What re-regulation does to player demand: Four natural experiments, 2021–2025
When a country changes its gambling law, do players actually move? That’s what we aim to find out in this study. Four European markets were re-regulated on four different dates between 2021 and 2025, making each one a natural experiment with the others as a comparison group.
In short
Denmark, the Netherlands, Sweden and Germany re-regulated on four different dates, making each one a natural experiment. Differenced against untreated markets, only two of the four moved players toward licensed operators.
Denmark, Jan 2025
+14.0pp
shift toward locally licensed operators
The largest of the four, and the only one visible to the naked eye in the raw series.
Netherlands, Oct 2021
+2.1pp
shift from a very low base
Small in points, but the licensed share roughly doubled. This is the cleanest of the four results.
Sweden, Jul 2023
+0.1pp
no measurable change
The closed supply loop reorganised the industry. It didn’t, however, meaningfully move players.
Germany, Jul 2021
−5.3pp
shift away from locally licensed operators
Licensing arrived alongside strict product rules, and demand kept leaving.
The design, and the trap it exists to avoid
The measure throughout is the share of observed demand in a market going to brands that hold that market’s own national licence. If a reform channelled players toward nationally-licensed operators, we would expect that share to rise. Each reform is plotted 12 months on either side of the month it took legal effect, so each comparison is with the same market shortly before and after legislation rather than being compared against a different market entirely.
There’s a serious problem with reading data like this naively, which is worth stating before any results. Our observations come from casino comparison sites, and that set changes over time. A comparison site opening or closing in a market moves the measured share even when nothing happens in the market itself. Read as raw data across seven years, the Dutch series shows almost nothing at the October 2021 market opening, and then it suddenly leaps from 1.4% in January 2024 to 60.9% that March, when no Dutch law changed whatsoever. That jump is an artefact of who was measuring, not a change in what was measured.
Everything below, therefore, holds the panel of contributing sites fixed, meaning only sites present in every month of a window count toward that window. Figure 8 shows what that correction is worth. The short answer is that, inside a tight window, it barely matters. This is precisely why these four estimates are trustworthy. It also shows why this dataset shouldn’t be used to draw a seven-year trend line.
Share of demand going to Dutch-licensed brands. Months relative to the reform on the horizontal axis; dashed lines are the mean either side. Vertical axis runs to 10%.
Flat for a year at around 2%, then a step up in the month the market opened, followed by a steady climb. The numbers are small in absolute terms, but still represent a doubling. It’s worth noting that this is the cleanest signal of the four studies.
Show data table Hide data table
| Months from reform | Locally licensed share (%) | Period |
|---|---|---|
| -12 | 2.4 | Before |
| -11 | 1.6 | Before |
| -10 | 1.8 | Before |
| -9 | 2.4 | Before |
| -8 | 2.1 | Before |
| -7 | 2.2 | Before |
| -6 | 1.8 | Before |
| -5 | 1.9 | Before |
| -4 | 1.8 | Before |
| -3 | 2.1 | Before |
| -2 | 1.8 | Before |
| -1 | 2.1 | Before |
| 0 | 3.4 | After |
| +1 | 2.6 | After |
| +2 | 2.4 | After |
| +3 | 3 | After |
| +4 | 3.1 | After |
| +5 | 3.3 | After |
| +6 | 3.9 | After |
| +7 | 6.3 | After |
| +8 | 5.2 | After |
| +9 | 4.7 | After |
| +10 | 6.6 | After |
| +11 | 4.2 | After |
Source: Kasinohai casino dataset.
Share of demand going to Danish-licensed brands. Months relative to the reform on the horizontal axis; dashed lines are the mean either side. Vertical axis runs to 100%.
The largest movement of the four, from a mean of 75.7% to 89.6%. Note that the pre-period is already rising, so some of this shift was underway before the rule took effect.
Show data table Hide data table
| Months from reform | Locally licensed share (%) | Period |
|---|---|---|
| -12 | 70.5 | Before |
| -11 | 71.5 | Before |
| -10 | 74.4 | Before |
| -9 | 74.6 | Before |
| -8 | 74.9 | Before |
| -7 | 67 | Before |
| -6 | 76.7 | Before |
| -5 | 74.6 | Before |
| -4 | 83.5 | Before |
| -3 | 78.5 | Before |
| -2 | 75.8 | Before |
| -1 | 85.9 | Before |
| 0 | 93.7 | After |
| +1 | 90.4 | After |
| +2 | 94.7 | After |
| +3 | 93.4 | After |
| +4 | 80.9 | After |
| +5 | 88.8 | After |
| +6 | 85.5 | After |
| +7 | 85.3 | After |
| +8 | 89.4 | After |
| +9 | 89.5 | After |
| +10 | 92.1 | After |
| +11 | 91.9 | After |
Source: Kasinohai casino dataset.
Share of demand going to Swedish-licensed brands. Months relative to the reform on the horizontal axis; dashed lines are the mean either side. Vertical axis runs to 20%.
Nothing. The data shows 9.6% before, 9.7% after. Sweden’s reform reorganised who may supply whom, but it didn’t change where Swedish players went.
Show data table Hide data table
| Months from reform | Locally licensed share (%) | Period |
|---|---|---|
| -12 | 16.6 | Before |
| -11 | 11.8 | Before |
| -10 | 9.6 | Before |
| -9 | 9.4 | Before |
| -8 | 7.2 | Before |
| -7 | 6.4 | Before |
| -6 | 10.1 | Before |
| -5 | 11.6 | Before |
| -4 | 7.8 | Before |
| -3 | 7.5 | Before |
| -2 | 7.9 | Before |
| -1 | 9.1 | Before |
| 0 | 10.7 | After |
| +1 | 8.8 | After |
| +2 | 8.9 | After |
| +3 | 9.4 | After |
| +4 | 10.5 | After |
| +5 | 8 | After |
| +6 | 10.2 | After |
| +7 | 10.9 | After |
| +8 | 13.8 | After |
| +9 | 11.4 | After |
| +10 | 10.8 | After |
| +11 | 2.3 | After |
Source: Kasinohai casino dataset.
Share of demand going to German-licensed brands. Months relative to the reform on the horizontal axis; dashed lines are the mean either side. Vertical axis runs to 25%.
The licensed share fell by two-fifths across the window. However, the decline was already underway before the reform. This is the one result that cannot be cleanly attributed to the law.
Show data table Hide data table
| Months from reform | Locally licensed share (%) | Period |
|---|---|---|
| -12 | 19.8 | Before |
| -11 | 14.3 | Before |
| -10 | 16.3 | Before |
| -9 | 13.9 | Before |
| -8 | 11 | Before |
| -7 | 15 | Before |
| -6 | 12.3 | Before |
| -5 | 9.9 | Before |
| -4 | 10.7 | Before |
| -3 | 12.5 | Before |
| -2 | 12.7 | Before |
| -1 | 9.7 | Before |
| 0 | 11.9 | After |
| +1 | 10.3 | After |
| +2 | 8.4 | After |
| +3 | 8.4 | After |
| +4 | 9.8 | After |
| +5 | 9.5 | After |
| +6 | 3.9 | After |
| +7 | 6.8 | After |
| +8 | 3.4 | After |
| +9 | 7.1 | After |
| +10 | 8.4 | After |
| +11 | 6.6 | After |
Source: Kasinohai casino dataset.
Change in locally licensed share of demand, percentage points.
There is no such thing as “the effect of re-regulation”. The same policy label produces a 14-point gain in one market and a 5-point loss in another.
Show data table Hide data table
| Market | Reform | Date | Before (%) | After (%) | Change (pp) | Relative change |
|---|---|---|---|---|---|---|
| Denmark | Supplier restrictions | Jan 2025 | 75.7 | 89.6 | +14 | +18% |
| Netherlands | Market opening (KOA) | Oct 2021 | 2 | 4.1 | +2.1 | +103% |
| Sweden | Closed supply loop | Jul 2023 | 9.6 | 9.7 | +0.1 | +1% |
| Germany | Nationwide licensing | Jul 2021 | 13.2 | 7.9 | -5.3 | -40% |
Source: Kasinohai casino dataset.
Does it survive a control group?
A before-and-after comparison can’t separate a reform from whatever else was happening that year. The standard remedy is to difference the treated market against markets that had no reform in the same window (here, Finland, Norway, Canada, Australia, and New Zealand). Because those markets have no national licence of their own, the outcome switches to the share of demand going to brands holding any EEA or UK national licence, which is defined consistently across markets.
Difference-in-differences, percentage points. Positive means demand moved toward licensed operators.
Denmark and the Netherlands survive the control. Germany’s result strengthens: relative to markets with no reform, the German licensed share fell by 17 points.
Show data table Hide data table
| Market | Treated change (pp) | Control change (pp) | Difference-in-differences (pp) |
|---|---|---|---|
| Denmark | +11.9 | -4.3 | +16.2 |
| Netherlands | -1.7 | -8.3 | +6.6 |
| Sweden | -1 | +1.5 | -2.5 |
| Germany | -23.3 | -5.9 | -17.4 |
Source: Kasinohai observed-demand data; control markets Finland, Norway, Canada, Australia, New Zealand.
- Actual reform date
- Placebo date, twelve months early
Change in locally licensed share, percentage points.
The Dutch result behaves as it should: the effect appears at the real date, not the fake earlier one. Denmark’s effect is partly present at the placebo date, and Sweden’s placebo swing (−11.6pp) dwarfs its real effect, so Sweden’s result can’t be told apart from noise. We advise reading both with caution.
Show data table Hide data table
| Market | Actual date, clicks (pp) | Actual date, deposits (pp) | Placebo date, clicks (pp) | Placebo date, deposits (pp) |
|---|---|---|---|---|
| Denmark | +14 | +5.4 | +5.1 | -5.8 |
| Netherlands | +2.1 | +1.7 | -1.2 | -0.3 |
| Sweden | +0.1 | -0.1 | -11.6 | -4.2 |
| Germany | -5.3 | -6.8 | -1.6 | -6.3 |
Source: Kasinohai casino dataset.
- All sites, uncorrected
- Fixed site panel
Change in locally licensed share, percentage points.
Inside a 12-month window, the correction changes nothing by more than 0.9pp. That agreement provides a sound basis for these estimates over that period, but it doesn’t hold true over multi-year horizons.
Show data table Hide data table
| Market | All sites (pp) | Fixed panel (pp) | Difference (pp) |
|---|---|---|---|
| Denmark | +13.6 | +14 | 0.4 |
| Netherlands | +3 | +2.1 | 0.9 |
| Sweden | +0.1 | +0.1 | 0 |
| Germany | -5.3 | -5.3 | 0 |
Source: Kasinohai casino dataset.
Why the same policy produces opposite results
The four reforms are not four versions of the same thing. They differ in what they restrict, and the results pattern tracks that difference rather than the law’s strength.
| Market | What the regime restricts | Measured effect |
|---|---|---|
| Denmark | Permissive within a licensed perimeter. Licensed since 2012, with a broad product range. The 2025 change tightened who may supply licensed operators rather than restricting what players can do. | Largest measured shift onshore. |
| Netherlands | Opening, then advertising restrictions. The market opened to licensed operators in October 2021. Untargeted advertising was subsequently restricted, narrowing how licensed brands reach players. | Real but small shift; the licensed share roughly doubled from a very low base. |
| Sweden | Supply-side control. From July 2023, licensed game suppliers could serve only licensed operators, and licensed operators could take games only from licensed suppliers. | No measurable change in where demand goes. |
| Germany | Restrictive product rules. Nationwide licensing arrived in July 2021 alongside stake limits on virtual slots, a mandatory spin delay and a cross-operator monthly deposit ceiling. | Licensed share fell; demand moved further offshore. |
The two reforms that moved players the most (in Denmark and the Netherlands) worked on the licensed perimeter itself: who may operate, and who may supply. The reform that moved players the wrong way targeted the product instead, arriving alongside stake limits, a mandatory delay between spins and a cross-operator deposit ceiling. Sweden’s supply-side rule, which substantially reorganised the industry, left the demand side untouched.
That is the takeaway for a Finnish reader. The 2027 Gambling Act will be judged on channelisation, and the evidence suggests channelisation depends less on whether a country licenses its market than on what it then permits licensed operators to offer. In other words, a licence regime players don’t want to use doesn’t channel anyone.
Method, and what could still be wrong
Data. Daily click and first-time-deposit counts by brand, market and date, from 2017-01-01 to 2026-08-02 (3,322 days with data). Brand licences come from the same structured casino dataset used throughout this section. This study doesn’t include monetary totals, traffic volumes or deposit volumes. Therefore, every published number is a share of demand or a difference between two shares.
What the measure is. Demand observed through casino comparison sites doesn’t cover the whole market. In fact, it over-represents players who actively shop between operators and under-represents those who go directly to a brand that they already know. A reform that changes brand loyalty rather than comparison behaviour would be partly invisible here. These are shares of a particular slice of demand, and they shouldn’t be read as national channelisation rates.
Licences are observed as of today. The dataset records the licences a brand holds now, and not the ones it held in 2021. A brand that gained or lost a national licence during a window is classified by its present state throughout. This biases the results toward finding no effect when licences were gained mid-window, and it’s our study’s most significant measurement limitation.
Germany should be read as unresolved. The German licensed share falls sharply, and the fall survives the control group. However, it also appears at the placebo date, which means the decline was already in motion before July 2021. The honest reading, therefore, is that German licensing didn’t arrest an existing drift offshore, i.e., it didn’t cause one.
Denmark’s magnitude is uncertain. Its pre-period is already trending upward, and its placebo test is partly contaminated. The direction is reliable across every specification, but the effect size isn’t, and the deposit-based estimate is roughly a third of the click-based one.
Sweden’s null should be read with caution. Its real-date effect is essentially zero, but its placebo test is not clean: the same calculation returns −11.6pp twelve months early, when nothing happened. That failed placebo means the Swedish series carries movements the design does not fully absorb, so the no-effect finding is directionally safe but not a precise zero.
No significance tests are reported. With four reforms and 12 months either side, conventional standard errors would imply more precision than this design supports. Instead, Figures 7 and 8 offer robustness checks, and readers should weigh a result’s consistency across them rather than any single number in isolation.
Reproducibility. Every figure in this study comes from a single query against the demand dataset, holding one classification of markets, licences and reform dates throughout. The classification rules and the construction of the fixed panel are described in full above. We will supply the underlying tables on request. Figures may be reproduced with attribution and a link.
Changes to this page
-
Publication date moved from 3 August 2026 to 3 November 2026.
References
- [1] Kansspelautoriteit. Kansspelautoriteit — Netherlands Gambling Authority (accessed 3 August 2026).
- [2] Spillemyndigheden. Spillemyndigheden — Danish Gambling Authority (accessed 3 August 2026).
- [3] Spelinspektionen. Spelinspektionen — Swedish Gambling Authority (accessed 3 August 2026).
- [4] GGL. Gemeinsame Glücksspielbehörde der Länder (accessed 3 August 2026).