Opening week in the NHL is a minefield for bettors, but it is also one of my favorite puzzles. The trick is separating useful information from the noise. I start with last season’s 5-on-5 numbers, add a small preseason signal, then layer in roster changes, coaching, goaltending, travel, rest, and expected lineups. From there, I turn the projection into a fair price, compare it with the market, and size the risk without getting carried away by one weird game.
Building NHL Opening Week Projections
Opening week is full of hype, but hype is not a reliable input. I would rather anchor a model to stable numbers from last season and let new information move the projection gradually. The biggest starting point is 5-on-5 performance, especially expected goals share, expected goals for per 60 minutes, and expected goals against per 60.
Last season’s 5-on-5 xGF% gives you a useful picture of territorial control. I also separate xGF/60 from xGA/60 because roster changes do not always affect offense and defense in the same way. A team can add a top-six scorer without fixing its defensive structure, while losing a top-pair defenseman can hurt its suppression more than its offense.
Finishing should be included, but it needs heavy regression. I look at goals versus expected goals and give more weight to established elite shooters than to depth players who happened to run hot. The same idea applies to special teams. Power-play xGF/60, penalty-kill xGA/60, and actual scoring rates all matter, but expected rates should carry more weight early because the sample is tiny.
Preseason is useful mainly for deployment. I would generally give it around 10% to 20% weight rather than treating preseason box scores as a new truth. If a player moves onto the top power-play unit, earns a major increase in ice time, or makes the roster in a larger role, that information is valuable. A random three-game shooting spike is much less useful.
Roster churn is another major piece. For each meaningful addition or departure, I want an estimate of expected even-strength ice time, on-ice xGF/60, on-ice xGA/60, and special-teams usage. A top-six forward can change expected scoring, while a shutdown defenseman may have a larger effect on chances against. New rookies should usually get conservative ice-time assumptions until their role is confirmed.
Coaching changes deserve a similar level of caution. A new system can eventually change shot volume, defensive structure, or special-teams performance, but I would not make a huge numerical adjustment before seeing evidence. Small changes are easier to defend and easier to correct later.
The final piece is opponent adjustment. Raw team rates can look better or worse simply because of the opponents a team faced. I adjust expected scoring and defending toward the quality of the opposition and repeat that process a couple of times. The goal is not to build a complicated machine that memorizes every matchup. It is simply to avoid treating a soft or difficult schedule as pure team talent.
A simple opening-week team profile might look like this in my notes: Team A has a 2.55 5-on-5 xGF/60, 2.30 xGA/60, a small positive finishing adjustment, a strong power play, and a recent addition expected to play on the top six. Team B has a 2.40 xGF/60, 2.45 xGA/60, a small negative finishing adjustment, and a weaker penalty kill after losing an important defender. Those numbers are illustrative, but the process is the important part.
Schedule, Travel, and Environment
Schedule context can create small edges that are easy to miss. I tag every game for rest, back-to-back status, travel, time-zone changes, and unusual start times. I do not want a giant travel penalty baked into every road game. The adjustment should be modest and tied to the actual situation.
A team playing on zero days of rest can see its expected offense trimmed and its expected goals against increased slightly. In an opening-week model, something around a few percentage points can be enough to capture the idea without pretending we know exactly how tired every player will be.
Travel can also matter in the first period. A long trip followed by an immediate road opener is a reasonable spot for a mild first-period offensive dampener. Time-zone changes can be included as another small adjustment, especially when combined with a back-to-back. These effects should not be stacked repeatedly. If travel is already reflected in the rest adjustment, do not add the same penalty again.
Altitude is another environmental factor. A game in Denver can produce a different pace and fatigue profile than a game at sea level. I would keep the adjustment small and increase it slightly when the visiting team is already dealing with a difficult schedule.
Opening week also tends to bring more uncertainty around penalties, line combinations, and game pace. Rather than pretending the model knows everything, I widen the error range. If my normal model uncertainty is relatively tight later in the season, opening week should have a wider band. That also means smaller stakes.
Goaltending and Lineup Intel
Goaltending can move an NHL projection quickly, so I assign starter probabilities before the game and update them as reliable lineup information arrives. A simple starter model can consider recent workload, rest, coaching tendencies, injuries, and back-to-back scheduling.
A goalie who has started several games in a short window may deserve a small fatigue adjustment. I would rather make that adjustment cautiously than assume a goalie automatically becomes bad after a heavy workload. The defensive environment matters too. A goalie facing a high volume of dangerous chances should not be judged by save percentage alone.
My goalie prior blends recent seasons with career performance and regression toward league average. Then I adjust it for the quality of the defense in front of the goalie. A team that consistently allows high-danger chances can make a strong goalie look worse, while a team that controls shot quality can provide a better environment.
Lineup changes deserve similar attention. A top-six forward being ruled out can lower expected scoring and power-play efficiency. A top-four defenseman missing can increase expected goals against and hurt the penalty kill. The key is avoiding double counting. If one missing winger is already responsible for a lower team xGF/60, do not also make a full finishing reduction for the same player.
Confirmed power-play units are especially useful. A shooter moving from the second unit to the first can create a real change in expected opportunity. The same goes for a playmaker replacing a shooter. The projection should respond to the role, not just the player’s name.
After games begin, I log 5-on-5 xGF/60, xGA/60, high-danger chances, line and defensive-pair performance, special-teams results, and goalie performance. One game should barely move the prior. A learning rate around 5% to 10% for the first game is much more sensible than rebuilding the model after one unusual result.
Turning Projections Into Betting Decisions
Once the hockey projection is built, the next job is turning it into a price. Suppose the model gives the home team a 56% chance to win. Its fair moneyline is roughly -127. That number matters more than the team logo.
If the no-vig market price implies a 52.5% chance, the model has a meaningful difference to investigate. If the market already implies something close to 56%, there may be no betting edge even if the model strongly likes the team. The goal is not to predict winners in isolation. It is to find prices that are better than the probability suggests.
For American odds, positive prices can be evaluated with expected value using the probability of winning and the potential profit. Negative prices use the same idea, but the payout is smaller. I prefer setting a minimum edge threshold rather than forcing a bet every time the model differs from the market by a tiny amount. Early in the season, a 1.5% to 2% edge can be a reasonable starting threshold because uncertainty is higher.
Staking should be conservative. Reduced Kelly or flat units make more sense to me during opening week than aggressive sizing. A model can be directionally right while still being badly calibrated on a specific input. Keeping early bets small protects the bankroll while the new season provides information.
I also track the opening price, the price I bet, and the closing price. Closing-line movement is useful because it helps separate a bad process from a bad result. A losing bet can still have been based on a reasonable price, while a winning bet can be poorly priced. Over time, that distinction matters.
The model should also have a post-game review. I compare projected and actual goals, expected goals, special-teams minutes, and goalie performance. Then I look for repeated errors. If the model repeatedly overestimates a new power-play unit, that is a reason to adjust the input. One strange game is not.
The ATSwins Approach
ATSwins fits naturally into this workflow because the platform brings predictions, player props, betting splits, and profit tracking into one place. Instead of treating a projection as the end of the process, I use it as a starting point for comparing model expectations with the current market.
For opening week, the useful part is keeping the process organized. I can look at a model number, compare it with the available price, consider lineup and goalie information, and decide whether the edge is strong enough to act on. The point is not to blindly follow a pick. It is to have a consistent framework for deciding when a price is worth taking.
Player props can also become interesting when roles change. A player promoted to a top power-play unit or given more offensive ice time may have a different projection than his previous-season average suggests. At the same time, preseason results should not automatically be treated as proof that the new role will produce a huge statistical jump.
Profit tracking is just as important as the predictions themselves. Tagging plays by reason, such as a goalie edge, a special-teams mismatch, or a rest spot, makes it easier to see which assumptions are actually working. If a certain type of opening-week bet keeps losing value against the closing line, that is useful information for the next update.
Common Mistakes and a Practical Workflow
The biggest mistake is overreacting to preseason box scores. Deployment, role, and fitness information matter more than a small collection of exhibition goals. Another common problem is treating last season’s shooting or save percentage as permanent talent. Regression is essential, with stronger exceptions only for players and goalies who have shown the skill consistently.
It is also easy to ignore the penalty kill when projecting a power play. A team can have a great power-play system and still see its expected production fall if the opposing penalty kill has improved significantly. Travel can be double counted too, especially when rest and travel are modeled separately without checking whether both adjustments describe the same fatigue effect.
I like a simple daily routine. In the morning, I confirm the schedule, rest situation, possible back-to-backs, projected lines, and goalie probabilities. During the day, I update market prices and flag meaningful differences between the model and the no-vig market. Before puck drop, I confirm the goalie and major lineup news, then finalize the price and stake.
After the game, I record the result and the underlying numbers. I do not immediately rewrite the model because a team scored four goals on a small number of chances. Instead, I update the inputs gradually and watch for patterns across several games.
The same process works for a hypothetical opening-night matchup. Imagine Team A has better 5-on-5 numbers, a stronger power play, and a rested starting goalie, while Team B is traveling across two time zones and has a weaker penalty kill. The model might produce a 56% home win probability and a fair price around -127. If the market is -120, the edge is small enough that the decision depends on the rest of the inputs. If the market moves to -135 without a meaningful new piece of information, the value may disappear. The important point is that the number changes the decision, not the name of the team.
Conclusion
Opening week is about managing uncertainty, not pretending uncertainty does not exist. Start with last season’s 5-on-5 performance, use preseason mainly to understand roles, make roster and coaching adjustments carefully, and give goaltending, rest, travel, and special teams the attention they deserve.
From there, turn probabilities into fair prices and compare those prices with the no-vig market. Keep stakes smaller while the model calibrates, track closing-line movement, and review the underlying numbers instead of judging the process from wins and losses alone.
Atswins can help keep that workflow in one place by acting as a comprehensive sports prediction engine alongside player props, betting splits, and profit tracking across NFL, NBA, MLB, NHL, and NCAA. The goal is a cleaner process: build the number, compare the price, understand the risk, and keep learning as the season develops.