This project implements a complete research-grade pipeline for medeling high-frequency trade arrivals using the hawkes process, and compares it rigorously against a standard poisson process.
The goal is to quantify and analyze self-excitation, clustering , and endogenous market dynamics in tick-level financial data.
Financial trades do not arrive independently. Market activity tends to cluster - bursts of trades trigger further trades.
A poisson process assumes independent arrivals. A Hawkes process models self-excitation::
- Implements Hawkes simulation from scratch(Ogata's thinnin algorihtm).
- Derives and codes the full log-likelihood function
- Performs MLE parameter estimation
- Compares Hawkes vs Poisson statistically.
- Analyses branching ratio and stability.
- Provides an interactive dashboard for visualization.
- Temporal point processes
- Self-excitation intensity models
- Log-likelihood derivaiton
- Stabilty condition : $ n = α/β < 1 $1 $$
- Branching proces interpretation
- A|C model comaparison
- Exponential inter-arrival simulation
- Analytical likelihood -Distrubution validationn
- Ogata's thinnin algorithm
- Intensity tracking over time
- Event clustering visualization
- Stabilty Validation
- Efficient log-liklihood implementation
- Constrained optimization
- Parameter recovery testion on synthetic data -Likelihood surface analysiss
- Tick-Level timestamp preprocessing
- Poisson vs Hawkes log-likelihood comparison
- A|C- Based model selection
- Branching ratio interpretation
- Simualation panel (μ, α, β controls)
- CSV upload for fitting
- Likelihood comparison output
- Interactive intensity visualization
- Python
- NumPy
- SciPy
- Pandas
- Matplotlib
- FastAPI
- React / Next.js (TBD)
- plotly.js
- Minimal quant-style UI
- Strong Understanding of Stochastic Processes
- Practical implementation of MLE under constraints
- Efficient numerical computation
- Model comparison rigor
- Clean mathematical documentation
- Financial market microstructure
- Poisson Process : Explains what exactly is poisso process , inter-arrival distribution , intensity definition , why memoryless property fails for trade data Along with all of these , it also has the explaination to why Poisson assumes independence and why does trade center clusterss
Hardik Runwal
MIT , Manipall
.png){kind=link}
