Step 04: Robustness Analysis

This step evaluates base candidates and SL/TP-modified candidates across rolling windows and fixed OOS periods. There is no train/test selection inside this step: every parameter set is tested on every validation segment, then aggregated into robustness diagnostics.

InputFull candidate universe after parameter optimization and SL/TP optimization, rolling validation windows, and fixed OOS periods.

OutputRolling robustness, OOS robustness, aggregated robustness score, component diagnostics, and the shortlist passed downstream.

Used laterThe shortlist is used by Regime Analysis, Scoring, Shortlist Benchmark, Leverage Suitability, and Summary.

Robustness diagnostics

Rolling validation details

Configuration

Robustness period	01.01.2022-01.01.2025
Research period	01.01.2022-01.01.2024
Validation period	01.01.2024-01.01.2025
Window / Shift	12 months / 3 months
Parameter space	buy_ema_short: 8-100 step 3 buy_ema_long: 20-200 step 3
Parameter combinations	1797

Rolling validation windows

Window Id	Window Timerange	Window Start	Window End

Rolling validation results (top 500)

Candidate Id	Buy Ema Short	Buy Ema Long	SL	TP	Robustness Scorei	Tradesi	Profit Totali	Win ratei	Profit Drawdowni	Max Drawdowni	Drawdown Duration Pcti	Sharpei	Sortinoi	Profit Factori	Expectancyi	Robustness Window Efficiency Meani	Robustness Window Efficiency Worsti	Robustness Window Efficiency Stdi	Robustness Positive Window Ratioi	Robustness Consistency Scorei	Robustness Drawdown Stability Scorei

OOS validation details

Out-of-sample validation: every candidate from the robustness grid is backtested on the fixed OOS periods (01.01.2025-01.07.2025, 01.07.2025-01.01.2026) using the same parameter sets selected from the full grid. This is the primary fixed-period forward-looking performance check and is later combined with rolling robustness. Equity curves show how each candidate would have performed in these unseen periods.

Configuration

OOS 1	01.01.2025-01.07.2025
OOS 2	01.07.2025-01.01.2026
Backtested candidates	1797
Displayed candidates	277

OOS 1: 01.01.2025-01.07.2025

Candidate Id	Buy Ema Short	Buy Ema Long	SL	TP	Tradesi	Profit Totali	Profit absi	Win ratei	Final Balancei	Sharpei	Sharpe Trade Basedi	Sortinoi	Sortino Trade Basedi	Profit Factori	Expectancyi	Profit Drawdowni	Max Drawdowni	Max Drawdown Relativei	Drawdown Duration Pcti	Underwater Meani	Time Below Entryi	Efficiency Scorei	Balanced Scorei	Profit Quality Scorei	Benchmark Profit Totali	Alpha vs Benchmarki	Beta Adjusted Alphai	Correlation to Benchmarki	Beta to Benchmarki	Upside Capturei	Downside Capturei	Candidate Typei	Account Profit Totali	Account Max Drawdowni

OOS 2: 01.07.2025-01.01.2026

Candidate Id	Buy Ema Short	Buy Ema Long	SL	TP	Tradesi	Profit Totali	Profit absi	Win ratei	Final Balancei	Sharpei	Sharpe Trade Basedi	Sortinoi	Sortino Trade Basedi	Profit Factori	Expectancyi	Profit Drawdowni	Max Drawdowni	Max Drawdown Relativei	Drawdown Duration Pcti	Underwater Meani	Time Below Entryi	Efficiency Scorei	Balanced Scorei	Profit Quality Scorei	Benchmark Profit Totali	Alpha vs Benchmarki	Beta Adjusted Alphai	Correlation to Benchmarki	Beta to Benchmarki	Upside Capturei	Downside Capturei	Candidate Typei	Account Profit Totali	Account Max Drawdowni

Shortlist to continue

Builds the smaller finalist set passed to Regime Analysis, Shortlist Benchmark, and Scoring. Candidates are selected by the following rules:

Top robustness: top 10 base candidates ranked by combined rolling + OOS robustness score.
Metric winners (Unfiltered): best parameter set per metric (Efficiency, Sharpe, Sortino, Profit, etc.) from Step 02 unfiltered winners.
Metric winners (Filtered): same as above but from the Step 02 filtered winners (passing winner_filters).
Static params: all parameter sets explicitly listed in selection.static_params in the config.
Top SL/TP by robustness: top 10 SL/TP candidates from Step 03, ranked by their rolling robustness score from Step 04.
SL/TP metric winners: top 2 SL/TP variants per metric, selected by their train-period improvement value.

Duplicates are removed after merging all sources.

Total shortlist rows	36
Top robustness	7
Metric winners (Unfiltered)	13
Static params	9
Top SL/TP by robustness	1
SL/TP metric winners	13

Candidate Id	Buy Ema Short	Buy Ema Long	SL	TP	Shortlist Reason	Windows Evaluated	Robustness Scorei	Rolling Robustness Scorei	OOS Robustness Scorei	Robustness Window Efficiency Meani	Robustness Window Efficiency Worsti	Robustness Positive Window Ratioi	Robustness Consistency Scorei	Robustness Drawdown Stability Scorei	OOS Window Efficiency Meani	OOS Window Efficiency Worsti	OOS Positive Window Ratioi	OOS Consistency Scorei