If you browse GitHub for trading bots or algorithmic finance repositories, you will find thousands of projects promising automated riches. Yet, when you dive into the source code, an alarming pattern emerges. Almost without exception, retail quantitative codebases suffer from critical architectural flaws that guarantee long-term account ruin in live markets.
First, unintentional look-ahead bias: inadvertently slicing time-series data such that a bar accesses prices or statistics from the future.
Second, naive lot sizing: allocating 10% of equity into an asset regardless of whether its annualized volatility is 12% (like short-term bonds) or 80% (like silver, tech breakouts, or crypto).
Third, uncapped or static leverage: using fixed broker margin multipliers that amplify drawdown during market shocks.
Fourth, missing circuit breakers: zero automated, broker-side liquidation logic when unexpected market gaps occur.
To address these vulnerabilities, we developed STRIDE (Systematic Trend and Risk-Integrated Daily Execution) as an open-source, production-grade reference architecture for quantitative risk management and multi-asset trend following.
Here is the breakdown of the engineering principles, mathematical formulations, and testing methodologies behind the system.
1. Zero Look-Ahead Bias: The Pure Functional Imperative
Look-ahead bias is the silent killer of backtests. A model that looks brilliant in simulation frequently blows up on day one of live execution because of a subtle off-by-one error in a rolling window.
In STRIDE, we eliminated this entire class of bugs by designing the strategy and risk layers as pure mathematical functions: the function receives an immutable slice of past bars, and returns the signals without any side effects.
Notice what is absent: No network calls to fetch missing data on the fly. No filesystem access during calculation. No system clock lookups like datetime.now() or time.time().
Because historical time is injected explicitly as an immutable sequence, the engine is 100% deterministic. To prove this invariance, we utilize Hypothesis for property-based testing. We feed thousands of randomized price histories into the strategy and mathematically verify that appending arbitrary future price bars leaves past signals completely unchanged.
2. Volatility Parity: Equalizing Dollar Risk (1N ATR)
A fundamental rule of quantitative trend following (originally popularized by Richard Dennis and Ed Seykota) is that you trade the volatility, not the instrument.
If you purchase $10,000 of intermediate treasuries (IEF) and $10,000 of silver (SLV), you do not have a balanced portfolio. Silver’s higher volatility means it will dictate 90% of your portfolio variance.
STRIDE enforces Volatility Parity: each position is sized such that a price move of 1N (the 20-day Average True Range) represents exactly 1% of total portfolio equity:
Position Size = (Account Equity x 1%) / (20-day ATR)
By equalizing volatility risk across 10 liquid macro ETFs (equities, treasuries, precious metals, broad commodities, USD, and crypto trusts), the portfolio captures macro trends without concentration risk.
3. Dynamic Leverage Formulation: The 5-Limit Minimum
Most retail scripts treat leverage as a static toggle (such as 2x or 3x). In volatile markets, static leverage inevitably leads to forced liquidations.
In STRIDE, effective portfolio leverage is dynamically calculated every single trading day as the mathematical minimum of five independent boundary conditions:
Effective Leverage = Minimum of (Kelly Limit, Volatility Target, Absolute Cap, Broker Limit, Capital Limit)
Kelly Criterion: Bounded fraction derived from empirical win-rates and payoff ratios to maximize long-term geometric compounding while preventing overbetting ruin.
Macro Volatility Target: Scales exposure inversely to macro market turbulence.
Absolute Ceiling: Hard structural ceiling capped between 1.5x and 2.0x (absolute maximum 4.0x).
Broker Limit: Constrained by strict broker buying power.
Capital Floor: Margin borrowing is disabled; additional exposure is achieved exclusively via structured leveraged ETFs, strictly bounding loss to invested capital.
4. Dual-Brake Circuit Breakers
When black swan events strike, trading algorithms cannot rely on human intervention. STRIDE incorporates an autonomous two-stage emergency brake system:
Stage 1 (Drawdown exceeds 20%): Automatic de-leveraging. Target position sizes are immediately halved across all assets.
Stage 2 (Drawdown exceeds 40%): The emergency kill switch triggers. All open broker positions are liquidated to cash, and live execution halts until human post-mortem audit.
5. Software Quality and Modern Python 3.12 Tooling
Financial software demands the same rigor as safety-critical engineering: Full type safety with mypy strict mode and zero unchecked dynamic attributes. Sub-second linting, formatting, and import organization with Ruff. Pydantic v2 SecretStr so broker API keys and tokens cannot leak into logs, exceptions, or terminal outputs. 791 automated tests executing locally in under 5 seconds with zero external dependencies.
Conclusion and Code Access
Quantitative trading does not require opaque black-box neural networks; it requires mathematical discipline, strict position sizing, and defensive software engineering.
STRIDE is 100% open source under the MIT License:
GitHub Repository: https://github.com/JAbarca74/stride
Clone it, run the tests (pytest), and adapt the risk engine for your own quantitative research. If you find the architecture useful, consider giving it a star on GitHub! ⭐
Top comments (0)