Hull Tactical Market Prediction Challenge Explained?

The Hull Tactical Market Prediction Challenge Explained: Can You Beat the Market?

Every investor knows the mantra: “you can’t time the market.” It’s the cornerstone of the Efficient Market Hypothesis (EMH) — the idea that all available information is already baked into prices, making consistent outperformance impossible. But what if that’s wrong? What if, with the right data and machine learning, you could predict market moves and build a better investing strategy?

That’s exactly the question posed by the Hull Tactical Market Prediction Challenge, a Kaggle competition that ran from September 2025 to June 2026. Sponsored by Hull Tactical, a quantitative asset management firm, this challenge asked participants to forecast daily S&P 500 returns while managing volatility — essentially, to do what EMH says can’t be done.

In this guide, you’ll learn about the competition’s structure, the data, the evaluation metric, and what it took to win. And if you missed the chance to compete, we’ll point you toward a current, pure‑Python forecasting challenge with a $50,000 prize pool.

Who Is Hull Tactical?

Hull Tactical is a quantitative investment firm that runs the Hull Tactical US ETF (HTUS) — an actively managed ETF that uses a proprietary algorithm to adjust its market exposure daily. Instead of staying fully invested at all times, HTUS can dial exposure up or down (even using modest leverage) based on market signals. The goal is to capture upside while limiting drawdowns.

This competition directly mirrors that real‑world strategy. Participants were given a rich dataset of market features and asked to build a model that predicts the S&P 500’s daily excess return, then uses that prediction to determine how much of the portfolio to allocate to the market each day.

Competition Overview: Predicting Predictability

The Hull Tactical challenge was a code‑only Kaggle competition — submissions were notebooks that ran without internet access, up to 8 hours for the training phase and 9 hours during the forecasting phase.

The Task

Predict the daily excess return of the S&P 500 (i.e., the return above the risk‑free rate).
Use that prediction to determine an allocation fraction between 0 and 2 (0 = fully in cash, 1 = fully invested, 2 = 2x leveraged).
Stay within a 120% volatility constraint relative to the market.

This wasn't just a point‑forecasting problem; it’s a risk‑adjusted portfolio allocation task. Participants were judged not only on how well they predicted returns, but on how wisely they used those predictions to manage exposure.

Prizes

A total of $95,000 was up for grabs:

1st Place – $50,000
2nd Place – $25,000
3rd Place – $10,000
4th–6th Place – $5,000 each

With 17,926 registrations and over 3,200 teams, the competition was fiercely contested.

The Data: A Broad Set of Market Features

The dataset provided daily observations spanning decades, with multiple feature groups:

M* (Market Dynamics / Technical)
E* (Macro Economic)
I* (Interest Rates)
P* (Price / Valuation)
V* (Volatility)
S* (Sentiment)
MOM* (Momentum)
D* (Dummy / Binary)

The target variables included forward_returns, risk_free_rate, and market_forward_excess_returns. A mock test set was provided, but the true evaluation used a hidden forecasting period after the submission deadline. Crucially, the public leaderboard score during the training phase was not meaningful — it used a historical window that was part of the training data, so overfitting was a huge risk.

For quants who enjoy diving into feature engineering, this dataset was a playground. The challenge was to identify which signals actually had predictive power out‑of‑sample — a task that aligns closely with the philosophy behind our walk‑forward testing guide and time‑series cross‑validation tutorial .

Evaluation: A Sharpe‑Based Metric with Penalties

The evaluation metric was a custom variant of the Sharpe ratio that penalized two things:

Excessive volatility — if your strategy’s volatility exceeded 120% of the market’s volatility, you were penalized.
Failing to outperform — if your annualized return didn’t beat the market’s, you were penalized.

In essence, the metric rewarded strategies that delivered high risk‑adjusted returns while staying within a reasonable volatility envelope. This forced participants to think about risk management as an integral part of the prediction problem — not an afterthought.

For a broader look at how to think about risk‑adjusted performance and avoid false discovery, see our articles on the Deflated Sharpe Ratio and walk‑forward validation .

How to Win (or What We Can Learn)

Because the competition’s leaderboard was intentionally uninformative during the training phase, winners had to rely on rigorous out‑of‑sample validation rather than chasing public scores. A few strategies likely separated the top performers:

Time‑series cross‑validation: Using expanding‑window or sliding‑window splits to simulate real‑time forecasting, as covered in our time‑series CV guide .
Feature engineering from lagged data: Since the test set provided lagged forward returns (one day behind), careful alignment of targets and features was critical to avoid look‑ahead bias.
Volatility targeting: Instead of just predicting returns, successful models likely incorporated volatility forecasts to scale positions appropriately — staying within the 120% constraint while maximizing Sharpe.
Ensemble of simple models: Given the complexity of the feature set, blending multiple models (e.g., LightGBM, neural nets, and linear models) often improves robustness.

Many of these skills overlap with those used in other quant competitions, such as the Jane Street Kaggle challenge or the IMC Prosperity challenge .

Lessons for Aspiring Quants

Even if you didn’t enter, the Hull Tactical challenge teaches a valuable lesson: markets may not be perfectly efficient, but they’re not easy to beat either. With a rich dataset and modern ML, it’s possible to find signals — but translating them into robust, low‑volatility outperformance requires discipline and sound risk management.

And if you enjoyed the challenge of building a volatility‑constrained market‑timing model, you might enjoy our current AlphaNova Competition 5 — a pure‑Python, walk‑forward cross‑sectional forecasting competition with a $50,000 USD prize pool. Open until July 31, 2026, it’s a perfect next step to test your skills with obfuscated financial data and a greedy quality‑selection process that rewards originality.

Where to Compete Next

The Hull Tactical competition may be over, but plenty of other quant challenges are still open or returning annually. Check out our comprehensive guide to the top quant hackathons and data science competitions for a full rundown, including:

WorldQuant BRAIN & IQC – Alpha mining with paid consulting opportunities
Citadel Datathon – Team‑based data sprints with interview eligibility
Rotman International Trading Competition – Live simulated trading for university students

And if you’re looking for a current, open challenge that will leave you with a portfolio‑worthy Python project, AlphaNova Competitions are waiting.

The Hull Tactical Market Prediction Challenge asked a bold question: can you time the market with machine learning? While only a few walked away with prizes, every participant walked away with a deeper appreciation for the complexity of financial markets — and the skills to keep trying. The challenge may be over, but the quest for alpha isn’t.