Development Log: Complete Fixation of Stop-Loss (SL) in a Single Position and Linear Risk Management

The development criteria for automated trading architecture at Semura Lab. aim to eliminate uncertainty in the market to the greatest extent and to prove mathematical advantage (edge) as objective factual data. In system trading, the installation of automated trading stop loss (SL) is recognized as an essential defensive measure, but many developers adopt ambiguous approaches regarding its design philosophy.

This article points out the uncertainties in system evaluation introduced by dynamic stop losses such as commonly preferred trailing stops. It records a structural design to rigorously evaluate resilience in walk-forward analysis (WFA) by completely fixing the SL for a single-position operation and linearizing risk.

1. Redefinition of “Stop-Loss” in Systems Engineering

In automated trading systems, a stop loss (SL) is not merely a fail-safe for protecting capital. From the perspective of systems engineering and process design, the SL must function as an “absolute constant” used to calculate the system’s expected value (EV).

In the basic equation for calculating expected value, if the loss amount (Risk) is a variable, it becomes extremely difficult to statistically predict and evaluate the system’s future performance. The long-term failure of no-stop loss, or of an averaging-down/martingale architecture that adjusts with account balance as a percentage, arises from this nonlinearity of risk, which makes the maximum drawdown mathematically impossible to compute.

2. Mathematical Deception Embedded in Dynamic Stop Losses (Trailing Stops)

Many developers incorporate dynamic stop losses (Dynamic SL) such as trailing stops and scaling-out into their logic to maximize profits while limiting losses. These features can visually smooth backtest results (profit factor and equity curves) and produce aesthetically pleasing outcomes.

However, dynamic stop losses introduce significant uncertainty in assessing system durability. The timing of trailing stop activation and the extent to which the stop level moves depend on momentary tick volatility, a highly random noise. As a result, the loss amount per trade or even small gains may not stay within a fixed range, and the dispersion (variance) can expand dramatically.

This expansion of dispersion hides the system’s true maximum drawdown. A logic overfitted to past markets (in-sample) by dynamic SLs may suffer stop losses at unexpected timings when faced with unseen markets (out-of-sample), leading to drawdowns that were not measured in backtests. Consequently, the calculation of the “worst-case scenario,” which underpins system evaluation, may not be feasible.

3. Advantages of Fully Fixed Risk and Linear Management

To eliminate the above uncertainties and design a robust system based on facts, an architecture that assumes a single position at a time and fixes the SL width as a completely fixed value (Fixed SL) is required.

Introducing a fixed SL makes the maximum loss in a single trade an absolute constant. This fully visualizes risk and enables linear management.

The greatest mathematical advantage of linearizing risk is that statistical significance becomes easier to prove. If the loss amount is always constant, the system’s edge concentrates only on the independent variables of win rate and average profit per trade. By eliminating noise-dependent dynamic settlement logic, you can objectively determine whether the logic truly captures the market’s fundamental bias or merely benefits from incidental volatility.

4. Rigor in Durability Evaluation with WFA (Walk-Forward Analysis)

A system with risk fixed and linearized proves its value in walk-forward analysis (WFA).

WFA subdivides historical data into multiple periods and iteratively optimizes (learns) and forward-tests (validates) to measure the universal durability of the logic. If a system includes dynamic variables like trailing stops, the optimal trailing width learned during the learning period collapses easily as volatility changes during the validation period, a typical trap of curve fitting (over-optimization).

In contrast, a model with fully fixed SL has risk as a constant, so optimization targets the “advantage of entry” itself. Through WFA, you measure whether fixed-risk (SL) maintains a consistently positive EV even in long-term unseen data environments. This makes the calculated maximum drawdown and losing-streak probability not just a record from past backtests but a highly reliable absolute withdrawal criterion (threshold) for stopping or discarding the system in future operations.

5. Equalizing Friction Costs and Infrastructure Perspectives

Moreover, fixing risk is also a prerequisite when incorporating real-market physical friction costs (spreads and slippage) into the system. When dynamic SL is used, the slippage impact varies per trade due to latency at the moment of settlement, making precise measurement of friction costs impossible. If SL is fixed, you can easily run stress tests that add a maximum slippage as a stress value.

Accurately measuring the friction between the system and the market infrastructure and keeping it within tolerable ranges is also a fundamental prerequisite for architectural design, where the base risk remains linear (constant).

6. Conclusion: Abandonment of Prediction and Complete Control of Risk

In Semura Lab.’s process design philosophy, defining that the market’s future cannot be predicted. Dynamic stop losses aimed at increasing profits are nothing more than excessive expectations and interference with the system in this realm of “unpredictable territory.”

Truly robust algorithmic trading starts by making the only controllable factor, the losses (risk), completely constant and linear, and managing it as a fact of systems engineering. A thorough fixed SL for a single position abandons the deceptive smoothness of asset curves and provides a highly rational, inorganic design criterion that endows the system with “fact-based robustness” capable of withstanding severe stress tests and WFA in future real-world operations.