Quantitative Trading | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Classic Quantative Hedge Fund Method 1 reply

Day Trading vs. Swing Trading vs. Position Trading 86 replies

Attachments: Quantitative Trading
Exit Attachments

Quantitative Trading

  • #21
  • May 27, 2026 3:25pm May 27, 2026 3:25pm
Quoting triphop
same here, I'm all over it. The differential is basically the training data- quant is a known discipline so it draws on the best of academic research and spits out damn near perfect specs. Decent trading strats - especially quant ones - are not generally published so it's got nothing to go on. it'll change but for now, it's down to us.
That is exactly it. Nobody is publishing their proprietary strategies online to be included in training data and shared with their competitors. Even at my level I wouldn't publish my strats, I mean I wouldn't even share them with friends or family.
  • #22
  • May 27, 2026 3:45pm May 27, 2026 3:45pm
Quoting Guy.Baker
{quote} That is exactly it. Nobody is publishing their proprietary strategies online to be included in training data and shared with their competitors. Even at my level I wouldn't publish my strats, I mean I wouldn't even share them with friends or family.
100%. The only thing that's genuinely surprised me though is that all this stuff - vol, gating, regime etc, it's proper gnarly to do for sure. But it's as important if not more so than strategy. Which is better - a breakout trading every breakout every morning, or one where there's 30% (real stat from my regimes) increased chance of continuation than reversal? All adds up. And it's not proprietary. And now we can all have a bash.
  • #23
  • May 29, 2026 3:08am May 29, 2026 3:08am
Quoting triphop
{quote} 100%. The only thing that's genuinely surprised me though is that all this stuff - vol, gating, regime etc, it's proper gnarly to do for sure. But it's as important if not more so than strategy. Which is better - a breakout trading every breakout every morning, or one where there's 30% (real stat from my regimes) increased chance of continuation than reversal? All adds up. And it's not proprietary. And now we can all have a bash.

It is fascinating stuff, and the concept of using AI filters for gating is very interesting,I really must find the time and the mood to look into it properly.

What is your background by the way?
  • #24
  • May 29, 2026 3:48am May 29, 2026 3:48am
Very cool thread , its a very interesting area -


Im building a newer quant model - its based on some work originally by jim simmons and extended to newer discovery of analysis.

I’ve been developing a new approach to market analysis that looks at price movement through what I call Mathematical Encapsulation and Distortion Mapping. The basic idea is that markets don’t just move in straight technical patterns — they compress, expand, distort, and re-balance around hidden mathematical zones. My model attempts to quantify this by measuring the “distortion factor” between price, time, volatility, and directional pressure, then encapsulating those movements into repeatable market structures. It’s still early-stage, but the goal is to build a more advanced quant-style framework that can identify when a market is stretched, coiled, or preparing for a timing-based shift. I’ll be sharing some observations, charts, and forward-looking analysis here, mainly across forex and major liquid markets. Not financial advice — just a new research framework and i hope worth watching as i post -

i've been offline for 20 odd years developing this

Interesting point - AI doesn't know how to deal with the math i use nor can it help me build it ( so in my opinion AI has a long way to go in real world coding at this level ) .
  • #25
  • May 29, 2026 4:32am May 29, 2026 4:32am
Quoting Guy.Baker
{quote} It is fascinating stuff, and the concept of using AI filters for gating is very interesting,I really must find the time and the mood to look into it properly. What is your background by the way?
Many fields prior to trading but mainly consultancy. Full time trading around 15 years I guess.
  • #26
  • May 29, 2026 4:38am May 29, 2026 4:38am
Quoting ShadowKeeper
Very cool thread , its a very interesting area - Im building a newer quant model - its based on some work originally by jim simmons and extended to newer discovery of analysis. I’ve been developing a new approach to market analysis that looks at price movement through what I call Mathematical Encapsulation and Distortion Mapping. The basic idea is that markets don’t just move in straight technical patterns — they compress, expand, distort, and re-balance around hidden mathematical zones. My model attempts to quantify this by measuring the “distortion...
Great stuff and an excellent first post in FTI, we need more of that in here to drown out the dilberts
I've been doing similar, also for 20 years or so so you really are the shadow keeper. Most of it was based on some heavy esoteria, I've only recently started to dovetail it with traditional quant. Adding vol in is so ridiculously obvious to trading, it's madness I was so late to it.

One thing I've noticed - past couple of days noticed - is that mapping deceleration and acceleration phases onto trends give you future swing points. Basically markers within a trend for future S&R

anyways look forward to your posts
  • #27
  • May 29, 2026 4:44am May 29, 2026 4:44am
Quoting ShadowKeeper
Very cool thread , its a very interesting area - Im building a newer quant model - its based on some work originally by jim simmons and extended to newer discovery of analysis. I’ve been developing a new approach to market analysis that looks at price movement through what I call Mathematical Encapsulation and Distortion Mapping. The basic idea is that markets don’t just move in straight technical patterns — they compress, expand, distort, and re-balance around hidden mathematical zones. My model attempts to quantify this by measuring the “distortion...

Very cool stuff, I suspect your math skills are in a different universe to mine, I'm a software engineer of many years, primarily office automation systems and a couple of internet startups, I have a reasonable level of probability theory understanding which is essential in trading of course. Which LLM do you prefer? I am a massive fan of Deepseek in Deep thinking mode, incredible really, it has helped me develop stuff that I could conceptualise but would have taken months to code and test manually.

I am very interested in reading about your work and I hope we can grow this thread into something useful, and attract more like minded people into our conversations.
  • #28
  • May 29, 2026 5:49am May 29, 2026 5:49am
I’m also heavily investigating AI as part of the ongoing development process.

I like Claude for reasoning and structure, and DeepSeek looks promising in certain areas, but I’m still cautious about trusting any single model too much, especially when proprietary code and trading logic are involved. At the moment I’m running my own hybrid setup using several Qwen-based models, Claude-assisted reasoning, and self-contained machine learning components. The goal is to keep the core logic private, controlled, and independent, while still using AI to help test, refine, and expand the research framework. For me, AI is not the system itself — it’s a tool around the system, helping accelerate the analysis without exposing the underlying edge.
  • #29
  • May 29, 2026 8:07am May 29, 2026 8:07am
Quoting ShadowKeeper
I’m also heavily investigating AI as part of the ongoing development process. I like Claude for reasoning and structure, and DeepSeek looks promising in certain areas, but I’m still cautious about trusting any single model too much, especially when proprietary code and trading logic are involved. At the moment I’m running my own hybrid setup using several Qwen-based models, Claude-assisted reasoning, and self-contained machine learning components. The goal is to keep the core logic private, controlled, and independent, while still using AI to help...
I understand your reticence, I've only used the LLMs to develop infrastructure projects so no "secret sauce" is disclosed, I do think it is probably important to code the key ingredients manually
  • #30
  • May 30, 2026 7:34am May 30, 2026 7:34am
I agree that AI is better as a coding and research assistant than as the trader itself. In my testing, the useful part is letting it help structure data and filters, then the execution rules must still be simple and fully controlled
  • #31
  • Edited 9:22am May 30, 2026 8:58am | Edited 9:22am
Quoting pizd0s
I agree that AI is better as a coding and research assistant than as the trader itself. In my testing, the useful part is letting it help structure data and filters, then the execution rules must still be simple and fully controlled

as touched on above, LLMs are essentially reformatting known information, they are not creative, I'd argue most people aren't either so I have no problem believing that these technologies will replace a large percentage of the workforce. Successful traders are simply not publishing their trading secrets, because if they did they might get killed, LLMs have no data on good trading strategies to work on, and if they did, almost by definition those strategies would be now worthless.
  • #32
  • May 30, 2026 9:58pm May 30, 2026 9:58pm
PART A Intro to Quant config


I’m currently working on what I call the V5 — the five highest-volatility markets I’m testing for my quant/algo system.
My current primary test markets are:
BTC
Nasdaq / NAS
DAX
GBPJPY
Gold
Secondary markets I’m also watching/testing:
USDCAD
AUDUSD
ES / E-mini
EURUSD
Wheat
Crude Oil
I’m interested to hear what others here class as the best markets or the hardest markets to model.
Strangely, I’m finding that higher-volatility markets are giving me better accuracy once the model locks in. That surprised me at first, but it’s also been one of the most exciting parts of the research.
The system is built around a few core model layers:
1. Time Model — Date Pivot Turn
This identifies key days where a market is more likely to turn, accelerate, or set up a directional trading day.
2. Price Model D — Daily Hit Zone
This projects the daily price zone where the market is expected to react or reach.
3. Price Model DH — Speed / Daily Distortion Measurement
This measures how much the market can distort away from the expected price path on the day.
4. Price Model I — Hour / Intraday Capture
This is the intraday layer used to refine the actual entry point and price capture.
The idea behind the time model is what I call Time Units of Encapsulation. In simple terms, it creates a defined window of time where distortion is allowed to occur before the market either confirms or invalidates the setup.
My working rule is:
The daily time model should be no more than one day out from the projected pivot turn.
The price model should be no more than the allowed Distortion Factor for that market.
BTC was the hardest market by far. Initially, the distortion was huge — around 400 points/pips per move with almost a two-day encapsulation window. I spent nearly two years just on BTC, and at one point it delayed the whole process because I thought the model wasn’t working properly.
A close friend even told me to try it on a “non-traditional” market because, in his view, mathematics can’t properly solve digital markets. That frustrated me enough that I wanted to prove him wrong. Haha.
Over time, I managed to reduce the BTC distortion factor from extremely wide levels down to around 10 or below, and occasionally the model hits almost perfectly, which was a big milestone.
Current working distortion tolerances:
BTC: less than 10
Nasdaq: less than 2 points
DAX: less than 2 points
GBPJPY: less than 5 pips
Gold: less than 1.5 points
This then led me into a different way of thinking about risk.
The standard trading mindset is usually: risk 1–2% per trade, aim for 3R or 4R, and manage exposure carefully. That makes sense for most strategies.
But I started asking a different question:
If the distortion factor is reduced close to the perfect entry point, should risk still be fixed at the same level?
For example, if a setup has a much tighter distortion allowance and much stronger model confirmation, does it deserve the same risk as a weaker setup? Probably not.
That led me into building a dynamic risk model based on precision, distortion, and leverage allocation.
I’m currently calling this idea:
RIPLE — Risk Into Precision Leverage Engine
The basic idea is that risk is not just based on account percentage. It is based on how precise the model is at the point of entry, how low the distortion factor is, and how strong the time/price alignment is.
So instead of treating every trade equally, the model can decide when to hit hard, when to scale lightly, and when to stay out completely.
I’ll post Part 2 later, but I’d be interested in feedback.

Which markets do you think are the best or hardest to model?
Do you find higher volatility improves or reduces accuracy in your own systems?
  • #33
  • Jun 1, 2026 2:54am Jun 1, 2026 2:54am
Let's see some charts "BTC less than 10 pips man" - genuinely curious
  • #34
  • Edited 3:14am Jun 2, 2026 2:43am | Edited 3:14am
Quoting ShadowKeeper
PART A Intro to Quant config I’m currently working on what I call the V5 — the five highest-volatility markets I’m testing for my quant/algo system. My current primary test markets are: BTC Nasdaq / NAS DAX GBPJPY Gold Secondary markets I’m also watching/testing: USDCAD AUDUSD ES / E-mini EURUSD Wheat Crude Oil I’m interested to hear what others here class as the best markets or the hardest markets to model. Strangely, I’m finding that higher-volatility markets are giving me better accuracy once the model locks in. That surprised me at first, but...

This all looks very interesting indeed, are you able to show examples of any of your tests so far?


On your risk estimate I think this varies at lot depending on the tolerance of the individual trader, personally I'm only using 0.45% per trade currently, but I do like your idea of assigning risk based on the probability of the trade's success.
  • #35
  • Jun 2, 6:32am (47 hr ago) Jun 2, 6:32am (47 hr ago)
Quant trading sounds clean, but the hard part is still proving the edge survives real spreads, slippage, regime changes, and enough sample size. Automation helps with consistency, but bad assumptions just get automated faster. The model matters, but so does the testing discipline behind it
  • #36
  • Jun 2, 6:36am (47 hr ago) Jun 2, 6:36am (47 hr ago)
Quoting pizd0s
Quant trading sounds clean, but the hard part is still proving the edge survives real spreads, slippage, regime changes, and enough sample size. Automation helps with consistency, but bad assumptions just get automated faster. The model matters, but so does the testing discipline behind it

Surely that would apply to any approach to finding an edge in the market ?
  • #37
  • Jun 2, 6:44am (47 hr ago) Jun 2, 6:44am (47 hr ago)
That’s a fair question, and I probably need to explain the distinction better.

The quant side of what I’m building is not simply about taking more risk because I “think” the trade is right. It is about trying to reduce the uncertainty around entry, timing, price distortion, and market structure. If the model can bring the distortion factor down and define the entry area with greater precision, then the risk profile changes. In that case, leverage is not being used randomly — it is being allocated based on measured precision, timing alignment, and the expected distortion allowance for that specific market.

For me, the goal is better accuracy first, then better risk allocation second. Higher leverage only makes sense if the entry location, stop logic, and invalidation point are extremely clear. Otherwise, leverage becomes dangerous very quickly.
On the automation side, I’m not trying to build a fully automated system that makes every decision for me. My focus is more on making entries, exits, alerts, slippage control, and trade management simpler and cleaner. Automation helps remove delay, emotion, and execution errors, but I still believe logic and human oversight are required.

No one can realistically program every possible market event, news shock, liquidity issue, or structural change. I’ve seen enough from previous automated systems and fund-style environments to know that full automation can become a risk in itself if it is trusted blindly.
So the system is more of a hybrid approach: quant modelling for precision, automation for cleaner execution, and human control for final judgement and risk oversight.

hope that helps
  • #38
  • Jun 2, 7:14am (46 hr ago) Jun 2, 7:14am (46 hr ago)
Quoting Guy.Baker
{quote} This all looks very interesting indeed, are you able to show examples of any of your tests so far? On your risk estimate I think this varies at lot depending on the tolerance of the individual trader, personally I'm only using 0.45% per trade currently, but I do like your idea of assigning risk based on the probability of the trade's success.

Just to clarify where I’m at with the system.
There are several models I’m currently finishing and testing. The core structure is built around multiple timing and pricing layers working together, rather than one single signal.

At the moment, the framework includes:

1. Timing Model — short-term daily
This identifies key daily pivot windows and potential turn dates.
2. Swing Timing Model
This loads into the short-term daily model and helps define the broader swing structure behind the move.
3. Additional timing layer
This helps confirm whether the daily and swing timing models are aligned.

These timing models then connect into separate price models, which are designed to project where the market should react, hit, distort, or reverse.
So in simple terms, it is three timing-based models linking into three price-based models.

I’m being a little careful about posting the actual pricing model at this stage, because some of the outputs are honestly surprising, even to me. I understand people probably won’t believe it properly until it is shown live, verified, or tracked in real time, and that is fair.

For now, I’m mainly sharing the concept and structure behind the output, not claiming perfection or asking anyone to blindly accept it.
But I do want to be clear: this is not something I have made up after the fact or randomly added onto charts. These are repeatable, calculated mathematical points that can be tested, measured, and compared across markets.

I’m not asking people to believe the model — I’m working toward making the output measurable enough that belief becomes irrelevant.
  • #39
  • Jun 2, 8:19am (45 hr ago) Jun 2, 8:19am (45 hr ago)
Quoting m4ni4c
Let's see some charts "BTC less than 10 pips man" - genuinely curious
Note - The dax ( X ) - did not give me a SELL point here in timing - failure (but the trend model swing kept in the same direction)

Overall, the model is working toward one main objective: capturing key timing points with greater consistency and precision. To give some context, these calculations used to take me weeks to work through manually. The mathematics behind it is complicated, heavy, and at times extremely onerous, but the more I automate and structure the process, the more valuable the output becomes. The goal is not just to find a trade, but to compress weeks of mathematical timing work into a repeatable framework that can highlight when a market is approaching an important turning point or high-probability reaction window.

It has taken a long time to get here, but I genuinely believe it will be worth it.
Attached Image(s) (click to enlarge)
Click to Enlarge Name: Screenshot 2026-06-02 211929.png Size: 26 KB
Click to Enlarge Name: Screenshot 2026-06-02 212953.png Size: 29 KB
Click to Enlarge Name: Screenshot 2026-06-02 213039.png Size: 34 KB
  • #40
  • Last Post: Jun 4, 3:47am (2 hr 26 min ago) Jun 4, 3:47am (2 hr 26 min ago)
Quoting ShadowKeeper
{quote} Just to clarify where I’m at with the system. There are several models I’m currently finishing and testing. The core structure is built around multiple timing and pricing layers working together, rather than one single signal. At the moment, the framework includes: 1. Timing Model — short-term daily This identifies key daily pivot windows and potential turn dates. 2. Swing Timing Model This loads into the short-term daily model and helps define the broader swing structure behind the move. 3. Additional timing layer This helps confirm whether...
For sure seeing a forward test on a demo account would be useful, anything based on historical data is subject to potential coding or curve fitting issues
0 traders viewing now