General Betting
Welcome to Live View – Take the tour to learn more
Start Tour
There is currently 1 person viewing this thread.
I’ve spent the last few days moving away from traditional "Form-based" horse racing analysis and diving deep into **Market Microstructure**. Specifically, I’m building an engine in F# that ignores the horses entirely and focuses purely on the physics of the Betfair Matching Engine.
The goal is to identify "Syndicate Signatures" by analyzing the **Historic Traded Price Data** (the sequence of every match, not just the current LTP).
Here is the general architecture I’m exploring:
1. **Defining the Anchor:** Using a combination of Volume Weighted Average Price (VWAP) and Volume Profile Nodes to find where the "Fair Value" actually sits.
2. **Detection of Aggressive Capital:** Implementing algorithms to catch "Price Sweeps"—where a single participant clears 3-5+ price levels in a single 50ms matching cycle.
3. **The Probability vs. Tick Dilemma:** I've moved all my internal logic to Implied Probability (%) rather than Price Ticks. It seems to be the only way to get a consistent "Edge" that doesn't fall apart when a horse moves across the 4.0 boundary, where the tick size jumps from **0.05** to **0.10**.
4. **Initiator Inference:** Looking at the sequencing of matches within a second to determine if the "Aggressor" was a Backer (pushing the price down) or a Layer (pushing it up), and then slipstreaming behind that momentum.
The goal is to identify "Syndicate Signatures" by analyzing the **Historic Traded Price Data** (the sequence of every match, not just the current LTP).
Here is the general architecture I’m exploring:
1. **Defining the Anchor:** Using a combination of Volume Weighted Average Price (VWAP) and Volume Profile Nodes to find where the "Fair Value" actually sits.
2. **Detection of Aggressive Capital:** Implementing algorithms to catch "Price Sweeps"—where a single participant clears 3-5+ price levels in a single 50ms matching cycle.
3. **The Probability vs. Tick Dilemma:** I've moved all my internal logic to Implied Probability (%) rather than Price Ticks. It seems to be the only way to get a consistent "Edge" that doesn't fall apart when a horse moves across the 4.0 boundary, where the tick size jumps from **0.05** to **0.10**.
4. **Initiator Inference:** Looking at the sequencing of matches within a second to determine if the "Aggressor" was a Backer (pushing the price down) or a Layer (pushing it up), and then slipstreaming behind that momentum.
Show More
There's lots of internet gold that can be mined from studying the market action whilst ignoring the runners.
IMHO, it is more profitable to base an automated system on averages rather than outliers.
Looking for outliers such as "detecting aggressive capital" is going to yield far fewer opportunities per day
than analysing a market to determine it's average state and then placing bets that will be value relative to that average if they get matched.
Automated systems that rely on the rule of large numbers is the holy grail to chase. Outliers, by their very definition fall outside of the "Rule of large numbers"
Good luck with your ideas and programming.
IMHO, it is more profitable to base an automated system on averages rather than outliers.
Looking for outliers such as "detecting aggressive capital" is going to yield far fewer opportunities per day
than analysing a market to determine it's average state and then placing bets that will be value relative to that average if they get matched.
Automated systems that rely on the rule of large numbers is the holy grail to chase. Outliers, by their very definition fall outside of the "Rule of large numbers"
Good luck with your ideas and programming.
I remember, way back when, in betting shops, price movements on horses
reported over the old extel system would sometimes stand out.
A horse would jump from 7/1 to 5/1 and would be accompanied by "that show is OK"
You knew it was worth a bet, and especially if it concerned a horse in specific ownership.
I guess your system is looking for that sort of fingerprint, on
a more regular basis.
But anonymising the data you risk not picking up specific patterns
that would be gold dust.
Good luck, and maybe, with less liquidity these days your system may
find things easier to spot.
I've done some manual tracking, of certain meetings, and found
reward in early price movements
Anyway once again, good luck,
reported over the old extel system would sometimes stand out.
A horse would jump from 7/1 to 5/1 and would be accompanied by "that show is OK"
You knew it was worth a bet, and especially if it concerned a horse in specific ownership.
I guess your system is looking for that sort of fingerprint, on
a more regular basis.
But anonymising the data you risk not picking up specific patterns
that would be gold dust.
Good luck, and maybe, with less liquidity these days your system may
find things easier to spot.
I've done some manual tracking, of certain meetings, and found
reward in early price movements
Anyway once again, good luck,
I have tested two different data types: the first is raw time-series data, with time, price, and volume; the second is traditional summary data used in finance, such as candlesticks. I am not actually coding but testing an agentic approach using Gemini and Raptor Mini.
So the code is generated and I only integrate it into my existing framework; I must find a better way to integrate it, but for now, since I want LLMs to work with data directly when creating algorithms, the first integration is not so direct, because agents code on JSON data samples while my code infrastructure uses direct application domain types.
So the code is generated and I only integrate it into my existing framework; I must find a better way to integrate it, but for now, since I want LLMs to work with data directly when creating algorithms, the first integration is not so direct, because agents code on JSON data samples while my code infrastructure uses direct application domain types.
Post Your Reply
Please login to post a reply.
Wonder
Instance ID: 13539