Quant Analyst Strategist

An agent for quantitative finance work covering algorithmic trading strategy development, risk modeling, portfolio optimization, and statistical analysis of financial markets using Python-based toolchains.

When to Use This Agent

Choose Quant Analyst Strategist when:

• Developing and backtesting algorithmic trading strategies
• Building risk models (VaR, CVaR, stress testing, scenario analysis)
• Optimizing portfolio allocations with Markowitz or Black-Litterman models
• Analyzing time series data for financial forecasting
• Calculating options pricing, Greeks, and volatility surfaces

Consider alternatives when:

• Building trading platform infrastructure (use a backend engineering agent)
• Creating financial dashboards without quant analysis (use a BI agent)
• Doing general data science without financial domain focus (use a data science agent)

Quick Start

Copy
# .claude/agents/quant-analyst-strategist.yml
name: Quant Analyst Strategist
model: claude-sonnet-4-20250514
tools:
  - Read
  - Write
  - Bash
  - Glob
  - Grep
prompt: |
  You are a quantitative analyst specializing in algorithmic trading
  and financial modeling. Develop strategies with rigorous backtesting,
  proper risk management, and statistical validation. Always account
  for transaction costs, slippage, and survivorship bias.

Example invocation:

Copy
claude --agent quant-analyst-strategist "Backtest a mean-reversion
  pairs trading strategy on SPY/IWM using a 20-day lookback window.
  Include transaction costs, calculate Sharpe ratio, and compare
  against buy-and-hold."

Core Concepts

Strategy Development Pipeline

Research → Backtest → Validate → Paper Trade → Live Deploy
   │          │          │           │             │
  Hypothesis  Historical Walk-forward Simulated    Monitored
  generation  simulation out-of-sample real-time   execution
  Signal      Transaction Statistical  Latency     Risk
  design      costs       significance test        controls

Risk Metrics Reference

Metric	Formula Concept	Target
Sharpe Ratio	(Return - Rf) / StdDev	> 1.5
Sortino Ratio	(Return - Rf) / DownsideDev	> 2.0
Max Drawdown	Peak-to-trough decline	< 20%
VaR (95%)	5th percentile loss	Context-dependent
Calmar Ratio	Annual Return / Max Drawdown	> 1.0
Win Rate	Profitable trades / Total trades	> 50% for trend
Profit Factor	Gross Profit / Gross Loss	> 1.5

Backtesting Framework

Copy
import pandas as pd
import numpy as np

class Backtest:
    def __init__(self, data, strategy, initial_capital=100_000):
        self.data = data
        self.strategy = strategy
        self.capital = initial_capital

    def run(self, commission=0.001, slippage=0.0005):
        signals = self.strategy.generate_signals(self.data)
        positions = self._calculate_positions(signals)
        returns = self._calculate_returns(
            positions, commission, slippage
        )
        return self._compute_metrics(returns)

    def _compute_metrics(self, returns):
        return {
            'total_return': (1 + returns).prod() - 1,
            'sharpe': returns.mean() / returns.std() * np.sqrt(252),
            'max_drawdown': self._max_drawdown(returns),
            'win_rate': (returns > 0).mean()
        }

Configuration

Parameter	Description	Default
risk_free_rate	Risk-free rate for Sharpe calculation	0.05
commission_rate	Per-trade commission	0.001 (10bps)
slippage_model	Slippage estimation method	Fixed (5bps)
position_sizing	Position size method	Kelly criterion
lookback_period	Default lookback window	252 days
confidence_level	VaR confidence level	95%
rebalance_frequency	Portfolio rebalance period	Monthly

Best Practices

1. Always account for transaction costs and slippage in backtests. A strategy that returns 15% annually before costs may return 2% after accounting for commissions, spread, slippage, and market impact. Model costs explicitly: fixed commission per trade, proportional spread cost, and market impact as a function of trade size relative to average volume. Strategies that only work with zero costs will fail in production.

2. Use walk-forward validation, not in-sample optimization. Optimizing parameters on the full dataset and reporting those results guarantees overfitting. Split data into training (optimize) and test (validate) periods chronologically. Better yet, use rolling walk-forward: optimize on the first 2 years, test on year 3, shift forward, repeat. Only strategies that perform consistently across out-of-sample windows are worth pursuing.

3. Test for statistical significance before trusting backtest results. A strategy with a 12% return over 5 years may be entirely due to luck. Calculate the t-statistic of your Sharpe ratio: t = Sharpe × sqrt(N_years). A t-statistic below 2 means the result is not statistically significant at the 95% level. Also run the strategy on randomized data to establish a null distribution and verify your actual results are unlikely to occur by chance.

4. Implement risk limits as hard constraints, not guidelines. Set maximum position size, maximum portfolio drawdown, maximum sector exposure, and maximum correlation between positions as enforceable limits in your execution system. When a limit is hit, the system should automatically reduce exposure, not send an alert that humans may ignore. Risk limits should never be overridden without formal approval.

5. Watch for survivorship bias and lookahead bias in data. Historical databases often exclude delisted companies (survivorship bias), making strategies appear more profitable than they would have been. Lookahead bias occurs when using data that wasn't available at the time of the trade decision—like using revised GDP figures instead of initial estimates. Use point-in-time databases and include delisted securities.

Common Issues

Backtest shows great returns but live trading loses money. The most likely cause is overfitting: the strategy was optimized to fit historical noise rather than genuine market patterns. Reduce the number of free parameters, use simpler models, and require out-of-sample validation. Also check for data quality issues: adjusted vs unadjusted prices, dividend handling, and corporate action adjustments can create phantom signals.

Strategy performance degrades over time. Markets are adaptive—profitable strategies attract capital that arbitrages the opportunity away. Monitor strategy performance in real time and compare against backtest expectations. If the Sharpe ratio drops below a predefined threshold for a sustained period, reduce position sizes or pause the strategy. Build a pipeline of new strategy ideas to replace degrading ones.

Risk metrics look acceptable but the strategy has concentrated exposure. Aggregate risk metrics can hide dangerous concentration. A portfolio with low overall VaR might have 80% exposure to a single sector or factor. Decompose risk by sector, factor, and position. Set maximum concentration limits. Stress test against historical events (2008 financial crisis, COVID crash) to verify the portfolio survives tail events.