Philadelphia finished last season with 98 points. Four games into this one, the Flyers have two points. They have also generated only 27% of the even-strength expected goals in those games.
No team in the last five seasons had an even-strength xG share that low over its first four games. The lowest was 33%.
Early October produces stories like that every year. We wanted to know which early numbers actually tell us something about the rest of a season, and when they start to tell us more than last season's standings do.
How we tested it
We took every regular season from 2021–22 through 2025–26: 160 team-seasons. For each team, we measured its first N games and compared those numbers with its points percentage over the remaining games. For example, a team's first five games were compared with games 6 through 82.
We tested five early measurements:
- Points percentage: the standings
- Goal share: goals for divided by all goals
- Shot-attempt share: shot attempts for divided by all shot attempts
- xG share: expected goals for divided by all expected goals, in all situations
- Even-strength xG share: the same calculation, using only even-strength play
The baseline was the team's points percentage from the previous season. A team's record from last season is known before opening night, so an early-season number is only useful if it adds to that baseline.
Last season is hard to beat
The table shows each measurement's correlation with points percentage over the rest of the season. Higher is better.
| Games played | Points % | Goal share | Even-strength xG share | Last season's points % |
|---|---|---|---|---|
| 1 | 0.17 | 0.15 | 0.30 | 0.56 |
| 3 | 0.30 | 0.28 | 0.45 | 0.56 |
| 5 | 0.33 | 0.32 | 0.45 | 0.55 |
| 10 | 0.45 | 0.43 | 0.56 | 0.55 |
| 15 | 0.50 | 0.49 | 0.60 | 0.55 |
| 20 | 0.56 | 0.56 | 0.65 | 0.55 |
| 25 | 0.63 | 0.62 | 0.64 | 0.54 |
| 40 | 0.68 | 0.69 | 0.64 | 0.50 |
After three games, last season's points percentage predicts the rest of the new season almost twice as well as the new standings do.
Even-strength xG share passes last season's record at about ten games. Points percentage does not catch up until somewhere between 20 and 25 games, around American Thanksgiving. Goal share follows almost the same path as the standings, because goals decide the standings.
By 40 games, the standings have overtaken xG share. Over half a season, a team's results include repeatable skills that our xG measure does not fully capture, such as shooting talent and goaltending.
Combining the signals
The best early-season estimate combines all three: last season's record, this season's points percentage and this season's even-strength xG share. We fitted that combination with one season held out at a time, so each season was predicted without using its own results.
| Games played | Error, last season only | Plus current points % | Plus even-strength xG share |
|---|---|---|---|
| 3 | 0.082 | 0.081 | 0.078 |
| 10 | 0.085 | 0.083 | 0.077 |
| 20 | 0.089 | 0.082 | 0.075 |
The errors are root-mean-square errors in rest-of-season points percentage. Over a 70-game remainder, 0.01 is about 1.4 standings points.
Adding the current record barely helps after three games. Adding xG share helps more. After ten games, xG share contributes more to the estimate than the standings do.
Hot and cold starts
How did teams do after very good or very bad starts? These are small groups, so treat them as examples rather than rules.
| Start after five games | Teams | Rest-of-season points % | Average final points |
|---|---|---|---|
| 10 points (5–0–0) | 9 | .652 | 110 |
| 8 points | 22 | .571 | 96 |
| 5 or 6 points | 60 | .553 | 91 |
| 2 points | 9 | .506 | 80 |
| 0 or 1 point | 6 | .381 | 59 |
The average team earns about .553 of available points under the NHL's loser-point system. Perfect five-game starts did point to good teams. Very bad starts usually pointed to bad teams. The middle tells us almost nothing.
Three games tell us even less. The eleven teams that lost their first three in regulation went on to earn .524 of their remaining points. That was almost the same as the .537 earned by teams with two points from their first three games.
This season's loudest starts
The table applies the combined estimate to every team as of October 5, 2026. It uses each team's prior points, current record and even-strength xG share. Most teams have played two or three games. The projections stay close to last season by design.
| Team | GP | Record | Even-strength xG share | Last season's points | Projected points |
|---|---|---|---|---|---|
| COL | 2 | 2–0–0 | 59% | 121 | 111 |
| TBL | 3 | 2–1–0 | 61% | 106 | 106 |
| DAL | 3 | 1–2–0 | 57% | 112 | 101 |
| CAR | 3 | 1–1–1 | 50% | 113 | 99 |
| FLA | 3 | 1–0–2 | 63% | 84 | 99 |
| NYR | 4 | 3–1–0 | 48% | 77 | 87 |
| TOR | 3 | 1–2–0 | 47% | 78 | 81 |
| CHI | 3 | 0–3–0 | 36% | 72 | 68 |
| PHI | 4 | 0–2–2 | 27% | 98 | 73* |
*Philadelphia's xG share is below anything in the historical sample, so its projection extends the fitted relationship beyond the range we observed. Treat it as a direction, not a number.
Some readings:
- Dallas is 1–2, but probably fine. The Stars have controlled 57% of even-strength expected goals. The early losses do not erase a 112-point team.
- The Rangers' 3–1 start looks better than their play. New York has controlled 48% of even-strength expected goals after a 77-point season. The estimate gives them a modest bump, not a new identity.
- Florida has two overtime losses but has controlled 63% of the expected goals. It is the best early share in the league, and it moves the estimate more than the record does.
- Philadelphia is the real warning sign. Four losses are not much evidence. Being outchanced this badly over four games has no precedent in our five-season sample.
What to take from it
Before about ten games, trust last season's standings more than this season's standings. Expected-goals share becomes useful sooner, and the record catches up later. A team's early-season record mostly tells us about the schedule it played and a few bounces. If a team is badly outchanced, though, start paying attention.
Method notes: regular-season games only; points include overtime and shootout losses; xG uses our score-neutral model. Arizona's 2023–24 season is the Utah baseline for 2024–25. The combined estimate is a linear fit across 2022–23 through 2025–26 and was validated one held-out season at a time.
Get NHL forecasts in your inbox
Occasional emails with NHL forecasts, analysis, and model updates.