I definitely agree! Machine learning is becoming the 'wild wild west' for FX trading. At least for us at RagingFX.com, it has been a game-changer.

Our approach is primarily focused on attributing dynamically-changing probabilities of success to our FX signals. Our algo learns instantly what to accept and not accept based on numerous internal and external market factors.

On any given moment, one can watch our algo change its mind based on, for example, market sentiment or any other key influencing factors, (i.e. news breakout, etc.)

Good to see others are seeing the benefits from ML too!

hi Tom, sounds impressive, especially as it appears you have visibility into why your ML predictors are doing what they are doing. Is it along the lines of lime (arXiv:1602.04938) or is it some other approach to interpretability? Do you have any white papers or such you can share with the group or privately? Thank you too, for the kind comments.

Hello All , I am not first hand familiar with AI or machine learning but from a layman point of view : All have the same data all have the same fast computers shouldn't all of them get to the same out put ?The machine is doing the learning it is the same computer the same software the differences are insignificant. So all trading the same idea/s right ? so all selling buying at the same time right ....so nothing really change the mass always wrong. :D

Eli Weiss, the same fact patterns can suggest different theories to different people; that very thing happens in the inferencing automation they design too -- that different programs can trade the same market data differently.

Mr. Harish, As I mention I do not know programing and do not have knowledge nor I have intimate experience with ML AI , but What I understand from the hype around it that human have no interaction with the market the "only" action they do is put data in and expect the computer to find pattern/s out with an executable possibility, where the risk reward is expectable .

Than if I am right than "all" computers / Programs will eventually will come to find the same pattern right ?

What you are talking about is the human interaction that each one define the pattern and the action follow base on his own agenda a point that the same machine you are talking about is trying to eliminate.

Harish - In short we are taking a less conventional approach to ML by treating probability values as vectors. The parallel with physics has opened up an interesting range of theories/equations for us, which have allowed us to construct surprisingly accurate predictions on a relatively consistent basis.

Our work is in progress as we speak. Earlier this year we developed a subscriber interface to keep us accountable and excited. Our goal is to offer predictive signals on a consistent basis with a success rate of 70-80%. ...with a reasonable lead-time between 10 minutes to 3 hours from delivery.

There are no white papers to share. We are currently tweaking our theories on five systems simultaneously. One of the five systems runs our subscriber service. Whenever we confirm a breakthrough, we add the changes to the subscriber system.

We update the results from that system weekly at RagingFX.com. You are welcome to follow our progress there.

Eli, I can certainly understand your reservation behind the ML hype and that is why we embarked on our project at RagingFX.com to see if we could identify points of solid differentiation. To start, we didn't read the rags or follow the trends. Instead we gathered a bunch of really smart people from various unrelated disciplines (i.e. traders, physicists, statisticians, mathematicians, etc). and worked through alternative ways to apply probability theory.

In our world, we can't be right all the time, but we can be right more times that we are wrong. Add a predictive Machine Learning component that changes dynamically just like you might change your mind when crossing a busy street, then the odds of earning successful outcomes can increase. Finally add standard risk management techniques to these odds and...

Based on your thinking, all humans would function identically. Fact is they don't because they all think differently.

Human stock pickers at hedge funds are actually beating their computer competition so far this year

"Like all things, performance go in waves, but the computers will need to do a bit better to justify all this interest."

Mr. Wolfgang ,

All use the same data. Market price, Volume, government statistics and alike.

All use the same computers trying to find Predictive patterns.

"Does this make the issue more transparent to you?"

Eli Weiss, You're right if indeed the endpoints are your goal. In our case we identify patterns that have nothing to do with tops or bottoms. Yes, we use the same data just like one would use a map to navigate throughout a city. The difference is that we might arrive at different places and at different times. While you focus on the 'tops' and 'bottoms', we spot key areas where our algo can tell us from past and current data that there's a good chance of reaching a successful outcome. By the time we reach your 'top' or 'bottom' area, we've made plenty to call it a day.

Mr. Kadala, I agree with you that there are point to trade without the need to buy the low of the day and sell the high of the day.

My point was that the mass "always" wrong and "always" buy at the top and sell at the bottom.

This is a fact.

Now if your team will come with an idea, their idea to profit in the market and use "algo" math to "locate" the idea during the trading session than ...

Is this is a ML or AI ?

I do not think so. You use fast computers and fancy math to help you trade your !!!idea. This is not a machine smart this is human smart that get help in execution with high math, fast connection and fast computers.

But the original idea the core thinking is your team. How to execute the idea ?

This is not an ML or AI.

Eli Weiss, I think what you are confusing is the difference between psychology of trading (i.e. herd instinct, etc) versus ML predictive trading. These are two entirely different approaches that use the same data just like you and I use the same highways to reach our destinations. Aside from the common data conduit, there's nothing else in common.

ML predictive data applies computational math, statistics, probability theory, optimization modeling and in our case some clever physics to essentially simulate the market on a minute by minute basis. You might compare what we do to a conversation with a good friend or sibling where you are so familiar with how they think that you often will finish a sentence for them in a conversation.

Mr. Weiss, if you have like you say no knowledge of the area of ML and AI, why would you participate in a discussion about this, I would like to learn from those who do.

Mr. Kadala do I understand correct that you have applied the world of quant trading to the Forex market, if so how does 'trading alphas' translate to the terms used in the Forex trading as the retail traders see it? Is that the world of technical indicators of something different?

- or something different? -

Mr. Kadala, it goes like this:

If you provide the computer data and the computer find by him self a pattern that can be execute and make money upon. Than this is ML , AI or any other name you want to call it.

All other options were you give the computer instruction of where the pattern is what to look for how to search and alike than at that point you are the system developer and one of your tools is a computer that helps you find what YOU ARE LOOKING for in a fast and a reliable way that other wise will take you "years" to do.

Henk: I am not the Focus of this conversation the topic is.

If you can not answer my arguments do not look for a way out by pointing at me personally.

To attack the person who debate personally and not to being able to answer the arguments the he present is a sign of someone who was defeated in the debate.

You welcome to disengage from the conversation it is your choice, or you can contribute to the topic if you are able to do so.

Have a nice day.

Eli Weiss, Pattern matching is only one aspect of predictive machine learning. Granted it is an important one but it is not the one that governs the final outcome. It's similar to the bricks that make up the wall but by themselves they do not offer any architectural value or structural integrity. Where the rubber meets the road with our approach at RagingFX.com is in our ability to apply probabilities to these pattern events. So, if your approach created five signals based on pattern matching alone, ours will too but in addition will also tell you which of the five has the greatest chance of success. Done daily, then your outcome along with proper risk management techniques can render very much above average results.

Henk Vedder, Trading alphas (i.e. equities) usually pegs an index against a fund's performance or your own. With Forex, pairs of currencies are pitted against each other, hence EURUSD is euro against the dollar. Since it is a zero-sum game, profits and losses are distributed equally the same way that a water level remains constant in a lake despite any wind-generated wave activity. Unlike stocks that can grow from $1 to $1,000 per share, currencies tend to swing up and down within a specified set of ranges. Great question!

Eli, sorry, it was not a critical remark, just genuine wondering what your stake is in the area of machine learning and trading. What is your experience in the area?

Tom Kadala I agree with your last statement at the end of the day everything is Pattern Recognition the computer bundle will add the math to the package and will add speed in recognize the pattern in real time and the back testing of how reliable the pattern is.

With proper money management that should produce a winning system.

My ? from start is: Who came up with / recognize the winning pattern ?

the Human or the machine ?

If I am as a trader seating all day long for hours no end in front of the monitors and discover an opportunity to trade but bcs of the short live / existence of this opportunity I as a trader / Human can not execute fast enough to take advantage then , Comes the computer that will recognize the pattern and not only in one market but across the board and will execute faster and more efficient than I can do.

If this is the process than in my humble opinion it is not a ML or AI.

np Henk we all adult here nothing is personal

Eli Weiss, Trading is both a science and an art. As a science it follows much of what ML and AI have to offer, however, as an art, it can adhere to popular opinions. For example, one indicator often used in trading is called the Fibonacci levels. Its clever math has caught the attention of most traders, hence everyone uses it to determine their support and resistance levels after a breakout. From a scientific perspective, there's no hard core rule that directly supports the four basic levels, however, since the majority of traders use these levels to set their entries and exits, they actually work for most of the times. Many ML and AI designs will incorporate these 'human-driven' nuances in their code as does ours.

BTW, ML and AI are not necessarily related to High Frequency Trading. That's a bunch of media hype to sell stories and books. With RagingFX.com, our signals are deliberately designed with a 10min to 3 hour shelf life. No need to be glued to your screens.

Tom, I wonder if you could use AI systems (like black box neural networks) to see if they can validate human chosen models based on for instance Fibonacci trading. If in a certain period Fibonacci would work good (maybe because the concept is used a lot in trading) then independent AI systems would come up with about the same risk/resultgraph?

Machine Learning is the way forward! But you will need some good quality and comprehensive data to feed it, that's where the headache begins... Lucky for us, there's Dow Jones DNA! https://www.dowjones.com/dna/

Tom Kadala, you chose wise words. I know years ago when checked on the fib. levels time and price we came to 30% at best not even as working levels to base trade upon.

Base on what you say you are confirming my thinking the art side will be for the Human to handle the back testing , validating and the executing will be in the hands of the computers.

So far we are on the same page.

Henk Vedder, Your question is one of the Holy Grails in AI and ML right now. Currently we are writing code that matches pair and aggregate probability values for a specific pattern. We've been doing it manually now for a few weeks (along with back testing). So far it's looking pretty promising. Promising for us means above 55% success on a consistent basis. We plan to continue manual testing for another few weeks before releasing the task to our algo.

Our rational is that each pair has its primary traders and these groups tend to move prices in ways that allow us to capture their respective consistent yet subtle defining 'preferences'. One way to look at our approach is that we define cultural differences using binary. It's similar to capturing subjective thinking in binary terms, then applying probabilities to give it the sway binary analysis tends to lack. It's pretty cool to watch in action.

Very interesting, I must say this area of thought gives pleasure in system thinking that goes above the aim of earning money with the systems themselves...