The Quantitative Trading Playbook

Most systematic trading strategies fail for one of three reasons: they overfit to historical data, they ignore execution realities, or they have no coherent risk model. This playbook addresses all three.

It’s written from the perspective of someone who has built and run systematic options strategies managing ₹100+ crore portfolios. Not theoretical — this is what actually works.

Start With Goals, Not Strategies

Before writing a single line of code, define what you’re trying to achieve in measurable terms:

• Target annual return (net of costs)
• Maximum acceptable drawdown
• Target Sharpe ratio
• Liquidity constraints (how quickly do you need to be able to exit?)
• Capital constraints (minimum and maximum position sizes)

Van Tharp’s position on this is correct: you’re not trading a market, you’re trading a system. A system has explicit objectives. Without explicit objectives, you have no way to evaluate whether your strategy is doing what it’s supposed to do.

The SMART framework applies here more cleanly than in most domains: Specific (exact metrics), Measurable (trackable in real time), Achievable (realistic given market dynamics), Relevant (connected to your actual business goals), Time-bound (evaluated over a defined horizon).

The Signal Hierarchy

A systematic strategy is a set of signals aggregated into a position. Before building anything, be explicit about what you believe generates alpha and why:

Types of signals:

• Structural: arise from market microstructure, position constraints, or behavioral patterns that have economic explanations
• Statistical: arise from empirical regularities with or without strong theoretical backing
• Fundamental: arise from information about underlying business value or economic conditions

The best signals have economic explanations. Pure statistical arbitrage without a reason is pattern-matching on noise that will eventually disappear when too many people discover it.

Signal validation checklist:

1. Is the edge statistically significant over a sufficiently long out-of-sample period?
2. Is the Sharpe stable across parameter variations? (Parameter sensitivity analysis)
3. Does the edge exist in different market regimes, or only in favorable conditions?
4. Is the edge large enough to survive realistic transaction costs?
5. Is there an economic explanation for why this edge exists?

A signal that fails any of these tests should be treated as suspect until proven otherwise.

Walk-Forward as Discipline

Walk-forward testing is not a technique — it’s a discipline. The purpose is to make it structurally impossible to optimize a strategy on the same data you’re evaluating it on.

The mechanics: divide your historical data into sequential periods. Train on period 1, test on period 2. Train on periods 1+2, test on period 3. Continue forward. Evaluate only on the out-of-sample periods.

Common mistakes:

• Look-ahead bias: Using information that wasn’t available at the time of the decision. Subtle forms include using future-adjusted prices, survival-biased universe construction, and parameter selection informed by future data.
• Overfitting through multiple testing: Testing hundreds of parameter combinations and selecting the best one inflates the expected Sharpe of your in-sample result. Use the overfitting score.

The overfitting score is a measure of how sensitive your strategy’s Sharpe is to parameter perturbations. A strategy with Sharpe 3 that only works at specific parameter values is more overfit than a strategy with Sharpe 2 that works across a range. Operationally: test your strategy across a parameter grid and measure the standard deviation of the Sharpe distribution. Low variance = robust.

Transaction Cost Modeling — Where Most Backtests Lie

Most backtests underestimate costs. The complete cost model for options strategies:

Explicit costs:

• Brokerage commission (per lot or per turn)
• Exchange fees (NSE / BSE fees by segment)
• STT (Securities Transaction Tax — applies on sell side for options, on both sides for equity)
• Stamp duty
• GST on brokerage

Implicit costs:

• Bid-ask spread: you never get the mid-price. Model this as a fixed fraction of the spread on each fill.
• Market impact: for large orders relative to market depth, your order moves the price
• Slippage: the difference between the price when you decided to trade and the price when you filled
• Roll cost for carrying positions into the next expiry

The right approach: build a fill simulator that applies realistic costs at every simulated trade. Track the difference between theoretical P&L (based on mid prices) and realistic P&L (with full cost model) explicitly. If the difference is too large relative to your edge, the strategy isn’t viable.

For options specifically: the Greeks matter. Delta hedging frequency is a cost driver. Gamma exposure creates convexity that must be managed. Vega positions have theta decay that compounds daily.

Capital Efficiency and Position Sizing

Kelly criterion and its variants: The Kelly criterion gives the theoretically optimal bet size to maximize long-run geometric growth. Full Kelly is too aggressive for most implementations — use fractional Kelly (25–50% of Kelly sizing) to reduce variance at the cost of some long-run growth.

For options strategies, Kelly sizing is more complex because position size isn’t a simple scalar — you’re sizing in terms of delta, vega, and gamma exposure, not just capital at risk.

Practical position sizing rules:

• Never risk more than 2% of portfolio on a single trade (for directional strategies)
• Adjust position size based on current volatility regime — scale down in high-vol environments
• Account for correlation across positions — two strategies that both lose in high-VIX environments aren’t diversified in the ways that matter

Capital utilization efficiency: Not the same as return on capital. A strategy using only 30% of available capital is leaving return potential on the table. Optimize for capital efficiency alongside risk-adjusted return.

Execution and Market Microstructure

Understanding microstructure is not optional for systematic options trading:

Order types and when to use them:

• Market orders: guaranteed fill, uncertain price — only for highly liquid instruments where the spread is trivial relative to the trade size
• Limit orders: certain price, uncertain fill — better for options where spreads can be 0.5–2% of the premium
• Stop orders: execution logic, not order type — model the slippage around stop execution carefully

Options-specific execution considerations:

• Open interest and volume at your strike and expiry matter — low-volume options have wider spreads and higher market impact
• Expiry proximity increases gamma and affects fill quality — avoid executing large positions in the last few days of expiry for non-directional strategies
• IV levels at execution versus at backtest time: if your strategy was designed in a low-IV environment and you’re deploying in high IV (or vice versa), the position sizes need to adjust

The Live Deployment Protocol

Never go from backtest to full capital immediately:

Stage 1: Paper trading — Run the strategy live with no actual orders. Track the difference between simulated fills and what you would have gotten in reality. This surfaces microstructure assumptions you got wrong.

Stage 2: Minimum size — Trade with the minimum lot size (or minimum viable capital). Collect real fill data. Compare to paper trading. Are your cost assumptions accurate?

Stage 3: Half size — Scale to 50% of target capital. Monitor drawdown, Sharpe, and cost metrics. Look for regime-sensitivity that wasn’t visible in backtesting.

Stage 4: Full deployment — Only after stages 1–3 have produced results consistent with backtest expectations.

Kill switches matter: define the conditions under which you will stop a strategy before you start it. A drawdown limit (e.g., stop if 3 standard deviations worse than expected), a regime change detector, an execution anomaly threshold. These should be pre-committed, not evaluated in the moment.

Monte Carlo Risk Analysis

Backtests show you the historical path. Monte Carlo simulations show you the distribution of possible futures.

The setup: take the return series from your backtest, resample with replacement (bootstrapping) to generate thousands of alternative paths. From this distribution, compute:

• Expected drawdown at various confidence levels (95%, 99%)
• Probability of hitting your drawdown limit over 1, 3, 12 months
• Distribution of Sharpe ratios across paths — what’s the range of plausible outcomes?

This is important because a backtest can show a Sharpe of 3 while the true expected Sharpe is 1.5 and you got lucky in the historical period you tested on. Monte Carlo makes this uncertainty visible.

Risk of ruin: the probability that the strategy depletes capital beyond the point of recovery. Even with a Sharpe of 2, if position sizing is too aggressive, risk of ruin is non-negligible. Compute it explicitly before going live.

Regime Detection and Adaptive Deployment

No strategy works in all regimes. Systematic strategies are better for knowing this explicitly.

Volatility regime classification: Classify each trading day or week into low/medium/high volatility using realized volatility, VIX level, or IV percentile. Most strategies have regime-dependent performance — know which regime yours is designed for.

Adaptive position sizing: Scale position size inversely with current volatility to maintain constant risk exposure. In high-vol regimes, positions are smaller; in low-vol regimes, positions are larger.

Strategy switching: Some traders maintain multiple strategies with different regime affinities and switch allocation based on current regime classification. This requires careful handling of transaction costs during switches.

What Separates Strategies That Last

After years of doing this, the patterns are clear:

Strategies that survive have:

1. An economic reason for the edge
2. Robust out-of-sample performance (not just in-sample optimization)
3. Realistic cost modeling from the start
4. Defined risk management with pre-committed rules
5. A deployment protocol that starts small and scales on evidence

Strategies that fail have:

1. Overfitting disguised as statistical significance
2. Backtests that ignore costs or use unrealistic fill assumptions
3. No drawdown management — “it’ll recover” is not a risk model
4. Position sizing based on maximum historical return rather than risk-adjusted optimization

The market doesn’t reward complexity. Simple strategies with real edges, honest cost models, and disciplined deployment outperform complex strategies that only exist in the researcher’s backtest.