Park Factors: Do They Actually Improve Predictions?

What Are Park Factors?

Every MLB ballpark is different. Coors Field sits a mile above sea level where the thin air lets fly balls carry. Petco Park has a marine layer that knocks them down. Fenway has the Green Monster. These aren't cosmetic differences — they directly affect how many runs are scored.

A park factor quantifies this: 1.0 is neutral, above 1.0 favors hitters, below 1.0 favors pitchers.

Calculating Our Own

We calculated park factors from 7,292 games across 2023-2025 using our Retrosheet data. For each park, we compared average runs scored in home games vs. that team's road games.

These numbers align with established baseball wisdom and correlate well with FanGraphs' published park factors.

Applying to Markov Chains

The tricky part: how do you inject park factors into a 25-state transition matrix simulation? We tested three approaches:

1. Simple run scaling — multiply predicted runs by the park factor
2. Conservative (50% effect) — apply half the adjustment to reduce overcorrection
3. Full environmental — adjust both teams' scoring expectations

The Results

We backtested all three methods on full seasons. The results were... mixed.

The per-park breakdown tells a more interesting story:

Ironically, the most extreme parks (Coors, T-Mobile) where park factors are largest actually saw worse accuracy with adjustments. The parks that benefited most were mid-range venues.

The Weather Question

Park factors capture the average effect of a venue. But day-to-day weather — temperature, wind, humidity — creates game-specific variation that static park factors miss.

A 95°F day at Wrigley with the wind blowing out is a completely different park than a 55°F night with wind blowing in. Services like BallparkPal provide game-time weather analysis, but they require paid subscriptions.

This remains an open area for future work — and potentially the biggest untapped environmental signal.

Conclusion

Park factors are real physics — there's no question that Coors Field produces more runs than Petco Park. But translating that physical reality into prediction accuracy is harder than it sounds.

The home team's park advantage is already partially captured by Elo ratings (which reflect a team's full home record). Adding explicit park factors on top provides small, inconsistent, and sometimes negative marginal value.

Our recommendation: Use park factors for score/total predictions (where they genuinely help) but not for win prediction (where Elo already accounts for home environment). Dynamic weather data, when available, may unlock more consistent gains.