Predatory High Frequency Trading Machine Learning Thread

[Esterincertis]

Hi, like I said on my personal trading journal found here: (https://www.cliqforex.com/trading-sy...g-journal.html) I am creating a new stat-arb egy. I'll be focusing on finding patterns in tick volume in order to track and predict large flows of capital between currencies. I believed it would be suitable to form a discussion on predatory high frequency trading. I myself am not a market maker/scalper, as a result of broker I use, but if you're a market maker/scalper don't hesitate to discuss the way your trading is going/provide personal consciousness.

I am exploring the use of neural networks when finding patterns in data. I am currently sampling and modeling data on 5m pubs, but trading on 333ms increments. The reason for rounding to 5m pubs is computational. It only requires to computing that is much to do tick data. 1 day I hope to try tick data, but it is likely I will need to rent a cloud established vm. I have used cheap vms in the past to scrape and record encode info, but it can require 20-60 secs to backtest just 16 hours of tick data. That makes it impractical to networks.

I'm performing some visualizations of the data as it investigates, therefore I can find a clearer idea about what a few of the connections plot out looking like. Here is an example.

[Esterincertis]

Added a survey

[Esterincertis]

I am finding success in my egy researching ai using order flow as a data input signal.


Is anyone else finding success also? If so let us trade insights

[Rororo93]

What's your objective? To see if a run of unlimited regressions can do a better job than you or many people at trading? Obviously. The human head is a series of regressions.

Or can it be to see how much you are able to make from letting a computer do the dirty work of figuring things out because you do not understand how to do it yourself?

If you knew how to think like a computer then you would not be wasting your time on this.

[Esterincertis]

what's your objective? To see if a series of unlimited regressions can do a better job than you or most people at trading? Obviously. Or is it to see just how much you are able to create out of letting a computer do the dirty work of figuring things out because you do not know how to do it yourself? If you knew how to think like a computer then you wouldn't be wasting your own time on this.

Great question, my purpose is to build a time series forecasting algorithm. I enter data I think may be applicable (in this scenario order flow) and the machine finds an extremely high number of signals using a high positive/negative correlation to the asset price. These signals are combined to a linear factor model and estimate expected value by finding bizarre patterns in quantity. However if I guess for example google search tendencies can predict commodity price, I can shoot connections in search results, and price assets with them.

Now if you have a look at my commerce explorer, then I developed a macro egy which I came up with manually (the hard way). It's mildly effective with a sharpe ratio ~1.25 on percentage yields. I used 3 macro indiors, forecasting them utilizing auto regression models and thus forecast future exchange rates.

This computer software is quite arguably far far stupider than most people, it is intended to perform a few calculations, and consequently can perform them considerably quicker. It invents 3-4 egies a second, and then examines them in OOS (from sample information) to make certain that it isn't cherry picked.

[Rororo93]

quote Good question, my objective is to create a time series forecasting algorithm. I input data I think could be applicable (in this case order flow) along with also the machine finds a very large number of signals with a high positive/negative correlation to the asset price. These signals are combined to a linear factor model and estimate expected value by discovering bizarre patterns in quantity. But if I guess for example google search tendencies can predict commodity price, I can take relationships in search results, and price assets with them. If you...

Well, then, you are likely to need a lot more hardware plus quite a few quantum computers. A minimum of one. You'd need an AIG to conduct the process... and cash wouldn't matter anymore at that point. The moment a single human being makes a single irrational decision (buy high, sell low) it begins to fudge the information. When you begin considering a massive group of irrational investors that the information gets very fuzzy. Sure; you may discover statistical correlations and define a few collection of outputs with reasonable probabilities but you need to continuously re-define the parameters and translate input that only an AIG is effective at analyzing.

Basically; nothing short of an AIG can translate black swan events because they kind of violate all present models. If you've got a computer that prediction Trump's presidency and the market's response then you understand what you need to do... destroy it. Immediately. It kills us all.

[Esterincertis]

quote Well, then, you're going to require a lot more hardware plus quite a few quantum computers. A minimum of one. You'd want an AIG to conduct the process... and money wouldn't matter anymore at that point. The second a single human being makes a single irrational choice (buy high, sell low) it begins to fudge the data. When you begin contemplating a huge group of irrational investors that the data gets very fuzzy. Sure; you may discover statistical correlations and define a few collection of outputs with decent probabilities but you need to constantly...

LOL AT TRUMP.

Sentiment is a element that has been measured. Lots of organizations concentrate on mining for sentiment over a companies favlity with consumers, and these very same services can be applied to FX, commodities, etc.. I composed an EA for a client who employed bitcoin sentiment measured via google search results that discovered a means of predicting black swan/6 sigma occasions. He made 200% weekly intraday stat-arb, Just regression models. Have a look at my YT station, I've a movie showing a number of the research I did.

Also, do you have any highschool/college instruction in statistics or computer science? High school doesn't count if it is not AP or IB.

[Rororo93]

quote LOL AT TRUMP. Sentiment is. Lots of organizations concentrate on mining for sentiment over a businesses favlity with customers, and these very same services can be implemented to FX, commodities, etc.. I composed an EA for a client who employed bitcoin sentiment measured via google search results that discovered a method of calling black swan/6 sigma events. He created 200% weekly intraday stat-arb, Just regression models. Have a look at my YT channel, I have a video showing some of this research I did for him. Also, do you have...

College for programming(C ). My highschool sweetheart moved into Bryant University at the United States for Actuarial Mathematics and graduated only shy of Suma Cum Laude. I was still helping her motive her way through several missions before she graduated(I had a passion for physics).

I worked through this justifiion for over a decade for the explicit purpose of doing all you have spoken of BEFORE I put one trade. The issue came to light when I began studying plogy. The answer came from quantum mechanics. In the end of the day we are still bound by time. If, for instance, 911 were to happen again then what's the sentiment? You can't obtain accurate data for black swans until they occur. Some? Sure. But most are freak occurances that result from people being assinine. Quantum entanglement's a real bitch. There is a point where an AI can just perceive so much statistical precision. Nothing is omniscient.

My 'schooling' is minimal. I always preferred believing and assimilating useful information over regurgitating lectures.

My question is; if you're already profiting then why not just go work at Two Sigma? It is probably a lot more entertaining(or boring depending upon your view) to trade alongside other quants.

The purpose being that you can refine the process indefinitely but the dimension of time will always invalidate any computer's capacity to accurately simulate the universe and it's impact on individual plogy. The butterfly effect is entirely combined with chaos theory.

In the event that you actually need the answer then it's 42.

[staticbangs]

... If you knew how to think like a computer then you would not be wasting your own time on this.

Good line.

However, no comment on topic either way.

[Esterincertis]

quote College for programming(C ). My highschool sweetheart went to Bryant University in the US to get Actuarial Mathematics and graduated only shy of Suma Cum Laude. I was helping her reason her way through some assignments until she graduated(I had a passion for physics). I worked through all this justifiion for more than a decade for the explicit purpose of doing everything you have talked of BEFORE I put a single trade. When I started studying plogy, the issue came to light. The answer came from quantum mechanics. At the end of the afternoon...

I can't work at two sigma since Iam seventeen and haven't finished highschool. As for black swans, the predatory egies find 30-40 opportunities per month each pattern recognition model. You dont need to be right all of the time together with neural networks, that I measure the signal against price for a correlation coefficient. Thus the versions presume some mistake. The pc jus keeps detecting better and better versions. It probably won't be ideal, and doesnt need to be. Just good enough to net a profit.