Surviving Market Friction: Evolving a Quant Project on the Hypothesis Canvas

Verdict: robust · Asset/TF: BTCUSDT/SOLUSDT Portfolio · Sample: 60,000 bars (2019-2026)

Rationale & Background

Every quant researcher eventually faces the "friction shock": the transition of backtested ideas from zero-fee environments to the brutal realities of live slippage and execution costs. In this case study, we outline the exact lifecycle of our trading hypotheses as tracked by the RLXBT Hypothesis Canvas (Idea Map). We document the transition from naive zero-fee strategies to execution-optimized limit-order models, culminating in a multi-pattern portfolio ensemble.

Part 1: The Great Friction Shock (Zero Fees to Taker Costs)

Our early strategies on the Hypothesis Canvas showed exceptional results under version 0.2.6 (e.g., S1 Flagship returning +170.8%, Sharpe 1.47). However, a commission-capping bug was undercharging fees by 100x.

When re-verified on version 0.2.7 under a realistic Hyperliquid taker fee model (4.404 bps per side, measured from 184 live fills):

• 5m HFT Scalp Strategies (e.g., hyp_1784303072649_18): Promptly demoted to rejected. Because the average profit per trade was smaller than the round-trip fee, the strategy lost -48.28% (Sharpe -11.83) at real costs.
• 1h Flagship Volatility Regime (hyp_1783794452336_0): Demoted to watch. Standalone returns flipped to -8.52% (Sharpe -0.23), proving that a high-turnover strategy cannot survive taker market orders.

Part 2: The Execution Pivot (Maker-Exit Policy)

Rather than discarding the underlying alpha, we shifted our focus from the signal to the execution model. We replaced market-order exits with resting limit orders (maker exits) starting 15 bars prior to the max hold time.

We verified this on SOL 1M BTC-Triggered Spike Reversal (hyp_1785824592951_186):

• Taker Exit: Sharpe 0.722
• Maker Exit: Sharpe 0.986 (with blended fee of 1.91 bps/side).
• Out-of-Sample Verification: WFE = +2.10, Monte Carlo Risk of Ruin = 0.0%.
  This proved that order-routing alone can transform a rejected signal into a robust, tradeable strategy.

Part 3: The Portfolio Ensemble (Zero Correlation)

The final stage of our Canvas evolution resolved the problem of "rejected" low-frequency patterns. An individual pattern (such as the BTC Capitulation Reversal) is rejected because it trades rarely, creating massive "cash drag" that underperforms a simple bull market baseline.

By ensembling the Hurst Trend Breakout and the BTC Capitulation Reversal into a single portfolio (hyp_1785845291447_1), we achieved:

1. Zero Correlation: Cross-strategy correlation of -0.042 (complete risk independence).
2. Drawdown Reduction: Drawdown compressed to -3.78% (compared to Hurst's standalone -6.86%).
3. No Cash Drag: Capital remains allocated to the active trend model, and is only utilized by the capitulation model during market panics (acting as an uncorrelated hedging booster).

graph TD
    A[Zero-Fee Backtests v0.2.6] -->|Friction Shock: Taker 4.4bps| B(5m HFT: REJECTED)
    A -->|Friction Shock: Taker 4.4bps| C(1h Vol Regime: Watch/Rejected)
    C -->|Execution Pivot: Maker 1.5bps| D(SOL Spike Reversal: Sharpe 0.72 -> 0.99)
    D -->|Portfolio Integration| E[Multi-Pattern Ensemble: Correlation -0.042, DD -3.78%]
    C -->|Portfolio Integration| E

Reproduce

All configs are mapped on the Hypothesis Canvas. Backtests can be run in the RLXBT app by loading the combo datasets:

• rlxbt_custom_features_BTCUSDT_1h_combo_v1.csv
• rlxbt_feature_btc_triggered_1m_SOLUSDT.csv