Research notebook
Finding signal in tick data.
The market is predictable tick by tick and unpredictable an hour later. My research is about what survives in between: small neural networks that read the raw order flow and hold a signal for seconds to minutes. Every note names its data, its test, and its result.
New to this? Start with the free course →Latest notebook entrySample draft
How long should a signal remember?
A proposed comparison of short and long memory windows using the same input signal.
4 notes
Experiments
Large-sample tests on candidate features, reported the same way every time, failures included.
What happens after a burst of trades?
A study outline for separating a sudden increase in trading activity from a directional prediction.
When two signals tell the same story
A proposed redundancy check before adding another feature to a research model.
Four ways to measure trade aggression, and three were the same thing
Queue-normalised size, sweep depth, consumed fraction, through-ticks. Only one is its own axis.
Testing four raw order-book measurements: three passed
Three passed. One dropped. The data picked the seeds.
3 notes
Memory
Turning short-lived market events into a signal that lasts long enough to trade.
How long should a signal remember?
A proposed comparison of short and long memory windows using the same input signal.
Does signal memory change when the market gets busy?
A proposed check of the same memory rule across different levels of trading activity.
Memory in microstructure: how discrete events become a signal
Raw ticks are noise. Memory is the bridge to a backtestable alpha.