Temporal Structural Forecasting: A Comprehensive Empirical Refutation of Weak-Form Market Efficiency

This registered report presents the most comprehensive empirical challenge to weak-form market efficiency ever assembled. We introduce Temporal Structural Forecasting (TSF), a methodology that identifies exploitable patterns in historical price data through multi-cycle seasonal decomposition. The study tests 44 preregistered hypotheses across four papers using 346 S&P 500 stocks spanning 11 GICS sectors over 20 years (2006–2025). We establish four independent refutation paths: (1) predictable structure exists in price data, (2) entry timing is exploitable after transaction costs, (3) exit timing is independently exploitable regardless of entry methodology, and (4) temporal structure improves factor portfolio returns. Preliminary results from a 30-stock pilot study (2015–2024) confirm all four refutation paths with 27/44 hypotheses (61%) supported. Most notably, applying TSF exit signals to benchmark strategies improved excess returns in 72.7% of cases (p = 1.19 × 10⁻⁵), demonstrating that exit timing constitutes an independent market inefficiency—a finding that uses zero TSF entry signals. The methodology, hypotheses, and analysis plan are locked via Zenodo deposit prior to primary data analysis on the 346-stock universe.Version 3.0 (Benchmark Exposure Categorization): January 9, 2026. This revision categorizes benchmarks by exposure type and restricts TSF comparisons to valid exposure-matched benchmarks, prior to primary data analysis. No results from the 346-stock universe have been analyzed. Change: Benchmarks are now categorized as Variable Exposure (signal-based, 16 strategies) or Scheduled Exposure (calendar-based, 6 strategies). TSF comparisons are restricted to variable exposure benchmarks only. Rationale: Calendar strategies deploy 100% capital during scheduled windows while signal strategies deploy capital only when signals fire (typically 30-50% average exposure). Comparing variable-exposure TSF against scheduled-exposure calendar benchmarks is methodologically invalid regardless of exposure-matched benchmark adjustments. The exposure-matched benchmark corrects for REALIZED average exposure differences, but cannot correct for fundamentally different exposure PROFILES (continuous variable vs binary scheduled). Affected sections: Section 1.5 Key Terms (exposure type definitions added), Section 2.3 Benchmark Strategies (exposure categorization added), Study 2b H2.7-H2.15 (DUAL EVALUATION removed; TSF compared to signal benchmarks only), Study 3a H3.0a-H3.0d (split by exposure type for independent validation), Study 3b (signal benchmarks only notation added). Version 2.0 (Pre-Analysis Methodology Revision) This revision corrects the benchmark methodology prior to primary data analysis. No results from the 346-stock universe have been analyzed. Change: The buy-and-hold benchmark comparison has been revised to use an exposure-matched benchmark. The original methodology compared strategy returns against a 100% continuously invested buy-and-hold benchmark, creating an invalid comparison for timing strategies that hold cash between signals. Revised methodology: Excess returns are now calculated against an exposure-matched benchmark: EMB = (avg_exposure × B&H_return) + ((1 − avg_exposure) × rf_return), where avg_exposure is the strategy's realized average capital deployment and rf_return is 5% annualized. This ensures valid comparison between strategies with different exposure profiles. Affected sections: Section 1.5 Key Terms: "Excess Return" definition updated Section 2.8 Performance Metrics: Exposure-matched benchmark formula added Hypotheses H2.3 and H2.13: "buy-and-hold" replaced with "exposure-matched benchmark" Rationale: Identified during external review prior to primary data analysis. The correction ensures that a strategy averaging 40% exposure is compared against 40% B&H + 60% risk-free, not 100% B&H.