Validation Methodology

Quantitative Research & Validation

We apply academic-grade rigor to strategy validation, ensuring that our systems are built on statistical evidence rather than curve-fitted anomalies.

Backtesting Integrity

Beyond Historical Modeling

Walk-Forward Analysis

Using rolling optimization windows to validate performance across evolving market conditions, preventing overfitting.

Monte Carlo Simulation

Testing robustness by running thousands of randomized trade sequences to assess drawdown probability.

Impact Modeling

Every research output incorporates conservative slippage models to reflect real-world execution friction.

Failure Mode Modeling

We explicitly model how a strategy behaves when it fails. By identifying the specific market conditions (regime shifts, liquidity dry-ups) that lead to performance decay, we build more effective kill-switch protocols.

Our goal is to ensure that losses occur within predefined, modeled expectations rather than through structural ignorance.

Trust and Credibility

Elite research is for sophisticated participants who value methodology over outcome. We prioritize the integrity of the research cycle above all else.

"A profitable outcome from a flawed process is a dangerous anomaly."

Applied Research

Quantitative Case Studies

Specific examples of our research methodology applied to Indian market structures.

Ornstein-Uhlenbeck Mean Reversion (Indian Equities)

Strategy Overview

  • Market: NSE Equities (Nifty 50 Constituents)
  • Timeframe: Intraday to Multi-day (Mean Holding: 2.4 Days)
  • Hypothesis: Co-integrated pairs in the same sector (e.g., HDFCBANK vs. ICICIBANK) exhibit mean-reverting spread behavior due to temporary liquidity imbalances.

Modeling Approach

We utilize a Kalman Filter to dynamically estimate the hedge ratio ($\beta$) and spread mean, allowing for time-varying co-integration parameters.

  • Model: Linear State-Space Model (Kalman Filter)
  • Entry/Exit: Z-Score of the spread residuals. Long spread at Z < -2.0, Short at Z > 2.0. Exit at Z = 0.
  • Risk Constraint: Half-life of mean reversion must be < 4 hours. Stop-loss at Z > 4.5 (regime break).

Performance Metrics (Backtest)

Annualized Return (CAGR) +28.4%
Sharpe Ratio 2.14
Max Drawdown -12.8%
Win Rate 62.3%
Profit Factor 1.85

*Includes transaction costs (0.03% slippage + taxes). Benchmarked against Nifty 50.

XGBoost Directional Swing (Derivatives)

Strategy Overview

  • Market: Bank Nifty Futures & Options
  • Timeframe: Swing (2-5 Days)
  • Hypothesis: Option Chain open interest shifts and implied volatility skews precede significant directional moves in the underlying futures.

Data & Feature Engineering

Trained on 5 years of 5-minute OHLCV and snapshots of the Option Chain (PCR, Max Pain shifts).

  • Input Features: OI Change (Atm +/- 5 strikes), IV Skew, VWAP Distance, FII/DII Net Flow.
  • Model: XGBoost Classifier (Gradient Boosting).
  • Validation: Walk-Forward optimization with a 3-month rolling window to adapt to changing volatility regimes.

Performance Metrics (Backtest)

Annualized Return (CAGR) +42.1%
Sharpe Ratio 1.92
Max Drawdown -18.5%
Win Rate 54.7%
Sortino Ratio 2.44

*High beta strategy. Stop-loss dynamic based on ATR (Average True Range).

LSTM-CNN Intraday Momentum

Strategy Overview

  • Market: Nifty 50 Futures (Intraday)
  • Timeframe: High Frequency (1-minute bars)
  • Hypothesis: Micro-structural patterns in order flow (tick data) contain predictive signals for short-term price momentum that linear models miss.

Architecture

Hybrid architecture using CNN layers to exact spatial features from order book snapshots and LSTM layers for temporal sequence dependency.

  • Input: Level 2 Order Book (Bid/Ask Depth), Trade Velocity, Order Imbalance.
  • Infrastructure: GPU-accelerated inference (TensorRT) to minimize latency (< 50ms).
  • Risk: Hard per-trade stop loss (0.15%) and daily max loss limit.

Performance Metrics (Paper Trade)

Projected CAGR +55.6%
Sharpe Ratio 2.85
Max Drawdown -9.2%
Avg Win / Avg Loss 1.4:1
Capacity ~5 Cr INR

*Limited scalable capacity due to intraday liquidity constraints.

PPO Agent for Regime Adaptation

Strategy Overview

  • Market: Multi-Asset (Equities, Gold, USD/INR)
  • Timeframe: Positional (Weekly Rebalancing)
  • Goal: Dynamic asset allocation that adapts to changing inflationary and growth regimes.

Agent Design

We use Proximal Policy Optimization (PPO), a policy gradient method, to train an agent to maximize risk-adjusted returns (Sharpe) rather than raw profit.

  • State Space: Macro indicators (Bond Yields, VIX, Crude Oil), Trend strengths of asset classes.
  • Action Space: Discrete allocation weights (0%, 25%, 50%, 75%, 100%) per asset.
  • Reward Function: Sortino Ratio over a rolling 6-month window.

Performance Metrics (Simulation)

Annualized Return (CAGR) +18.5%
Sharpe Ratio 1.45
Max Drawdown -8.4%
Correlation to Nifty 0.32
Volatility 9.8%

*Designed as a defensive, low-volatility portfolio anchor.