Hot days and rainfall against reported organic volume growth · built 30 Oct 2026, 21:34
Hot days are counted per city, then weighted into one score for Western Europe and one for China. A hot day is a daily high above 25°C in Europe and above 30°C in China. Counts are compared with the same quarter in other years, so Q3 is judged against Q3s. A hot day is assumed to be worth the same in Q2 and Q3, so one summer slope is applied to both. The two quarters differ in how many hot days they carry, not in what a hot day does.
Rainfall is counted alongside heat, a rain day being 1mm or more. In Europe it adds almost nothing once heat is in the model; in China it is the only weather variable that moves anything at all.
Carlsberg's own growth prints carry pricing, channel and portfolio effects, so this tests national volume instead: each series year-on-year against May–September hot days, same calendar month on both sides. Tick the series you trust; the estimate below follows.
—
| Use | Series | Scope | Summer months | % per hot day | t | Hot days / month | Rain-day effect | Not tested because |
|---|
Series in the same country overlap (Germany has three), so tick one per country or the estimate counts that country twice. Nordic summers have only 1–2 hot days a month, so those slopes rest on very few warm days. These are production indices, which follow output, not necessarily what was drunk.
The same hot-day gap priced two ways: with Carlsberg's own Q3 slope, and with the national slope from the ticks above. The third column averages them. Western Europe, against the 2026 base year.
| Quarter | Normal hot days | 2026 hot days | Carlsberg slope | Drag, Carlsberg | National slope | Drag, national | Drag, average |
|---|
Drag is in percentage points of regional organic volume growth: (normal − 2026 hot days) × slope. Both slopes are measured on Western Europe only.
The whole exercise points at Q2 and Q3 on the assumption that those are the big quarters. They are, and by more than the calendar would suggest — but the ranking is not the one people expect. Q2 is the largest quarter in every series here, a little ahead of Q3, so the quarter where Carlsberg's own data showed nothing is the quarter that sells the most beer. That is what makes the next section matter.
From whole years of national monthly data only. An even split would be 25% each.
China's share is from only — whole years, because its series has holes, so read it as indicative. Germany rests on —.
Bars are hot days in the quarter; the line is Carlsberg's reported organic beer volume growth for the region. The two biggest hot summers in the series, 2018 and 2026, are also the two best Q3 volume prints. Q2 prints do not line up with Q2 heat, but the next section shows why that is a measurement problem and not a different kind of beer drinker.
Hollow points are volume figures derived from half-year arithmetic rather than stated by Carlsberg.
One point per year. The line is fitted on these points only.
Hot days above or below normal, translated through the fitted sensitivity.
On Carlsberg's own figures Q2 shows nothing. Including the derived figures it comes out mildly negative (t —); on the stated Q2 figures it is flat (t —). An earlier version of this page read that as a real absence on the grounds that two independently built series agreed. That reasoning was wrong: both series are the same thirteen low-variance quarters, so their agreement says nothing about whether the test had the power to find an effect.
The German monthly data does have that power, and it finds one. Across — April-to-June months an extra hot day is worth —% of volume, t — — against —% in July to September. The gap between the two slopes is —, which is to say they are indistinguishable. A hot day in Q2 is worth the same as a hot day in Q3.
What differs is how many of them there are, and how much that number moves. A German Q2 averages — hot days against — in Q3, and the year-on-year swing in that count is about half as wide. Half the variation in the regressor buys roughly half the t-statistic for the same underlying slope — and on thirteen quarterly observations that is exactly the difference between detectable and invisible. Carlsberg's Q2 null is a sample-size result, not a statement about beer.
Pooling the two quarters does not rescue Carlsberg's data. Taking each year's average Q2/Q3 hot days against the average of its two growth prints (13 years) gives —pp per hot day, t — — worse than Q3 alone, because the Q2 prints (mostly derived from half-year arithmetic and rounded to whole percent) pull it towards zero. So the honest position is: the physics and the German monthly data say one slope of about 0.6% per hot day; Carlsberg's reported Q3 agrees at about 0.5pp; Carlsberg's Q2 neither confirms nor contradicts it. The outlook applies the Q3 figure to both quarters.
Effect per extra hot day, fitted on national monthly data. The two quarters agree; their hot-day counts do not.
Each month fitted on its own, Germany and Poland. May and June are Q2 months and they behave like the Q3 months beside them.
All 13 quarters, derived figures included. Hollow points are derived. This is the picture that has too few points to settle anything.
The fitted effect per hot day in each quarter, with its t-statistic. Only Q3 clears the bar here — which is the limitation being described above.
For the record, a Western European Q2 averages — hot days against — in Q3, on the weighted regional score.
Carlsberg blamed wet weather for China in the third quarter of 2026, and called out bad weather in Finland in the same statement, so the question is whether rain carries information that heat does not. Rain days are added alongside hot days in the same year-on-year monthly model.
In Europe, not really. The rain coefficient has the right sign but does not clear the bar, and it is strongly negatively correlated with heat — — in the German summer — so most of what rain measures is a cool summer measured a second time. Adding it moves the fit from — to —. Counting hot days already captures the European weather effect; a separate rain term is not worth carrying.
In China it is the opposite, and this is the one place in the whole exercise where China produces a signal. Heat does nothing (t —), while a rain day is worth —% of volume at t —, lifting the fit from nothing to —. Carlsberg's own Chinese quarters agree in direction: adding rain takes the Asia Q3 fit from — to —. Neither reaches significance on these sample sizes, so this is a direction worth tracking rather than a number to use — but it does mean the Chinese beer market may be rain-driven where the European one is heat-driven.
| Country | n | % per hot day, heat alone | t | % per hot day, with rain | t | % per rain day | t | corr(heat, rain) |
|---|
Summer months only. Because heat and rain move together, the two slopes in a joint fit split one effect between them and neither should be read on its own.
Carlsberg discloses 13 usable Q3s. Germany's statistics office has published monthly beer sales since 1993, which is a far better-powered test of whether the effect is real at all — at the cost of measuring the wrong thing: a whole country's beer, not Carlsberg's. Volume and hot days are both taken year-on-year against the same calendar month, so seasonality, trend and month length cancel without fitting anything for them.
Germany says yes, emphatically: — on — summer months, an extra hot day worth —% of volume. In — random reshufflings of the weather against the same volumes, — produced a t-statistic this large; the largest any shuffle reached was —.
The number that matters is the agreement. German national monthly data puts the effect at — hot days per 1% of volume. Carlsberg's own Western European Q3 disclosures, an entirely separate dataset at a different frequency, put it at —. Two unrelated sources landing in the same place is the best reason on this page to believe the Q3 figure.
Each point is one month against the same month a year earlier: change in hot days horizontally, change in beer sales vertically.
The German summer result re-run on subsets. Effect in % of volume per extra hot day.
| Sample | n | % per hot day | t | R² |
|---|
It halves in the recent decade, which is worth watching: a more premium, more off-trade market may simply be less weather-driven than it was.
Poland leans the right way but cannot clear the bar; China is flat, which is the same verdict its Carlsberg data gives.
| Country | Months used | n | Correlation | % per hot day | t | Span |
|---|
These are not the same measure and should not be stacked: Germany is tax-warehouse sales including export, Poland is production at firms of 50 or more staff, China is production at above-designated-size firms. Poland's history has a unit break at 2021 and misses small brewers; China's has holes through 2016 to 2018 and no January at all, because its statistics office publishes January and February combined.
The heat test on China runs the wrong way — more hot days sit alongside slightly weaker volumes, at every cut and on both hot-day definitions. Heat does not reduce beer demand, so this is the fit catching something else: the Wusu growth curve, COVID and the property slump. Reweighting the provinces does not rescue it, and the national production data is flat rather than negative, which is the better reading.
Rain is the more promising variable, as the section above sets out and as Carlsberg's own commentary on the June and July peak season implies. It is not significant on the data available, but it is the only thing here that moves China's fit at all. The practical reading: the Chinese hot-day score is not a usable volume indicator, and if a weather variable is going to work for China it is likely to be a wet one.
China's own volume series is disclosed in only 16 quarters, so Asia is shown here; China is the bulk of it.
The Europe result strengthens as the data gets cleaner, which is the pattern you want to see: the weakest cut mixes in winter quarters, the strongest uses only the Q3 figures Carlsberg states in a trading statement. The China result does the opposite.
| Test | Quarters used | n | Correlation | Days per 1pp | t | R² |
|---|
Days above 25°C / 30°C only. A t-statistic above about 2 is the conventional bar for calling a relationship real; on 13 to 18 observations, treat the size of the effect as a range, not a point.
The same test re-run under different market weightings, to separate what the weights decide from what they don't. The Europe slope stays positive in every one, so the direction is safe before a single weight is entered. The size is not: it runs from under two hot days per point to over forty. China stays negative in every province weighting, including the two extremes, which rules the weights out as the explanation there.
Q2 and Q3, stated figures only. Shorter bars mean a stronger weather effect. Each weighting is shown only for the region whose weights it changes.
Europe spans — to — hot days per point, positive in — of — weightings. China is negative in — of —. The scale is logarithmic, because one weighting is a long way out.
Summer hot days per market, Q2 and Q3 added together. The European weights now come from a ranking — Nordics and the UK high, France medium, Germany low — rather than from the Capital Markets Day deck; Poland and Switzerland were not ranked and are set from a volume estimate. The Chinese provinces still count equally. Replacing these with the real disclosed shares changes the size of the European effect by a few tenths of a hot day and nothing else: the weighting chart above shows the ranking and an independent volume triangulation landing close together.
| Market | Region | 10-yr mean | 2026 | vs mean | Weight |
|---|
France, Switzerland and the UK ran far hotter than normal in 2026; the Nordics ran cooler. Which of those dominates the European score is entirely a question of the weights.
A six-month ensemble cannot reach next summer, so there is no forecast of Q3 2027 worth printing. What there is instead: 2026 was hot, and a normal 2027 is therefore a drag on the comparison before a single decision is made.
| Region | Quarter | Normal hot days | 2026 hot days | Gap | n | t | Volume pp if 2027 is normal |
|---|
Each quarter uses its own fitted sensitivity, so only Q3 in Western Europe carries a volume figure. The other rows are blank because no link was measurable in them, not because their weather is unknown — the hot-day counts are there.
Daily highs come from Open-Meteo's reanalysis archive, 51 cities, 2004 to date. A day's normal is the mean high for that calendar day over the previous ten years, taken from a ±7-day window so one freak day cannot define it; the 5-year normal is computed the same way. The tracker also counts days more than 2°C above that normal, but those run at roughly 30 per quarter in every season by construction, so the absolute thresholds are what the charts above use. Daily rainfall comes from the same archive; a rain day is 1mm or more. Volume growth is Carlsberg's reported organic beer volume growth by region. Regressions are run on Q2 and Q3 with a COVID dummy for 2020Q2 to 2021Q2.
The national monthly tests take volume and weather both year-on-year against the same calendar month, which removes seasonality, trend and month length without estimating anything for them. That is what makes them the better-powered check: hundreds of observations instead of thirteen, at the cost of measuring a whole country's beer rather than Carlsberg's.
Three things would change the answer. The European weights are a triangulation of a ranking, not disclosed shares, and the Chinese provinces count equally even though a desert province and a subtropical one are not alike. Most Q2 and Q4 volume figures had to be derived from half-year arithmetic. And weather is one driver among several — price, the channel mix, and the 2020 and 2021 closures all sit in the residual.
The German effect halves after 2016, from —% to —% per hot day. It is still significant, but premiumisation and the shift to off-trade are untested candidates, and if the halving is real the recent-decade figure is the one to use going forward.