Skip to content

Vision Weekly Three-Dimensional Trading Optimization: Walk-Forward Entry Strategies

Date: October 2nd, 2025 Author: Daniel Shanklin, Director of AI and Technology, AIC Holdings Tags: Trading Strategy, Walk-Forward Analysis, Risk Management, Vision Research

Sources: - Vision Research LLC weekly short recommendations (2024-2025) - Internal analysis of 72 realized Vision Weekly trades - Walk-forward methodology research


Our previous analysis revealed that systematic exit strategies dramatically outperform entry timing approaches, but highlighted an important learning opportunity: look-ahead bias in entry strategies. The "local high" entry strategy required knowing future price movements—challenging in real trading.

This follow-up study addresses this by testing implementable walk-forward entry methods against our original immediate entry + systematic exit approach.

The Look-Ahead Bias Challenge

Previous Approach: Wait for "local highs" before entering short positions Challenge: You can only identify a local high after the subsequent decline

Example: - Vision recommends shorting XYZ on Monday at $100 - "Wait for 5-day local high" strategy says enter at $105 on Wednesday - But you couldn't know Wednesday was the 5-day high until Friday's close

Walk-Forward Solution: Williams Fractals

Williams Fractal Algorithm provides implementable signals without future knowledge: - Bearish Fractal: Middle candle has highest high, with two lower highs on each side - Confirmation: Signals confirmed 2 bars after formation (implementable lag) - Real-Time Application: Provides actionable short entry signals within 2-5 trading days

Testing Entry Strategies

We tested three entry approaches across 60 trades: 1. Immediate Entry: Enter at Vision's recommendation price 2. Williams Fractal Entry: Wait for bearish fractal confirmation (2-5 day lag) 3. Fixed Delays: Traditional 1, 3, 5, 10, 15, 20 day delays

Each strategy was combined with various stop-loss (1%, 3%, 5%, 10%, 20%) and stop-gain levels (5%, 10%, 20%, 50%, none).

Key Results

Actual Optimal Strategy: - Entry: 3-day delay - Stop-Loss: 1% - Stop-Gain: None - Average Return: +1.9%

Entry Strategy Performance: - 3-day delay + 1% stop: +1.9% (optimal) - Immediate entry + 1% stop: +1.7% - Immediate entry + 3% stop: -0.2% - Vision's original approach: -5.9%

Key Finding: Short delays with tight stop-losses provide modest but meaningful improvements over Vision's original approach. Williams Fractals were not detected in sufficient quantity for reliable testing.

Why Short Delays Can Help

Contrary to expectations, short delays (1-3 days) can provide modest improvements: 1. Volatility Timing: Brief waiting periods can avoid immediate post-recommendation volatility spikes 2. Sentiment Confirmation: Short delays allow negative sentiment to develop and stabilize 3. Technical Entry: 1-3 day delays often coincide with minor technical resistance levels

However, longer delays (10+ days) erode performance as information advantages fade.

Conclusions

Testing implementable walk-forward entry methods reveals that systematic risk management provides meaningful improvements:

Best No Stop-Gain Strategies (preserving unlimited upside): 1. 3-day delay + 1% stop-loss: +1.9% average return 2. Immediate entry + 1% stop-loss: +1.7% average return 3. Vision's original approach: -5.9% average return

Best Overall Strategy (allowing profit caps): - 20-day delay + 20% stop-loss + 30% stop-gain: +3.8% average return

Key Insights: - Systematic stop-loss management provides a 7.8 percentage point improvement over Vision's original approach when preserving unlimited upside - Allowing stop-gains can achieve 9.7 percentage point improvement but caps profitable trades - Short delays (1-3 days) slightly outperform immediate entry with proper risk management

For institutional investors: Choose between preserving unlimited upside (3-day + 1% stop-loss) or accepting profit caps for higher average returns (20-day + 20% stop-loss + 30% stop-gain). Williams Fractal methods require further refinement for this dataset.