Dylan Ricciardi
Analyst Intern · PredictionMarketsPicks
Analyst intern building and backtesting NFL simulation models. Pitched and validated the 20-year Madden ratings backtest, then authored the 10,000-season Madden 27 simulation with a full 2010-2024 calibration record.
Dylan Ricciardi is an analyst intern at PredictionMarketsPicks, working on the NFL simulation and backtesting stack. He pitched the 20-year Madden ratings backtest — a walk-forward test across 4,237 real NFL games that found measurable preseason signal in a data source nobody treats as analysis — and the result held up well enough to publish.
He then authored the Madden 27 season simulation: ten thousand runs of the 2026 season built from launch ratings alone, played across the real schedule with real NFL tiebreakers, and calibrated against every season from 2010 to 2024. The work is reproducible from a seed, publishes full distributions rather than single numbers, and reports where it disagrees with the Kalshi win-total board instead of only where it agrees.
A senior at High Technology High School, class of 2027, he is headed toward data analytics.
- Coverage
- NFL Analytics · Monte Carlo Simulation · Prediction Markets
Frequently Asked Questions
Who is Dylan Ricciardi?
Dylan Ricciardi is an analyst intern at PredictionMarketsPicks who builds and backtests NFL simulation models. He pitched and validated the 20-year Madden ratings backtest and authored the 10,000-season Madden 27 simulation, which carries a full 2010-2024 calibration record.
What does Dylan Ricciardi work on?
NFL simulation and model validation — Monte Carlo season simulations, walk-forward backtesting, and calibration. His published work covers the 20-year Madden ratings backtest and the Madden 27 ten-thousand-season simulation, including where those projections diverge from the Kalshi win-total board.
Is the Madden 27 simulation reliable?
It is one independent opinion, not a forecast. The average error on a projected team win total is 2.27 wins, which is why the full distribution is published rather than a single number. The more useful property is calibration: replayed across every season from 2010 to 2024, when the model says something happens 60% of the time, it happens about 60% of the time.
Articles by Dylan
1 article published on PredictionMarketsPicks.