Iceberg Order Detection Algorithm
Budget: €30 – €250 EUR
I need an algorithm that can sift through live market data feeds and flag potential iceberg orders in real-time stock prices. The incoming stream will be tick-by-tick quotes and trades (أسعار الأسهم – الأسعار اللحظية). No images are involved; the focus is exclusively on numerical market data delivered via a low-latency API.
Scope
• Consume and parse the live feed I provide (FIX/FAST or WebSocket; final choice is yours as long as latency stays minimal).
• Identify hidden-size (iceberg) patterns by analysing order book depth, trade print sequences, and quote replenishment behaviour.
• Produce a confidence score plus the key supporting metrics for each detection event.
• Output should arrive within milliseconds so the signals can be routed straight to my downstream execution logic.
Preferred Stack
Python or C++ with efficient data structures, Pandas / NumPy for rapid prototyping, then numba, cython or plain C++ for speed. If you favour another high-performance setup—Rust, Go, Java—state it and justify the expected latency.
Deliverables
1. Well-documented source code with unit tests
2. A short benchmark report on sample replay data showing throughput and detection accuracy
3. Setup instructions to run on a Linux server
Acceptance Criteria
• Handles at least 100,000 messages per second on a single core.
• Flags at least 90 % of synthetic iceberg patterns in the supplied test set with fewer than 5 % false positives.
Feel free to suggest additional statistical or machine-learning techniques if they improve precision without sacrificing speed.
Scope
• Consume and parse the live feed I provide (FIX/FAST or WebSocket; final choice is yours as long as latency stays minimal).
• Identify hidden-size (iceberg) patterns by analysing order book depth, trade print sequences, and quote replenishment behaviour.
• Produce a confidence score plus the key supporting metrics for each detection event.
• Output should arrive within milliseconds so the signals can be routed straight to my downstream execution logic.
Preferred Stack
Python or C++ with efficient data structures, Pandas / NumPy for rapid prototyping, then numba, cython or plain C++ for speed. If you favour another high-performance setup—Rust, Go, Java—state it and justify the expected latency.
Deliverables
1. Well-documented source code with unit tests
2. A short benchmark report on sample replay data showing throughput and detection accuracy
3. Setup instructions to run on a Linux server
Acceptance Criteria
• Handles at least 100,000 messages per second on a single core.
• Flags at least 90 % of synthetic iceberg patterns in the supplied test set with fewer than 5 % false positives.
Feel free to suggest additional statistical or machine-learning techniques if they improve precision without sacrificing speed.