Data Download
Select exchange, symbol, date range and multiple timeframes. Smart download: Binance monthly bulk archives first, CCXT fallback. Real-time progress via SSE.
Case Study / Quant Research
Quant Research
Platform
An end-to-end quantitative research system: data acquisition from 111+ exchanges, vectorized backtesting with real execution costs, 8 strategy families, interactive Plotly dashboard and advanced robustness analysis — all in Python.
Python 3.12+
FastAPI
CCXT · 111 Exchanges
Parquet · Vectorized
scikit-learn · LightGBM
Plotly Interactive
Project Stats
111+
Exchanges via CCXT
8
Strategy families implemented
40+
Technical & statistical features
9
Timeframes: 1m through 1d
Why this was built
From hypothesis to reliable execution
Most quant systems have a large gap between idea and execution: insufficient data, backtests without realistic costs, strategies without market regime awareness. This platform closes that gap with a fully reproducible end-to-end pipeline.
From incremental Binance data download using monthly bulk archives, to vectorized backtesting with execution delay and real costs (10 bps fee + 2 bps slippage), to an interactive Plotly dashboard — all implemented in Python with type annotations throughout.
Three-Layer Architecture
Browser Dashboard
FastAPI + SSE
↓
FastAPI Web Server
src/web/app.py
↓
Research Library
src/ modules
↓
Parquet Data Store
data/processed/*.parquet
Binance Bulk Archives
CCXT Fallback
Nobitex UDF
Incremental Merge
Core Capabilities
5 dashboard sections
The research dashboard is split into 5 independent sections, each covering one phase of the quantitative research cycle.
Data Inventory
Visual grid of all downloaded datasets with full metadata: exchange, symbol, timeframe, row count, file size and quality status.
Research & Backtesting
Multi-select datasets and strategies, configure capital/fees/slippage, run vectorized backtests with live progress tracking.
Interactive Report
Plotly charts: equity curve with market regime shading, drawdown panel, monthly return heatmap, rolling 30-bar Sharpe and a color-coded sortable metrics table.
Automatic Insights
90-day rolling analysis across all datasets with regime detection, strategy ranking and a momentum gauge.
Data Validation
Automatic checks across 6 categories: timestamp gaps, duplicates, OHLCV consistency, zero-volume bars, outliers and overall quality score.
Strategy Library
8 implemented strategy families
Each strategy is implemented as a frozen dataclass — immutable parameters, fully reproducible results. Signals are binary: Long or Cash. No leverage or short selling.
| Strategy | Type | Entry Logic | Market Sensitivity |
|---|---|---|---|
| EMA Trend | Trend | Fast EMA(20) > Slow EMA(100) | Best in trending markets; poor in choppy conditions |
| RSI Mean Reversion | Mean-Rev | RSI < 30 entry, RSI > 50 exit | Range-bound and sideways markets |
| Bollinger Band | Mean-Rev | z-score < −2σ entry, mean reversion exit | Moderate volatility, mean-reverting regimes |
| Donchian Breakout | Trend | 55-period channel breakout (Turtle Trading) | Trending markets with clear breakouts |
| ATR Breakout | Trend | Close > MA + 1.5×ATR(20) | High volatility with adaptive breakout threshold |
| MACD Crossover | Trend | Fast EMA crosses signal line | Trending markets with momentum |
| Stochastic Rev. | Mean-Rev | %K at oversold/overbought levels | Ranging markets with defined oscillation |
| ML Signal | ML | Gradient Boosting on 6 technical features | Chronological 65/35 train/test split |
Advanced Analysis
Beyond simple backtesting
The platform includes tools that eliminate look-ahead bias and measure genuine out-of-sample robustness.
Walk-Forward Validation
Rolling out-of-sample validation with configurable train/test windows. Prevents overfitting to historical data.
Monte Carlo Analysis
Bootstrap resampling of return series to estimate worst-case drawdown distributions and strategy robustness.
Regime Detection
Identifies 4 market regimes: Trending Up/Down, Ranging, High/Low-Vol. Recommends best strategy per regime.
Factor Research
Information Coefficient (IC), Rank IC, factor decay across 1–24 bar horizons, cross-factor correlation matrix.
Portfolio Analysis
Cross-dataset correlation, portfolio-level Sharpe aggregation, drawdown analysis across multiple assets.
Parameter Stability
Automatic testing of nearby parameter variations — prevents curve-fitting over narrow parameter ranges.
ML Research Baseline
Random Forest for return and volatility prediction. Strictly chronological splits, no shuffling — zero data leakage.
Feature Engineering
40+ technical and statistical features: Trend, Momentum, Volatility, Volume, Structure and Statistical categories.
Performance Metrics
Computed backtest metrics
CAGRCompound Annual Growth Rate
SharpeRisk-adjusted return
SortinoSharpe using downside deviation only
Max DDMaximum drawdown depth
CalmarCAGR divided by Max Drawdown
Win RateFraction of profitable trades
Profit FactorGross profit / gross loss
1-bar DelayExecution lag for realism
10bp + 2bpFee + slippage per side
Technology Stack
Production-grade dependencies
Every component was chosen for production standards: no unnecessary dependencies, full type annotations throughout.
Python 3.12+
FastAPI 0.136
pandas 2.2+
numpy 1.26+
CCXT 4.2+
pyarrow (Parquet)
Plotly 5.19+
scikit-learn 1.4+
LightGBM 4.3+
XGBoost 2.0+
scipy 1.12+
Typer CLI
Design Principles
Architecture decisions
Full Reproducibility
Given identical Parquet inputs, all backtests and reports are fully deterministic — no hidden randomness.
Zero Data Leakage
Chronological train/test splits, never shuffled — look-ahead bias is eliminated by design.
Frozen Dataclasses
Immutable strategy parameter containers — parameter sweep experiments are fully traceable.
Async Job Management
Thread-based background jobs with SSE for real-time progress — no Redis or Celery required.
Open source — MIT License
This project is publicly available on GitHub. Full documentation, modular architecture and a CLI for batch processing — ready to use and extend.
MIT License
Python Type Annotations
pytest test suite
ruff + mypy
CLI + Web API
demo mode (synthetic data)
GitHub: Quant_research
git clone https://github.com/0xh0551/Quant_research
Project Lesson
A quant system starts with a hypothesis, not code
Most quant projects get lost in scattered scripts. This platform demonstrates that when the pipeline from data to report is designed with a clear architecture, trading decisions are grounded in evidence rather than intuition.
Walk-forward validation and Monte Carlo analysis answer the critical question: does this strategy actually work, or is it just curve-fitted to historical data?
Three core principles
Data quality first — validated across 6 layers before any analysis
Real execution costs — fees + slippage + 1-bar delay
Mandatory robustness — walk-forward and Monte Carlo before any deployment decision
Want to build a quant system for your work?
If you need a data pipeline, reliable backtesting, risk management or a research dashboard, start with an architecture diagnosis session.