NHL Game Totals: Why Goalies Are the Whole Story

The Thesis

In hockey, one player touches the puck on nearly every single play against their team. The goalie. No other major sport has a single position with this much influence over game outcomes.

NHL game totals — the over/under on combined goals — are primarily a function of who's in net. Our model is built around that fact.

What We're Predicting

For each NHL game, we project the total number of goals scored by both teams combined. We compare our projection to the market's over/under line and generate OVER, UNDER, or HOLD signals depending on where we see edge.

The Model

Step 1: Base Projection

We start with team-level offensive and defensive metrics, measured per 60 minutes of play:

• GF/60: Goals scored per 60 minutes (offensive strength)
• GA/60: Goals allowed per 60 minutes (defensive strength)

The base projection combines both teams:

home_expected = f(home_GF60, away_GA60)
away_expected = f(away_GF60, home_GA60)
base_total = home_expected + away_expected

This gives us a neutral starting point — what we'd expect if both teams played with league-average goaltending.

Step 2: Goalie Tier Adjustment

This is where our model diverges from most. We classify every NHL goalie into one of six tiers:

These multipliers adjust the expected goals against for each team. When an elite goalie starts, we expect 15% fewer goals on that side. When a backup gets the nod, we expect 15% more.

The spread matters. A game between two elite goalies could see our projection drop by nearly a full goal compared to baseline. A game with two backups could push the other direction just as hard.

Step 3: The Back-to-Back Edge

This is the single biggest factor in our model and the market's most consistent blind spot.

When a goalie starts on the second night of a back-to-back, they allow roughly 12% more goals than their baseline. The data is unambiguous:

Why does this create edge? Because the market often doesn't know who's starting until hours before puck drop. When a starting goalie is announced on a B2B, the line adjusts — but not always enough.

Our model flags B2B goalie starts and adjusts projections before the market fully prices them in.

Step 4: Team Rest

Goalie fatigue isn't the only rest factor. Teams on back-to-backs also play slightly differently:

Note the asymmetry: goalie fatigue increases goals (pushing totals up), while team fatigue decreases offense (pushing totals down). A B2B game often has both effects partially canceling out, which is why the goalie factor is more predictive.

Step 5: Pace Adjustment

Some teams play fast. Some play trap hockey. Pace directly affects the number of scoring chances per game.

We measure opponent pace and adjust the total projection accordingly. Fast-paced matchups generate more shots and more goals. Defensive grinders suppress volume.

Step 6: Clamping

We bound our final projection between 4.0 and 9.0 total goals. This prevents extreme combinations of adjustments from producing unrealistic projections. An NHL game effectively can't end 1-0 in regulation (it can, but the probability is negligible for modeling purposes), and 10+ goal games are exceedingly rare.

Converting to Probability

Once we have an adjusted projection, we convert it to a probability using a normal distribution with a standard deviation of 1.4 goals:

P(over the line) = 1 - CDF((line - projection) / 1.4)

The 1.4 standard deviation captures the inherent variance in hockey scoring. Even accurately projected games can swing by a goal or two based on bounces, power plays, and empty-net situations.

Edge Calculation

We compare our probability to the market's implied probability:

edge = model_prob - market_prob

Signals are generated at three confidence levels:

The higher threshold for HIGH confidence (12% vs 10% for NBA) reflects the inherent variance in hockey. Goals are rare events — a single lucky bounce changes the outcome. We need wider edge to overcome that noise.

Goalie Confirmation

One critical feature of our dashboard: the goalies confirmed flag.

NHL starting goalies aren't officially announced until warmups, roughly 30 minutes before puck drop. Before confirmation, our projections use expected starters. After confirmation, the projections firm up significantly.

We flag unconfirmed goalie situations on the dashboard because:

1. If the expected starter gets scratched and the backup goes in, the projection changes materially
2. B2B situations are especially volatile — coaches sometimes surprise with their goalie decisions
3. Signals generated before goalie confirmation carry more uncertainty

Real Example: B2B Backup vs. Elite Starter

Matchup: Team A (on B2B, backup goalie) at Team B (2 days rest, elite goalie)

Base projection: 6.2 goals

Adjustments:

• Team A backup goalie: × 1.15 (on goals against Team A)
• Team A B2B rest: × 1.12 (goalie fatigue) × 0.96 (team fatigue)
• Team B elite goalie: × 0.85 (on goals against Team B)
• Team B rested: × 0.98 (goalie) × 1.01 (team)

This pushes the projection in both directions. Team A's side sees more goals (backup + fatigued). Team B's side sees fewer (elite + rested).

The net effect depends on the magnitudes, but often these B2B-backup-vs-elite situations produce UNDER signals — the elite goalie's suppression outweighs the backup's leakiness because the rested team also plays better defense.

Performance Tracking

We track every signal with full transparency, broken down by direction:

• OVER signals: Win rate and P&L for games we signaled over
• UNDER signals: Win rate and P&L for games we signaled under
• Overall: Combined performance across all signals
• Average edge: The mean edge at signal generation

All performance data is visible on the hockey dashboard.

What Makes This Work

1. Goalie-Centric Modeling

Most public totals models focus on team stats. We focus on the individual who has the most influence: the goalie. A bad team with an elite goalie is a fundamentally different proposition than the same team with their backup.

2. B2B Detection

The back-to-back adjustment is our single largest edge source. The market knows B2Bs matter, but it consistently underprices the impact — especially when the B2B goalie start isn't announced until late.

3. Six-Tier Granularity

Binary "starter vs. backup" isn't enough. The difference between an elite and above-average goalie matters. The difference between a below-average starter and a backup matters. Six tiers capture these distinctions.

Current Limitations

1. Goalie tier classification: Tiers are based on save percentage and historical performance. Mid-season breakouts or slumps can lag in our classification.
2. Score effects: We don't yet model how game state (leading vs. trailing) affects late-game goalie pulls and empty-net goals.
3. Power play impacts: Specific penalty kill and power play efficiencies aren't factored into the base projection.
4. Playoff adjustments: Playoff hockey is structurally different (tighter checking, better goaltending) but we don't yet adjust for it.

What's Next

• Score-state modeling: Adjusting projections based on expected game flow (close game vs. blowout)
• Special teams integration: Power play and penalty kill rates as explicit model inputs
• Goalie trend detection: Faster tier updates when a goalie is trending up or down mid-season
• Period-level projections: Breaking the game into periods for more granular signals

The Bottom Line

NHL game totals are driven by goaltending quality and rest patterns. Our model captures these factors with a six-tier goalie system, explicit back-to-back fatigue modeling, and pace adjustments — then converts the projection into calibrated probabilities to find market mispricings.

The goalie is the story. We read it better than the market.

---

See live NHL signals on the Hockey Dashboard. Customize parameters in Settings.