# Backtesting Event-Driven Strategies

> Go beyond factors, trade world events. Backtest against 30 years of dated shocks in 95 languages on the open web with zero look-ahead bias by construction.

**URL:** https://nosible.com/backtesting-event-driven-strategies

## Hero

For event-driven quants and academic backtesters.

**Go beyond factors, trade world events.**

The factor zoo is saturated.

Six hundred-plus published factors. Two-thirds fail at a t-stat of 1.96. Eight in ten fail under multiple-testing discipline.

CTAs:
- **Start Trial** → https://nosible.com/start-trial
- **See the research** → #evidence

## 01 · The saturation — The factor zoo is saturated.

Carhart momentum. Fama-French five. Quality. Low volatility. Defensive. Profitability. Investment. Liquidity. The published library is fixed. Replication studies converge: most factors do not survive out-of-sample.

A side-by-side split makes the saturation visible at a glance: one library is exhausted, the other compounds with the world.

### Factor zoo (Fixed library)

One library. Crowded by replication. Decays on publication.

| Value | Label |
| --- | --- |
| 600+ | Published factors |
| 65% | Fail at t ≤ 1.96 |
| 82% | Fail under multiple-testing |
| ~50% | Sharpe lost post-publication |

### Event universe (Compounding)

Many dimensions. Open-ended supply. No fixed inventory to crowd.

| Value | Label |
| --- | --- |
| 100M+ | Indexed events |
| 300K+ | Sources |
| 95 | Languages |
| 30 years | Point-in-time |

// Fixed library on the left. Open-ended supply on the right.

## 02 · Factor decay and event alpha · 2024–2025 — Papers on factor decay, backtest methodology, and event-driven alpha.

The first row benchmarks factor saturation and backtest methodology. The second is recent LLM-on-text work for event-driven alpha.

// Index: 01 to 03: factor-zoo saturation. 04 to 05: backtest methodology. 06 to 08: LLM-on-text work for event-driven alpha.

1. **Not All Factors Crowd Equally: Modeling, Measuring, and Trading on Alpha Decay** — Lee, 2025, arXiv. [arxiv.org/abs/2512.11913](https://arxiv.org/abs/2512.11913)
   Derives a hyperbolic alpha-decay form K/(1+λt); finds 7 of 8 Fama-French factors are systematically crowded under the resulting test.

2. **Why Has Factor Investing Failed? The Role of Specification Errors** — López de Prado & Zoonekynd, 2024, SSRN. [papers.ssrn.com/sol3/papers.cfm?abstract_id=4697929](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4697929)
   Factor strategies fail from associational rather than causal specification choices; the factor mirage is separate from data-mining failures.

3. **Causal Inference in Financial Event Studies** — Goldsmith-Pinkham & Lyu, 2025, arXiv. [arxiv.org/abs/2511.15123](https://arxiv.org/abs/2511.15123)
   Shows standard event-study factor models give inconsistent abnormal returns when misspecified, especially during volatile periods and long horizons.

4. **The Three Types of Backtests** — Joubert, Sestovic, Barziy, Distaso & López de Prado, 2024, SSRN. [papers.ssrn.com/sol3/papers.cfm?abstract_id=4897573](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4897573)
   Practitioner taxonomy of walk-forward, resampling, and Monte Carlo backtests with explicit guidance on Sharpe inflation under multiple trials.

5. **A Test of Lookahead Bias in LLM Forecasts** — Gao, Jiang & Yan, 2025, arXiv. [arxiv.org/abs/2512.23847](https://arxiv.org/abs/2512.23847)
   Statistical test detecting whether an LLM was trained on the very headlines a researcher feeds it during backtest.

6. **Profit Mirage: Revisiting Information Leakage in LLM-based Financial Agents** — Li, Zeng, Xing, Xu & Xu, 2025, arXiv. [arxiv.org/abs/2510.07920](https://arxiv.org/abs/2510.07920)
   Quantifies how LLM trading agents post in-sample returns that collapse to zero once the model's knowledge window ends.

7. **Event-Aware Sentiment Factors from LLM-Augmented Financial Tweets** — Wang & Wei, 2025, arXiv. [arxiv.org/abs/2508.07408](https://arxiv.org/abs/2508.07408)
   Event-tagged categories from LLM-augmented tweets clear Sharpe magnitudes near 0.38 across 1 to 7 day horizons, significant at 95 percent.

8. **Uncertain Regulations, Definite Impacts: SEC Regulatory Interventions on Crypto Assets** — Saggu, Ante & Kopiec, 2024, FRL. [arxiv.org/abs/2412.02452](https://arxiv.org/abs/2412.02452)
   Hand-curated SEC enforcement events drop crypto prices around 5 percent in three days and as much as 17 percent over a month.

## 03 · The asymmetry — The event supply has no fixed inventory.

Factors saturate because the published library is finite. Researchers mine the same eight decades of CRSP across the same characteristics. Every new sanctions wave, tariff list, central-bank pivot or supply-chain shock adds to the event supply.

Three reasons event-driven alpha does not decay the way factors do:

- **01 — The supply grows.** Tomorrow's events are not in today's archive. New sanctions, tariffs and central-bank pivots add to the supply each week.
- **02 — The signal is dated.** Every event has a verified publication minute. There is no point-in-time ambiguity to mine away after the fact.
- **03 — The reaction is multilingual.** First disclosure happens in 95 languages. The English wire usually arrives twenty minutes later, by which time the move is in.

## 04 · NOSIBLE WORLD — An archive of open-web events, replayed at the publication minute.

Events are dated to first public disclosure. Revisions are preserved alongside the original. The same query run today and six months from now returns the same result from a frozen snapshot.

| Value | Label |
| --- | --- |
| 100M+ | Events |
| 300K+ | Sources |
| 95 | Languages |
| 30 years | Point-in-time |

### Named cases

Three real, dated shocks used as backtest stress cases. Each is verifiable against the public record on the date given.

| Risk | Place | Date | Event |
| --- | --- | --- | --- |
| Geopolitical | United Kingdom | 2016·06 | UK votes to leave the EU; sterling and risk assets repriced in hours. |
| Bank run | Silicon Valley Bank | 2023·03 | Silicon Valley Bank collapses inside 48 hours of the first public disclosure. |
| Carry unwind | Japan | 2024·08 | Yen carry unwind drives a global cross-asset selloff in a single session. |

## 05 · What you build — Backtests on NOSIBLE WORLD.

Wire NOSIBLE WORLD into your existing event-study framework, factor library, or bias-diagnostic stack.

- **§01 Event-driven backtests** — Anchor every retrieval to the verified publication minute. Cross-validate with purged folds and defend the result against any referee.
- **§02 Sanctions and tariff alpha** — Trade triggers fire on first public disclosure across 95 languages. The English wire usually arrives twenty minutes later, by which time the move is in.
- **§03 Crisis-period robustness** — Replay your strategy across frozen historical snapshots. Pin the API version and reproduce the same audit a year from now.

> **Field note:** A backtest is reproducible only if its snapshot is frozen.

## 06 · Get started — Run event-driven backtests on NOSIBLE WORLD.

Talk to us about API access and historical snapshots.

CTA: **Start Trial** → https://nosible.com/start-trial

## Related
- [Home](https://nosible.com)
- [NOSIBLE Search API](https://nosible.com/search-api)
- [Research](https://nosible.com/blog)
- [Start Trial](https://nosible.com/start-trial)