Comsefaz estimates families' 'losses' due to gambling at R$ 62,5 billion, but the methodology is flawed and ignores official SPA data.
The number is circulating in newsrooms, public policy forums, and parliamentary speeches: R$ 62,5 billion in "losses" for Brazilian families due to sports betting in 2025. This figure comes from the 3rd Update of the Fiscal Bulletin of Brazilian States , prepared by Comsefaz (National Committee of Finance Secretaries) in partnership with the Celso Furtado International Center for Development Policies, and has been widely reproduced by the press without the caveat that the authors themselves registered: it is an estimate of the order of magnitude, not a precise causal measurement. In fact, it's just another dubious study graced with the BNLData Narrative Seal.
The difference between what the report estimates and what the official data records is 69%. For the same year, 2025, the Prizes and Betting Secretariat of the Ministry of Finance – SPA-MF – recorded gross revenue of R$ 36.959.783.379,70 declared by licensed operators. The discrepancy reveals not only a difference in scale, but a fundamental methodological problem.
How the number was constructed
The Comsefaz estimate is not based on data declared by betting companies. It is derived from Pix transfers from individuals to legal entities classified by the Central Bank in the category "arts, culture, sports and recreation," a designation of the Economic Activity Payment Statistics system – EPAE, which goes far beyond the world of betting. Gyms, sports clubs, event producers, cinemas and streaming platforms are included in the same classification of Section R * of the National Classification of Economic Activities – CNAE. (see at the end of the text)
The label “arts, culture, sports and recreation” in the Central Bank’s payment statistics system (EPAE) does not exclusively correspond to betting houses. The category also includes gyms, sports clubs, event producers, cinemas, streaming platforms, and other businesses unrelated to betting. The model attributes all abnormal growth in the aggregate of this category from 2025 onwards to betting—it does not statistically isolate the specific betting component within the sector. Any organic expansion of other segments of this category during the period (for example, recovery of the events and tourism sector post-pandemic, or growth in streaming) would automatically be accounted for as a “betting effect”.
The methodological procedure described on pages 38 to 40 of the bulletin works as follows: the researchers trained an ARIMA model with data up to August 2024, before the entry into force of the betting regulations in January 2025, to project what the “normal” behavior of the payment series in this category would be. Everything that exceeded this projection was attributed to the betting market. The problem is immediate: any organic expansion of other segments of the same category during the period, such as the recovery of the events sector post-pandemic or the growth of streaming, would automatically be accounted for as a “betting effect”.
There is also an additional complication in the choice of the cutoff point itself. The authors acknowledge that the series already shows an inflection point starting in August 2024, months before the regulation comes into effect, attributing this movement to companies that anticipated adjustments to the new rules. This means that the period used to "train" the model may already contain part of the phenomenon that was intended to be isolated, which tends to artificially inflate the effect measured from January 2025 onwards.
Four misconceptions that distort the debate.
Analysis of the methodology points to at least four problems that compromise the comparability of the number with official data.
The first is the imprecision of the data category. The classification "arts, culture, sports and recreation" encompasses a variety of activities beyond betting, and the model does not statistically isolate the specific betting component within the aggregate. Any growth in other areas of the sector inflates the number attributed to betting.
The second misconception lies in the confusion between gross and net. The report produces two distinct figures for 2025: an additional gross volume of R$ 350,97 billion and a net transfer of R$ 62,5 billion, defined as the value that does not return to bettors in the form of prizes. The media even treated the gross value as if it were the "loss of families." The number comparable to the gross revenue declared to SPA-MF is the net; but even it is not directly equivalent to the GGR (Gross Gaming Revenue) calculated by licensed operators.
The third problem is structural: comparing regulatory data with a macroeconometric inference as if they were equivalent quantities. The gross revenue of R$ 36,96 billion reported by SPA-MF is administrative data declared by licensed operators, subject to specific accounting rules and oversight. The figure of R$ 62,5 billion is a statistical inference about aggregate payment flows, subject to the specification of the ARIMA model, the chosen training period, and the quality of the Central Bank's data series. Mixing the two without qualification is equivalent to comparing a direct measurement with an indirect estimate, ignoring the margins of error inherent in each approach.
The fourth misconception concerns the very assertion of causality. The report treats the difference between the observed and the counterfactual as a transfer from households to the sector, but a statistical discrepancy between series is, at most, a temporal association. To support causality, administrative data from the betting houses themselves would be necessary, with entries and exits of prizes, analysis of individual consumption habits, or quasi-experiments that isolated the effect of regulation from other coincident variables in the same period.
Gaps that the bulletin does not address.
The document does not cite, at any point, the official 2025 revenue figures released by SPA-MF. This absence is significant for two reasons. The first is that comparing the data with official figures is precisely what would allow for calibrating the estimate. The second is that the 69% discrepancy between the two figures is far greater than what could be attributed to the illegal betting market, whose size does not accommodate this difference, which points to problems in the model's own specification.
The report also lacks sensitivity tests that would allow the reader to assess the stability of the result: there are no variations in the estimation window, alternative models, confidence intervals, or controls for seasonality. Without these demonstrations of robustness, it is impossible to know whether the R$ 62,5 billion would withstand slightly different methodological choices.
The problem is not the existence of the study. Indirect statistical models are valuable for understanding phenomena that still lack consolidated data series. The problem is the circulation of the number without the caveats that the authors themselves included in the document, at a time when the regulatory debate on sports betting is still ongoing and impact estimates directly influence institutional narratives and decisions.
(*) Section R of the CNAE (National Classification of Economic Activities) of IBGE Concla is divided into four major divisions (codes 90 to 93):
Division 90: Artistic, Creative and Entertainment Activities
Performing arts and shows: Theatrical, musical, dance and circus productions.
⇒ Infrastructure services: Sound and lighting services for events.
⇒ Artistic creation: Work of visual artists, independent journalists, and writers.
⇒ Space management: Administration of theaters, performance halls, and cultural centers.
Division 91: Activities Related to Cultural and Environmental Heritage
⇒ Cultural facilities: Operation of museums and conservation of historical sites and buildings.
⇒ Collections and records: Activities of libraries and documentation archives.
⇒ Environmental preservation: Operation of botanical gardens, zoos, and ecological reserves.
Division 92: Gambling and Betting Activities
⇒ Lotteries and betting: Operation of authorized gambling games and sale of lottery tickets.
⇒ Casinos and bingo halls: Operation of physical or virtual gambling games permitted by law.
Division 93: Sports, Recreation and Leisure Activities
⇒ Clubs and leagues: Management of official or amateur sports clubs and federations.
⇒ Physical facilities: Operation of gyms, stadiums, courts and swimming pools.
⇒ Leisure and entertainment: Discotheques, dance clubs, amusement parks and theme parks.
