Research / Working Paper

Monte Carlo Filter Evaluation in Walk-Forward Strategy Selection

A reproducibility-first investigation into Monte Carlo filtering as a robustness signal for walk-forward strategy selection — applied to the firm’s full 30-asset corpus spanning crypto majors, large- and mid-cap altcoins, and three forex pairs, totalling roughly 23 million strategy-windows. Detailed per-window aggregates published for the four most-traversed assets — BTC, DOGE, BNB, SOL — illustrate the broader pattern.

Read on SSRN

abstract_id=6636018

Reproducibility package

Monte-Carlo-paper — Python, Rust & R cross-verified

Abstract

We evaluate Monte Carlo filtering as a candidate robustness signal in a walk-forward backtesting pipeline applied to a large universe of systematic strategies. The empirical setting is deliberately exhaustive: 30 asset/timeframe partitions spanning crypto majors (BTC, ETH, BNB, SOL), large altcoins (DOGE, XRP, LTC, AVAX, LINK, BCH, DOT, TRX), 15 mid-cap names, and three forex pairs (AUDUSD, USDCAD, USDCHF), with roughly 38,000 parameterised strategies per asset evaluated under a sliding walk-forward optimisation regime across multiple lookback horizons. The total corpus contains approximately 23 million strategy-windows once IS / OOS samples and Daru Finance’s proprietary perturbation suite are unrolled.

At the strategy level, in-sample profitability is found to carry near-zero predictive power for next-window out-of-sample profitability — a finding that is robust across assets, timeframes and indicator families. At the portfolio level, however, applying a robustness filter to the same population produces durable improvements in strong regimes; the filter’s own internal signal is shown to telegraph regime breakdown in advance.

The paper’s methodological contribution is the demonstration that the locus of detectable edge in this empirical setting is the evaluation process, not the strategies it operates upon — and that practical risk management for systematic populations is an evaluation-process problem, not a strategy-search problem. The specific operational construction of the filter used by Daru Finance in consulting engagements is proprietary and is not disclosed in the public paper.

Methodology summary

Walk-forward optimisation across multiple lookback horizons.
Monte Carlo robustness assessment situated within the published filter literature.
Three-language reproducibility: Python (analysis & figures), Rust (compute), R (verification).
Pixel-level reproducibility for every figure in the paper from committed aggregates.

Cite this paper