Research Paper
Quant Research
The mathematical foundations behind probabilistic risk analysis for trading strategies. Understanding why distribution-based modeling outperforms intuition-based position sizing.
Contents
Why Single Path Forecasting Fails
Traditional trading analysis relies on a single historical backtest, one path through past data. This creates a dangerous illusion of certainty. A strategy with a 55% win rate and 2:1 reward-to-risk ratio sounds profitable, but the same mathematical edge produces wildly different equity trajectories depending on the sequence of wins and losses.
This is the path dependency problem. Two traders with identical strategies can experience completely different drawdowns, peak equity, and final outcomes, simply due to the random ordering of trades. A single backtest captures only one realization of an infinite probability space.
Key Insight
A positive expectancy strategy can still lead to ruin. The variance of outcomes, not just the mean, determines whether a trader survives long enough to realize their edge.
Interactive Demo: Same Strategy, Different Outcomes
Each path uses 55% win rate, 2:1 R:R, 2% risk. Click "Run" to see how variance creates wildly different equity trajectories from the same edge.
Click "Run 5 Paths" to visualize divergent outcomes
Monte Carlo Simulation
Monte Carlo simulation solves the single-path problem by generating thousands of randomized outcome sequences from the same strategy parameters. Each simulation shuffles the order of wins and losses, creating a distribution of possible equity curves rather than a single line.
From this distribution, we extract percentile bands: the P5 (pessimistic), P50 (median), and P95 (optimistic) trajectories. The spread between these percentiles reveals the true variance of your strategy, information that a single backtest completely obscures.
Process Flow
Define Edge
Win rate, R:R, risk %
Simulate N Paths
Randomize trade sequences
Aggregate Stats
Percentiles, drawdowns, ruin
Risk Decision
Size positions for survival
The mathematical foundation is the Law of Large Numbers. As simulation count increases, the percentile estimates converge to their true values. RangeSight runs simulations across thousands of randomized outcomes to balance computational cost with statistical reliability.
Interactive Demo: Monte Carlo Fan Chart
Aggregating 500 paths reveals the distribution envelope. The shaded band between P5 and P95 shows where 90% of outcomes fall.
Adjust simulations and click "Simulate"
Probability of Ruin
The probability of ruin is the likelihood that a trading account will decline to a critical threshold (typically 90% loss from peak) at any point during a sequence of trades. Unlike average return, ruin probability captures the worst-case path dependency of a strategy.
Classical ruin theory provides closed-form approximations for simple cases. For a strategy with win probability p, loss probability q = 1 - p, and equal bet sizes, the probability of ruin starting with n units is:
Classical Gambler's Ruin Formula
P(ruin) = (q/p)n when p ≠ q
Where p = win probability, q = 1 - p, n = starting units. This simplifies for equal bet sizes; real strategies with variable R:R require Monte Carlo estimation.
In practice, traders don't use equal bet sizes. They use percentage-based risk (e.g., 2% of current equity). This compounding effect makes closed-form solutions intractable. Monte Carlo simulation becomes the only practical method to estimate ruin probability for real trading strategies.
Critical Relationship
Ruin probability is exponentially sensitive to risk per trade. Doubling risk from 2% to 4% can increase ruin probability by 5–10x depending on your edge. The demo below lets you observe this non-linear relationship directly.
Interactive Demo: Ruin vs. Risk Sizing
Strategy: 50% WR, 1.8:1 R:R, a marginally positive expectancy. Adjust risk per trade to see how quickly ruin probability escalates.
Adjust risk and click "Run" to see ruin probability
Tail Risk Modeling
Standard financial models assume returns follow a normal (Gaussian) distribution. In reality, markets exhibit leptokurtosis, heavier tails than a normal distribution predicts. This means extreme moves (crashes, gaps, flash events) occur far more frequently than standard models account for.
A normal distribution predicts a 4-sigma event occurring roughly once every 31,560 trading days (~126 years). In real markets, 4-sigma events occur multiple times per decade. This discrepancy is why traders who size positions based on normal-distribution assumptions are chronically under-prepared for tail events.
RangeSight's simulation engine can model fat-tailed distributions by injecting occasional extreme events into the simulation. This stress-tests strategies against the kind of market conditions that destroy accounts sized for normal-distribution assumptions.
Visual: Fat-Tailed Return Distribution
Real markets exhibit fat tails: extreme events occur far more often than a normal distribution predicts. The red bars show tail events that standard models undercount.
Click "Generate" to visualize fat-tailed distributions
Simulation Methodology
RangeSight's simulation engine is designed for statistical reliability while maintaining computational efficiency across devices. Below are the key principles and assumptions underlying the simulation methodology.
Simulation Scale
Simulations are run across thousands of randomized outcomes to model variability in results. The number of paths is calibrated to produce stable percentile estimates without sacrificing responsiveness.
Trade Generation
Each trade outcome is generated probabilistically based on the specified win rate and reward-to-risk ratio. Gains and losses compound against current equity, reflecting real percentage-based risk sizing.
Ruin Definition
An account is considered 'ruined' when equity falls below a critical drawdown threshold during the simulation. This threshold is intentionally conservative — most traders experience psychological capitulation well before reaching it.
Path Independence
Each simulation path is independently generated. Trades within a path are sequentially dependent (compounding equity), but paths are independent of each other, enabling unbiased distribution estimates.
Percentile Estimation
Outcomes are aggregated to estimate probabilities across a range of scenarios. The P5 (pessimistic), P50 (median), and P95 (optimistic) trajectories capture the spread of possible results from a given strategy.
Limitations
The model assumes independent trades with a fixed win rate, and does not account for slippage, commissions, or market impact. Real trading introduces additional variance not captured by this simplified model.
On Statistical Convergence
By the Central Limit Theorem, percentile estimates converge as simulation count increases. RangeSight calibrates simulation volume to achieve practical accuracy for trading decisions without requiring computationally expensive larger samples.
These models are implemented and extended within RangeSight to allow repeated testing, comparison, and scenario evaluation.
Understanding Quantitative Trading Research
What Is Quantitative Trading Research?
Quantitative trading research applies mathematical and statistical methods to analyze financial markets. Rather than relying on intuition or qualitative judgment, quant research uses data-driven models to identify patterns, estimate probabilities, and evaluate strategy performance under thousands of simulated conditions.
Why Variance Matters More Than Average Returns
Most traders focus on expected returns, but variance determines whether a strategy survives long enough to realize those returns. A strategy with high expected value but extreme variance can still lead to account ruin. Understanding the distribution of outcomes, not just the mean, is the foundation of robust quantitative analysis.
The Role of Monte Carlo Methods
Monte Carlo simulation is a core tool in quantitative research. By generating thousands of random paths from the same statistical parameters, researchers can map the full probability space of trading outcomes. This reveals tail risks, drawdown distributions, and survival probabilities that single-path backtests miss entirely.
From Research to Risk Management
The goal of quant research is not to predict the market. It's to quantify risk. By understanding the statistical properties of a strategy, traders can make informed decisions about position sizing, stop placement, and portfolio allocation. Research without risk management is speculation; research with risk management is professional trading.
Apply The Research
Run deeper simulations in the full RangeSight platform.
The concepts on this page power the simulation engine inside RangeSight. Download to run Monte Carlo analysis on real ticker data with extended simulation capabilities.