's time to move on. When I got involved with Kaggle and Numerai it was to improve my python and data science skills. I wanted to integrate my machine learning models into the currency scan dashboard but it was going to require an entirely different approach to the modeling. That work is finally finished, the ML models are FINALLY in there! Couldn't be happier to have that out of the way.

Along the way, I've made quite a few changes to the CS visually. To keep it from getting too busy, I changed how the technical models "report" and have hidden the majority of that information. Any collective bias that occurs in those underlying rules (seven models) is ensembled into the watchlist as a simple red or green fill.

Next to that, I have the reporting for the first batch of machine learning models (the first 24 hours). There are actually a large group of additional models on the back end that will slowly get integrated, but this is what I need right to get back to trading... You heard that right, I'm wading back into the market with a very small account ($5k). I'll be linking it into MyFxBook and maybe here as well if I'm able (the last time I couldn't get the explorers to work). I have my eye on possible providing signals to the new Erasure project when it launches and I want to start building a track-record again for PoW.

If all goes well I'll be trading by the first of the year and will be back here writing on a regular basis. I hope so, I'm ready to get back with my trading people and cry about the markets again. Happy Holidays everyone! - JP

Godluck this time around

Thank you, now if I can just avoid broker bankruptcies/melt-downs maybe I'll have a fighting chance???
Definitely looking forward to getting back in the grind.

Trade Explorer is up and running, feel free to follow along as I kick off the slow grind... Super excited to get back at it, little nervous too (can't lie), but looking forward to making that long march towards a five-year track record and see how it plays out and what I'm able to accomplish. If all goes well, maybe at the ripe old age of 56/57 I'll throw my hat back into the managed funds ring for ten more years of torture. Updates to follow this week on the scan, got a little more done and you know I have to talk about it, just my nature.

Not a lot going on this morning on my end. We see a signal starting to develop in the GBPJPY, still early on that though. A little bit of price action as well in the AUDNZD, CADJPY, NZDUSD and USDJPY. We'll see how this plays out through the day, there could be some reversals here to capitalize on later.

Drilling just a little deeper into the pound, ML forecast is for a move lower from current levels. With a mixed forecast there I'm pretty much on the sidelines. The upside is, if it does push lower it will continue to look more asymmetrical from a risk profile, getting long will make a little more sense internally.

Quick update, fixed the composite forecast, signaling sensitivity was incorrect. All good now.

As a rule, the only time my machine learning forecasts show up is if a trade is signaled. Realized today there are times where I want to take a peek at those, especially if I'm writing about something specific (like the GBPJPY today). So I'm adding a feature to the management side of things to allow me to manually call n number of pairs, all pairs, or just let it default to "trade only" pairs. Should have that knocked out today then it's back on to position sizing.

Position/Bet sizing is incredibly important when trading. Been thinking about this a lot the last few months. Most of us are familiar with Kelly Criterion, but we also know that using it can be the shortest distance to the exit if there is a run of bad trades (regime shifts, market cycles, changes in model efficiency all create havoc). The trick is to water down the known optimal exposure to the unknown survivable exposure. A lot of traders take a flat reduction on KC (like half). I'm thinking about taking that as a baseline and adjusting based two additional metrics, signal strength and meta-label probability. Hopefully, the combination of the three give me a more dynamic approach to risk and will allow me to survive the unknown while optimizing for portfolio grow... Sounded good there, we'll see how it works.

The decision matrix (aka Currency Scan) is finally finished. Lot of updates and improvements already planned, but it's where it needs to be to trade. Looking forward to 2019! Except for some small trades here and there for stress testing the system, I'm probably taking the rest of the year off. Time for a break and some R&R with the family.