Published · · Last reviewed · · 7 min read
English · 한국어
When Is Lotto Bought More? Weekly Sales Across 1,110 Rounds
The clearest result: In the Seollal week, sales were 7.6% higher (95% interval: 6.1% to 9.1%). It is the only one of the 7 effects tested that held in both the first and the last part of the data.
Lotto 6/45 sales change from week to week. Some changes have a plain reason, such as a holiday; others are said to follow the news, for example a very large first-prize payout the week before. This page measures which of those stories show up in the official weekly sales of every round from number 141 to number 1,250, and which do not.
This is research on buying habits and the calendar. It says nothing about which numbers will be drawn, and it does not change the chance of any ticket winning.
In short
- A big first-prize amount per winner in the previous round: looked real at first, then did not repeat in the later rounds.
- Few first-prize winners in the previous round: looked real at first, then did not repeat in the later rounds.
- The last two weeks of the year: no sign of an effect.
- The Seollal week: confirmed, sales 7.6% higher (95% interval: 6.1% to 9.1%).
- The Chuseok week: no sign of an effect.
- The first week of a month: no sign of an effect.
- The last third of a month: no sign of an effect.
The data
The source is the official result of every round, one draw per week on Saturday: the draw date, the total sales, and the first-prize winners and the amount each one received. From round 88 a game costs KRW 1,000, so sales divided by 1,000 is the number of games sold. Round 87 sold KRW 83.8 billion and round 88 sold KRW 51.9 billion: the price cut at that point removed about 38% of weekly sales at once, which is why the study starts there. The model needs a year of earlier rounds to define "the usual first-prize amount", so the fitted rounds run from 141 (2005-08-13) to 1,250 (2026-11-14), 1,110 rounds in all.
How we measured it
Weekly sales move with a slow trend, so the trend has to be taken out before any effect can be seen. The model describes the logarithm of weekly sales as a smooth curve over time (a cubic spline with a knot about every year) plus the effects below. Using the logarithm turns every effect into a percentage of that week's sales. Standard errors allow for the fact that neighbouring weeks resemble each other (Newey-West, 8 weeks).
We tested 7 effects at once, so some would look real by luck alone. To guard against that we split the rounds. The effects that stand out in the first 60% of rounds (the first 666 rounds, up to round 807) are checked on the last 40% (the last 444 rounds), where the p-values are corrected for the number of effects checked (Holm). An effect counts as confirmed only if it passes there with the same sign.
The seven effects
- Previous round's first-prize amount per winner: how many doublings it is above the usual amount (the median of the 52 rounds before). It is 0 when nobody won.
- Few first-prize winners in the previous round: at most half the number that its sales imply, so the prize rolled on. This includes rounds with none.
- Year-end: the draw is on 18-31 December.
- Seollal week: the seven days ending on the draw date contain Seollal or the day before or after.
- Chuseok week: the same for Chuseok.
- First week of a month: the draw is on day 1-7.
- Last third of a month: the draw is on day 22-31. The reference for the last two is day 8-21.
The dates of Seollal and Chuseok follow the Korean lunisolar calendar (the first day of the first lunar month and the fifteenth day of the eighth). LottoFlick computed them from new moons and solstices and compared all 46 dates of 2004 to 2026 with the official special-day service of the Korea Astronomy and Space Science Institute; all of them agree (file kr-holidays.json).
Results at a glance
| What was tested | Change in sales | 95% interval | Rounds affected | Verdict |
|---|---|---|---|---|
| a big first-prize amount per winner in the previous round | 1.1% lower per doubling | -2.2% ~ -0.1% | 1,107 | not confirmed |
| few first-prize winners in the previous round | 4.4% higher | 0.5% ~ 8.4% | 107 | not confirmed |
| the last two weeks of the year | 0.8% higher | -0.5% ~ 2.1% | 42 | no sign |
| the Seollal week | 7.6% higher | 6.1% ~ 9.1% | 29 | confirmed |
| the Chuseok week | 0.3% higher | -1.2% ~ 1.9% | 28 | no sign |
| the first week of a month | 0.2% higher | -0.3% ~ 0.7% | 255 | no sign |
| the last third of a month | 0.6% higher | -0.1% ~ 1.3% | 344 | no sign |
Does big prize news raise next week's sales?
A big first-prize amount per winner in the previous round: not confirmed. It looked like an effect in the first 60% of rounds (1.6% lower), but the last 40% did not repeat it (0.1% lower, p = 0.69). Over all rounds sales were 1.1% lower (95% interval: 0.1% to 2.2%).
The effect is measured per doubling of the amount compared with the usual first-prize amount. The later rounds do not show sales rising after big prize news, so the story that a large payout sends crowds to the counter the next week is not supported by the weekly totals.
After few or no first-prize winners
Few first-prize winners in the previous round: not confirmed. It looked like an effect in the first 60% of rounds (6.1% higher), but the last 40% did not repeat it (0.3% higher, p = 0.70). Over all rounds sales were 4.4% higher (95% interval: 0.5% to 8.4%).
When nobody wins the first prize, its money goes into the next round, so the next first prize is larger. There were only 3 such rounds among those fitted, too few to measure on their own, which is why the test uses "few winners" (107 rounds) instead. A prize rolling on is the setting in which people say they buy more, and the weekly totals show how far that holds.
Holidays and the calendar
The Seollal week: confirmed. Sales were 7.6% higher (95% interval: 6.1% to 9.1%). The first 60% of rounds say 7.6% higher and the last 40% say 7.6% higher, so the two parts agree.
The Chuseok week: no sign of an effect. Sales were 0.3% higher (95% interval: 1.2% lower to 1.9% higher), and the interval includes no change.
The last two weeks of the year: no sign of an effect. Sales were 0.8% higher (95% interval: 0.5% lower to 2.1% higher), and the interval includes no change.
The first week of a month: no sign of an effect. Sales were 0.2% higher (95% interval: 0.3% lower to 0.7% higher), and the interval includes no change.
The last third of a month: no sign of an effect. Sales were 0.6% higher (95% interval: 0.1% lower to 1.3% higher), and the interval includes no change.
Of the two big family holidays of the lunar calendar, only Seollal moves lottery sales. A holiday week is short (29 rounds for Seollal and 28 for Chuseok across the fitted years), so those estimates have wider intervals than the monthly patterns, which cover hundreds of rounds each.
The long-term trend
Behind every weekly figure there is a long climb. The yearly median of weekly sales ran from KRW 42.6 billion in 2008, the lowest year, to KRW 121.2 billion in 2026, the highest: 2.8 times as much. Median sales rose in 18 of the 21 year-to-year comparisons, and the steepest rise was 11% in 2011. The model treats all of this as the trend and measures the effects around it; it does not try to explain it, and nothing here says why sales moved.
What it says about buying habits
Two habits of mind are often blamed for lottery buying: chasing a bigger prize and acting on the news. The weekly totals give no support to the story that a bigger prize pulls in more buyers, either as news (a large amount per winner) or as a prize that rolls on. The calendar does leave a mark: the Seollal week. Read this next to the sunk-cost article (what a buyer gets back in the long run), the present-bias article (why a large number now beats a larger number later) and the negative expected value article.
Limits
- Totals, not people. The data are national weekly sales. They cannot tell who buys more or why, only that the total moves.
- Association, not cause. A pattern in the calendar shows that sales move together with it; this page does not show what in the holiday or the date causes it.
- Choices that could change the numbers. The knot spacing of the trend, the "at most half the expected first-prize winners" definition of few winners and the holiday window are fixed choices; other reasonable choices would move the estimates a little.
- Dates. The holiday dates agree with the official special-day service, but the window around them (the week, the day before and the day after) is our own choice.
How this page updates
The figures are refitted every week from the latest official data (now through round 1,250). If a refit changes which effects are confirmed or flips a direction, the page keeps its text until the same new result has come out three weeks in a row; then it is published and listed on the corrections page as an update.
LottoFlick is independent and not an official lottery operator. Numbers are random and cannot be predicted.
Sources and verification
- LottoFlick model output used on this page (trend, effects, validation) — computed by LottoFlick
- Seollal and Chuseok dates computed by LottoFlick and checked against the Korea Astronomy and Space Science Institute special-day service — verified
- LottoFlick dataset: one row per round with sales and winners — computed by LottoFlick
- Donghaeng Lottery: Lotto 6/45 round results (public data service) — verified