Multi-Asset Volatility Regime Model
Budget: ₹12,500 – ₹37,500 INR
Quant Developer — Volatility Forecasting & Regime Detection Model (Systematic Trading)
Project Overview
We run a systematic trading system with multiple signal sleeves, including trend-following and mean-reversion components. We're looking for a quantitative developer/researcher to build a shared volatility and regime-detection layer that feeds two things:
Position sizing — via a volatility forecast used to scale exposure.
Dynamic sleeve weighting — via a regime classifier that adjusts the relative allocation between our trend and mean-reversion sleeves based on the current market regime.
This is not a new alpha signal — it's infrastructure that sits underneath our existing signals and directly addresses a known risk: trend and mean-reversion sleeves tend to become dangerously correlated during strong trending regimes. We need a systematic way to detect that condition and de-weight accordingly, rather than running both sleeves at static weights.
Scope of Work
1. Volatility Forecasting Module
Implement and backtest a volatility estimator for use in position sizing. Open to your recommendation on approach, but candidates include:
Realized volatility (rolling, various windows)
EWMA (RiskMetrics-style)
GARCH-family models (GARCH, EGARCH, GJR-GARCH)
Compare methods on forecast accuracy and stability, not just in-sample fit.
Output: a clean, reusable vol forecast series per instrument, with clear documentation on lookback/parameter choices and how it should be consumed downstream (e.g., target-vol position sizing formula).
2. Regime Detection Module
Build a model that classifies market conditions into distinct regimes — at minimum: trending, mean-reverting, and choppy/low-signal.
Open to methodology — options include (but aren't limited to) Hidden Markov Models, threshold/rule-based classifiers on trend-strength and autocorrelation metrics, or ML-based classifiers.
The core deliverable is a dynamic weighting scheme: a mechanism that adjusts trend vs. mean-reversion sleeve weights based on the current regime classification, rather than static 50/50 (or fixed) weights.
Must explicitly handle the sleeve-correlation problem: identify when trend and mean-reversion signals are likely to become highly correlated (typically strong trend regimes) and produce a de-weighting signal in advance or in real time.
3. Integration & Validation
Backtest the combined system (existing sleeves + dynamic weighting) against the static-weight baseline, with clear before/after performance and risk metrics (Sharpe, max drawdown, sleeve correlation over time, turnover impact from re-weighting).
Document assumptions, parameter sensitivity, and known limitations.
Code should be modular enough to swap in alternative vol/regime methods later.
Deliverables
Well-documented, tested code (Python preferred) for both modules
Backtest results and comparison report (static weights vs. dynamic weighting)
Written documentation covering methodology, parameter choices, and how to run/maintain the modules
Brief handoff walkthrough (call or written) covering how to extend or retrain the models
Ideal Candidate
Strong background in quantitative finance / systematic trading, ideally with direct experience in volatility modeling (GARCH/EWMA) and regime-switching models (HMM or similar)
Comfortable working with time series data, backtesting frameworks, and futures/equities market data
Prior experience with multi-strategy or multi-sleeve portfolio construction is a strong plus
Python proficiency (pandas, numpy, statsmodels, arch, hmmlearn or similar)
Able to explain modeling tradeoffs clearly, not just deliver a black box
What We'll Provide
Historical price/return data for relevant instruments
Existing signal outputs for the trend and mean-reversion sleeves (for correlation analysis and weighting integration)
Access for Q&A on system architecture as needed
Project Overview
We run a systematic trading system with multiple signal sleeves, including trend-following and mean-reversion components. We're looking for a quantitative developer/researcher to build a shared volatility and regime-detection layer that feeds two things:
Position sizing — via a volatility forecast used to scale exposure.
Dynamic sleeve weighting — via a regime classifier that adjusts the relative allocation between our trend and mean-reversion sleeves based on the current market regime.
This is not a new alpha signal — it's infrastructure that sits underneath our existing signals and directly addresses a known risk: trend and mean-reversion sleeves tend to become dangerously correlated during strong trending regimes. We need a systematic way to detect that condition and de-weight accordingly, rather than running both sleeves at static weights.
Scope of Work
1. Volatility Forecasting Module
Implement and backtest a volatility estimator for use in position sizing. Open to your recommendation on approach, but candidates include:
Realized volatility (rolling, various windows)
EWMA (RiskMetrics-style)
GARCH-family models (GARCH, EGARCH, GJR-GARCH)
Compare methods on forecast accuracy and stability, not just in-sample fit.
Output: a clean, reusable vol forecast series per instrument, with clear documentation on lookback/parameter choices and how it should be consumed downstream (e.g., target-vol position sizing formula).
2. Regime Detection Module
Build a model that classifies market conditions into distinct regimes — at minimum: trending, mean-reverting, and choppy/low-signal.
Open to methodology — options include (but aren't limited to) Hidden Markov Models, threshold/rule-based classifiers on trend-strength and autocorrelation metrics, or ML-based classifiers.
The core deliverable is a dynamic weighting scheme: a mechanism that adjusts trend vs. mean-reversion sleeve weights based on the current regime classification, rather than static 50/50 (or fixed) weights.
Must explicitly handle the sleeve-correlation problem: identify when trend and mean-reversion signals are likely to become highly correlated (typically strong trend regimes) and produce a de-weighting signal in advance or in real time.
3. Integration & Validation
Backtest the combined system (existing sleeves + dynamic weighting) against the static-weight baseline, with clear before/after performance and risk metrics (Sharpe, max drawdown, sleeve correlation over time, turnover impact from re-weighting).
Document assumptions, parameter sensitivity, and known limitations.
Code should be modular enough to swap in alternative vol/regime methods later.
Deliverables
Well-documented, tested code (Python preferred) for both modules
Backtest results and comparison report (static weights vs. dynamic weighting)
Written documentation covering methodology, parameter choices, and how to run/maintain the modules
Brief handoff walkthrough (call or written) covering how to extend or retrain the models
Ideal Candidate
Strong background in quantitative finance / systematic trading, ideally with direct experience in volatility modeling (GARCH/EWMA) and regime-switching models (HMM or similar)
Comfortable working with time series data, backtesting frameworks, and futures/equities market data
Prior experience with multi-strategy or multi-sleeve portfolio construction is a strong plus
Python proficiency (pandas, numpy, statsmodels, arch, hmmlearn or similar)
Able to explain modeling tradeoffs clearly, not just deliver a black box
What We'll Provide
Historical price/return data for relevant instruments
Existing signal outputs for the trend and mean-reversion sleeves (for correlation analysis and weighting integration)
Access for Q&A on system architecture as needed