Does Preseason Performance Predict Anything? What the Data Says

Every October the same ritual plays out: a team goes 4-0 in the preseason, the fanbase starts pricing in a playoff run, and a beat writer files eight hundred words on "what we learned." I wanted to know whether we learn anything at all. So I pulled preseason and regular-season net ratings across six seasons, lined them up, and ran the correlation. The verdict is not kind to October optimism — preseason net rating barely predicts the real thing.

The number, up front

Across 180 team-seasons, the pooled correlation between a team's preseason net rating and its eventual regular-season net rating is r = 0.257. Squared, that's an r² of 0.066 — preseason net rating explains about 6.6% of the variance in how teams actually perform once the games count. The other 93-plus percent is noise, roster changes, health, and basketball that hadn't happened yet.

I want to be precise about what that means, because "barely predicts" is not the same as "predicts nothing." A correlation of 0.257 is positive. It is real. It is also weak, and an intellectually honest read of a weak-but-nonzero signal is that it's still weak. If you handed me two teams and told me only their preseason net ratings, I'd have a faint, mostly-useless lean toward the better one — roughly the forecasting power of noticing which team's bus showed up first.

What r² meansr² is the share of one variable's variance explained by the other — here, the fraction of regular-season net rating you could account for from preseason alone.

Why preseason is mostly noise

The reasons are not mysterious, and once you list them the weak correlation looks less like a surprise and more like the only sane outcome.

The samples are tiny. A team plays four or five preseason games. Net rating over four games is a quantity with enormous error bars; one blowout against a road-tripping opponent's third unit can swing it by double digits. We'd never trust a four-game sample in November, but we'll happily build a narrative on one in October.

The rotations are experimental. Preseason is, by design, a laboratory. Coaches stagger starters, hand twenty minutes to a two-way contract they're evaluating, and run sets they have no intention of using in March. The lineups that produce the preseason net rating are frequently lineups that will never share the floor again.

The stars rest — and so does the other team. Front offices treat preseason as injury-avoidance with a scoreboard attached. Your best player sits the second night of a back-to-back, or plays twelve minutes and showers. The opponent is doing the exact same thing, which means you're often measuring your bench against their bench and pretending it says something about either team's ceiling.

Nobody is trying. There's no scouting, no game-planning, no late-game execution because there are no stakes. The competitive intensity that makes regular-season net rating meaningful is precisely the ingredient preseason lacks. You're timing a sprinter who's jogging.

It's even noisier than the pooled number suggests

The 0.257 figure pools all six seasons together, which flatters it by averaging out the year-to-year chaos. Look at the seasons individually and the instability is the real headline:

In 2021-22 the correlation reached 0.486, which is high enough that you could almost convince yourself preseason matters. Two seasons later, in 2023-24, it was -0.048 — functionally zero, and faintly negative, meaning preseason net rating that year pointed very slightly the wrong way. A signal that swings from 0.49 to below zero across adjacent seasons isn't a signal you can act on. It's a coin that happens to land heads a bit more than half the time when you stack enough flips together.

What actually carries the weight

If preseason explains 6.6%, the obvious question is where the rest comes from — and the honest answer is the boring stuff that doesn't make for fun October content. Prior-season results are a far sturdier starting point; a team that posted a strong net rating over 82 real games is, all else equal, likely to be good again, because 82 games is a sample and four is a rumor. Roster continuity compounds that: the more of last year's minutes a team returns, the more last year's performance carries forward. And health — who's available, and for how many games — swamps nearly everything, which is exactly why it's the variable preseason is least equipped to reveal.

I'm deliberately not putting numbers on those factors here, because I didn't run that regression and I'm not going to invent coefficients to sound authoritative. The point is directional: the durable inputs are the ones built on large samples and real stakes, and preseason is definitionally short on both. This is the same reason serious win-probability models lean on full-season priors rather than a handful of exhibition games, and why reasonable projection systems can disagree so sharply while all quietly agreeing that October box scores deserve almost no weight.

The takeaway

Enjoy the preseason. Watch the rookies, clock the new rotations, note who looks a step quicker than last spring. Just don't mistake any of it for evidence. A team that wins all five exhibitions has told you it can win five low-stakes games against opponents who weren't trying — which, per 180 team-seasons of data, explains 6.6% of what happens next. The other 93% is the season, and the season hasn't started. When someone shows you a glittering preseason net rating in October, the correct response is the same one I'd give for a single hot shooting night: interesting, and almost meaningless.